INTRODUCTION
Organic food consumption has become one of the most significant trends in the food
industry. The growth in the sales of organic foods has outpaced that of conventional foods
in recent years (Organic Trade Association, 2018), despite the premiums that organic foods
usually command. While increasing demand coupled with organic premiums have
encouraged growers to adopt organic production practices, some growers have experienced
challenges which has slowed down the supply growth needed to satisfy demand. Organic
wheat growers in U.S. western states face some of these challenges, which affects the
welfare of other actors along the supply chain including food manufacturers, retailers and
final consumers. The purpose of this dissertation is to address issues related to supply and
demand in the organic wheat market that will facilitate the decision-making process and
contribute to the welfare of actors along the supply chain.
The overall objective of the first essay is to evaluate the impact of organic premium
uncertainty on the profitability of organic wheat production and examine options that
growers may have to reduce uncertainty. Improving growers’ understanding of these issues
may benefit them as they make production decisions, and in turn contribute to increasing
the organic wheat supply. Specifically, we address three objectives: (1) examine the
uncertainty associated with organic wheat prices and organic premiums and how they
affect the profitability of organic wheat production relative to conventional wheat
production, (2) examine whether hedging in conventional wheat futures mitigates the
organic wheat price risk and whether conventional futures prices can be used to predict
organic prices, and (3) apply and evaluate different methods for imputing missing
observations in the historical organic wheat price data. This study is the first one to examine
2
the possibility to cross hedge price risk of an organic commodity using the futures market
for its conventional counterpart, in the absence of its own futures market. Further, previous
studies have typically examined dynamics between spot and futures prices for the same
commodity. This study adds to the current literature by examining the dynamics between
qualitatively differentiated commodities.
The overall objective for the second essay is to segment consumers into groups to
understand who “very likely,” “likely,” and “unlikely” consumers of organic wheat
products are and how they differ to identify what factors determine an interest in organic
wheat products. Past studies have examined the factors associated with preferences for
organic products, and there tends to be a consensus that characteristics such as lifestyle and
attitudinal factors may explain consumer behavior better than demographics (Gil, Gracia,
and Sanchez, 2000; Li, Zepeda, and Gould, 2007; Ureña, Bernabéu, and Olmeda, 2008;
Bitzios, Fraser, and Haddock-Fraser, 2011; Gracia and de Magistris, 2013). This essay adds
to the current literature by shedding light on what factors determine preferences for organic
foods in the context of wheat products. Further, it contributes to the existing literature by
performing the analysis and contrasting the results for two different wheat product
categories, bread and cookies, since past studies have found that preferences for and
perceptions of organic label depend on whether the product is classified as a virtue or a
vice (Van Doorn and Verhoef, 2011; Ellison et al., 2016). In this essay, bread is considered
a virtue product and cookies are considered a vice product (Hui, Bradlow, and Fader,
2009). In addition, this study is the first one to examine whether organic wheat products
are appealing to consumers who avoid wheat or gluten. Finally, we differentiate consumers
3
based on factors affecting their likelihood to prefer organic bread and cookies specifically,
which has not been addressed in the previous literature. This knowledge can be used to
assist manufacturers and marketers in making production and marketing decisions when
allocating the limited supply of organic wheat to match consumers’ needs, as well as gauge
the growth potential. In addition, it will allow wheat growers to better understand the
current economic conditions and perspective on the market growth potential in organic
wheat.
Specifically, the second essay addresses four objectives: (1) cluster consumers into
segments based on their attitudes toward organic products and production systems, their
past purchases of organic wheat products and importance of certified organic labels, (2)
examine the differences across the segments in terms of their preferences, reasons for
purchasing/not purchasing organic wheat products, socio-demographics, lifestyle
characteristics, and shopping and consumption behavior, (3) calculate willingness to pay
(WTP)
1
values for certified organic products for whole sample and for each segment, and
(4) examine the differences in findings across the two types of wheat products.
The overall objective of the third essay is to examine consumer WTP for the organic
label alone and in combination with other related labels to understand whether combining
the organic label with other labels is beneficial for consumers or whether it confuses them,
and what factors may affect it. Past studies have examined consumer WTP for the
combination of organic and non-GMO labels, but with somewhat contrasting findings.
1
Willingness to pay for a product attribute is the additional dollar amount that a consumer would pay to
obtain the product with the attribute, while keeping the overall utility of the consumer constant. The WTP
amount is determined by finding the utility that consumer gains from the additional attribute and then finding
the dollar amount (i.e., increase in cost to the consumer) that will decrease the utility by the same magnitude.
4
While Conner and Christy (2004) found that organic consumers specifically are willing to
pay more for this combination of labels than for the organic label alone, McFadden and
Lusk (2017) found out that consumers in general tend to perceive these labels as substitutes
and are not willing to pay extra for the combination. In the third essay, we add to the
existing literature by examining whether knowledge that organic is non-GMO by definition
plays a role in consumer valuation of the organic and non-GMO labels combination. Since
past studies found that consumers in general perceive organic products as healthier relative
to conventional products (Magnusson et al., 2001; Lea and Worsley, 2005; Krystallis,
Fotopoulos, and Zotos, 2006; Lee et al., 2013), we further examine whether combinations
of the organic label with some health promoting labels have any effect on consumer
preferences for organic bakery products. Gluten-free and low-carb or sugar-free labels are
chosen to reflect some of the recent trends in consumer food demand (Asioli et al., 2017),
and this study is the first one to examine consumer preferences for combinations of these
labels with organic label.
For the third essay we consider the following four objectives: (1) obtain consumers’
WTP for organic label on bread and cookies, (2) examine the impact of combining organic
and non-GMO labels on the WTP for organic products and whether it depends on
consumers’ prior knowledge that organic must be non-GMO, (3) examine the impact of
including gluten-free, sugar-free or low-carb labels on the WTP for organic products, and
(4) evaluate the effect of factors including overall knowledge and familiarity with organic
and gluten/wheat intolerance/avoidance on the WTP. The findings will benefit food
manufacturers and marketers as they develop new products and marketing strategies. The
5
number of labels on the products newly introduced to the U.S. market has greatly increased
in the recent years (U.S. Department of Agriculture Economic Research Service, 2017),
but consumers may or may not value specific label combinations. The insights regarding
multiple labeling that involves the organic label specifically may be of great importance to
manufacturers and marketers of products containing organic wheat.
References
Asioli, D., J. Aschemann-Witzel, V. Caputo, R. Vecchio, A. Annunziata, T. Næs, and P.
Varela. 2017. “Making Sense of the “Clean Label” Trends: A Review of
Consumer Food Choice Behavior and Discussion of Industry Implications.” Food
Research International 99:58-71.
Bitzios, M., I. Fraser, and J. Haddock-Fraser. 2011. “Functional Ingredients and Food
Choice: Results from a Dual-Mode Study Employing Means-End-Chain Analysis
and a Choice Experiment.” Food Policy 36(5):715-725.
Conner, D., and R. Christy. 2004. “The Organic Label: How to Reconcile its Meaning
with Consumer Preferences.” Journal of Food Distribution Research 35(1):40-43.
Ellison, B., B.R. Duff, Z. Wang, and T.B. White. 2016. “Putting the Organic Label in
Context: Examining the Interactions between the Organic Label, Product Type,
and Retail Outlet.” Food Quality and Preference 49:140-150.
Gil, J.M., A. Gracia, and M. Sanchez. 2000. “Market Segmentation and Willingness to
Pay for Organic Products in Spain.” The International Food and Agribusiness
Management Review 3(2):207-226.
6
Gracia, A., and T. de Magistris. 2013. “Organic Food Product Purchase Behaviour: A
Pilot Study for Urban Consumers in the South of Italy.” Spanish Journal of
Agricultural Research 5(4):439-451.
Hui, S.K., E.T. Bradlow, and P.S. Fader. 2009. “Testing Behavioral Hypotheses Using an
Integrated Model of Grocery Store Shopping Path and Purchase Behavior.”
Journal of Consumer Research 36(3):478-493.
Krystallis, A., C. Fotopoulos, and Y. Zotos. 2006. “Organic Consumers' Profile and Their
Willingness to Pay (WTP) for Selected Organic Food Products in Greece.”
Journal of International Consumer Marketing 19(1):81-106.
Lea, E., and T. Worsley. 2005. “Australians' Organic Food Beliefs, Demographics and
Values.” British Food Journal 107(11):855-869.
Lee, W.C.J., M. Shimizu, K.M. Kniffin, and B. Wansink. 2013. “You Taste What You
See: Do Organic Labels Bias Taste Perceptions?.” Food Quality and Preference
29(1):33-39.
Li, J., L. Zepeda, and B.W. Gould. 2007. “The Demand for Organic Food in the US: An
Empirical Assessment.” Journal of Food Distribution Research 38(3):54-69.
Magnusson, M.K., A. Arvola, U.K.K. Hursti, L. Åberg, and P.O. Sjödén. 2001.
“Attitudes towards Organic Foods among Swedish Consumers.” British Food
Journal 103(3):209-227.
McFadden, B.R., and J.L. Lusk. 2017. “Effects of the National Bioengineered Food
Disclosure Standard: Willingness to Pay for Labels that Communicate the
7
Presence or Absence of Genetic Modification.” Applied Economic Perspectives
and Policy 40(2):259-275.
Organic Trade Association [OTA]. 2018. “Maturing U.S. Organic Sector Sees Steady
Growth of 6.4 Percent in 2017.” Press release. Available at:
https://ota.com/news/press-releases/20236 [Accessed October 29, 2018].
Ureña, F., R. Bernabéu, and M. Olmeda. 2008. “Women, Men and Organic Food:
Differences in Their Attitudes and Willingness to Pay. A Spanish Case Study.”
International Journal of consumer Studies 32(1):18-26.
U.S. Department of Agriculture Economic Research Service [USDA ERS]. 2017. “New
Products.” Available at: https://www.ers.usda.gov/topics/food-markets-
prices/processing-marketing/new-products/ [Accessed August 21, 2018].
Van Doorn, J., and P.C. Verhoef. 2011. “Willingness to Pay for Organic Products:
Differences between Virtue and Vice Foods.” International Journal of Research
in Marketing 28(3):167-180.
8
CHAPTER 1
ORGANIC WHEAT PRICES AND PREMIUM UNCERTAINTY: DO CROSS
HEDGING AND FORECASTING PLAY A ROLE?
1.1 Abstract
We compare the volatility of organic wheat prices to that of conventional wheat prices
using historical measures. To reduce uncertainty associated with organic wheat prices and
organic premium, we examine the possibility of cross hedging using conventional wheat
futures and the ability of futures to forecast the organic prices. Results provide some
evidence that conventional futures can be used to cross hedge organic wheat price risk, but
results depend on the method used to impute the missing values and time period. Similarly,
we find a long-run equilibrium relationship between organic wheat prices and conventional
wheat futures prices. Finally, futures prices contain some information useful in predicting
organic prices in the short run.
1.2 Introduction
Organic foods have gained popularity among consumers in the US during the last two
decades, documented by the significant increase in consumers' demand for organic food.
For example, the sales of organic foods have increased more than 10-fold in the period
between 1997 and 2017, from $3.4 billion to $45.2 billion (Organic Trade Association,
2018), despite the significant organic premiums that most organic food products command.
Many studies have investigated what drives consumer demand for organic food. As
summarized in Hughner et al. (2007), the common reasons include beliefs that organic food
9
is healthier, safer, and tastes better than conventionally produced food. Also, consumers of
organic food believe that organic production is better for the environment, promotes animal
welfare, and supports local economies. The interest in organic foods coupled with organic
premiums has encouraged some producers to switch from conventional to organic
production, but in many markets the supply increases have not kept up with increases in
consumer demand.
The organic wheat market has experienced increased scarcity as demand for organic
wheat products has grown. Organic wheat acreage for both food and feed production
represented just 1% of total U.S. wheat acreage and 6.7% of total U.S. organic acreage in
2016 (U.S. Department of Agriculture, 2017a), but bread and grains accounted for 9% of
overall organic food consumption in 2012 (U.S. Department of Agriculture, 2017b). The
Organic Trade Association (OTA) found that growth in the organic grain market “could
have been even more robust in 2015 if greater supply had been available” (OTA, 2016),
suggesting that demand growth in the organic grain market has outpaced supply growth in
recent years.
Agricultural production is inherently risky since yields are largely affected by
factors outside of producers’ control, such as weather, pests, and diseases. In addition,
agricultural commodity prices and market conditions at harvest are unknown when
production decisions are made. Producers who adopt organic production practices face
additional challenges and restrictions from the National Organic Program (U.S.
Department of Agriculture, 2002), which defines national standards for the organic
production system. These restrictions include limited use of chemical inputs such as
10
fertilizers and pesticides and usually lead to reduced yields (Lotter, 2003; Korsaeth, 2008;
Seufert, Ramankutty, and Foley, 2012; De Ponti, Rijk, and Van Ittersum, 2012) and higher
total production costs per bushel as a result. While limits on fertilizer and pesticide use
may decrease the per acre operating costs of organic compared to conventional grain
production, the total per acre economic costs of producing organic grains may be higher
when other costs—such as labor and land—are included (McBride et al., 2015). Lower per
acre yields in organic production further increase total costs per bushel compared to
conventional production.
Looking at wheat specifically, McBride et al. (2012) find that the additional
operating, capital, and economic costs of producing organic wheat were $2–$4/bu over
conventional wheat in 2009,
2
while the organic wheat premium was $3.79/bu, indicating
that the higher costs of producing organic wheat can be offset with higher organic prices.
Thus, organic wheat production can be more profitable than the conventional production,
assuming the transition period to organic production has already been made. But it also
indicates that the relative profitability of the organic wheat production depends on the
organic premium, which in turn depends on how organic and conventional wheat prices
develop over time. Organic wheat prices have changed rapidly in past years, leading to an
overall increase in excess of 140% between 2010 and 2017 and positively affecting the
organic premium. Although current organic wheat prices allow for profitable organic wheat
production in the West, growers face uncertainties regarding the length of favorable market
2
The authors used data from 2009 Agricultural Resource Management Survey (ARMS) of organic and
conventional wheat growers. The higher cost to produce organic wheat was driven by lower yields of organic
wheat production (30 bu/acre) compared to those of conventional wheat production (44 bu/acre). The authors
also accounted for the cost of transitioning to organic wheat production.
11
conditions, which potentially affect their decision to begin or continue dryland organic
wheat production. This study evaluates the uncertainty associated with organic prices and
premiums and explores options that growers may have to manage this uncertainty.
This essay has three primary objectives. First, we compare the risks associated with
organic and conventional wheat prices by examining the historical volatilities and evaluate
the organic premium risk by calculating the probability that the organic premium falls
below additional costs of producing organic wheat. We hypothesize that organic wheat
prices are more volatile than conventional wheat prices, making organic wheat production
riskier from the perspective of growers, possibly affecting negatively growers’ perceptions
about organic wheat profitability and acting as a barrier to adopting organic wheat
production practices.
Second, we explore some options that growers considering conversion to or
maintaining organic wheat production have to manage the price risk associated with
organic wheat. More specifically, we investigate whether hedging in conventional wheat
futures mitigates the organic wheat price risk and whether conventional futures prices can
be used to predict organic cash prices. Since the number of cash transactions on the organic
wheat market is likely not large enough to support trading in organic wheat futures, we
consider the alternative of using conventional wheat futures to cross hedge organic price
risk. To estimate the optimal hedge ratio, we use a cointegration approach, which is based
on the concept of market integration.
Our analysis is complicated by the limited availability of historical organic wheat
price data and missing observations in the data that are available. Our third objective is to
12
simulate missing organic wheat prices. We use three methods to add robustness to our
analysis and to help determine whether our results are sensitive to the methods. This will
allow us to highlight possible limitations and provide more validity to our results. In
addition, we investigate which method is the optimal one.
1.3 Background and Literature Review
1.3.1 Price Volatility as a Measure of Price Uncertainty
Price volatility is defined as the deviation of a price from its mean value or price
movements within a short period of time (Balcombe, 2010). Higher volatility makes it
harder to predict future prices and creates more uncertainty associated with future price
expectations. In general, commodity prices are highly volatile (Deaton and Laroque, 1992;
Pindyck, 2004). The main factors that contribute to price volatility for agricultural
commodities are instability of supply due to weather, pests, and weeds; inelastic short-run
demand and supply; changes in food and agricultural policies; and current prices that affect
production decisions (Demeke et al., 2012). McKay (2016) compares the price volatility
of organic commodities and their conventional counterparts from 2007–2015 and finds that
organic corn, soybeans, and oats prices were less volatile than conventional prices, while
organic wheat and barley prices were more volatile. Higher volatility of prices for organic
commodities can affect the risk perceptions associated with the relative profitability of
organic production. We compare the volatilities of wheat prices using historical volatilities,
while McKay (2016) uses a coefficient of variation.
13
1.3.2 Organic versus Conventional Production Profitability and Risk Perceptions
While some studies have found that organic production is less profitable than conventional
production (Dobbs and Smolik, 1997), other studies have found the opposite. Mahoney et
al. (2004) find that net returns for selected organic crops are significantly larger than those
for conventional crops, and they are statistically equal when organic price premiums were
not considered. Delbridge et al. (2013) consider the possible differences in the size of
organic and conventional farms to evaluate whole-farm net returns for a corn–soybean
rotation and find that risk-averse growers would be better off adopting organic production
practices. However, this result is sensitive to the changes in the organic premium and
yields. Similarly, Archer et al. (2007) find that during the period of transitioning to
organic—when growers do not receive organic premiums for their crops—the rotation
systems of corn, soybean, and wheat generate lower net present values than do
conventional systems. However, results for organic production are more positive when
organic premiums are considered. These studies suggest that the profitability of organic
production depends on the price premiums, which in turn depend on how organic and
conventional prices develop over time. Thus, higher volatility and uncertainty of organic
prices, if found, can affect rates of adoption or continuation of organic production and
confirms the value of having tools to manage the uncertainty associated with organic price
premiums.
1.3.3 Hedging and Optimal Hedge Ratio
Investments in agricultural production generally occur well before harvest; in the interim,
prices usually change. Hedging is one tool used to mitigate the risk associated with price
14
changes in agriculture. Often, the most efficient hedge is not a one-to-one hedge. In other
words, not all the spot risk is hedged in the futures market. Instead, hedgers apply an
optimal hedge ratio (OHR). Traditionally, it is calculated as the ratio of the covariance
between spot and futures prices, 𝑆𝑡 and 𝐹𝑡, to the variance of the futures prices, 𝐹𝑡, (Myers
and Thompson, 1989) with the goal to minimize the variance of the portfolio, expressed as
(1-1) 𝜆∗=𝐶𝑜𝑣(𝑆𝑡,𝐹𝑡)
𝑉𝑎𝑟(𝐹𝑡).
OHR can be estimated using regression analysis, but several techniques using
different assumptions have been used in the literature. Some studies assume a constant
(static) OHR over time, which can be estimated using ordinary least squares (OLS)
estimation methods (e.g., Rolfo, 1980; Wilson, 1982; Benninga, Eldor, and Zilcha, 1984;
Figlewski, 1994). Other studies relax this assumption by allowing the distribution of spot
and futures prices to vary over time, making it possible to estimate a time-variant (dynamic)
OHR using variations of generalized autoregressive conditional heteroskedasticity
(GARCH) and stochastic volatility models (Cecchetti, Cumby, and Figlewski, 1988;
Baillie and Myers, 1991; Park and Switzer, 1995; Chang, McAleer, and Tansuchat, 2011;
Revoredo-Giha and Zuppiroli, 2014).
3
Although some studies show that assuming a time-
invariant OHR is not appropriate (Baillie and Myers, 1991), others show that using more
complex models to account for a time-variant OHR does not lead to significant reduction
in portfolio variance (Lien, Tse, and Tsui, 2002; Lien and Tse, 2002; Cotter and Hanly,
2012). Lien and Luo (1994) compare the hedging performance of a GARCH model and a
3
Lien and Tse (2002) provide a thorough review of the traditional (static) and recently developed (dynamic)
hedging frameworks.
15
vector error correction (VEC) model in the presence of conditional heteroskedasticity using
data from major foreign exchange markets. While the GARCH model has been found to
have better statistical performance when estimating OHRs, it has not been found to have
better hedging performance, which is what ultimately matters to a hedger. Hence, we start
with the assumption that the OHR is constant over time and examine the validity of that
assumption by performing specification tests.
Regardless of methodology, most studies find that the OHR is less than unity,
meaning that the naïve method of hedging all expected production using futures contracts
is usually not appropriate. Looking at wheat specifically, Wilson (1982) examines the
efficiency of the U.S. futures markets for several wheat varieties and finds that the time-
invariant OHR is less than unity and the risk is reduced more if nearby futures contracts
are used as opposed to those in the more distant future. Revoredo-Giha and Zuppiroli
(2014) compare the effectiveness of short-term hedging of wheat price risk using U.S. and
European futures markets, while considering time-varying OHRs. They find that U.S.
futures markets can reduce the price variance of the portfolio by 77% with the OHR close
to unity, while European markets reduce the variance by only 30% with the OHR
significantly less than unity.
1.3.4 Cross Hedging
The organic wheat market is considered thin. Because of the lack of liquidity, there is no
futures market for organic wheat. Thus, we evaluate the possibility of cross hedging, which
involves hedging in a futures market for a related commodity. The challenging task is to
find a related commodity. According to Anderson and Danthine (1981), the correlation
16
between prices of the hedged commodity and the related futures commodity must be
significantly different from 0.
Several studies have examined the possibility of cross hedging when no futures
contract is established for the commodity in question. Blake and Catlett (1984) simulate a
routine cross hedge and find that the use of corn futures to manage the price risk of hay
increased gross returns per ton of hay. Zacharias et al. (1987) apply a numerical simulation
approach and find that growers can benefit from cross hedging the price risk of rough rice
using wheat futures. On the other hand, Coffey, Anderson, and Parcell (2000) find that
cross hedging the price risk of grain by-products (corn gluten feed, hominy, distiller’s dried
grain) using corn futures fails to perform efficiently.
This study builds on the previous literature by examining the possibility of cross
hedging organic wheat price risk using conventional wheat futures. To estimate the OHR,
we use the cointegration approach, which is based on the concept of market integration.
Understanding market integration not only allows us to estimate cross-hedge OHR but also
investigate the dynamics between organic spot prices and conventional futures prices. This,
in turn, can be used to evaluate the potential of conventional futures prices to predict
organic spot prices.
1.3.5 Market Integration
If the same information is used to form expectations about supply and demand in two
different markets, these markets and their prices become linked. The strength of the linkage
between prices can be examined by investigating their long-run and short-run relationships.
If nonstationary prices share a stable long-run equilibrium, then the markets are said to be
17
cointegrated. In this case, if one of the prices deviates from this equilibrium due to a shock
in the market, an adjustment will take place to re-establish the equilibrium relationship.
For cointegration between two markets to exist, the prices need to be nonstationary
in levels. Given two nonstationary series, 𝑥𝑡 and 𝑦𝑡, the series are said to be cointegrated
if there exists a unique 𝛽1 that renders the difference 𝑦𝑡−𝛽0−𝛽1𝑥𝑡=𝑢𝑡 stationary. In
this case, 𝛽1 is a cointegrating parameter and the difference 𝑦𝑡−𝛽0−𝛽1𝑥𝑡 is a
cointegrating regression.
4
Traditionally, market integration has been examined between spot markets for the
same commodity connected horizontally across space (Gonzalez-Rivera and Helfand,
2001; Rapsomanikis, Hallam, and Conforti, 2006; Asche et al., 2012; Rosa, Vasciaveo, and
Weaver, 2014) or vertically along the supply chain (Cramon-Taubadel, 1998; Pozo,
Schroeder, and Bachmeier, 2013) and between different commodities acting as substitutes
(Campiche et al., 2007; Rosa, Vasciaveo, and Weaver, 2014). Some studies have examined
market integration specifically between spot markets for organic and conventional
commodities, which are qualitatively differentiated but can potentially act as substitutes to
some extent (Kleemann and Effenberger, 2010; Singerman, Lence, and Kimble-Evans,
2014; Würriehausen, Ihle, and Lakner, 2015; Nemati and Saghaian, 2016; Ankamah-
Yeboah, Nielsen, and Nielsen, 2017).
Other studies have used the cointegration concept to investigate whether futures
prices can be used to forecast spot prices and to examine the efficiency of futures markets
4
Maddala and Kim (1999) provide detailed review of cointegration. Rapsomanikis, Hallam, and Conforti
(2006) offer a brief review of cointegration and testing for cointegration.
18
in transmitting price signals to spot markets (Bessler and Covey, 1991; Lai and Lai, 1991;
Wahab and Lashgari, 1993; Beck, 1994; Fortenbery and Zapata, 1997; Aulton, Ennew, and
Rayner, 1997; McKenzie and Holt, 2002; Wang and Ke, 2005; Carter and Mohapatra,
2008). Understanding the relationships between futures and cash markets can be helpful in
determining how changes in futures markets can impact spot prices. If current futures
prices are unbiased forecasts of future spot prices, then the futures markets are said to be
efficient and can be used to forecast future spot prices. Fewer studies have used
cointegration specifically to estimate OHRs, and their overall aim was to compare the
effectiveness of cointegration and conventional approaches in the process of OHR
estimation (see the Methods section).
1.4 Data
We use monthly farm gate/FOB organic and conventional food grade wheat prices as spot
prices, which were collected from the Agricultural Marketing Service (AMS) and the
Economic Research Service (ERS) agencies of the U.S. Department of Agriculture
(USDA) between January 2008 and August 2017. In total, 116 pricing observations were
obtained for conventional wheat and 85 observations for organic wheat, with 26.7% of the
organic wheat prices missing.
To add robustness to our analysis and to examine whether the results are sensitive
to the methods used, we imputed values for the missing organic prices using three methods:
i) spline interpolation, ii) exponential weighted moving average, and iii) an expectation-
19
maximization with bootstrapping (EMB) algorithm.
5
While the first two methods consider
only observations in the proximity of the missing values, the EMB algorithm utilizes the
whole distribution of the data in the imputation process. In addition, it accounts for the time
series nature of the data.
Futures prices for conventional wheat correspond to the soft red winter variety
traded at the Chicago Board of Trade (CBOT) and are collected from the Commodity
Research Bureau (CRB) with monthly frequency. Futures contracts are available for 5
delivery months in each year—March, May, July, September, and December. The futures
price series is a collection of nearby futures prices between January 2008 and August 2017,
with a total of 116 observations. We roll over to the contract with the next available
delivery month in the month of an actual delivery period. For example, for the futures
contract with a maturity date in March, we record futures prices up to February. In March,
we use the price of the May contract.
All spot and futures prices used in the analysis are deflated using the seasonally
adjusted consumer price index for cereals and bakery products. Figure 1-1 shows the plot
5
The spline interpolation method fills in missing values by connecting observed values immediately before
and after the missing values using a smooth curve. The exponential weighted moving average method
calculates a missing value by taking the average of several observed values before and after the missing
value, with the observations immediately before and after the missing observation receiving the highest
weight. Weights decline exponentially with more distant observations. The EMB algorithm works under the
assumption that the complete data (observed and unobserved) follow a multivariate normal distribution with
the distribution parameters (μ,Σ) = θ and that the data are missing at random. First, the algorithm finds the
posterior distribution of the complete-data parameters θ given the observed data and then it takes m draws of
θ from this posterior distribution. In the next step, missing data are obtained by drawing values from the
complete-data distribution conditional on the observed data and the draws of θ, creating m sets of complete
data. In the last step, we combine m imputed values by taking a simple average using the Amelia II package
developed by Honaker, King, and Blackwell (2011). We choose m = 10 in our analysis, but the Honaker,
King, and Blackwell (2011) note that m = 5 is usually adequate.
20
of observed organic wheat spot prices and conventional wheat spot and nearby futures
prices. Organic and conventional wheat prices tend to move in the same direction,
suggesting a potential long-run relationship between the price series. But the difference
between the prices (i.e., the organic premium) varies over time. The plots show no clear
trend in the development of prices, as periods of price increases and decreases follow one
another. Lastly, the plots suggest that organic prices are less stable than conventional
prices. Also, as expected, conventional futures and spot prices follow each other closely.
Figure 1-2 depicts observed organic prices as well as prices obtained using the three
imputation methods. There are some differences in the imputed organic prices across the
three methods, particularly around 2016, when no data were observed for several
consecutive months.
Figure 1-1. Observed monthly organic wheat spot prices, and conventional wheat
spot and futures prices, 01/2008–08/2017 ($/bu)
21
Figure 1-2. Observed monthly organic wheat prices compared to complete organic
prices obtained using three imputation methods ($/bu)
Table 1-1 reports summary statistics for all price series and the organic premium,
calculated as the difference between organic spot prices (observed and imputed) and
conventional spot prices. Organic wheat prices are on average double conventional wheat
prices. Similarly, the range of the organic prices is double the range of conventional prices.
The standard deviation for each organic price series is relatively large compared to
conventional wheat (spot and futures) prices, indicating higher uncertainty associated with
organic prices. Using an F-test, we find that the differences in variance between organic
prices and conventional spot and futures prices are statistically significant.
22
Table 1-1. Summary Statistics for Conventional Wheat Futures and Spot Prices,
Organic Wheat Spot Prices, and Organic Premium, January 2008–August 2017
($/bu)
No.
of Obs.
Mean
St.
Dev.
Min.
Max.
Range
Conventional futures pricesa
116
5.41
1.44
3.38
10.70
7.32
Conventional spot prices
116
5.38
1.44
2.93
10.19
7.27
Organic spot prices
Observed
85
11.96
3.96
5.02
23.91
18.89
EMB algorithm
116
11.92
3.92
5.02
23.91
18.89
Spline interpolation
116
11.80
4.18
5.02
23.91
18.89
Exponential weighted
moving avg.
116
11.85
3.85
5.02
23.91
18.89
Organic premium
Observed
85
6.41
3.72
0.55
15.66
15.11
EMB algorithm
116
6.54
3.77
0.29
15.86
15.57
Spline interpolation
116
6.42
4.00
0.50
18.27
17.77
Exponential weighted
moving avg.
116
6.48
3.67
0.55
15.66
15.11
Note: a Nearby futures prices (soft red winter variety); that is, prices for the nearest futures contract. The
contract is rolled over to the second-nearest contract the day before an actual delivery period.
As mentioned previously, the price risk of a commodity can be cross hedged by
taking a position on the futures market for a related commodity, under the condition that
the correlation between the prices of the two commodities be significantly different from
0 (Anderson and Danthine, 1981). In general, stronger correlations create more effective
hedges. Table 1-2 reports the correlations between conventional futures prices and the three
organic spot price series. We find the correlations to be significant at the 90% confidence
level and between 0.15 and 0.17, depending on the method used to impute missing organic
prices. Positive correlations indicate that the spot and futures prices move in the same
direction more than half the time, implying that hedging could be risk-reducing.
23
Table 1-2. Correlations between Conventional Futures Prices and Organic Spot
Prices
Conventional
Futures
Organic EMB
Algorithm
Organic
S.I.a
Organic
E.W.M.A.b
Conventional Futures
1.000
0.152*
0.157*
0.166*
Organic EMB Algorithm
1.000
0.940***
0.968***
Organic S.I.a
1.000
0.963***
Organic E.W.M.A.b
1.000
Note: *, **, *** denote significance at the 10%, 5%, and 1% level, respectively.
a Spline Interpolation
b Exponential Weighted Moving Average
1.5 Methods
1.5.1 Organic Premium Risk and Price Risk Evaluation
To evaluate the risk associated with the organic premium, we first find the best-fitting
probability density. Since four organic wheat price series are available (one observed and
three imputed using the three imputation methods), we obtain four organic premium series.
We use kernel density to fit each organic premium set because it does not impose any
potentially limiting assumptions about the distribution of the data.
6
In the next step, we
sample 10,000 values from each kernel density. The values are drawn from each fitted
kernel density with the probability that is attached to each value of the fitted density, so
that the density of the simulated values comes close to the fitted kernel density. The
simulated values are then used to calculate the probability that the organic premium falls
below the additional costs of producing organic.
6
The Epanechnikov (quadratic) kernel is chosen for the kernel function, since it can be shown that it is an
optimal kernel, but in general the choice of kernel is not critical (Cameron and Trivedi, 2005). The unbiased
cross-validation method is used for the bandwidth selection, as it is entirely data-driven and minimizes the
integrated squared error, which is a global measure evaluating the performance of the kernel smoothing at all
data points (Cameron and Trivedi, 2005).
24
Next, we calculate historical volatilities of organic and conventional prices. This
enables us to compare how much prices change from one period to another and how much
uncertainty is associated with the change. Since we have monthly data, we calculate
monthly historical volatilities and we calculate the moving volatilities over a period of 12
months, following the standard procedure (described, e.g., in Figlewski, 1994).
1.5.2 Estimation of Optimal Hedge Ratio
Historically, an OLS regression of spot prices, 𝑆𝑡, on futures prices, 𝐹𝑡, at time t, with both
prices either in levels, differences, or as percentage changes, has been used to estimate the
OHR, expressed as
(1-2) 𝑆𝑡=𝜇+𝜆𝐹𝑡+𝜀𝑡,
where slope coefficient 𝜆 is the OHR. However, the effectiveness of this approach is
limited since the OHR obtained from equation (1-2) does not account for the past
information available to the hedger at time t (Myers and Thompson, 1989), and it likely
yields an unreliable OHR if the relationships between the spot and futures prices are not
specified correctly (Ghosh, 1993). Thus, we apply a method that extends this simple OLS
approach and include lags of futures and spot prices that may play a role in explaining the
movements in spot prices and capture the short-run relationships between the prices. We
also incorporate the cointegration relation, when it exists, between spot and futures prices,
to account for the long-run relationships between the prices. As summarized in Lien and
Tse (2002), several studies (e.g., Lien and Luo, 1994; Ghosh, 1993; Wahab and Lashgari,
1993; Chou, Denis, and Lee, 1996) have found that this cointegration approach performs
25
better than the simple OLS approach in equation (1-2). If cointegration is not found, we
estimate
(1-3) 𝛥𝑂𝑆𝑡=𝜇+𝜆𝛥𝐶𝐹𝑡+∑𝛽𝑖𝛥𝑂𝑆𝑡−𝑖
𝑘
𝑖=1 +∑𝛾𝑗𝛥𝐶𝐹𝑡−𝑗
𝑙
𝑗=1 +𝜀𝑡,
as proposed by Myers and Thompson (1989). If cointegration is found, we add the error
correction term to obtain
(1-4) 𝛥𝑂𝑆𝑡=𝜇+𝜆𝛥𝐶𝐹𝑡+∑𝛽𝑖𝛥𝑂𝑆𝑡−𝑖
𝑘
𝑖=1 +∑𝛾𝑗𝛥𝐶𝐹𝑡−𝑗
𝑙
𝑗=1 +𝛼𝑍𝑡−1 +𝜀𝑡,
as described in Lien and Tse (2002). In each case, the OHR is the estimate of the slope
coefficient 𝜆. In these equations, 𝛥𝑂𝑆𝑡 is the difference between the organic wheat spot
prices in two time periods 𝑂𝑆𝑡−𝑂𝑆𝑡−1, 𝛥𝐶𝐹𝑡 is the difference between the conventional
wheat futures prices in two time periods 𝐶𝐹𝑡−𝐶𝐹𝑡−1, 𝛥𝑂𝑆𝑡−𝑖 is the ith lag of the organic
spot price difference, and 𝛥𝐶𝐹𝑡−𝑗 is the jth lag of the conventional futures price difference.
The number of lags, k and l, is determined by the Akaike Information Criterion (AIC), and
𝑍𝑡−1 in equation (1-4) is the lagged error correction term, obtained from the regression
between 𝑂𝑆𝑡 and 𝐶𝐹𝑡,
(1-5) 𝑂𝑆𝑡=𝛼+𝛽𝐶𝐹𝑡+𝑧𝑡.
The regression analysis applied is a part of either a structural vector autoregressive
(SVAR) or a structural vector error correction (SVEC) model, depending on whether
equation (1-3) or (1-4) is estimated, respectively. Typically, estimating SVAR and SVEC
models in the case of a bivariate price analysis involves a simultaneous estimation of the
system of two equations, where each price variable is in function of its own lags and lags
26
of the other price variable, and the contemporaneous relationship between the two price
variables is captured in one of these two equations only. Given our interest in estimating
the OHR, we only consider equations with organic price set as the dependent variable.
Following the theory behind OHR calculation, we include the contemporaneous effect of
conventional futures prices in the equation.
1.5.3 Examination of Relationships between Prices
In addition to estimating the OHR, equations (1-3) and (1-4) also allow us to examine long-
run and short-run relationships between organic spot and conventional futures prices.
Understanding these relationships provides insights into the possibility of predicting
organic spot prices using conventional futures prices. Following Rapsomanikis, Hallam,
and Conforti (2006), we perform short-run and long-run causality tests to determine
whether futures prices can be used to predict organic prices (or vice versa). Short-run
causality is examined using Granger causality tests, following Toda and Yamamoto (1995).
Using this procedure, we apply a Wald test to determine whether prediction of one price
variable improves if lags of the other price variable are included in the vector
autoregressive (VAR) model. The model is estimated using prices in levels. If the joint
effect of past lags of price series 𝑥𝑡 is significantly different from 0 in the equation with
the price series 𝑦𝑡 as the dependent variable, then 𝑥𝑡 is said to Granger-cause 𝑦𝑡, and past
values of 𝑥𝑡 can be used to improve the prediction of 𝑦𝑡.
If cointegration is found between the prices, then we examine long-run causality by
applying a standard t-test to the coefficient of the error correction term estimated using
equation (1-4). If the coefficient on the error correction term is significantly different from
27
0 in the equation with price series 𝑦𝑡 as dependent variable, then the long-run causality
runs from 𝑥𝑡 to 𝑦𝑡.
1.6 Results
1.6.1 Risks Associated with Organic Premium and Prices
The best-fitting kernel densities for organic premiums are in figure 1-3. Densities of
organic premiums are far from being normal, as they are visibly skewed to the left,
indicating that an organic premium of lower value is more likely to occur. The fitted density
of organic premium calculated with organic prices imputed using the exponential weighted
moving average is closest to the fitted density of the observed organic premium.
Observed EMB Algorithm
Spline Interpolation Exponential Weighted Moving Avg.
Figure 1-3. Histogram of data and kernel density for organic premiums
28
In the next step, 10,000 draws are taken from each estimated kernel density to
obtain the probability of the premium being less than $4/bu as an upper limit and less than
$2/bu as a lower limit (the estimated increased costs for organic wheat, following McBride
et al., 2012).
7
We also obtain the probability of organic premiums above $8/bu. This allows
growers to cover organic costs across two periods. Table 1-3 reports the mean values,
standard deviations and probabilities, calculated using the simulated organic premiums.
With a simulated mean organic premium of $6.47–$6.63/bu, organic wheat growers would
more than offset the higher cost per bushel by producing organic wheat. However, the
calculated probabilities need to be examined to understand the risk associated with the
premium.
Table 1-3. Means, Standard Deviations, and Probabilities Calculated Using
Simulated Organic Premiums
Organic Premium
Mean
($/bu)
St. Dev.
($/bu)
Pr(<$2)
(%)
Pr(<$4)
(%)
Pr(>$8)
(%)
Observed
6.52
3.97
10.44
30.64
32.37
EMB algorithm
6.62
3.92
9.37
30.24
34.83
Spline interpolation
6.47
4.21
11.40
33.31
31.85
Exponential weighted moving avg.
6.63
3.98
10.86
30.45
35.18
The probability of observing an organic premium below the maximum additional
cost of $4/bu is 30.2%–33.3%. In other words, if the premium falls below $4/bu, which
happens approximately one-third of the time, the grower may be unable to cover the
additional costs, resulting in lower profitability of the organic wheat compared to
7
The probabilities were calculated as follows: First, all 10,000 values were ordered from the lowest to the
highest and then, to calculate the probability of organic premium being below $x/bu, the count of all drawn
values below or equal to $x was divided by 10,000.
29
conventional. Results also show that the probability of organic premiums below $2/bu is
9.4%–11.4%, which means that in 10% of the time, wheat growers will not receive organic
premiums sufficient to cover the additional costs. However, the lower relative profitability
in one period is likely compensated by higher profitability in other periods. As results in
table 1-3 show, the probability of organic premiums being above $8/bu, which is enough
to cover the additional costs of producing organic for two periods, is 31.9%–35.2%.
It is important to note that all calculated probabilities are unconditional, which
means they represent the probability of an event occurring over the entire observed period,
not taking into consideration specific values observed today. For example, if a premium
below $4/bu is observed today, the probability of observing a premium below $4/bu in the
next month is more than 30% due to the time series nature of the data and strong
dependence between observations in two adjacent time periods. However, as the time
passes, the dependence weakens, and higher premiums may be more likely to be observed.
Figure 1-4. Annualized historical volatilities
30
Next, we calculated monthly historical volatilities of organic and conventional
wheat prices and annualized them by multiplying each calculated volatility by √12, plotted
in figure 1-4. The plots provide visual evidence that organic prices are more volatile than
conventional prices. Annualized volatilities of conventional prices are 10%–40%, while
the annualized volatilities of organic prices are 20%–90%, double that of conventional
prices. This shows that organic prices tend to change more dramatically and are less stable
over a short horizon. The plots also show that periods of higher volatility are followed by
periods of lower volatility in the case of organic prices, while the volatilities of
conventional prices are relatively stable over time. This suggests that if more organic wheat
production is desired, more risk-averse growers may need tools that will enable them to
efficiently manage the risk associated with organic prices and premiums.
1.6.2 Time Series Properties of the Data
As a first step in any regression analysis involving time series, it is necessary to examine
whether the time series are stationary using unit root tests. We apply three commonly used
tests to determine whether the price series used in the analysis are stationary: the
Augmented Dickey–Fuller (1979) ADF test, the Phillips–Perron (1988) PP test, and the
Kwiatkowski–Phillips–Schmidt–Shin (1992) KPSS test. We use all three since some tests
perform better in certain circumstances.
8
8
For example, some studies suggest that the Augmented Dickey–Fuller test may perform poorly and tend to
accept the null of nonstationarity in the presence of serial correlation or heteroskedasticity (Rapsomanikis,
Hallam, and Conforti, 2006; Esposti and Listorti, 2013).
31
For each set of organic cash and conventional futures prices, we confirm that the
prices are nonstationary in levels and stationary after first differencing.
9
This leads us to
test for cointegration between cash and futures prices. We apply the maximum likelihood
method developed by Johansen (1988, 1991). We use the AIC to determine the number of
lags, k, to be used. We estimate the trace and maximum eigenvalue statistics using a
constant in the cointegrating equation. Table 1-4 reports trace test statistics, 𝜆𝑡𝑟𝑎𝑐𝑒, and
maximum eigenvalue test statistics, 𝜆𝑚𝑎𝑥, for each pair of prices. The null hypothesis of
no cointegrating relationship (𝑟=0) is rejected for all three pairs of conventional futures
prices with organic spot prices using at least one of the two estimated statistics; we
therefore estimate equation (1-4) for each pair.
Table 1-4. Johansen Cointegration Test Results
Number of Cointegrating Vectors =
Rank (r)
Null
Alterna-
tive
𝜆𝑡𝑟𝑎𝑐𝑒
𝜆𝑚𝑎𝑥
Conventional futures and organic spot
r =0
r =1
16.65
13.99*
prices (EMB algorithm)
r =1
r =2
2.65
2.65
Conventional futures and organic spot
r =0
r =1
22.07**
15.03*
prices (spline interpolation)
r =1
r =2
7.04
7.04
Conventional futures and organic spot
r =0
r =1
r =1
20.71**
5.85
14.87*
prices (exponential weighted moving avg.)
r =2
5.85
Note: * and ** denote significance at the 10% and 5% level, respectively.
9
Unit root test results are available from the authors upon request.
32
1.6.3 Cross Hedge for Organic Wheat Using Conventional Wheat Futures
Table 1-5 reports the results of estimating the OHRs. The AIC selected 3 lags, 1 lag, and 1
lag as optimal for the regressions involving organic prices imputed using the EMB
algorithm (Model 1), the spline interpolation (Model 2), and the exponential weighted
moving average (Model 3), respectively. Table 1-5 also reports results of misspecification
tests. We fail to reject the null of no autocorrelation using the Box–Ljung test for all three
models. This means the models are well specified in terms of the number of included lags.
Table 1-5. Regression Results
𝛥𝑂𝑆𝑡=𝜇+𝜆𝛥𝐶𝐹𝑡+∑𝛽𝑖𝛥𝑂𝑆𝑡−𝑖
𝑘
𝑖=1 +∑𝛾𝑗𝛥𝐶𝐹𝑡−𝑗
𝑙
𝑗=1 +𝛼𝑍𝑡−1 +𝜀𝑡
Model 1
(EMB Algorithm)
Model 2
(Spline Interpolation)
Model 3
(Exponential
Weighted Moving
Avg.)
Estimate
SE
Estimate
SE
Estimate
SE
𝜇
-0.041
0.194
-0.007
0.192
-0.062
0.180
𝛥𝐶𝐹𝑡
0.276
0.456
-0.841**
0.408
-0.879**
0.381
𝛥𝑂𝑆𝑡−1
-0.272***
0.097
0.024
0.092
-0.219**
0.090
𝛥𝑂𝑆𝑡−2
-0.290***
0.095
-
-
-
-
𝛥𝑂𝑆𝑡−3
-0.145
0.089
-
-
-
-
𝛥𝐶𝐹𝑡−1
0.793*
0.403
1.385***
0.382
1.092***
0.358
𝛥𝐶𝐹𝑡−2
0.518
0.408
-
-
-
-
𝛥𝐶𝐹𝑡−3
0.095
0.409
-
-
-
-
𝑍𝑡−1
-0.022
0.039
-0.008
0.017
0.000
0.014
Misspecification tests
Autocorrelation
Q-stat (lags = 2)
0.186
0.912
1.006
Conditional heteroskedasticity
𝑄(𝑚)
6.093
12.771
11.591
Rank test
18.653**
29.037***
24.369***
𝑄𝑘(𝑚)
45.049
53.059*
46.221
𝑄𝑘
𝑟(𝑚)
52.473*
46.040
45.799
Q-stat (lags = 2)
7.036**
3.238
0.803
Note: *, **, *** denote significance at the 10%, 5%, and 1% level, respectively.
33
The coefficient estimate on the differenced futures price in the current period, 𝛥𝐶𝐹𝑡,
is of primary interest because it represents the OHR. In Model 1, the coefficient estimate
is not statistically significant, implying that organic wheat price risk cannot be cross hedged
using conventional futures. In Models 2 and 3, the coefficient estimate is statistically
significant and large, but negative. Based on the calculation of OHR shown in equation
(1-1), the covariance between organic spot prices and conventional futures prices is
negative, after controlling for lags and the error correction term included in the estimated
regressions. Thus, if there is an increase in futures prices, the organic spot prices decrease
and vice versa. Typically, spot and futures prices for the same commodity are positively
correlated. In that case, growers first sell futures contracts and later, when both spot and
futures prices decline, losses in the spot market can be offset with a gain in the futures
market. But a negative OHR coefficient means that spot and futures prices move in
opposite directions, making a typical strategy of first selling futures contracts not
applicable. However, if growers purchase futures contracts first, then growers can offset
the loss in the spot market with gains in the futures market if spot prices decline and futures
prices increase. This means that cross hedging organic price risk using conventional futures
prices can be applied in practice, even if spot and futures prices move in opposite
directions. However, we find only limited evidence for the possibility of cross hedging
organic prices using conventional futures, since only two of the three estimated models
show an OHR significantly different from 0.
34
1.6.4 Relationships between Organic Spot and Conventional Futures Prices
The significance of the lagged price variables in table 1-5 suggests that there are short-run
relationships between organic spot and conventional futures prices. The results differ
slightly based on the method used to impute missing organic prices, but there is agreement
across the three models that past futures prices affect organic prices. The results of the
Wald test, applied to examine short-run Granger causality (reported in table 1-6), provide
some evidence that futures prices Granger-cause organic prices, meaning that past futures
prices contain information that helps predict current organic prices in the short run. On the
other hand, results show clearly that organic prices do not affect futures prices in the
Granger sense, regardless of the method used to impute the organic prices.
Table 1-6. Short-Run Granger Causality Tests
No. of Lags in
VAR Model
𝜒2
Statistic
p-Value
Futures prices Granger-cause organic prices
EMB algorithm
4
5.3
0.260
Spline interpolation
1
10.3***
0.001
Exponential weighted moving avg.
2
12.5***
0.002
Organic prices Granger-cause futures prices
EMB algorithm
4
2.0
0.740
Spline interpolation
1
0.8
0.380
Exponential weighted moving avg.
2
3.2
0.200
Note: *** denotes significance at the 1% level. H0: X does not Granger-cause Y (= dependent variable in
VAR model). Number of lags in VAR models (in levels) is determined based on AIC.
Results further show that coefficients from the error correction term in all three
models are not significant, although a cointegrating relationship has been found between
organic spot and conventional futures prices. The insignificance of the error correction term
in the models with organic prices as dependent variables suggests that if there is a shock to
35
the system, it is the futures price that adjusts to the deviation from the long-run equilibrium.
This has been confirmed by the significance of the error correction term in the regressions
with the conventional wheat prices as the dependent variable (not reported).
10
Although the
long-run relationship from organic prices to futures prices means that futures prices adjust
to the deviation from the long-run equilibrium, it happens slowly, over a longer period of
time. On the other hand, the relatively large short-run effect from futures prices to organic
prices means that information from the conventional futures market is passed to the organic
spot market quickly, in a relatively short time. Thus, we find some evidence that futures
prices can be used to predict organic prices, but only for short horizons.
11
1.6.5 Evaluation of Imputation Methods
Since we find that results are not robust to the methods used to impute missing organic
prices, we evaluate the performance of each imputation method based on how accurately
it predicts the values for the missing observations. First, 10% of the originally observed
organic prices (9 of 85 total observations) are dropped randomly. Then, each method is
applied to impute the values of the observations dropped from the dataset. Lastly, the root
mean squared error (RMSE)
12
is calculated using the imputed and observed values and
compared across the three methods.
10
These results are based on the specification of the cointegrating relationship in equation (1-5), where
organic spot price is set as dependent variable. However, we verified that the results are robust to the
specification of the cointegrating relationship.
11
We repeated the whole analysis using hard red winter futures, finding similar results in terms of negative
OHRs and the existence of dynamic relationships.
12
The RMSE is the root of the mean of the squared deviations between the imputed and observed values of
the organic prices: 𝑅𝑀𝑆𝐸=√1
𝑁∑(𝑃𝑖𝑚𝑝𝑢𝑡𝑒𝑑 −𝑃𝑜𝑏𝑠𝑒𝑟𝑣𝑒𝑑)2
𝑁
1.
36
The lowest RMSE (= 2.45) is found for the exponential weighted moving average
method. The RMSE value for the EMB algorithm method is 2.75, and the largest RMSE
(= 2.87) is found for the spline interpolation method. We therefore consider the results
obtained using the exponential weighted moving average method to have the highest
validity and to be the most appropriate to conclude with.
1.6.6 Further Examination of OHRs
Since we find negative OHR, which had not been expected given the positive unconditional
correlations between the conventional futures and organic spot prices (reported in table
1-2), we further examine the unconditional correlations and OHRs in selected subperiods
to determine the significance of the obtained result and whether it is driven by any
particular time period. We split the sample in two subperiods: i) January 2008–December
2012 and ii) January 2013–August 2017. We choose to split the data in this way for two
reasons. First, the initial, pooled sample is relatively small, with only 116 observations;
splitting it into two subsamples of similar sizes (60 and 56 observations) will yield
estimates of comparable statistical validity. Second, based on the plot of organic spot and
conventional futures prices in figure 1-1, the development of prices appears to become
more divergent starting in the second half of 2013. Table 1-7 reports the unconditional
correlations between the conventional futures and organic spot prices in the two
subperiods. All correlations are positive and statistically more significant than correlations
found for the pooled prices series.
37
Table 1-7. Correlations between Conventional Futures Prices and Organic Spot
Prices
EMB
Algorithm
Spline
Interpolation
Exponential Weighted
Moving Avg.
01/2008–12/2012
0.63***
0.64***
0.63***
01/2013–08/2017
0.35***
0.34***
0.36***
Note: *** denotes significance at the 1% level.
For each subperiod, we estimate the same models in terms of number of lags and
presence of cointegration as those estimated using the whole sample, with the justification
that all observations come from the same data-generating process. Table 1-8 reports models
estimated for the 2008–2012 subperiod. The magnitude of the estimated OHR remains
negative and significant in Models 2 and 3, despite strong positive unconditional
correlations. In contrast with results obtained for the whole period, 2008–2017, there is a
clear indication that there is no short-run relationship between organic spot and
conventional futures prices in 2008–2012, regardless of the methods used to impute
missing prices. However, organic prices are now found to adjust to the deviations from
long-run equilibrium, and the speed of adjustment is significantly high. Table 1-9 reports
the results of models estimated for the 2013–2017 subperiod. The OHR is positive but
insignificant, regardless of the method used to impute missing prices. Additionally, there
is no clear pattern of short-run and long-run relationships across the three methods.
To summarize, considering the results obtained using the exponential weighted
moving average method—since it has been found to be the most accurate in predicting
missing organic prices—the possibility of cross hedging organic wheat price risk is
dependent on the time period. However, there is evidence of short-run and/or long-run
relationships between the prices, depending on the time period.
38
Table 1-8. Regression Results, 01/2008–12/2012
Model 1
(EMB Algorithm)
Model 2
(Spline Interpolation)
Model 3
(Exp. Weighted
Moving Avg.)
Estimate
SE
Estimate
SE
Estimate
SE
𝜇
-0.005
0.152
-0.066
0.192
-0.065
0.190
𝛥𝐶𝐹𝑡
0.022
0.281
-0.692**
0.331
-0.654*
0.329
𝛥𝑂𝑆𝑡−1
-0.165
0.105
-0.092
0.111
-0.234**
0.111
𝛥𝑂𝑆𝑡−2
-0.063
0.098
-
-
-
-
𝛥𝑂𝑆𝑡−3
-0.159*
0.084
-
-
-
-
𝛥𝐶𝐹𝑡−1
0.278
0.298
0.415
0.396
0.035
0.382
𝛥𝐶𝐹𝑡−2
0.279
0.294
-
-
-
-
𝛥𝐶𝐹𝑡−3
0.066
0.294
-
-
-
-
𝑍𝑡−1
-0.207***
0.071
-0.264***
0.074
-0.276***
0.071
Misspecification tests
Autocorrelation
Q-stat (lags = 2)
1.012
0.442
0.151
Conditional heteroskedasticity
Q-stat (lags = 2)
0.496
1.032
0.464
Note: *, **, *** denote significance at the 10%, 5%, and 1% level, respectively.
Table 1-9. Regression Results, 01/2013–08/2017
Model 1
(EMB Algorithm)
Model 2
(Spline Interpolation)
Model 3
(Exp. Weighted
Moving Avg.)
Estimate
SE
Estimate
SE
Estimate
SE
𝜇
-0.363
0.420
-0.164
0.315
-0.149
0.289
𝛥𝐶𝐹𝑡
1.255
1.394
0.524
1.093
0.229
0.998
𝛥𝑂𝑆𝑡−1
-0.360**
0.148
0.140
0.135
-0.159
0.131
𝛥𝑂𝑆𝑡−2
-0.431***
0.141
-
-
-
-
𝛥𝑂𝑆𝑡−3
-0.165
0.148
-
-
-
-
𝛥𝐶𝐹𝑡−1
2.541**
1.253
1.640
1.090
2.185**
0.986
𝛥𝐶𝐹𝑡−2
0.812
1.372
-
-
-
-
𝛥𝐶𝐹𝑡−3
-1.827
1.424
-
-
-
-
𝑍𝑡−1
-0.032
0.025
-0.165**
0.069
-0.132*
0.068
Misspecification tests
Autocorrelation
Q-stat (lags = 2)
0.485
0.559
0.868
Conditional heteroskedasticity
Q-stat (lags = 2)
3.689
5.550*
1.389
Note: *, **, *** denote significance at the 10%, 5%, and 1% level, respectively.
39
1.7 Conclusions
In this study, we examine the profitability risk associated with organic wheat, focusing on
organic prices and premiums. As expected, we find organic prices to be more volatile than
conventional wheat prices, indicating there is more uncertainty associated with organic
wheat prices. Simulating organic premiums reveals that, depending on the method used to
impute missing prices, there is a 30%–33% probability of observing a premium below
$4/bu, assumed to be the maximum additional cost of producing organic wheat, and a 9%–
11% probability that the premium will be below $2/bu, assumed to be the minimum
additional cost of producing organic wheat. Thus, there are occasions when organic wheat
production is relatively less profitable per bushel than conventional wheat production. On
the other hand, there are more occasions when organic wheat production is more profitable
per bushel and the gains from organic premiums cover the additional costs. We find that
the probability of observing an organic premium above $8/bu is 32%–35%. However, these
probabilities are unconditional, not taking into consideration the observed premium in a
particular time period. For example, if the observed premium is low in one period, it is
likely to be low in the next time period as well.
The analysis suggests that tools to manage the risk associated with the organic
prices and premium may be needed if more organic wheat production is desired. Since the
organic premium and the organic wheat price are linked and there is no futures market
established for organic wheat, we examine the possibility of hedging the organic wheat
price risk using conventional wheat futures contracts. Results suggest that the coefficient
representing OHR is significantly different from 0 but negative. This means that there is
40
an inverse relationship between changes in organic spot and conventional futures prices.
In this case, growers looking to mitigate losses from a decrease in spot prices could cross
hedge using conventional futures prices, but they need to purchase conventional futures
contracts as their hedge. However, the statistical significance of the estimated OHRs is
sensitive to the methods used to impute the missing organic prices and the time period,
providing only limited evidence that organic price risk can be cross hedged using
conventional futures prices.
In addition to examining OHRs, the estimated models allow us to investigate the
short-run and long-run dynamics between the organic spot and conventional futures prices.
We find complex relationships between the two prices. Considering the entire studied time
period, tests of short-run Granger causality reveal that futures are weakly exogenous,
meaning that they contain some information to help predict organic spot prices in the short
run. Our analysis also provides some evidence of cointegration between organic spot and
conventional futures markets. However, we find organic prices to be weakly exogenous in
the long run, meaning that futures prices adjust to the deviations from the long-run
equilibrium relationship rather than organic prices, but the speed of adjustment is slow.
Further examination of the dynamic relationships in different time periods reveals that their
nature has been changing over time. The organic wheat market was developing in the
studied period and its lack of maturity and stability could have affected the dynamic
relationships.
We conclude that cross hedging the risk associated with organic prices using
conventional futures market might be useful to growers, but the evidence is limited. Recent
41
changes in the federal crop insurance program, which allow wheat growers to use prices
agreed to in a contract or organic wheat price election established by USDA in calculating
their compensation, make the crop insurance program likely a better option. Conventional
futures prices can be used to predict organic wheat prices, but only over a short timeframe,
based on our examination of the dynamic relationships in recent years.
This essay is the first one to examine dynamic relationships between futures and
spot prices of two qualitatively differentiated commodities—conventional and organic,
respectively. Further, it is the first study to examine the possibility of hedging the price risk
of an organically grown commodity using futures market of the same commodity but
grown conventionally. However, there are some limitations: First, we did not observe the
organic wheat prices completely and the imputed prices contain some error. Second, our
results are limited to the studied ten-year period. An extension of the dataset with
additional, future prices might affect the results, in particular since we find that the results
differ across the studied subperiods. Thus, the results need to be considered with caution.
Nevertheless, these findings are useful in providing direction for future research to examine
in more detail how conventional wheat futures prices might affect the development of
organic prices in the short run. This can be of great importance to growers and food
manufacturers as they attempt to predict the movement of organic wheat prices. Also,
future research can identify other commodities that might be more closely correlated with
organic wheat prices and could potentially be examined for cross-hedge possibilities.
42
References
Anderson, R.W., and J.P. Danthine. 1981. “Cross Hedging.” Journal of Political Econ-
omy 89(6):1182-1196. http://doi.org/10.1086/261028
Ankamah-Yeboah, I., M. Nielsen, and R. Nielsen. 2017. “Does Organic Supply Growth
Lead to Reduced Price Premiums? The Case of Salmonids in Denmark.” JEL 1:1-
17. Online at: http://orgprints.org/31425/.
Archer, D.W., A.A. Jaradat, J.M. Johnson, S.L. Weyers, R.W. Gesch, F. Forcella, and
H.K. Kludze. 2007. “Crop Productivity and Economics during the Transition to
Alternative Cropping Systems.” Agronomy Journal 99(6):1538-1547.
http://doi.org/10.2134/agronj2006.0364
Asche, F., L.S. Bennear, A. Oglend, and M.D. Smith. 2012. “US Shrimp Market
Integration.” Marine Resource Economics 27(2):181-192.
http://doi.org/10.5950/0738-1360-27.2.181
Aulton, A.J., C.T. Ennew, and A.J. Rayner. 1997. “Efficiency Tests of Futures Markets
for UK Agricultural Commodities.” Journal of Agricultural Economics 48(1-
3):408-424. http://doi.org/10.1111/j.1477-9552.1997.tb01162.x
Baillie, R.T., and R.J. Myers. 1991. “Bivariate GARCH Estimation of the Optimal
Commodity Futures Hedge.” Journal of Applied Econometrics 6(2):109-124.
http://doi.org/10.1002/jae.3950060202
Balcombe, K. 2010. “The Nature and Determinants of Volatility in Agricultural Prices:
An Empirical Study from 1962-2008.” FAO Commodity Market Review
(2009/2010):1-24.
43
Beck, S.E. 1994. “Cointegration and Market Efficiency in Commodities Futures
Markets.” Applied Economics 26(3):249-257.
http://doi.org/10.1080/00036849400000006
Benninga, S., R. Eldor, and I. Zilcha. 1984. “The Optimal Hedge Ratio in Unbiased
Futures Markets.” Journal of Futures Markets 4(2):155-159.
http://doi.org/10.1002/fut.3990040206
Bessler, D.A., and T. Covey. 1991. “Cointegration: Some Results on US Cattle Prices.”
Journal of Futures Markets 11(4):461-474. http://doi.org/10.1002/fut.3990110406
Blake, M.L., and L. Catlett. 1984. “Cross Hedging Hay Using Corn Futures: An
Empirical Test.” Western Journal of Agricultural Economics 127-134.
Cameron, A.C., and P.K. Trivedi. 2005. Microeconometrics: Methods and Applications.
Cambridge University Press. http://doi.org/10.1017/CBO9780511811241
Campiche, J.L., H.L. Bryant, J.W. Richardson, and J.L. Outlaw. 2007. “Examining the
Evolving Correspondence between Petroleum Prices and Agricultural Commodity
Prices.” Paper presented at AAEA annual meeting, Portland OR, 29 July —1
August.
Carter, C.A., and S. Mohapatra. 2008. “How Reliable are Hog Futures as Forecasts?.”
American Journal of Agricultural Economics 90(2):367-378.
http://doi.org/10.1111/j.1467-8276.2007.01122.x
Cecchetti, S.G., R.E. Cumby, and S. Figlewski. 1988. “Estimation of the Optimal Futures
Hedge.” The Review of Economics and Statistics 623-630.
http://doi.org/10.2307/1935825
44
Chang, C.L., M. McAleer, and R. Tansuchat. 2011. “Crude Oil Hedging Strategies Using
Dynamic Multivariate GARCH.” Energy Economics 33(5):912-923.
http://doi.org/10.1016/j.eneco.2011.01.009
Chou, W.L., K.F. Denis, and C.F. Lee. 1996. “Hedging with the Nikkei Index Futures:
The Conventional Model versus the Error Correction Model.” The Quarterly
Review of Economics and Finance 36(4):495-505. http://doi.org/10.1016/S1062-
9769(96)90048-4
Coffey, B., J.D. Anderson, and J.L. Parcell. 2000. “Optimal Hedging Ratios and Hedging
Risk for Grain By-Products.” Paper presented at AAEA annual meeting, Tampa
FL, 30 July —2 August.
Cotter, J., and J. Hanly. 2012. “Hedging Effectiveness under Conditions of Asymmetry.”
The European Journal of Finance 18(2):135-147.
http://doi.org/10.1080/1351847X.2011.574977
Cramon-Taubadel, S.V. 1998. “Estimating Asymmetric Price Transmission with the
Error Correction Representation: An application to the German Pork Market.”
European Review of Agricultural Economics 25(1):1-18.
http://doi.org/10.1093/erae/25.1.1
De Ponti, T., B. Rijk, and M.K. Van Ittersum. 2012. “The Crop Yield Gap between
Organic and Conventional Agriculture.” Agricultural Systems 108:1-9.
http://doi.org/10.1016/j.agsy.2011.12.004
Deaton, A., and G. Laroque. 1992. “On the Behaviour of Commodity Prices.” The
Review of Economic Studies 59(1):1-23. http://doi.org/10.2307/2297923
45
Delbridge, T.A., C. Fernholz, R.P. King, and W. Lazarus. 2013. “A Whole-Farm
Profitability Analysis of Organic and Conventional Cropping Systems.”
Agricultural Systems 122:1-10. http://doi.org/10.1016/j.agsy.2013.07.007
Demeke, M., D. Dawe, J. Tefft, T. Ferede, and W. Bell. 2012. “Stabilizing Price
Incentives for Staple Grain Producers in the Context of Broader Agricultural
Policies.” Debates and Country Experiences 12-05.
Dickey, D. A., and W.A. Fuller. 1979. “Distribution of Estimators for Time Series
Regressions with a Unit Root.” Journal of American Statistical Association
74:427-431. http://doi.org/10.2307/2286348
Dobbs, T.L., and J.D. Smolik. 1997. “Productivity and Profitability of Conventional and
Alternative Farming Systems: A Long-Term On-Farm Paired Comparison.”
Journal of Sustainable Agriculture 9(1):63-79.
http://doi.org/10.1300/J064v09n01_06
Esposti, R., and G. Listorti. 2013. “Agricultural Price Transmission across Space and
Commodities during Price Bubbles.” Agricultural Economics 44(1):125-139.
http://doi.org/10.1111/j.1574-0862.2012.00636.x
Figlewski, S. 1994. “Forecasting Volatility Using Historical Data.” Working Paper,
Finance Dept., New York University, New York City.
Fortenbery, T.R., and H.O. Zapata. 1997. “An Evaluation of Price Linkages between
Futures and Cash Markets for Cheddar Cheese.” Journal of Futures Markets
17(3):279-301. http://doi.org/10.1002/(SICI)1096-
9934(199705)17:3%3C279::AID-FUT2%3E3.0.CO;2-F
46
Ghosh, A. 1993. “Hedging with Stock Index Futures: Estimation and Forecasting with
Error Correction Model.” Journal of Futures Markets 13(7):743-752.
http://doi.org/10.1002/fut.3990130703
Gonzalez-Rivera, G., and S.M. Helfand. 2001. “The Extent, Pattern, and Degree of
Market Integration: A Multivariate Approach for the Brazilian Rice Market.”
American Journal of Agricultural Economics 83(3):576-592.
http://doi.org/10.1111/0002-9092.00179
Honaker, J., G. King, and M. Blackwell. 2011. “Amelia II: A Program for Missing Data.”
Journal of Statistical Software 45(7):1-47. http://doi.org/10.18637/jss.v045.i07
Hughner, R.S., P. McDonagh, A. Prothero, C.J. Shultz, and J. Stanton. 2007. “Who are
Organic Food Consumers? A Compilation and Review of why People Purchase
Organic Food.” Journal of Consumer Behaviour: An International Research
Review 6(2‐3):94-110. http://doi.org/10.1002/cb.210
Johansen, S. 1988. “Statistical Analysis of Cointegration Vectors.” Journal of Economic
Dynamics and Control 12(2-3):231-254. http://doi.org/10.1016/0165-
1889(88)90041-3
Johansen, S. 1991. “Estimation and Hypothesis Testing of Cointegration Vectors in
Gaussian Vector Autoregressive Models.” Econometrica: Journal of the
Econometric Society 1551-1580. http://doi.org/10.2307/2938278
Kleemann, L., and A. Effenberger. 2010. “Price Transmission in the Pineapple Market:
What Role for Organic Fruit?.” Working Paper, Kiel Institute for the World
Economy.
47
Korsaeth, A. 2008. “Relations between Nitrogen Leaching and Food Productivity in
Organic and Conventional Cropping Systems in a Long-Term Field Study.”
Agriculture, Ecosystems and Environment 127(3):177-188.
http://doi.org/10.1016/j.agee.2008.03.014
Kwiatkowski, D., P.C.B. Phillips, P. Schmidt, and Y. Shin. 1992. “Testing the Null
Hypothesis of Stationarity against the Alternative of a Unit Root.” Journal of
Econometrics 54:159-178. http://doi.org/10.1016/0304-4076(92)90104-Y
Lai, K.S., and M. Lai. 1991. “A Cointegration Test for Market Efficiency.” Journal of
Futures Markets 11(5):567-575. http://doi.org/10.1002/fut.3990110505
Lien, D., and X. Luo. 1994. “Multiperiod Hedging in the Presence of Conditional
Heteroskedasticity.” Journal of Futures Markets 14(8):927-955.
http://doi.org/10.1002/fut.3990140806
Lien, D., and Y.K. Tse. 2002. “Some Recent Developments in Futures Hedging.” Journal
of Economic Surveys 16(3):357-396. http://doi.org/10.1111/1467-6419.00172
Lien, D., Y.K. Tse, and A.K. Tsui. 2002. “Evaluating the Hedging Performance of the
Constant-Correlation GARCH Model.” Applied Financial Economics 12(11):791-
798. http://doi.org/10.1080/09603100110046045
Lotter, D.W. 2003. “Organic Agriculture.” Journal of Sustainable Agriculture 21(4):59-
128. http://doi.org/10.1300/J064v21n04_06
Maddala, G.S., and I.M. Kim. 1999. Unit Roots, Cointegration, and Structural Change.
Cambridge University Press. http://doi.org/10.1017/CBO9780511751974
48
Mahoney, P.R., K.D. Olson, P.M. Porter, D.R. Huggins, C.A. Perillo, and R.K.
Crookston. 2004. “Profitability of Organic Cropping Systems in Southwestern
Minnesota.” Renewable Agriculture and Food Systems 19(1):35-46.
http://doi.org/10.1079/RAFS200361
McBride, W.D., C.R. Greene, M.B. Ali, and L. Foreman. 2012. “The Structure and
Profitability of Organic Field Crop Production: The Case of Wheat.” Presented at
AAEA annual meeting, Seattle WA, 12-14 August.
McBride, W.D., C. Greene, L. Foreman, and M. Ali. 2015. The Profit Potential of
Certified Organic Field Crop Production. Washington DC: U.S. Department of
Agriculture, Economic Research Service, July.
http://doi.org/10.2139/ssrn.2981672
McKay, S.M. 2016. “Understanding Organic Prices: An Analysis of Organic Price Risk
and Premiums” MS thesis, Virginia Tech.
McKenzie, A.M., and M.T. Holt. 2002. “Market Efficiency in Agricultural Futures
Markets.” Applied Economics 34(12):1519-1532.
http://doi.org/10.1080/00036840110102761
Myers, R.J., and S.R. Thompson. 1989. “Generalized Optimal Hedge Ratio Estimation.”
American Journal of Agricultural Economics 71(4):858-868.
http://doi.org/10.2307/1242663
Nemati, M., and S. Saghaian. 2016. “Dynamics of Price Adjustment in Qualitatively
Differentiated Markets in the US: The Case of Organic and Conventional
49
Apples.” Paper presented at SAEA annual meeting, San Antonio TX, 6-9
February.
Organic Trade Association [OTA]. 2016. “U.S. Organic Sales Post New Record of $43.3
Billion in 2015.” Press release. Available at: https://www.ota.com/news/press-
releases/19031 [Accessed December 5, 2017].
Organic Trade Association [OTA]. 2018. “Maturing U.S. Organic Sector Sees Steady
Growth of 6.4 Percent in 2017.” Press release. Available at:
https://ota.com/news/press-releases/20236 [Accessed October 29, 2018].
Park, T.H., and L.N. Switzer. 1995. “Bivariate GARCH Estimation of the Optimal Hedge
Ratios for Stock Index Futures: A Note.” Journal of Futures Markets 15(1):61-67.
http://doi.org/10.1002/fut.3990150106
Phillips, P.C.B., and P. Perron. 1988. “Testing for a Unit Root in Time Series
Regressions.” Biometrica 75:335-346. http://doi.org/10.1093/biomet/75.2.335
Pindyck, R.S. 2004. “Volatility and Commodity Price Dynamics.” Journal of Futures
Markets 24(11):1029-1047. http://doi.org/10.1002/fut.20120
Pozo, V.F., T.C. Schroeder, and L.J. Bachmeier. 2013. “Asymmetric Price Transmission
in the US Beef Market: New Evidence from New Data.” In Proceedings of the
NCCC-134 Conference on Applied Commodity Price Analysis, Forecasting, and
Market Risk Management, St. Louis MO, 22-23 April.
Rapsomanikis, G., D. Hallam, and P. Conforti. 2006. “Market Integration and Price
Transmission in Selected Food and Cash Crop Markets of Developing Countries:
50
Review and Applications.” In Sarris, A., and D. Hallam. Agricultural Commodity
Markets and Trade. Edward Elgar, Cheltenham, UK, pp. 187-217.
Revoredo-Giha, C., and M. Zuppiroli. 2014. “Hedging Effectiveness of European Wheat
Futures Markets.” Paper presented at EAAE congress, Ljubljana Slovenia, 26-29
August.
Rolfo, J. 1980. “Optimal Hedging under Price and Quantity Uncertainty: The Case of a
Cocoa Producer.” Journal of Political Economy 88(1):100-116.
http://doi.org/10.1086/260849
Rosa, F., M. Vasciaveo, and R.D. Weaver. 2014. “Agricultural and Oil Commodities:
Price Transmission and Market Integration between US and Italy.” Bio-Based and
Applied Economics 3(2):93-117.
Seufert, V., N. Ramankutty, and J.A. Foley. 2012. “Comparing the Yields of Organic and
Conventional Agriculture.” Nature 485(7397):229-232.
http://doi.org/10.1038/nature11069
Singerman, A., S.H. Lence, and A. Kimble-Evans. 2014. “How Related Are the Prices of
Organic and Conventional Corn and Soybean?.” Agribusiness 30:309-330
Toda, H.Y., and T. Yamamoto. 1995. “Statistical Inference in Vector Autoregressions
with Possibly Integrated Processes.” Journal of Econometrics 66(1-2):225-250.
http://doi.org/10.1016/0304-4076(94)01616-8
U.S. Department of Agriculture. 2002. National Organic Program.
51
U.S. Department of Agriculture. 2017a. Certified Organic Survey 2016 Summary.
Washington DC, September. Available at:
https://www.nass.usda.gov/Publications/Todays_Reports/reports/census17.pdf
U.S. Department of Agriculture. 2017b. Organic Market Overview. Washington, DC:
U.S. Department of Agriculture, Economic Research Service. Available at:
https://www.ers.usda.gov/topics/natural-resources-environment/organic-
agriculture/organic-market-overview/
Wahab, M., and M. Lashgari. 1993. “Price Dynamics and Error Correction in Stock
Index and Stock Index Futures Markets: A Cointegration Approach.” Journal of
Futures Markets 13(7):711-742. http://doi.org/10.1002/fut.3990130702
Wang, H.H., and B. Ke. 2005. “Efficiency Tests of Agricultural Commodity Futures
Markets in China.” Australian Journal of Agricultural and Resource Economics
49(2):125-141. http://doi.org/10.1111/j.1467-8489.2005.00283.x
Wilson, W.W. 1982. “Hedging Effectiveness of US Wheat Futures Markets.” Department
of Agricultural Economics, North Dakota State University.
Würriehausen, N., R. Ihle, and S. Lakner. 2015. “Price Relationships between
Qualitatively Differentiated Agricultural Products: Organic and Conventional
Wheat in Germany.” Agricultural Economics 46:195-209.
http://doi.org/10.1111/agec.12151
Zacharias, T.P., M.D. Lange, W.J. Gleason, and H.D. Traylor. 1987. “A Producer-Level
Cross-Hedge for Rough Rice Using Wheat Futures.” Journal of Agricultural and
Applied Economics 19(2):75-82. http://doi.org/10.1017/S0081305200025346
52
CHAPTER 2
HOW DO CURRENT AND POTENTIAL CONSUMERS OF ORGANIC WHEAT
PRODUCTS DIFFER FROM NON-CONSUMERS? A MARKET SEGMENTATION
ANALYSIS
2.1 Abstract
We use latent class modeling to identify groups of “very likely,” “likely,” and “unlikely”
consumers of organic wheat products based on factors indicating their preferences for
organic wheat products and attitudes toward organics in general. We perform the analysis
for two types of wheat products—bread and cookies—using data from an online survey
conducted across the western United States in Summer 2017. The results show that
significant differences, which are not product-specific, exist across the segments. “Very
likely” consumers rank price and taste as less important compared to the other two groups,
who tend to believe that organic products are too expensive, regular products taste better,
and organic products are not better. These groups value local and natural labels over
organic label. In addition, significant differences in willingness to pay (WTP) for certified
organic, socio-demographics, lifestyles, and shopping and consumption habits are found
across the segments. Finally, we find that those with some wheat/gluten
intolerance/avoidance tend to be “very likely” consumers of organic wheat products.
2.2 Introduction
The demand for organic wheat in the US is currently stronger than domestic supply. There
is an initiative to stimulate and support organic wheat production in the U.S. western states
53
by addressing grower concerns regarding yields, soil quality, weed management and
productivity (U.S. Department of Agriculture – National Institute of Food and Agriculture,
2014-2020). However, it is also important to understand who current and potential
consumers of organic wheat products are, what product attributes are important to them,
and what determines or limits their interest in organic foods. This knowledge can help
manufacturers and marketers to make production and marketing decisions when allocating
the limited supply of organic wheat and to better understand market growth potential.
A large body of research characterizes consumers of organic foods and assesses the
determinants of their interest in and WTP for organics across a variety of products (e.g.,
Jolly, 1991; Gil, Gracia, and Sanchez, 2000; Lockie et al., 2004; Krystallis and
Chryssohoidis, 2005; Hughner et al., 2007). However, studies disagree in their assessments
of which consumer characteristics and factors play a role in identifying consumers of
organic foods versus non-consumers. For example, some studies find that demographics
do not explain organic shopping behavior well and that attitude and lifestyle factors tend
to explain it better (Li, Zepeda, and Gould, 2007; Gracia and de Magistris, 2013), while
others find significant relationships between demographics and organic food demand
(Govindasamy and Italia, 1999; Loureiro, McCluskey, and Mittelhammer, 2001; Onyango,
Hallman, and Bellows, 2007; Ngobo, 2011).
The overall purpose of this essay is to segment consumers into “very likely,”
“likely,” and “unlikely” consumers of organic bread and cookie products to examine the
differences between the segments and identify factors that determine their interest level
and preferences for organics. To our knowledge, this study is the first one to perform such
54
analysis with the focus on products containing organic wheat.
This study has four objectives: (1) We cluster consumers into segments based on
their attitudes toward organic products and production systems, their past purchases of
organic wheat products, and the importance they place on certified organic labels to
identify groups of “very likely,” “likely,” and “unlikely” consumers of organic wheat
products. We use factor analysis to reduce the number of attitudinal variables used in the
segmentation process, and then we apply latent class modeling to carry out the
segmentation. (2) We examine the differences among the groups in terms of their
preferences for labels and product characteristics, reasons for purchasing or not purchasing
organic wheat products, socio-demographics, lifestyle characteristics, and shopping and
consumption behavior. Finally, we (3) calculate WTP values for certified organic products
for the entire sample and for each segment; and (4) examine the differences in findings
across the two types of wheat products.
2.3 Background and Literature Review
2.3.1 Determinants of Organic Food Choice and Consumption
The literature on organic food consumption (Hughner et al., 2007) indicates that organic
consumers are in general female and have children living with them, while the effects of
age, education, and income are either insignificant or inconsistent across the studies. The
discrepancies may be a result of different methodologies, samples of respondents from
culturally different regions, or the category of product analyzed. For example, using a
sample of respondents in Sweden, Magnusson et al. (2001) found that respondents aged
18–25 demonstrated stronger intentions to purchase organic bread, while gender, education
55
level and presence of children played no role.
However, some studies find that other consumer characteristics such as lifestyle
and attitudinal factors can explain consumer behavior and sometimes better than socio-
demographics (Gil, Gracia, and Sanchez, 2000; Li, Zepeda, and Gould, 2007; Gracia and
de Magistris, 2013). Gil, Gracia, and Sanchez (2000) segmented Spanish consumers by
lifestyle attributes and found that consumers who were concerned about a healthy diet and
the environment were most likely to purchase organic food and pay a premium. On a
similar note, Gracia and de Magistris (2013) found that attitudes towards organic foods
regarding health and environmental benefits affected intention to purchase organic foods
and the final decision. Li, Zepeda, and Gould (2007) found that consumer beliefs that
organic foods are more nutritious was an important determinant of organic food purchases.
Attitudes toward organic products and production systems have been found to play
a role in motivating purchase of organic food products in general, but positive attitudes do
not necessarily translate into purchasing decisions. Some studies have found that many
respondents express positive attitudes toward organic production, but their purchase
frequency is still low (Magnusson et al., 2001; Bellows et al., 2008; Aertsens et al., 2009).
We use attitudes toward organic production systems in conjunction with the consumer
purchase history of organic wheat products to identify consumers who have positive
attitudes toward organic food but do not necessarily purchase organic wheat products, to
explore their characteristics and preferences and identify possible barriers to purchasing
organic wheat products.
56
2.3.2 Consumer Preferences Related to Wheat Products and WTP for Certified Organic
Few studies examine the attributes consumers consider when purchasing or consuming
bread products specifically. Magnusson et al. (2001) found that 97% and 94% of
respondents in Sweden selected taste and freshness, respectively, as important or very
important, while only 17% selected organic production. Using organic bread as the food
item presented to respondents, Lusk and Briggeman (2009) found that safety, nutrition,
taste, and price were among the most important food values for U.S. consumers. Bitzios,
Fraser, and Haddock-Fraser (2011) found that the key attributes that UK consumers
considered when purchasing bread were the type of flour, price, texture, taste, aroma, and
perceived healthiness, while the organic production method did not matter to them. Based
on these studies, it appears that taste is considered universally important.
Several studies examined also consumer WTP for organic bread and cookies,
although they were conducted either outside of the United States or they obtained WTP
estimates for a whole sample rather than by consumer segment. Krystallis, Fotopoulos, and
Zotos (2006) found that a sample of Greek consumers were willing to pay 75.5% more for
organic bread than for conventional. Ureña, Bernabéu, and Olmeda (2008) found that a
sample of Spanish consumers were willing to pay an average of 9.2% more for organic
bread, biscuits, and sweets, with significant differences between men and women, and
segments of regular, occasional, potential, and non-consumers, where regular male
consumers were willing to pay the highest premium (19.2%) and probable female
consumers the lowest (2.8%). Similarly, Zheng (2014) found that Canadian consumers
were willing to pay 9.5% for certified organic whole grain bread, while Hasselbach and
57
Roosen (2015) found that consumers in Germany were not willing to pay for organic bread
unless it was produced locally. Bitzios, Fraser, and Haddock-Fraser (2011) found that UK
consumers do not care much about the organic label on bread, but a combination of
functional ingredient and organic label yielded positive utility for consumers in one of the
three segments they identified. Finally, Lee et al. (2013) found that U.S. consumers were
willing to pay 50% more for organic cookies.
2.3.3 Motivations and Barriers to Purchasing Organic Products
Another topic that has been researched in the literature is motivations and barriers to
purchasing and consuming organic products. Health and environmental concerns are
among the main reasons that people purchase organic foods (Lea and Worsley, 2005; Padel
and Foster, 2005; Zepeda, Chang, and Leviten-Reid, 2006). Other reasons include concerns
about animal welfare, support for the local economy (Padel and Foster), and the belief that
organic foods taste better (Lea and Worsley). Price and familiarity/availability of organic
products are key barriers to purchasing organic foods (Lea and Worsley; Zepeda, Chang,
and Leviten-Reid; Bellows et al., 2008), in addition to quality issues and lack of trust
(Bellows et al.). For bread, price is a major barrier to purchasing organic (Magnusson et
al., 2001; Kihlberg and Risvik, 2007). Kihlberg and Risvik found that a majority of
consumers in Sweden think organic bread tastes better than conventional bread after a
sensory assessment, but they would not buy it if the price were too high relative to
conventional. In a sensory evaluation of whole grain bread, Teuber, Dolgopolova, and
Nordström (2016) found that once taste was accounted for, a sample of German
respondents were not willing to pay a premium for the organic label, suggesting that taste
58
is an important attribute for bread and consumers are not willing to compromise on it.
2.3.4 Effects of Product Type on Consumer Preferences for Organic Products
Past studies have found that consumers have different WTP for organic virtue and vice
foods (Van Doorn and Verhoef, 2011), as well as divergent taste and nutritional
expectations (Ellison et al., 2016). Thus, we perform the analysis using bread and cookies.
Following Hui, Bradlow, and Fader (2009), bread can be classified as a virtue food item
and cookies as a vice food item, in relation to long-term health outcomes and immediate
pleasure from consumption. Analyzing two product categories will also allow us to
examine in one study whether organic food consumers share common characteristics and
preferences across virtue and vice products.
2.4 Data and Survey Methodology
The data used in the analysis come from an online survey conducted in 16 U.S. western
states in June 2017. The same data are used in Essay 3; thus, the following survey
description is relevant to both essays. The survey was designed and administered using the
Qualtrics platform and a pretest of the survey was performed using a group of graduate
students at Utah State University. A link to the final survey was sent to Qualtrics panelists
in an email invitation to participate in the study. In return for the completion of the survey,
they received points that they could trade later for rewards including cash and gift cards.
In total, 1,009 valid responses were received. The respondents were selected so that
the sample was representative of the population in the western United States based on three
criteria: age group, gender, and state of residence. Only respondents at least 18 years of
59
age were allowed to participate. Table 2-1 reports the share of respondents by state
compared to the share of the population by state. Table 2-2 reports selected
sociodemographic characteristics for the sample and population. The sample is fairly
representative of the population in the western United States in terms of origin, gender, age
group, marital status, household size, presence of children, and unemployment rate. The
share of those with some college education or higher is greater in the sample than in the
general population, while the share of those with household income above $100,000 and
the labor force participation rate is lower. Organic consumers tend to be wealthier (e.g.
Dettmann and Dimitri, 2009; Ngobo, 2011; Dimitri and Dettmann, 2012), likely to have
busier schedules, and consequently have less time or interest to participate in online
surveys. Thus, it appears that the organic consumers are underrepresented in the sample
which needs to be considered when interpreting the results.
Table 2-1. Respondents and Population in the U.S. Western States by State
State
Sample count
Population count
Sample share
Population share
Arizona
92
5,206,537
9.1%
8.1%
California
512
30,028,400
50.7%
46.7%
Colorado
71
4,199,802
7.0%
6.5%
Idaho
21
1,222,749
2.1%
1.9%
Kansas
42
2,191,407
4.2%
3.4%
Montana
9
806,722
0.9%
1.3%
Nebraska
19
1,425,560
1.9%
2.2%
Nevada
21
2,222,290
2.1%
3.5%
New Mexico
26
1,585,693
2.6%
2.5%
North Dakota
11
585,614
1.1%
0.9%
Oklahoma
53
2,950,241
5.3%
4.6%
Oregon
52
3,167,825
5.2%
4.9%
South Dakota
3
648,789
0.3%
1.0%
Utah
32
2,083,586
3.2%
3.2%
Washington
40
5,558,381
4.0%
8.6%
Wyoming
5
446,607
0.5%
0.7%
All 16 states
1,009
64,330,203
100.0%
100.0%
Note: Population data are sourced from the U.S. Census Bureau, 2015 American Community Survey
1-Year Estimates, and exclude individuals less than 18 years of age.
60
Table 2-2. Socio-Demographic Characteristics of Respondents Compared to the
Population in U.S. Western States
Characteristic
Sample mean
Population in U.S.
western states
Female (%)
51.6
50.6
Age group (%)
18–24 yrs.
11.8
13.0
25–44 yrs.
35.2
35.8
45–64 yrs.
35.5
32.8
>64 yrs.
17.5
18.4
Annual household income (%)
<$10,000
5.9
6.1
$10,000–$49,999
44.7
36.6
$50,000–$99,999
32.9
30.3
$100,000–$149,999
10.7
14.4
>$149,999
5.7
12.6
Married (%)
50.1
52.2a
Household size (persons)
2.7
2.8
Presence of children <18 years (%)
33.6
32.9
Education attainment (%)
Less than high school
0.5
14.0
High school
16.7
24.2
Some college or associate’s degree
43.1
33.3
4-year college
26.6
18.5b
Graduate degree or higher
13.2
10.0
Labor force participation rate (%)
56.2
63.3c
Unemployment rate (%)
6.8
6.5c
Ethnic background (%)d
African American
5.4
4.8
American Indian
1.0
1.7
Asian
6.3
8.8
Hawaiian
0.2
0.3
Other, one
0.5
7.6
Other, two or more
3.4
3.1
White (non-Hispanic)
72.2
58.6
White (Hispanic)
9.2
15.1
Note: Population data are sourced from the U.S. Census Bureau, 2015 American Community Survey
1-Year Estimates, and exclude individuals less 18 years of age, unless indicated otherwise.
a Sample includes persons 18+, population data for persons 20+.
b Includes graduate degree for the population between 18–24 years.
c Sample includes persons 18+, population data for persons 16+.
d Data source: U.S. Census Bureau, 2011–2015 American Community Survey 5-Year Estimates.
Respondents were asked a variety of survey questions about their shopping and
61
consumption habits, preferences related to labels and characteristics of bread and cookies,
reasons for purchasing or not purchasing organic bread and cookies in the past, attitudes
toward organic food and organic production in general, knowledge of organic production
system and products, and lifestyle and sociodemographic questions. To determine
consumer WTP and further examine preferences for selected labels (including organic), we
employed a hypothetical choice experiment in which we repeatedly asked respondents to
choose which alternative they would purchase in a real shopping scenario, considering their
budget.
2.4.1 Choice Experiment Design
In each choice question, a respondent could choose from three alternatives: “conventional,”
“organic,” and “none.” Alternatives labeled “conventional” and “organic” varied in four
attributes—“price” and the presence or absence of three labels: “non–GMO,” “gluten-
free,” and “low-carb” for bread or “sugar-free” for cookies.
Price was drawn randomly from a range based on the national average price for
each combination of product (bread or cookies) and production method (conventional or
organic). The basis for the price range determination was the April 2017 U.S. average price
point of $2.00/lb for high quality bread and $3.40/lb for chocolate chip cookies (Bureau of
Labor Statistics, 2017). Following Carlson and Jaenicke (2016), who found an organic
premium for bread of around 30% over conventional bread, we applied a 30% premium to
conventional prices to obtain the price points for organic bread and organic cookies. Using
these prices, we then established the price ranges based on an approximately 50% discount
and a 100% premium above the obtained price points. Table 2-3 lists the attributes with
62
their levels or ranges.
Table 2-3. Attributes and Their Levels or Ranges
Attributes
Levels and ranges
Non–GMO label
Present/Absent
Gluten-free label
Present/Absent
Low-carb or sugar-free label
Present/Absent
Price
Conventional bread
$1.00–$5.00
Organic bread
$2.00–$7.00
Conventional cookies
$2.00–$7.00
Organic cookies
$3.00–$8.00
Not considering the price attribute, each product alternative (conventional and
organic) in each choice set could vary in three attributes with two levels per each attribute,
resulting in 23=8 possible specifications of each product alternative. Since two product
alternatives were available in each choice set, there were 8∗8=64 total possible
combinations of product alternatives and thus 64 possible choice sets, which we split into
eight blocks with eight choice sets per block, while preserving the balance and
orthogonality, which is necessary for optimal design with maximum statistical efficiency.
Then, we added the price attribute to each alternative in each choice set by drawing the
price randomly from the determined price ranges. Each respondent received one block of
the choice sets for bread and one block for cookies, assigned randomly and independently
of each other. In total, each respondent evaluated 16 choice sets. An example of the
complete questionnaire is in Appendix A; the questionnaires differed across respondents
only in terms of the assigned blocks of the choice sets.
63
2.5 Model Specification and Methodology
2.5.1 Latent Class Modeling
We employed a latent class model (LCM) within a discrete choice modeling framework.
LCM combines the multinomial logit (MNL) model (McFadden, 1974; Train, 2009), which
is a basic approach to modeling consumer preferences, and latent class analysis, which is
an alternative to clustering techniques (Magidson and Vermunt, 2002). As such, LCM
serves two purposes (Swait, 1994; Boxall and Adamowicz, 2002; Greene and Hensher,
2003): First, it allows to cluster consumers into latent (unobserved) classes (segments)
based on selected segmentation characteristics. Second, during the segmentation process,
a MNL model is estimated for each segment simultaneously. This procedure results in
segments with homogeneous consumer preferences and characteristics within a segment
and heterogeneous preferences and characteristics across segments, allowing to capture
some of the heterogeneity in preferences which is not accounted for when a MNL model
is estimated for the whole sample (see Ruto and Garrod, 2009; Hidrue et al., 2011).
Respondent n’s utility from bread alternative i among j=1,…,J alternatives, 𝑈𝑛𝑖, is
(2-1) 𝑈𝑛𝑖 =𝑉𝑛𝑖 +𝜀𝑛𝑖 =𝛽𝑋𝑛𝑖 +𝜀𝑛𝑖,
where 𝑉𝑛𝑖 is part of the utility that is observed by the researcher and 𝜀𝑛𝑖 is unobserved and
random. 𝑉𝑛𝑖 is linear in parameters and comprises a vector of attributes of bread alternative
i (labels and price) faced by respondent n, 𝑋𝑛𝑖, and a vector of taste parameters, 𝛽,
associated with the attributes and assumed not to vary across respondents within the MNL
framework.
64
A utility-maximizing respondent will choose bread alternative i among j=1,…,J
bread alternatives if the utility from this alternative, 𝑈𝑛𝑖, is greater than utility from all
other available alternatives, 𝑈𝑛𝑗, for all 𝑗≠𝑖. Following McFadden (1974) and Train
(2009), the probability that the respondent n will choose bread alternative i is
(2-2) 𝑃𝑛𝑖 =𝑃(𝑉𝑛𝑖 +𝜀𝑛𝑖 >𝑉𝑛𝑗 +𝜀𝑛𝑗 ∀𝑗≠𝑖)=𝑃(𝜀𝑛𝑗 <𝜀𝑛𝑖 +𝑉𝑛𝑖 −𝑉𝑛𝑗 ∀𝑗≠𝑖).
Under the assumption that the unobserved 𝜀′s in equation (2-2) are i.i.d. type I extreme
value, equation (2-2) can be rewritten as the product of cumulative density functions,
𝐹(𝜀𝑛𝑗), for each 𝜀𝑛𝑗 ∀𝑗≠𝑖, where 𝐹(𝜀𝑛𝑗)=exp(−𝑒𝑥𝑝(−𝜀𝑛𝑗))=exp(−𝑒𝑥𝑝(−(𝜀𝑛𝑖 +
𝑉𝑛𝑖 −𝑉𝑛𝑗))). The resulting expression is further integrated over all possible values for 𝜀𝑛𝑖,
and it can be manipulated to calculate the logit choice probability of choosing alternative i
by respondent n, 𝑃𝑛𝑖, as (Train, 2009)
(2-3) 𝑃𝑛𝑖 =exp(𝑉𝑛𝑖)
∑exp(𝑉𝑛𝑗)
𝐽
𝑗=1 =exp(𝛽𝑋𝑛𝑖)
∑exp(𝛽𝑋𝑛𝑗)
𝐽
𝑗=1 .
Next, assuming that respondent i belongs to a latent class q (q=1,…,Q, where Q is
the number of latent classes), the logit choice probability is calculated as (Boxall and
Adamowicz, 2002)
(2-4) 𝑃𝑛𝑖|𝑞 =exp(𝑉𝑛𝑖|𝑞)
∑exp(𝑉𝑛𝑗|𝑞)
𝐽
𝑗=1 =exp(𝛽𝑞𝑋𝑛𝑖)
∑exp(𝛽𝑞𝑋𝑛𝑗)
𝐽
𝑗=1 ,
where 𝛽𝑞 becomes a class-specific vector of taste parameters, allowing the taste parameters
to vary across classes. Further, assuming a vector of observed respondent-specific
characteristics, 𝑍𝑛, that determines class membership and a vector of the parameters for
the respondent-specific characteristics in class q, 𝜃𝑞, the probability of assigning
65
respondent n to latent class q can be calculated as (Boxall and Adamowicz, 2002; Greene
and Hensher, 2003)
(2-5) 𝑃𝑛𝑞 =exp(𝜃𝑞𝑍𝑛)
∑exp(𝜃𝑞𝑍𝑛)
𝑄
𝑞=1 .
In this essay, the vector 𝑍𝑛 contains attitudes toward organic products and production
systems, past purchases of organic breads/cookies, and ranked importance of the organic
label in bread/cookies. The respondent-specific characteristics that are not observed but
affect the assignment of respondent n to a latent class q, 𝜀𝑛𝑞, are assumed to be i.i.d. type I
extreme value, as in the case of a MNL model (Boxall and Adamowicz, 2002; Greene and
Hensher, 2003). Finally, the probability that any respondent n chooses a bread alternative
i is calculated as (Boxall and Adamowicz, 2002)
(2-6) 𝑃𝑛𝑖 =∑𝑃𝑛𝑞
𝑄
𝑞=1 𝑃𝑛𝑖|𝑞 =∑[ exp(𝜃𝑞𝑍𝑛)
∑exp(𝜃𝑞𝑍𝑛)
𝑄
𝑞=1 ]
𝑄
𝑞=1 [exp(𝛽𝑞𝑋𝑛𝑖)
∑exp(𝛽𝑞𝑋𝑛𝑗)
𝐽
𝑗=1 ].
The number of latent classes Q must be specified but is usually not known in
advance. Therefore, we estimate models with two to eight classes and evaluate each model
using several fit criteria, including the Akaike information criterion (AIC), Bayesian
information criterion (BIC), Hannan-Quinn criterion (HQC), log-likelihood value, and
pseudo-R2. Lower AIC, BIC, and HQC and higher log-likelihood and pseudo-R2 values
indicate better overall fit. However, BIC has been found to perform relatively better
compared to the other criteria considered (Nylund, Asparouhov, and Muthén, 2007; Jung
and Wickrama, 2008).
66
2.5.2 Willingness to Pay
After the estimation of LCM, we calculate mean WTP values for the organic label within
each class using two methods. Applying the first method, mean WTP for the organic label
within a latent class q (q = 1,…,Q) is calculated as the negative ratio of the coefficient
estimate for the organic label in class q, 𝛽𝑜𝑟𝑔𝑎𝑛𝑖𝑐,𝑞, to the coefficient estimate for price in
class q, 𝛽𝑝𝑟𝑖𝑐𝑒,𝑞,
(2-7) 𝑊𝑇𝑃𝑜𝑟𝑔𝑎𝑛𝑖𝑐,𝑞 =−𝛽𝑜𝑟𝑔𝑎𝑛𝑖𝑐,𝑞
𝛽𝑝𝑟𝑖𝑐𝑒,𝑞 .
For the WTP values calculated using equation (2-7), we use Delta method to construct
confidence intervals around the mean WTP estimates and determine the significance of the
mean WTP within each class.
Applying the second method, we use individual WTP values obtained for each
respondent in the sample and calculate mean of the respondents’ WTP values within each
class. First, each respondent is assigned to one of the Q classes, so that ∑𝑀𝑞
𝑄
𝑞=1 =1009,
where 𝑀𝑞 is the number of respondents assigned to class q. For each respondent 𝑚𝑞 (𝑚𝑞=
1,…,𝑀𝑞) in class q, we obtain the coefficient estimates for the organic label, 𝛽𝑜𝑟𝑔𝑎𝑛𝑖𝑐,𝑚𝑞,
and price, 𝛽𝑝𝑟𝑖𝑐𝑒,𝑚𝑞.
13
The individual WTP is then calculated as
(2-8) 𝑊𝑇𝑃𝑜𝑟𝑔𝑎𝑛𝑖𝑐,𝑚𝑞=−𝛽𝑜𝑟𝑔𝑎𝑛𝑖𝑐,𝑚𝑞
𝛽𝑝𝑟𝑖𝑐𝑒,𝑚𝑞.
Finally, the WTP within each class q is calculated as the average of the individual WTP
values for all respondents 𝑚𝑞 assigned to class q
13
More details on how these individual level estimates are calculated can be found in Greene (2012).
67
(2-9) 𝑊𝑇𝑃𝑜𝑟𝑔𝑎𝑛𝑖𝑐,𝑞 =∑𝑊𝑇𝑃𝑜𝑟𝑔𝑎𝑛𝑖𝑐,𝑚𝑞
𝑀𝑞
𝑚𝑞=1 𝑀𝑞.
We do not obtain standard errors for the mean WTP values which are calculated using this
method, but the individual WTP values are useful for evaluating the statistical significance
of the differences in WTP values across the groups of respondents.
2.5.3 Factor Analysis and Factor Scores
In total, respondents answered nine attitudinal questions, but some of the attitudes may be
correlated. Before the estimation of LCM, we employed factor analysis to extract
underlying latent factors describing the attitudes, with the aim to reduce the number of
variables in vector 𝑍𝑛 in equation (2-6) while explaining sufficient amount of variance
contained in the originally observed variables. First, we subjected the first half of the
sample (n = 504) to exploratory factor analysis (EFA). The minimum residual method was
used to extract potential factors
14
and the number of factors was determined using parallel
analysis, scree test, selected fit indices,
15
and following common practice in the literature.
16
To enable possible nonzero correlations between factors, we applied the oblique rotation
to the solution with desired number of factors. Next, to verify the EFA solution, we
14
Revelle (2017) suggests the minimum residual method as a good alternative to maximum likelihood
method for factor extraction. Fabrigar et al. (1999) and Costello and Osborne (2005) cite maximum likelihood
method as the best choice, but it assumes the data are multivariate normal. In this study, the assumption of
multivariate normality does not hold.
15
We used the Tucker–Lewis index (TLI), root mean squared error of approximation (RMSEA), and root
mean square of the residuals (RMSR).
16
1) We retain variables with a minimum loading of 0.32 (Costello and Osborne, 2005; Tabachnick and
Fidell, 2007), but prefer 0.50 for a strong factor (Costello and Osborne, 2005). 2) At least three variables
need to load on a factor, and two variables only may load on a factor only if their correlation is above 0.70
(Yong and Pearce, 2013). 3) Variables with multiple correlations below 0.30 may need to be removed from
analysis (Yong and Pearce, 2013).
68
subjected the second half of the sample (n = 505) to confirmatory factor analysis (CFA)
(Fabrigar et al., 1999) and evaluated selected fit indices.
17
In the next step, we proceeded to obtain each respondent’s scores associated with
the factors—factor scores—which were to be used in the LCM instead of the original
attitudinal variables. There are several approaches to obtaining factor scores but no clear
direction in the current literature regarding a preferred method (DiStefano, Zhu, and
Mindrila, 2009; Yong and Pearce, 2013). First, factor scores can be obtained after either
EFA or CFA. We opted to calculate factor scores after CFA (using the whole sample)
because only the variables that load significantly on the factor are used in the calculation
of the factor score, which eliminates the noise from weakly loading variables. Second,
several methods of calculating factor scores are available (DiStefano, Zhu, and Mindrila
discuss advantages and disadvantages). A major issue with factor scores is their
indeterminacy (i.e., the possibility to obtain infinite number of scores that would be
consistent with the same correlation pattern between the variables and underlying factors;
Grice, 2001; DiStefano, Zhu, and Mindrila). Considering this issue, we chose to apply the
Bartlett method,
18
which produces unbiased factor scores with high validity (Grice;
DiStefano, Zhu, and Mindrila).
17
We used the Tucker–Lewis index (TLI), root mean squared error of approximation (RMSEA), standardized
root mean squared residual (SRMR), and comparative fit index (CFI).
18
We found that the correlation between factor scores produced using Bartlett method after CFA (-0.44)
came close to the correlation found between latent factors after EFA (-0.49). The approximation is referred
to as correlational accuracy and is one of the measures that can be used to assess the degree of factor
indeterminacy (Grice, 2001; DiStefano, Zhu, and Mindrila, 2009).
69
2.5.4 Differences across Groups
After the classes were identified using LCM, we organized them into groups of “very
likely,” “likely,” and “unlikely” consumers based on the similarities and differences in
terms of the preferences for the organic label and past purchases of organic bread and
cookies. From now on, “class” or “segment” is used to refer to one of the latent classes
obtained from LCM, and “group” to refer to the grouping of classes. After identifying the
groups of respondents, we examined the statistical significance of the differences in
selected characteristics between the respondents in the “very likely” group on one side and
the respondents in the “likely” and “unlikely” groups on the other to determine which
factors play a role in differentiating consumers from non-consumers. Where applicable, we
also compared groups across product categories and/or contrasted the results within the
bread and cookies product category to determine whether differences are product specific.
We examined differences in these characteristics: WTP for organic label in bread
and cookies—calculated as mean of the individual WTP values of respondents in each
group, based on equation (2-9); preferences for labels and characteristics of bread and
cookies; interest in organic versions of bread, cookies, and other wheat product categories;
possible motivations and barriers to purchasing organic bread and cookies; lifestyles; and
consumption behavior. We used Welch’s t-test and Wilcoxon (Mann-Whitney) test to
evaluate the significance of the differences. Both tests allowed us to compare the means or
proportions for two groups under the assumption of unequal variance, but while the Welch
test assumes that the data are normally distributed, the Wilcoxon test does not.
70
2.6 Results
2.6.1 Factor Analysis and Factor Scores
We performed Bartlett’s test of sphericity (𝜒2(36) = 2,095, p-value < 0.001) and the
Kaiser–Meyer–Olkin (KMO) test of factor adequacy (overall measure of sampling
adequacy MSA = 0.80, MSA per variable > 0.67) and concluded that the factor analysis
was appropriate. Following recommendations in the literature during EFA and using only
the first half of the sample, we found that two attitudinal statements needed to be removed
from analysis. Consequently, we obtained a satisfactory 2-factor solution. To confirm that
this solution was appropriate, we performed CFA using the second half of the sample.
Obtained fit statistics (Tucker–Lewis index TLI = 0.922, comparative fit index CFI =
0.951, root mean squared error of approximation RMSEA = 0.077, and standardized root
mean squared residual SRMR = 0.048) indicated a moderately good fit .
19
Table 2-4 summarizes the 2-factor solution after performing EFA on the whole
sample and seven attitudinal statements. The first factor is strongly and positively
correlated with beliefs that organic products are healthier and fresher than conventional
ones, do not contain harmful substances, and the organic production is better for the
environment. In summary, this factor represents a positive view of organic products and
production; a belief that organic products and production are differentiated from
conventional products and production.
19
We followed Hooper, Coughlan, and Mullen (2008), who summarized fit indices and their recommended
cut-off values. Reported TLI and RMSEA indices were found to miss the recommendations; however, these
indices were obtained using only one half of the sample. After performing EFA using the whole sample, all
recommendations were met (see table 2-4 notes).
71
Table 2-4. Rotated 2-Factor Solution Applying EFA on the Whole Sample (n=1009)
Statement
Factor
1
Factor
2
Uniqueness
Organic products are healthier than conventional
0.75
-0.12
0.34
Organic products are fresher than conventional
0.72
0.15
0.57
Organic production is better for the environment
0.64
-0.11
0.50
Organic products do not contain harmful substances
0.63
0.04
0.63
Organic products are not safer than conventional
-0.09
0.67
0.49
Buying organic food does not benefit local farmers
0.09
0.64
0.64
Organic products do not taste better than
conventional
-0.05
0.62
0.58
Variance explained
0.27
0.18
-
Note: TLI = 0.962 (>0.95), RMSEA = 0.058 (<0.07), RMSR = 0.02 (<0.05). Cut-off values for fit indices,
in parentheses, are based on Hooper, Coughlan, and Mullen (2008).
The second factor is strongly and positively correlated with beliefs that organic
products are not safer and do not taste better than conventional products and that buying
organic food does not benefit local farmers. This factor represents a neutral view toward
organic products and production, since it indicates a belief that organic products and
production do not differentiate from conventional products and production. The two factors
together account for 46% of total variance in the data.
The correlation between the factors is -0.49, indicating a relatively strong negative
relationship. This exceeds the ±0.32 threshold, which is used to warrant an oblique as
opposed to orthogonal rotation (Tabachnick and Fidell, 2007). This finding of nonzero
correlation is also supported by the theoretical reasoning that positive and neutral beliefs
toward organic products and production may be negatively correlated. In addition,
examining the correlation pattern between variables (see table 2-5) shows that correlations
between statements loading on factors 1 and 2 are negative and the correlations involving
72
the two statements dropped from the factor analysis are relatively weak. Finally, we
performed CFA to obtain the factor scores.
Table 2-5. Correlations between Statements
Healthier
Fresher
Better for
environment
No harmful
substances
Not
safer
No benefits for
local farm
Not better
taste
Too expensive
Selection not
good
Healthiera
1.00
0.51
0.57
0.49
-0.39
-0.23
-0.36
-0.13
0.02
Freshera
0.51
1.00
0.45
0.38
-0.17
-0.06
-0.20
-0.05
0.08
Better for
environmenta
0.57
0.45
1.00
0.42
-0.35
-0.26
-0.27
-0.09
0.03
No harmful
substancesa
0.49
0.38
0.42
1.00
-0.28
-0.10
-0.16
-0.08
0.05
Not saferb
-0.39
-0.17
-0.35
-0.28
1.00
0.41
0.47
0.22
0.12
No benefits for
local farmb
-0.23
-0.06
-0.26
-0.10
0.41
1.00
0.38
0.15
0.19
Not better tasteb
-0.36
-0.20
-0.27
-0.16
0.47
0.38
1.00
0.31
0.19
Too expensive
-0.13
-0.05
-0.09
-0.08
0.22
0.15
0.31
1.00
0.26
Selection not good
0.02
0.08
0.03
0.05
0.12
0.19
0.19
0.26
1.00
Note: a,b The same superscript indicates that the statements load on the same factor. Statements without
superscripts were dropped from factor analysis and used in the subsequent analysis as is.
2.6.2 Latent Class Modeling
We estimated the LCM using NLOGIT 5.0 (Greene, 2012). Table 2-6 reports values for
the criteria used to evaluate the model fit. For both bread and cookies, starting with a 2-
class model all fit criteria improved with every additional class up to the 7-class model.
Statistically, 7-class models appear to be best. However, considering 5% minimum size of
class membership (Nasserinejad et al., 2017) and our focus on segments of current and
potential consumers of organic products and interpretability of results (Jung and
Wickrama, 2008), we chose 6-class models for bread and cookies.
73
Table 2-6. Criteria Used in Determining Number of Latent Classes (LC) in the LCM
2 LC
3 LC
4 LC
5 LC
6 LC
7 LC
8 LC
Bread
AIC
13,061
11,623
11,135
10,919
10,798
10,651
10,631
BIC
13,194
11,847
11,450
11,324
11,295
11,239
11,310
HQC
13,106
11,700
11,243
11,057
10,968
10,852
10,863
Log-Likelihood
-6,511
-5,780
-5,522
-5,401
-5,328
-5,242
-5,219
Pseudo R2
0.266
0.348
0.377
0.391
0.399
0.409
0.412
Parametersa
19
32
45
58
71
84
97
Cookies
AIC
12,046
11,249
10,715
10,586
10,435
10,286
10,256
BIC
12,179
11,473
11,029
10,992
10,931
10,874
10,935
HQC
12,092
11,325
10,822
10,725
10,604
10,487
10,488
Log-Likelihood
-6,004
-5,592
-5,312
-5,235
-5,146
-5,059
-5,031
Pseudo R2
0.323
0.369
0.401
0.410
0.420
0.430
0.433
Parametersa
19
32
45
58
71
84
97
Note: Sample size is 8,072 choices from 1,009 individuals.
a Number of parameters in the estimated model.
2.6.2.1 Latent Class Modeling for Bread
Table 2-7 reports results of the LCM for bread. The results are organized in two sections:
1) MNL model estimates, and 2) latent class parameter estimates, obtained for each latent
class. In the section with the MNL model estimates, the focus is on the stated preferences
for the organic attribute, but all attributes presented to respondents in the choice experiment
were included in the estimation process to account for the effect of these attributes on
respondents’ choices. The MNL model estimates for each segment are interpreted relative
to a base product—a conventionally produced bread (“organic” = 0) without any labels
(“non–GMO” = 0, “gluten-free” = 0, “low-carb” = 0). “No choice” measures the utility of
choosing no bread relative to the base product. Table 2-7 also reports the mean WTP values
for organic label for each segment, calculated based on equations (2-7) and (2-9).
74
Table 2-7. Latent Class Model Parameter Estimates, Bread
Unlikely
Likely
Very likely
Variable
Class 1
Class 2
Class 3
Class 4
Class 5
Class 6
Multinomial Logit Model Estimates
Price
-1.02***
(0.08)
-1.50***
(0.12)
-2.88***
(0.31)
-0.63***
(0.09)
-0.32***
(0.04)
-0.48***
(0.10)
Organic
-0.79***
(0.19)
-0.93***
(0.13)
1.96***
(0.40)
1.72***
(0.27)
0.49***
(0.09)
4.01***
(0.42)
Non-GMO
0.00
(0.12)
0.25*
(0.14)
0.77**
(0.35)
-0.01
(0.21)
0.35***
(0.08)
0.62***
(0.23)
Gluten-free
-0.80***
(0.14)
0.11
(0.13)
0.28
(0.28)
-0.46**
(0.23)
0.33***
(0.09)
0.72***
(0.27)
Low-carb
-0.14
(0.13)
0.11
(0.17)
0.34
(0.30)
-0.15
(0.20)
0.38***
(0.08)
0.13
(0.20)
No choice
-3.38***
(0.30)
-9.70***
(0.62)
-10.39***
(1.18)
1.50***
(0.40)
-2.89***
(0.27)
0.20
(0.46)
WTP for organica
-$0.77
***
-$0.62
***
$0.68
***
$2.74
***
$1.52
***
$8.33
***
WTP for organicb
-$0.54
-$0.50
$0.57
$2.56
$1.50
$7.99
Latent Class Parameter Estimates
Intercept
-1.20***
(0.42)
-1.02***
(0.37)
-1.54***
(0.55)
0
-0.98**
(0.39)
-0.93*
(0.49)
Factor 1c
(organic better)
0.14
(0.16)
0.13
(0.14)
0.31
(0.21)
0
0.55***
(0.16)
1.11***
(0.23)
Factor 2c
(organic not
better)
0.60***
(0.17)
0.33**
(0.15)
0.07
(0.22)
0
0.52***
(0.15)
0.08
(0.19)
Organic food is
too expensivec
0.17
(0.14)
0.15
(0.13)
0.60***
(0.21)
0
-0.28**
(0.12)
-0.50***
(0.15)
Selection of
organic is not
goodc
-0.01
(0.12)
0.17
(0.11)
0.05
(0.15)
0
0.10
(0.12)
0.12
(0.14)
Rank of organic
label in breadd
0.26***
(0.08)
0.28***
(0.07)
0.05
(0.10)
0
0.12
(0.07)
-0.19*
(0.10)
Past purchase of
organic bread
-0.75**
(0.36)
-0.39
(0.29)
-0.08
(0.38)
0
1.63***
(0.29)
1.33***
(0.39)
Class probability
18.2%
24.2%
9.2%
17.6%
20.8%
9.9%
% purchased
organic bread
13.3%
18.6%
31.8%
35.8%
76.1%
83.7%
Note: Standard errors are in parentheses. *, **, *** denote significance at the 10%, 5%, and 1% level,
respectively.
a Mean WTP calculated according to equation (2-7); significance determined using Delta method.
b Mean WTP calculated according to equation (2-9); significance not available.
c The variables represent attitudes toward organic food and production systems either as a composite
measure (“Factor 1” and “Factor 2”) or as an individual attitude if they did not load on any factor during
factor analysis.
d 1 = highest, 7 = lowest.
75
The section with latent class parameter estimates in table 2-7 reports differences
across segments with respect to the covariates used in the segmentation process.
Significance and magnitude of each coefficient within a segment indicate whether and how
the covariate affects the likelihood of a respondent being assigned to the segment, relative
to the reference segment. Here, segment 4 is the reference segment, with its coefficients
normalized to 0 to allow identification. The reference segment was selected arbitrarily by
the software during the estimation process, but it did not affect choice and class
probabilities. Ultimately, the interest is to identify differences in preferences for organic
label on bread and how they relate to attitudes and past purchases of organic bread. Table
2-7 also reports the probability of assigning a respondent to each class and the percentage
of respondents in each class who actually purchased organic bread in the past month.
In the next step, the six classes or segments were split into three groups, labeled
“unlikely,” “likely,” and “very likely.” The classes were grouped based on their similarities
with respect to the combination of three criteria: “past purchase of organic bread”, “%
purchased organic bread in past month”, and mean WTP values, given the interest to
differentiate between respondents based on their likelihood to purchase organic bread.
Looking at table 2-7, consumers in segment 1 (18.2%) are very unlikely to be
interested in organic bread. Among all segments, they tend to hold the strongest beliefs that
organic products in general are not better than conventional products and they are least
likely to have purchased organic bread in the past month. They also rank importance of the
organic label in bread lower than the “very likely” and “likely” segments. On average, they
chose the organic alternative only in one of eight choice scenarios. Looking at the results
76
of the MNL models, they do not show interest in any other labels offered to them in the
choice scenarios and appear to be satisfied with basic bread products.
Consumers in segment 2 (24.2%) are also unlikely to be interested in organic bread,
choosing the organic alternative only once in eight choice scenarios. Compared to segment
1, they tend to rank the importance of an organic label on bread similarly low, but they are
less convinced that organic products are not better than conventional products and they
show an interest in the non–GMO attribute, which is one of the components of organic.
Similarly, as consumers in segment 1, they chose the organic alternative only once out of
eight choice scenarios.
Among all segments, consumers in segment 3 (9.2%) are most likely to believe that
organic products are too expensive. They derive positive utility from the organic attribute,
but their perception that organic products are too expensive might act as a barrier to actual
purchase. On average, in comparison to the “very likely” segments, they tend to believe
less that organic products are better than conventional products. In our survey, they chose
organic bread in three of the eight choice scenarios. They are also interested in the non–
GMO label.
Segment 4 (17.6%) also constitutes a segment of “likely” organic bread consumers.
They chose organic bread once in eight choice scenarios, not because they prefer
conventional bread but rather due to their strong preference for no bread at all. Consumers
in this segment are the only ones who prefer no bread to conventional; if they choose bread
at all, they prefer organic.
Consumers in segment 5 (20.8%) are highly interested in organic bread. In
77
comparison to the two “likely” segments, they tend to hold more positive attitudes toward
organics and disagree more that organic products are too expensive, but not as strongly as
consumers in segment 6. In comparison to segment 6, they also tend to believe more that
organic products are not better than conventional products. On average, they selected
organic bread in four of eight choice scenarios. Besides organic, they gain positive utility
from all labels offered to them in the choice scenarios.
Consumers in segment 6 (9.9%) are very enthusiastic about organic bread; on
average, they selected organic bread in seven of eight choice scenarios. Among all
segments, they tend to hold the strongest beliefs that organic products are better than
conventional products, give the highest importance to the organic attribute, and disagree
the most with the statement indicating that organic products are too expensive. They are
also interested in non–GMO and gluten-free labels.
Among “very likely” and “likely” segments of organic bread consumers, segment
6 stands out as the one with the highest utility for organics. In terms of estimated WTP for
an organic label on a loaf of bread (approx. 1 lb.), based on equation (2-7), segment 6 (WTP
= $8.33) is followed by segments 4 (WTP = $2.74), 5 (WTP = $1.52), and 3 (WTP = $0.68).
Segment 4 (“likely”) has higher WTP for the organic label than segment 5 (“very likely”),
but since it contains a smaller share of respondents who purchased organic bread in the
past month and they have positive utility from no bread at all, they are classified as “likely”
rather than “very likely” consumers. Both segments in the “unlikely” group have negative
WTP for the organic label.
78
2.6.2.2 Latent Class Modeling for Cookies
Table 2-8 reports results from the estimation of LCM for cookies—the MNL model
estimates and latent class parameter estimates. Following the convention established for
bread, we label segments 1, 2 and 3 “unlikely,” segments 4 and 5 “likely,” and segment 6
“very likely” consumers of organic cookies.
Looking at table 2-8, consumers in segments 1, 2, and 3 hold, on average,
significantly weaker beliefs that organic products are better than conventional products,
agree more that organic products are too expensive, rank importance of the organic label
in cookies lower, and are less likely to have purchased organic cookies in the past month,
relative to consumers in segment 6. In addition, consumers in these segments derive
negative utility from the organic attribute in cookies (although this is not statistically
significant for segment 2). Consumers in segment 1 (25.5%) do not like any of the cookie
labels provided to them in the choice scenarios, preferring a basic product. On average,
they did not select organic cookies in any of the eight choice scenarios. These consumers
are very unlikely to be interested in organic cookies. Among “unlikely” segments, segment
2 (15.5%) differs in that they gain some positive utility from the non–GMO label (a
component of organic). They chose the organic alternative in one of eight choice scenarios
and their WTP for the organic label in cookies is negative, but insignificant. Consumers in
segment 3 (26.5%) chose the organic cookie alternative in two of eight choice scenarios,
but their estimated WTP for the organic label is negative and significant.
79
Table 2-8. Latent Class Model Parameter Estimates, Cookies
Unlikely
Likely
Very
likely
Variable
Class 1
Class 2
Class 3
Class 4
Class 5
Class 6
Multinomial Logit Model Estimates
Price
-0.98***
(0.13)
-1.69***
(0.20)
-1.27***
(0.10)
-0.57***
(0.07)
-0.22***
(0.06)
-0.08
(0.14)
Organic
-0.86***
(0.28)
-0.24
(0.28)
-0.41***
(0.09)
0.46*
(0.25)
0.29**
(0.12)
3.30***
(0.38)
Non-GMO
-0.51**
(0.21)
0.26*
(0.15)
0.04
(0.10)
0.17
(0.15)
0.33***
(0.11)
0.02
(0.30)
Gluten-free
-1.00***
(0.28)
-0.15
(0.18)
0.16
(0.11)
-1.05***
(0.27)
0.40***
(0.12)
0.74**
(0.33)
Sugar-free
-1.12***
(0.34)
-0.15
(0.20)
-0.28***
(0.11)
-2.30***
(0.42)
0.61***
(0.16)
-0.12
(0.39)
No choice
-2.33***
(0.44)
-7.33***
(1.07)
-10.88***
(0.70)
-3.15***
(0.45)
-2.85***
(0.43)
0.48
(0.87)
WTP for organica
-$0.88
***
-$0.14
-$0.32
***
$0.81
**
$1.35
***
$42.88
WTP for organicb
-$0.80
-$0.13
-$0.30
$0.58
$1.19
$34.89
Latent Class Parameter Estimates
Intercept
1.61***
(0.59)
0.75
(0.66)
1.04*
(0.58)
2.19***
(0.59)
0.96
(0.62)
0
Factor 1c
(organic better)
-1.32***
(0.30)
-1.14***
(0.30)
-0.93***
(0.30)
-1.06***
(0.32)
-0.17
(0.31)
0
Factor 2c
(organic not better)
-0.21
(0.22)
-0.25
(0.23)
-0.10
(0.21)
-0.21
(0.21)
0.42**
(0.21)
0
Organic food is too
expensivec
0.95***
(0.18)
0.99***
(0.21)
0.70***
(0.17)
0.67***
(0.20)
0.38**
(0.18)
0
Selection of
organic is not goodc
-0.23
(0.18)
-0.03
(0.18)
-0.18
(0.17)
-0.09
(0.19)
-0.20
(0.18)
0
Rank of organic
label in cookiesd
0.22*
(0.12)
0.27**
(0.13)
0.38***
(0.12)
-0.06
(0.13)
0.16
(0.13)
0
Past purchase of
organic cookies
-3.63***
(0.51)
-2.56***
(0.49)
-1.96***
(0.42)
-1.94***
(0.44)
-0.79*
(0.45)
0
Class probability
25.5%
15.5%
26.5%
13.4%
13.3%
5.8%
% purchased
organic cookies
4.6%
12.7%
19.6%
26.9%
56.8%
79.0%
Note: Standard errors are in parentheses. *, **, *** denote significance at the 10%, 5%, and 1% level,
respectively.
a Mean WTP calculated according to equation (2-7); significance determined using Delta method.
b Mean WTP calculated according to equation (2-9); significance not available.
c The variables represent attitudes toward organic food and production systems either as a composite
measure (“Factor 1” and “Factor 2”) or as an individual attitude if they did not load on any factor during
factor analysis.
d 1 = highest, 7 = lowest.
80
Consumers in segments 4 (13.4%) and 5 (13.3%) are “likely” consumers of organic
cookies. In comparison to consumers in segment 6, they agree more that organic products
are too expensive, are less likely to have purchased organic cookies in the past month, and
either agree less that organic products are better (segment 4) or agree more that organic
products are not better (segment 5). But in contrast to “unlikely” segments, they gain
positive utility from organic cookies. Their ranking of importance of the organic label is
not significantly different on average from the ranking of “very likely” segment 6.
Consumers in segment 4 have a significant mean WTP of $0.81 for an organic label on a
1 lb. bag of cookies. They do not care for non–GMO label, but they clearly dislike gluten-
free and sugar-free labels. In the survey, they chose the organic alternative in two of eight
choice scenarios. Consumers in segment 5 have a significant mean WTP of $1.35 for the
organic label on cookies. In contrast to consumers in segment 4, they also gain positive
utility from other labels. On average, they chose organic cookies in four of eight choice
scenarios.
The “very likely” consumers in segment 6 (5.8%) love organic cookies and they
prefer them to conventional ones. Their high utility from the organic label on cookies,
combined with their low sensitivity to price, results in extremely high mean WTP values
using both equations (2-7) and (2-9). However, since the estimated price coefficient is
insignificant, the WTP measure is not meaningful (Hensher, Rose, and Greene, 2005).
Further, the applied Delta method provides evidence that the mean WTP based on equation
(2-7) is insignificant. Nevertheless, consumers in segment 6 have the highest utility from
organic cookies among all segments and in the survey, they chose organic cookies in seven
81
of eight scenarios. Although we do not obtain a meaningful mean WTP value for
consumers in this segment, we can conclude that they are willing to pay any reasonable
amount for the organic cookies and more than any other segment. In addition to organic,
they also have relatively strong preferences for the gluten-free label.
2.6.3 Comparisons across “Very Likely,” “Likely,” and “Unlikely” Groups
2.6.3.1 WTP for the Organic Label
Table 2-9 summarizes average WTP values for the organic label on bread and cookies
across all groups and for the whole sample. The mean WTP values were calculated (on a
group level) based on equation (2-9). We find significant differences in average WTP
values across groups within a product category and between groups across the two
products. Within bread category, the “very likely” group has a mean WTP for the organic
label $4.00 higher than that of the “unlikely” group. Within cookie category, note that “the
very likely” group corresponds to segment 6. As discussed in the previous section, we did
not find a meaningful mean WTP estimate for the organic label for this segment; however,
we report their WTP value for the illustration purposes, considering the previous discussion
which indicates that they are WTP more for the organic label than the other two groups.
Looking at the mean WTP values for the entire sample hides the differences found
across groups, indicating that it can be more appropriate to obtain WTP estimates for
consumer segments than for the whole sample. Also, this shows that consumer
segmentation can be very important, and further analysis of differences among groups is
necessary to understand how segments with high WTP differ from those with low WTP.
82
Table 2-9. WTP Values by Group and Product Category
Very likely
Likely
Unlikely
Whole sample
Bread
$3.57b
$1.91***b
-$0.52***^^^b
$1.37a(b)
Cookies
$34.89b
$0.88***b
-$0.46***^^^b
$2.06a(b)
Note: *** denote significance of the differences between “very likely” group (reference group) and “likely”
and “unlikely” groups at the 1% level within each product category. ^^^ denote significance of the
differences between “likely” group and “unlikely” group at the 1% level within each product category. a,b
denote significance of the differences between particular groups across product categories at the 5% and
1% level, respectively.
If a difference in the significance is found between the Welch and Wilcoxon tests, the result obtained using
the Wilcoxon test is reported in parentheses.
2.6.3.2 Preferences for Wheat Products
Next, we examined how preferences for selected product labels and characteristics differed
across the groups within and across product categories. We asked respondents to rank the
importance of seven labels, where 7 = “most important” and 1 = “least important.” Figure
2-1 plots the average ranking of each label within each group, and table 2-10 provides
additional details.
Figure 2-1. Average preferences for labels on bread and cookie products
Note: Respondents were asked to rank the importance of the labels, where 7 = “most important” and 1 =
“least important.”
1
2
3
4
5
6
7
Bread
Very likely/regular
Likely/occasional
Unlikely/very occasional
1
2
3
4
5
6
7Cookies
Very likely/regular
Likely/occasional
Unlikely/very occasional
83
Table 2-10. Average Importance Ranking of Labels across Groups
Bread
Cookies
Label
Very
likely
Likely
Unlikely
Very
likely
Likely
Unlikely
Natural
4.78ab
5.07ac
5.30bc
4.68ab
5.14ac
5.46*bc
Whole grain
4.70ab
5.91a(c)
5.72b(c)
4.08*b
4.15*c
4.76*bc
Organic
4.57ab
4.05ac
3.00bc
5.21*ab
4.41*ac
3.25*bc
Non–GMO
4.01ab
3.60a
3.46b
3.98
4.09*c
3.75*c
Local
3.47ab
3.91ac
4.34bc
3.37ab
4.23*ac
4.63*bc
Low-carb/sugar-free
3.31b
3.24c
3.81bc
3.36a
2.83*ac
3.40*c
Gluten-free
3.16ab
2.22a(c)
2.38b(c)
3.32b
3.16*c
2.76*bc
Note: Top three labels within each group are in bold. * denotes significant differences in rank for cookies
compared to bread within the group at the 10% level or better. a,b,c denote significant differences between
group of “very likely” and “likely,” “very likely” and “unlikely,” and “likely” and “unlikely,” respectively,
at the 10% level or better. If a difference in the significance is found between the Welch and Wilcoxon
tests, the result obtained using the Wilcoxon test is reported in parentheses.
In the “very likely” group, “natural,” “whole grain,” and “organic” labels are the
most important labels, regardless of product category. As expected, “organic” is ranked
highest by this group and lowest by the “unlikely” group. Interestingly, the “natural” and
“local” labels are both ranked significantly higher in the “likely” group and highest in the
“unlikely” group for both bread and cookies. As the importance of “organic” increases, the
importance of “natural” and “local” appears to decline, and vice versa.
20
This could be a
result of how the question was formulated, forcing those who find “organic” very important
to rank the other labels lower, but not necessarily meaning that they do not care for “local”
and “natural.”
21
“Gluten-free” and “low-carb” or “sugar-free” are ranked least important
20
Using a sample of consumers in Germany, Hempel and Hamm (2016) found that consumers who do not
regard organic production as very important had higher WTP values for local than for organic. On the other
hand, consumers who regard organic production as very important had overall higher WTP values for organic
than for local if the alternative product was from Germany, but their WTP for local increased above WTP for
organic when the alternative product was from a neighboring country.
21
For example, Lusk and Briggeman (2009) found a positive correlation between importance of naturalness
and WTP for organic bread. Similarly, Onyango, Hallman, and Bellows (2007) found a positive association
between naturalness and organic food purchases.
84
by “very likely” and “likely” groups for both products. The “unlikely” group ranks “gluten-
free” as least important but “organic” as second least important for both products. Finally,
average rankings for “whole grain” are significantly lower and for “organic” are
significantly higher for cookies than for bread across all three consumer groups. “Whole-
grain” is important for bread within the “likely” group in particular, and these consumers
might find bread labeled both “organic” and “whole-grain” appealing.
Table 2-11 reports the share within each group of those who had purchased the
product with the specific label in the previous month. For bread, “whole grain” and “non-
organic” are among the top three labels in all three groups. For cookies, only “local” is
among the top three labels across all three groups. “Organic” is among the top three labels
in the “very likely” and “likely” groups for both bread and cookies.
Table 2-11. Labels and Types of Purchased Bread and Cookie Products
Bread
Cookies
Label or type
Very
likely
Likely
Unlikely
Very
likely
Likely
Unlikely
Organic
79%
34%***
16%***
79%
41%***
12%***
Whole grain
54%b
58%
45%**
29%b
15%**(***)
6%***
Non-organic
41%
39%
38%
32%
36%
31%
Local
38%
27%***
27%***
34%
32%
21%**
Non–GMO
31%
15%***
9%***
40%
21%***
8%***
Gluten-free
31%
7%***
6%***
27%
20%
7%***
Home-baked
22%
9%***
16%**
31%
30%
28%
Low-carb/
sugar-free
20%(a)
6%***
8%***
31%(a)
20%(*)
15%**(***)
Note: Values represent share of those who purchased bread/cookies with the specific label or characteristic
in the past month. *, **, *** denote significance of the differences between “very likely” group (reference
group) and “likely” and “unlikely” groups at the 10%, 5%, and 1% level, respectively, within each product
category. a,b denote significance of the differences between “very likely” groups across product categories
at the 10% and 1% level, respectively. If a difference in significance is found between the Welch and
Wilcoxon tests, the result obtained using the Wilcoxon test is reported in the parentheses.
Although the “very likely” group ranks “local” as significantly less important
85
compared to the other two groups, a higher share of them actually purchased “local” bread
and/or cookies. Further, the share of those within the “very likely” group who purchased
bread with “organic,” “non–GMO,” “gluten-free,” “home-baked,” and “low-carb” labels,
and cookies with “organic,” “non-GMO,” and “whole-grain” labels, is significantly higher
than in the other two groups, although the “very likely” consumers rank some of these
labels as less or equally important in comparison to the other two groups. This suggests
that they are overall more likely to be interested in various labels. Finally, considering the
shares of those who purchased products with specific labels in the “very likely” group,
there are no significant differences across the two product types, with the exception of the
“whole grain” label, which appears to be more important for bread than for cookies.
“Whole grain” is also found to be more important for bread among the other two groups.
Figure 2-2. Preferences for characteristics of bread and cookie products
Note: Respondents were asked to rank the importance of the labels, where 8 = “most important” and 1 =
“least important.”
1
2
3
4
5
6
7Bread
Very likely/regular
Likely/occasional
Unlikely/very occasional
1
2
3
4
5
6
7Cookies
Very likely/regular
Likely/occasional
Unlikely/very occasional
86
Next, we asked respondents to rank the importance of eight product characteristics
(where 8 = “most important” and 1 = “least important”). Figure 2-2 summarizes the average
rankings of the product characteristics, and table 2-12 provides additional details.
Table 2-12. Average Importance Ranking of Product Characteristics across Groups
Product
characteristic
Bread
Cookies
Very likely
Likely
Unlikely
Very likely
Likely
Unlikely
Freshness
5.82b
6.06c
6.35bc
6.24*(ns)ab
5.76*a
5.81*b
Taste
5.71ab
6.09a
6.23b
5.84a(ns)b
6.42*a(ns)c
6.77*bc
Nutrition
5.34b
5.37c
4.12bc
5.03ab
3.93*ac
3.37*bc
Price
4.14ab
5.12ac
5.73bc
3.73ab
4.29*ac
5.55bc
Safety
4.11ab
3.33a
3.20b
3.87b
3.94*c
3.05bc
Appearance
3.84
3.67c
4.03c
3.95b
4.26*(c)
4.47*b(c)
Brand
3.80
3.77
3.99
3.84b
4.27*
4.39*b
Origin
3.24ab
2.59ac(ns)
2.35bc(ns)
3.50(a)b
3.13*(a)c
2.60*bc
Note: Top three characteristics within each group are in bold. * denotes significant differences in rank for
cookies compared to bread within the group at the 10% level or better. a,b,c denote significant differences
between group of “very likely” and “likely,” “very likely” and “unlikely,” and “likely” and “unlikely,”
respectively, at the 10% level or better. If a difference in the significance is found between the Welch and
Wilcoxon tests, the result obtained using the Wilcoxon test is reported in parentheses. ns denotes “not
significant”.
All groups rank “freshness” and “taste” as two most important, regardless of
product category. The “likely” and “unlikely” groups rank “taste” as more important for
cookies compared to bread. Also, the average ranking of “taste” in these groups tends to
be higher compared to the “very likely” group. If they find organic versions to be less tasty
compared to conventional, this may affect their choice.
Among all groups, the “very likely” group ranks “price” as least important, while
the “unlikely” group ranks “price” as most important, regardless of the product category.
This indicates that consumers of organic products tend to be less price-sensitive than non-
consumers, and the higher price of organic may act as a barrier to consumption for some
potential consumers.
87
2.6.3.3 Motivations and Barriers to Purchase Organic Bread and Cookies
Within each group identified for bread/cookies, some respondents had purchased organic
bread/cookies in the past month, and some had not. We examined motivations for
purchasing organic bread/cookies (table 2-13), potential barriers to purchasing (table 2-14),
and any significant differences across consumer groups.
Table 2-13. Reasons for Purchasing Organic Bread (B) and Cookies (C)
Bread
Cookies
Reason
Very
likely
Likely
Unlikely
Very
likely
Likely
Unlikely
Organic B (C) is (are)
healthier.
61%
61%
49%*
57%
51%
53%
Organic B (C) do(es) not
contain harmful substances.
44%
53%
35%
47%
38%
44%
Organic B (C) taste(s)
better.
37%
37%
17%***
41%
32%
22%**
My family members like
organic B (C).
31%
29%
34%
33%
28%
33%
Organic food production is
better for the environment.
30%b
38%
23%
51%b
33%**
31%**
I like to try new food
alternatives.
29%
29%
37%
29%
34%
41%
Organic B (C) is (are) more
visually attractive.
14%
7%**(*)
10%
16%
18%
18%
Organic food is trendy or in
fashion.
10%(a)
7%
10%
18%(a)
9%
7%*(**)
Purchased organic B (C) in
the past month
241
(79%)
92
(34%)
71
(16%)
49
(79%)
107
(41%)
85
(12%)
Note: Values represent share of those who selected a specific reason among those who purchased organic
bread/cookies. The total number of those who purchased organic bread/cookies per each group and their
share in the group are reported in brackets in the last row. Top three reasons per each group are in bold. *,
**, *** denote significance of the differences between “very likely” group (reference group) and “likely”
and “unlikely” groups at the 10%, 5%, and 1% level, respectively, within each product category. a,b denote
significance of the differences between “very likely” groups across product categories at the 10% and 1%
level, respectively. If a difference in the significance is found between the Welch and Wilcoxon tests, the
result obtained using the Wilcoxon test is reported in parentheses.
Across all six groups, among the top three reasons for purchasing organic bread
88
and cookies are that these options are “healthier” and that they “do not contain harmful
substances.” For “very likely” consumers, the belief that “organic bread tastes better” is in
the top three for bread and “organic food production is better for the environment” for
cookies. Compared to the “very likely” group, a significantly smaller share of consumers
in the “unlikely” group believes that “organic bread/cookies taste better”. As found
previously, all groups rank “taste” as one of the most important characteristics, but among
all groups, the “unlikely” group tends to care most about taste. As it turns out, they are also
less likely to believe that organic bread/cookies taste better.
Next, we examined why consumers did not purchase organic bread and cookies
(see table 2-14). “I did not think about it” is among the three most cited reasons across all
six groups, without any significant differences. “Organic bread (cookies) is (are) too
expensive” is also among the top three reasons for “likely” and “unlikely” groups
regardless of the product category, while the share of those who selected this reason in the
“very likely” group is significantly smaller. This further indicates that higher prices of
organic bread and cookies may act as a barrier to growth in the organic market. Similarly,
the share of those in the “likely” and “unlikely” groups, who selected that “organic bread
(cookies) is (are) not better than regular” and “regular bread (cookies) taste(s) better than
organic”, is higher compared to the “very likely” group. However, we did not examine
whether these negative expectations regarding taste are based on an actual experience or a
belief. Depending on that, organic bread/cookie tastings in stores might help increase the
sales.
89
Table 2-14. Reasons for Not Purchasing Organic Bread (B) and Cookies (C)
Bread
Cookies
Reason
Very
likely
Likely
Unlike-
ly
Very
likely
Likely
Unlike-
ly
I did not think about it.
30%
31%
31%
23%
30%
33%
Organic B (C) is (are) too
expensive.
27%
b(ns)
54%
***
52%
***
8%
b(ns)
29%
**(*)
41%
***(**)
I am not familiar with organic B
(C).
20%
11%(*)
23%
23%
25%
26%
Organic B (C) is (are) not
available in the store where I shop.
15%
14%
7%
*(**)
8%
16%
11%
I do not think organic B (C) is
(are) better than regular B (C).
14%
c(ns)
28%
***(**)
33%
***
0%
c(ns)
28%
***(**)
26%
***(**)
It is difficult to find the variety I
like.
12%
a(b)
9%
7%
38%
a(b)
13%
(**)
11%
*(***)
Organic B (C) has (have) a shorter
shelf life.
11%
c(ns)
10%
11%
0%
c(ns)
7%
***(ns)
7%
***(ns)
I am not familiar with organic
foods.
8%
b(ns)
6%
14%
0%
b(ns)
7%
***(ns)
8%
***(ns)
Regular B (C) taste(s) better than
organic.
6%
14%
**(*)
20%
***
8%
26%
**(ns)
28%
**(ns)
I do not trust that it is really
organic.
6%
b(ns)
12%
11%
0%
b(ns)
9%
***(ns)
10%
***(ns)
My family members do not like
organic food.
5%
a(ns)
6%
8%
0%
a(ns)
8%
***(ns)
6%
***(ns)
Organic B (C) is (are) not visually
attractive.
2%
1%
4%
0%
4%
**(ns)
4%
***(ns)
Did not purchase organic B (C) in
the past month
66
(22%)
175
(66%)
364
(84%)
13
(21%)
152
(59%)
603
(88%)
Note: The values represent share of those who selected a specific reason among those who did not purchase
organic bread/cookies. The total number of those who did not purchase organic bread/cookies per each
group and their share in the whole group in the brackets are reported in the last row. Top three reasons per
each group are in bold. *, **, *** denote significance of the differences between “very likely” group
(reference group) and “likely” and “unlikely” groups at the 10%, 5%, and 1% level, respectively, within
each product category. a,b,c denote significance of the differences between “very likely” groups across
product categories at the 10%, 5%, and 1% level, respectively. If a difference in the significance is found
between the Welch and Wilcoxon tests, the result obtained using the Wilcoxon test is reported in
parentheses. ns denotes “not significant”.
Overall, the share of those who hold a somewhat negative view toward organic
bread and cookies within the “very likely” group tends to be smaller. Instead, they select
the lack of familiarity with organic bread and cookies as one of the most important reasons
for not purchasing, suggesting that promoting organic bread and cookies to these groups
90
may increase sales. For cookies, they select difficulty in finding the variety they like as
most important, suggesting that further research examining their preferences in more detail
may be necessary.
2.6.3.4 Consumer Interest in Organic Versions of Wheat Products
Next, we asked respondents to rank importance of having a certified organic option for
different wheat product categories, where 9 = “most important” and 1 = “least important.”
The results in table 2-15 reveal that the importance of having a certified organic option for
different products is similar across all six groups, with only a few differences. In summary,
across all groups identified within bread and cookies, “specialty bread,” “white bread,” and
“pasta” are the three most important products in terms of having a certified organic version,
while “pastries,” “breadsticks,” and “pies” (and, also “cookies” among “likely” consumers
within bread product category) are ranked least important. This provides an evidence that
consumers view organic versions of virtue and vice products differently.
Table 2-15. Importance of Certified Organic Option by Wheat Product Category
Organic wheat
product category
Bread
Cookies
Very likely
Likely
Unlikely
Very likely
Likely
Unlikely
Specialty bread
6.87
7.03
6.18***
7.02
6.79
6.51(*)
White bread
6.85
6.95
6.60
6.52
6.69
6.82
Pasta
5.35
5.35
5.41
5.15
5.28
5.43
Bagels
4.83
4.94
4.82
5.02
4.81
4.86
Cookies
4.77
4.13***
4.53
5.13
4.68
4.37**
Crackers
4.50
4.90**
4.54
4.47
4.65
4.63
Pastries
4.29
4.13
4.52
4.23
4.39
4.34
Breadsticks
3.95
3.76
4.15
4.27
3.92
3.98
Pies
3.60
3.83
4.25***
3.21
3.78(**)
4.06**(***)
Note: Reported values represent average ranking of importance for each product category within a group of
consumers. The three most important product categories per group are in bold and the three least important
product categories are in bold and italic. *, **, *** denote significance of the differences between “very
likely” group (reference group) and “likely” and “unlikely” groups at the 10%, 5%, and 1% level,
respectively, within each product category. If a difference in the significance is found between the Welch
and Wilcoxon tests, the result obtained using the Wilcoxon test is reported in parentheses.
91
2.6.3.5 Socio-Demographic Characteristics
Table 2-16 reports selected sociodemographic characteristics for all groups. We also
compared the characteristics of “very likely” consumers across bread and cookies, but no
significant differences were found.
Table 2-16. Sociodemographic Characteristics across Groups for Bread and Cookies
Bread
Cookies
Characteristic
Very
likely
Likely
Unlikely
Very
likely
Likely
Unlikely
Female
52%
53%
51%
47%
54%
51%
Married
52%
52%
48%
57%
50%
50%
Income above $60k
47%
43%
31%***
53%
46%
35%***
4-year college or higher
48%
47%
30%***
50%
48%
36%**
Employed (full and part-time)
73%
49%***
49%***
69%
66%
51%***
Resident of California
56%
47%**
49%**
65%
56%
48%***(**)
Age below 45 years
66%
31%***
43%***
58%
57%
42%**
Children 0–17 years
82%
44%***
56%***
86%
72%
54%*(**)
Household size
2.89
2.42***
2.74(*)
2.92
2.79
2.65
Ethnicity (=1 if “white,” 0
otherwise)
66%
80%***
72%*
66%
69%
74%
N
307
267
435
62
259
688
Note: The values reported are either shares or means. *, **, *** denote significance of the differences
between “very likely” group (reference group) and “likely” and “unlikely” groups at the 10%, 5%, and 1%
level, respectively, within each product category. If a difference in the significance is found between the
Welch and Wilcoxon tests, the result obtained using the Wilcoxon test is reported in parentheses.
The differences in proportion of females and those who are married are not
statistically significant across the groups for bread and cookies, suggesting that gender and
marital status do not determine interest in organic bread and cookies. Past studies have also
found no connection between gender and decision to purchase organic foods (Zepeda and
Li, 2007; Nasir and Karakaya, 2014), but others have found a positive relationship for
female gender (Govindasamy and Italia, 1999) or married (Dimitri and Dettmann, 2012)
and likelihood to pay more for or buy organic food. However, higher income and education
levels positively impact interest in organic bread and cookies. While the differences
92
between the “very likely” and “likely” groups are not statistically significant, they are
significant between “very likely” and “unlikely” groups. This finding is consistent with the
previous findings in the literature that higher income and education positively affect
likelihood of purchasing organic foods (Dettmann and Dimitri, 2009; Ngobo, 2011; Dimitri
and Dettmann, 2012), although some studies have found no effects of income and
education on the intention to purchase (Nasir and Karakaya, 2014). Our results further
suggest that being employed (full-time or part-time) and California residence increase
interest in organic bread and cookies, but the significance of the differences between
groups depends on product type.
The effects of age and presence of children in the household are more ambiguous,
but they appear to be related. The proportion of respondents less than 45 years old and with
children in the household are lowest in the “unlikely” group for cookies, which is in line
with the previous findings in the literature that younger consumers (Zepeda and Li, 2007)
and consumers with younger children in the household (Hughner et al., 2007) are more
likely to purchase organic products. But the shares of younger consumers and consumers
with children in the household are lowest in the “likely” group for bread, suggesting that
the effects of age and the presence of children depend on the product category. Older
respondents who are less likely to have children living with them appear to be more
interested in organic versions of staple products, such as bread, which might be due to their
financial situation or health concerns.
22
Household size appears to have a small effect on interest in organic bread and
22
Health concern has been identified as a major reason for consumers’ positive attitudes toward or choice of
organic foods (Magnusson et al., 2003; Honkanen, Verplanken, and Olsen, 2006; Hughner et al., 2007).
93
cookies. Only for bread, the average household size of respondents in the “likely” group is
significantly smaller from the household size of the “very likely” group, which makes sense
since the “likely” group contains the lowest share of respondents with children in the
household.
Since the majority of our respondents selected “white” as their ethnic background,
we further examined whether there are differences across groups in the proportion of
whites. Significant differences are found only for bread, where the proportion of whites in
the “very likely” group is lower than in the other two groups. Overall, it appears that the
effect of ethnicity is weak, which was also found by Dimitri and Dettmann (2012).
2.6.3.6 Lifestyle Choices
Respondents were asked to evaluate their level of agreement with 16 lifestyle statements
on a 5-point scale (see table 2-17). On average, all groups somewhat agree that they “eat
grains daily,” with no significant differences across the groups, regardless of the product
category. The “very likely” group agrees more that they control their “fat consumption”
and that “supporting local farmers is important” to them compared to the other two groups
for both bread and cookies. Similarly, respondents in this group also agree significantly
more that they avoid eating “processed foods” and “food products with additives,” they are
more concerned about the “safety” and “origin” of their food, “physical activity and
exercise” and “agricultural open space” are more important to them, and they “buy
products with low environmental impact” more, but the “very likely” group within the
cookies category agrees with these statements significantly more than the same group
within the bread category.
94
Table 2-17. Average Level of Agreement with Lifestyle Statements by Group
Bread
Cookies
Statement
Very
likely
Likely
Unlikely
Very
likely
Likely
Unlikely
I control my salt and sugar intake.
3.85
3.84
3.59
***
4.02
3.78
3.70
**
I control my fat consumption.
3.78(a)
3.61
*
3.34
***
4.02(a)
3.67
**
3.45
***
I follow a vegetarian or vegan
diet.
2.32
1.63
***
1.48
***
2.37
2.15
1.58
***
I eat fresh produce daily.
4.02
3.72
***(**)
3.52
***
4.19
3.89
**(ns)
3.62
***
I eat grains daily.
3.71
3.83
3.69
3.92
3.79
3.69(*)
I avoid eating processed foods.
3.48a
3.16
***
2.62
***
3.76a
3.32
***(**)
2.85
***
I avoid eating food products with
additives.
3.66b
3.22
***
2.65
***
4.03b
3.43
***
2.91
***
I am concerned about my health.
4.10
4.01
3.88
***
4.26
4.08
(*)
3.91
**(***)
I am concerned about the safety
of my food.
4.18
a(b)
3.82
***
3.66
***
4.40
a(b)
4.06
**(***)
3.73
***
I am concerned about the origin
of my food.
3.81c
3.51
***
3.27
***
4.23c
3.66
***
3.38
***
I eat out infrequently.
3.25
3.15
3.10
3.44
3.20
3.12*
Physical activity or exercise is an
important part of my routine.
3.96
c(b)
3.64
***
3.43
***
4.32
c(b)
3.75
***
3.55
***
I buy products with low
environmental impact.
3.48c
3.04
***
2.82
***
3.85c
3.27
***
2.94
***
Recycling is a priority for me.
4.07
3.93
3.70
***
4.19
3.93
*(ns)
3.82
***(**)
Supporting local farmers is
important to me.
4.07
3.76
***
3.68
***
4.24
3.95
**
3.73
***
Agricultural open space is
important to me.
3.88b
3.71
*
3.51
***
4.18b
3.86
**
3.56
***
N
307
267
435
62
259
688
Note: The reported values represent group means of the responses ranging from 1 to 5, where 1 = “strongly
disagree,” 2 = “somewhat disagree,” 3 = “unsure,” 4 = “somewhat agree,” 5 = “strongly agree.”
*, **, *** denote significance of the differences between “very likely” group (reference group) and “likely”
and “unlikely” groups at the 10%, 5%, and 1% level, respectively, within each product category. a,b,c denote
significance of the differences between “very likely” groups across product categories at the 10%, 5%, and
1% level, respectively. If a difference in the significance is found between the Welch and Wilcoxon tests,
the result obtained using the Wilcoxon test is reported in parentheses. ns denotes “not significant”.
Further, respondents in the “very likely” group agree more, on average, that they
control their “salt and sugar intake” and that they are concerned about their “health” when
95
compared to the “unlikely” group, but the level of agreement with these statements is not
different statistically when compared to the “likely” group. They also agree more that they
“follow a vegetarian or vegan diet,” “eat fresh produce daily,” and “recycling is a priority”
compared to the other groups. When it comes to frequency of eating out, overall there are
no major differences except that consumers in the “unlikely” group agree less that they
“eat out infrequently” at the 10% level.
Examining lifestyle statements across consumer groups confirms that lifestyle
plays an important role in distinguishing consumers of organic food from non-consumers,
which was previously found in Gil, Gracia, and Sanchez (2000) and Sanjuán et al. (2003).
The authors in these studies segmented Spanish consumers based on lifestyle and found
that those who were concerned about healthy diet and environment were more likely to
purchase organic food (Gil, Gracia, and Sanchez, 2000), and those with higher health
awareness and balanced lifestyles consumed organic food more frequently. Although we
have not segmented survey respondents based on lifestyles, we find comparable results and
clear differences in lifestyle across the three groups of consumers, regardless of the
examined product. It appears that “very likely” consumers are, on average, more inclined
toward making lifestyle choices considered to be healthier and better for the environment,
and they care more about the origin, safety, and production aspects of their food. Our results
also support previous findings in de Magistris and Gracia (2008) and Chen (2009) that
healthy lifestyle plays a role in positively affecting consumer attitudes toward organic
foods.
96
2.6.3.7 Shopping and Consumption Behavior
We asked respondents to select the store type(s) where they typically shop for bread and
cookies. Table 2-18 reports the share of respondents in each group who selected each store
type. “Very likely” consumers are significantly more likely to purchase bread/cookies in
specialty stores, bulk stores, and local bakeries in comparison to “likely” and “unlikely”
consumers for bread and “unlikely” consumers for cookies.
Table 2-18. Group Percentage Shares and Means Related to Consumer Shopping
Habits
Bread
Cookies
Very
likely
Likely
Unlikely
Very
likely
Likely
Unlikely
Store type
Grocery store
71%a(b)
72%
76%
53%a(b)
59%
69%**
Multipurpose store
49%
35%***
49%
40%
46%
46%
Specialty store
38%
23%***
11%***
34%
30%
16%***
Bulk store
30%
17%***
15%***
32%
23%
16%**(***)
Local bakery
23%
9%***
11%***
29%
23%
10%***
Discount store
11%
10%
12%
11%
17%
14%
Reviewed informationc
Nutrition facts panel
69a
62***
50***
78a
66***
55***
Ingredient list
68b(a)
61***(**)
48***
77b(a)
66***(**)
53***
Serving size
55
49**
46***
62
53**
47***
Package size
58
57
61
62
59
58
Allergy warnings
47
25***
25***
54
41**(***)
26***
Front labels
71b(a)
56***
40***
79b(a)
65***
47***
Other
Bread/cookies
purchase frequencyd
2.79b
2.45***
2.74
2.31b
2.02*
1.79***
Bread/cookies cost
$9.10
$4.40***
$5.72***
$8.95
$7.99
$5.00***
Household grocery
shopping
83%
81%
82%
85%
83%
81%
Note: *, **, *** denote significance of the differences between “very likely” group (reference group) and
“likely” and “unlikely” groups at the 10%, 5%, and 1% level, respectively, within each product category. a,b
denote significance of the differences between “very likely” groups across product categories at the 5% and
1% level, respectively. If a difference in the significance is found between the Welch and Wilcoxon tests,
the result obtained using the Wilcoxon test is reported in parentheses.
c 0 = “never,” 25 = “sometimes,” 50 = “about half the time,” 75 = “most of the time,” 100 = “always.”
Reported values are group means.
d 0 = “never,” 1 = “once a month or less,” 2 = “several times a month,” 3 = “once a week,” 4 = “several
times a week.” Reported values are group means.
97
We also asked respondents to indicate on a 100-point scale how often they review
specific product information, where 0 = “never” and 100 = “always.” Table 2-18 also
reports the averages per group and information type. As expected, we find that “very
likely” consumers review nutrition facts panels, ingredient lists, serving sizes, allergy
warnings, and front labels significantly more frequently than the other two groups,
regardless of product category. We also find that, on average, “very likely” consumers
within each product category purchase bread/cookies as frequently as or more than and
spend as much as or more per purchase than the other two groups. The average percentage
of household grocery shopping that each respondent is responsible for is 81%–85%, and
there are no significant differences across the groups. This indicates that the results are not
affected by the overall involvement in grocery shopping.
We also asked respondents how often they consume sustainable foods in general
(e.g., labeled as organic, locally grown, GMO–free, natural, grass-fed, free-range, etc.). As
shown in table 2-19, “very likely” consumers of both bread and cookies consume
sustainable foods significantly more frequently than consumers in the other two groups, as
expected, and significant differences are found between these two groups as well (not
reported).
Finally, we asked respondents whether they or any member of their household
suffers from wheat/gluten intolerance, celiac disease, or avoids wheat/gluten for other
reasons. The aim was to determine whether consumers with these limitations, which
certainly affect their choice and consumption of wheat products, found organic versions of
wheat products appealing. Table 2-19 reports the share of consumers within each group
98
who indicated a specific limitation. As it turns out, for bread, the share of respondents with
limitations is significantly higher in the group of “very likely” consumers of organic bread
than in the other two groups. For cookies, there are differences in share of respondents with
these limitations between the groups as well, but these are not necessarily significant. These
results suggest that indeed those who need to or choose to avoid wheat/gluten products
might be substituting the regular wheat products with organic versions, as they are
concentrated in the “very likely” group. However, it appears that this may hold more for
staple wheat product categories like bread and less for “optional” products, like cookies.
Table 2-19. Group Percentage Shares and Means Related to Consumer
Consumption Behavior and Limitations
Bread
Cookies
Very
likely
Likely
Unlikely
Very
likely
Likely
Unlikely
Sustainable food
consumption frequencyb
3.02a
2.13***
1.35***
3.55a
2.71***
1.69***
Wheat intolerance/allergy
21%
4%***
5%***
27%
15%**
6%***
Wheat avoidance
22%
6%***
4%***
23%
19%
5%***
Gluten intolerance/allergy
19%
7%***
6%***
13%
13%
8%
Gluten avoidance
28%
9%***
6%***
29%
19%*
10%***
Celiac disease
4%
1%***
1%***
2%
4%
1%
Note: *, **, *** denote significance of the differences between “very likely” group (reference group) and
“likely” and “unlikely” groups at the 10%, 5%, and 1% level, respectively, within each product category.
a denotes significance of the differences between “very likely” groups across product categories at the 1%
level.
b 0 = “never,” 1 = “once a month or less,” 2 = “several times a month,” 3 = “once a week,” 4 = “several
times a week.” Reported values are group means.
2.7 Conclusions
This study attempts to understand the factors that determine consumer interest in organic
wheat products, the product characteristics and labels that matter to current and potential
consumers, and how potential consumers (who represent growth potential in the market)
99
differ from current organic consumers by examining the differences across consumer
segments. The analysis is performed for bread (a staple or virtue product) and cookies (a
hedonistic or vice product) to assess whether the findings may be broadly applicable or
product specific. We use latent class modeling for discrete choice analysis to cluster
respondents based on their attitudes toward organics in general and their preferences for
the organic label in selected organic wheat products, resulting in three groups: “very
likely,” “likely,” and “unlikely” consumers, each indicating the likelihood of purchasing a
wheat product with an organic label. Welch tests and Wilcoxon tests are used to evaluate
the significance of the differences between the groups. Data were collected using an online
survey administered in 16 U.S. western states during Summer 2017 that yielded 1,009 valid
responses.
The analysis reveals significant differences that are not product specific among
groups. First, all groups have significantly different WTP for organic labels, indicating that
focusing on WTP among the whole sample may be misleading; in some cases, it may be
more appropriate to obtain WTP estimates by segment. Second, as the importance of the
organic label on bread and cookies increases, the importance of local and natural labels
decreases (and vice versa). However, we also find that a higher share of “very likely”
consumers purchased local bread and cookies compared to “unlikely” consumers. Apart
from the local label, we find that they have a higher interest in labels in general, since an
equal or a larger share of “very likely” consumers purchased bread and cookies with some
labels, although they rank them equally or less important as the other two groups.
Third, those classified as “very likely” consumers of organic bread and cookies
100
rank price and taste overall as less important than the other two groups. Among those who
had purchased organic bread and cookies in the past month, “unlikely” consumers selected
that “organic tastes better” less frequently than consumers in the “very likely” and “likely”
groups. Among those who did not purchase organic versions of the products, “likely” and
“unlikely” consumers believe more frequently that organic products are too expensive,
organic is not better than regular (conventional), and regular tastes better. In summary, it
appears that the higher price of organic wheat products and expected inferior taste
compared to non-organic versions may act as a barrier to the consumption of organic wheat
products. This is in line with findings by Ellison et al. (2016), who also find that the organic
versions of vice products are perceived to be more nutritious than conventional ones, which
provides some background for our finding that the “very likely” group ranks nutritional
value in cookies (a vice product) as significantly more important than the other two groups.
Fourth, consumers in all three groups, on average, select specialty bread, white
bread, and pasta as the three wheat products they would most like to have a certified organic
option. Fifth, higher income, higher education, employment, and California residence are
related to higher interest in organic bread and cookies, while gender and marital status play
no role. Sixth, as expected, “very likely” consumers tend to have lifestyles that are
considered healthier and more environmentally responsible, be more concerned about the
safety and origin of their food, and support local agriculture more compared to other
consumers, particularly in the “unlikely” group.
Finally, “very likely” consumers review nutrition fact panels, ingredient lists,
serving sizes, allergy warnings, and front labels, and they purchase bread and cookies in
101
specialty stores, bulk stores, and local bakeries significantly more frequently than
consumers in the other two groups. Also, we find that the share of respondents who need
to or choose to avoid gluten/wheat is the highest in the “very likely” groups, indicating that
they find organic wheat products appealing and that they may be substituting regular wheat
products with organic versions.
We draw several conclusions to assist food manufacturers and marketers in
production and marketing decisions: First, it may be worth investing in research and
development to improve the taste of organic wheat products, particularly for hedonistic
products such as cookies, which may help justify the higher price tag. All groups rank
“organic” as more important for cookies than bread, yet smaller share of respondents
purchased organic cookies than bread and expected taste is one of the reasons. At the same
time, “likely” and “unlikely” consumers rank “taste” as more important in cookies than
bread. However, we did not examine whether consumers’ expectations that organic
products are less tasty are based on an actual experience or a belief. Marketers may also
want to conduct organic cookie sample tastings, which will allow consumers to experience
and personally evaluate the taste of organic cookies, and it may help convince “likely”
consumers, who appear to be more price conscious than “very likely” consumers, that
organic cookies are worth the extra cost.
Second, our results suggest that organic bread produced with whole-grain flour may
appeal to “likely” consumers, who rank the whole-grain label in bread as most important.
To further attract “likely” consumers, organic bread should be advertised as fresh, tasty,
and nutritious, and organic cookies as fresh and tasty, since nutrition does not play an
102
important role for them.
One of our contributions is that we identify consumer segments for bread and
cookies based on their likelihood of purchasing organic bread and cookies, which had not
been done previously. Second, this study contributes to the vast literature examining which
consumer characteristics determine consumption of organic products in the context of
wheat products, which have not received much attention in the literature so far.
Specifically, we identify and discuss differences in socio-demographics and lifestyles
across consumer segments. Third, we provide recommendations for food manufacturers
and marketers regarding preferred product characteristics and labels. Fourth, we contrast
the findings for two wheat product categories to determine whether our findings are broadly
applicable or product specific. However, there are some limitations which need to be
considered while interpreting the results. First, we suspect that organic food consumers are
underrepresented in our sample and the groups of “very likely” and “likely” consumers
may be larger. Second, since the employed choice experiment was hypothetical, i.e. it did
not involve actual purchases, there is a possibility that respondents evaluated the attributes,
in particular price, less thoroughly. Consequently, the calculated WTP values may be
higher than the actual WTP.
This study is one of few examining consumer segments and/or their preferences for
organic wheat products in more detail, but more work can be done. Future research may
build on our findings and examine whether consumers’ beliefs that organic wheat products
are less tasty are based on expectations or on actual experience, and what aspects of taste
they would like to improve, if any. Depending on the outcome, sample tastings or careful
103
improvements to taste could be undertaken, while considering the needs and preferences
of current consumers. It would also be worth investigating whether consumers value the
organic label in whole-grain bread significantly more than in white bread. Finally, it would
be interesting to examine why respondents with gluten/wheat consumption limitations are
interested in organic versions of wheat products, particularly bread, to provide further
recommendations for marketing organic wheat products to the growing segment of
gluten/wheat-avoiding consumers, which appears to find organic versions of wheat
products appealing.
References
Aertsens, J., W. Verbeke, K. Mondelaers, and G. Van Huylenbroeck. 2009. “Personal
Determinants of Organic Food Consumption: A Review.” British Food Journal
111(10): 1140-1167.
Bellows, A.C., B. Onyango, A. Diamond, and W.K. Hallman. 2008. “Understanding
Consumer Interest in Organics: Production Values vs. Purchasing Behavior.”
Journal of Agricultural & Food Industrial Organization 6(1).
Bitzios, M., I. Fraser, and J. Haddock-Fraser. 2011. “Functional Ingredients and Food
Choice: Results from a Dual-Mode Study Employing Means-End-Chain Analysis
and a Choice Experiment.” Food Policy 36(5):715-725.
Boxall, P.C., and W.L. Adamowicz. 2002. “Understanding Heterogeneous Preferences in
Random Utility Models: A Latent Class Approach.” Environmental and Resource
Economics 23(4):421-446.
104
Bureau of Labor Statistics [BLS]. 2017. Average Retail Food and Energy Prices, U.S.
and Midwest Region. Available at: https://www.bls.gov/regions/mid-
atlantic/data/AverageRetailFoodAndEnergyPrices_USandMidwest_Table.htm
[Accessed May 20, 2017].
Carlson, A., and E. Jaenicke. 2016. Changes in Retail Organic Price Premiums from
2004 to 2010. U.S. Department of Agriculture, Economic Research Service, May.
Chen, M.F. 2009. “Attitude toward Organic Foods among Taiwanese as Related to
Health Consciousness, Environmental Attitudes, and the Mediating Effects of a
Healthy Lifestyle.” British Food Journal 111(2):165-178.
Costello, A.B., and J.W. Osborne. 2005. “Best Practices in Exploratory Factor Analysis:
Four Recommendations for Getting the Most from Your Analysis.” Practical
Assessment, Research & Evaluation 10(7):1-9.
de Magistris, T., and A. Gracia. 2008. “The Decision to Buy Organic Food Products in
Southern Italy.” British Food Journal 110(9):929-947.
Dettmann, R.L., and C. Dimitri. 2009. “Who's Buying Organic Vegetables? Demographic
Characteristics of US Consumers.” Journal of Food Products Marketing
16(1):79-91.
Dimitri, C., and R.L. Dettmann. 2012. “Organic Food Consumers: What Do We Really
Know about Them?.” British Food Journal 114(8):1157-1183.
DiStefano, C., M. Zhu, and D. Mindrila. 2009. “Understanding and Using Factor Scores:
Considerations for the Applied Researcher.” Practical Assessment, Research &
Evaluation 14(20):1-11.
105
Ellison, B., B.R. Duff, Z. Wang, and T.B. White. 2016. “Putting the Organic Label in
Context: Examining the Interactions between the Organic Label, Product Type,
and Retail Outlet.” Food Quality and Preference 49:140-150.
Fabrigar, L.R., D.T. Wegener, R.C. MacCallum, and E.J. Strahan. 1999. “Evaluating the
Use of Exploratory Factor Analysis in Psychological Research.” Psychological
Methods 4(3):272-299.
Gil, J.M., A. Gracia, and M. Sanchez. 2000. “Market Segmentation and Willingness to
Pay for Organic Products in Spain.” The International Food and Agribusiness
Management Review 3(2):207-226.
Govindasamy, R., and J. Italia. 1999. “Predicting Willingness-to-pay a Premium for
Organically Grown Fresh Produce.” Journal of Food Distribution Research
30:44-53.
Gracia, A., and T. de Magistris. 2013. “Organic Food Product Purchase Behaviour: A
Pilot Study for Urban Consumers in the South of Italy.” Spanish Journal of
Agricultural Research 5(4):439-451.
Greene, W.H. 2012. NLOGIT Version 5 Reference Guide. New York: Econometric
Software, Inc.
Greene, W.H., and D.A. Hensher. 2003. “A Latent Class Model for Discrete Choice
Analysis: Contrasts with Mixed Logit.” Transportation Research Part B:
Methodological 37(8):681-698.
Grice, J.W. 2001. “Computing and Evaluating Factor Scores.” Psychological Methods
6(4):430-450.
106
Hasselbach, J.L., and J. Roosen. 2015. “Consumer Heterogeneity in the Willingness to
Pay for Local and Organic Food. Journal of Food Products Marketing 21(6):608-
625.
Hempel, C., and U. Hamm. 2016. “How Important is Local Food to Organic-Minded
Consumers?.” Appetite 96:309-318.
Hensher, D.A., J.M. Rose, and W.H. Greene. 2005. Applied Choice Analysis: A Primer.
Cambridge University Press.
Hidrue, M.K., G.R. Parsons, W. Kempton, and M.P. Gardner. 2011. “Willingness to Pay
for Electric Vehicles and Their Attributes.” Resource and Energy Economics
33(3):686-705.
Honkanen, P., B. Verplanken, and S.O. Olsen. 2006. “Ethical Values and Motives
Driving Organic Food Choice.” Journal of Consumer Behaviour 5(5):420-430.
Hooper, D., J. Coughlan, and M. Mullen. 2008. “Structural Equation Modelling:
Guidelines for Determining Model Fit.” The Electronic Journal of Business
Research Methods 6(1):53-60.
Hughner, R.S., P. McDonagh, A. Prothero, C.J. Shultz, and J. Stanton. 2007. “Who are
Organic Food Consumers? A Compilation and Review of Why People Purchase
Organic Food.” Journal of Consumer Behaviour 6(2‐3):94-110.
Hui, S.K., E.T. Bradlow, and P.S. Fader. 2009. “Testing Behavioral Hypotheses Using an
Integrated Model of Grocery Store Shopping Path and Purchase Behavior.”
Journal of Consumer Research 36(3):478-493.
107
Jolly, D.A. 1991. “Differences between Buyers and Nonbuyers of Organic Produce and
Willingness to Pay Organic Price Premiums.” Journal of Agribusiness 9(1):97-
111.
Jung, T., and K.A.S. Wickrama. 2008. “An Introduction to Latent Class Growth Analysis
and Growth Mixture Modeling.” Social and Personality Psychology Compass
2(1):302-317.
Kihlberg, I., and E. Risvik. 2007. “Consumers of Organic Foods – Value Segments and
Liking of Bread.” Food Quality and Preference 18(3):471-481.
Krystallis, A., and G. Chryssohoidis. 2005. “Consumers' Willingness to Pay for Organic
Food: Factors that Affect it and Variation per Organic Product Type.” British
Food Journal 107(5):320-343.
Krystallis, A., C. Fotopoulos, and Y. Zotos. 2006. “Organic Consumers' Profile and their
Willingness to Pay (WTP) for Selected Organic Food Products in Greece.”
Journal of International Consumer Marketing 19(1):81-106.
Lea, E., and T. Worsley. 2005. “Australians' Organic Food Beliefs, Demographics and
Values.” British Food Journal 107(11):855-869.
Lee, W.C.J., M. Shimizu, K.M. Kniffin, and B. Wansink. 2013. “You Taste What You
See: Do Organic Labels Bias Taste Perceptions?.” Food Quality and Preference
29(1):33-39.
Li, J., L. Zepeda, and B.W. Gould. 2007. “The Demand for Organic Food in the US: An
Empirical Assessment.” Journal of Food Distribution Research 38(3):54-69.
108
Lockie, S., K. Lyons, G. Lawrence, and J. Grice. 2004. “Choosing Organics: A Path
Analysis of Factors Underlying the Selection of Organic Food Among Australian
Consumers.” Appetite 43(2):135-146.
Loureiro, M.L., J.J. McCluskey, and R.C. Mittelhammer. 2001. “Assessing Consumer
Preferences for Organic, Eco-Labeled, and Regular Apples.” Journal of
Agricultural and Resource Economics 404-416.
Lusk, J.L., and B.C. Briggeman. 2009. “Food Values.” American Journal of Agricultural
Economics 91(1):184-196.
Magidson, J., and J. Vermunt. 2002. “Latent Class Models for Clustering: A Comparison
with K-Means.” Canadian Journal of Marketing Research 20(1):36-43.
Magnusson, M.K., A. Arvola, U.K.K. Hursti, L. Åberg, and P.O. Sjödén. 2001.
“Attitudes towards Organic Foods among Swedish Consumers.” British Food
Journal 103(3):209-227.
Magnusson, M.K., A. Arvola, U.K.K. Hursti, L., Åberg, and P.O. Sjödén. 2003. “Choice
of Organic Foods is Related to Perceived Consequences for Human Health and to
Environmentally Friendly Behaviour.” Appetite 40(2):109-117.
McFadden, D. 1974. “Conditional Logit Analysis of Qualitative Choice Behavior.” In P.
Zarembka, ed. Frontiers in Econometrics. New York: Academic Press, pp. 105–
142.
Nasir, V.A., and F. Karakaya. 2014. “Underlying Motivations of Organic Food Purchase
Intentions.” Agribusiness 30(3):290-308.
109
Nasserinejad, K., J. van Rosmalen, W. de Kort, and E. Lesaffre. 2017. “Comparison of
Criteria for Choosing the Number of Classes in Bayesian Finite Mixture Models.”
PLoS ONE 12(1)
Ngobo, P.V. 2011. “What Drives Household Choice of Organic Products in Grocery
Stores?.” Journal of Retailing 87(1):90-100.
Nylund, K.L., T. Asparouhov, and B.O. Muthén. 2007. “Deciding on the Number of
Classes in Latent Class Analysis and Growth Mixture Modeling: A Monte Carlo
Simulation Study.” Structural Equation Modeling 14(4):535-569.
Onyango, B.M., W.K. Hallman, and A.C. Bellows. 2007. “Purchasing Organic Food in
US Food Systems: A Study of Attitudes and Practice.” British Food Journal
109(5):399-411.
Padel, S., and C. Foster. 2005. “Exploring the Gap between Attitudes and Behaviour:
Understanding Why Consumers Buy or Do Not Buy Organic Food.” British Food
Journal 107(8):606-625.
Revelle, W.R. 2017. “psych: Procedures for Personality and Psychological Research.”
Software.
Ruto, E., and G. Garrod. 2009. “Investigating Farmers' Preferences for the Design of
Agri-Environment Schemes: A Choice Experiment Approach.” Journal of
Environmental Planning and Management 52(5):631-647.
Sanjuán, A.I., M. Sánchez, J.M. Gil, A. Gracia, and F. Soler. 2003. “Brakes to Organic
Market Enlargement in Spain: Consumers’ and Retailers’ Attitudes and
Willingness to Pay.” International Journal of Consumer Studies 27(2):134-144.
110
Swait, J. 1994. “A Structural Equation Model of Latent Segmentation and Product
Choice for Cross-Sectional Revealed Preference Choice Data.” Journal of
Retailing and Consumer Services 1(2):77-89.
Tabachnick, B.G., and L.S. Fidell. 2007. Using Multivariate Statistics, 5th. ed. Boston:
Allyn & Bacon.
Teuber, R., I. Dolgopolova, and J. Nordström. 2016. “Some Like It Organic, Some Like
It Purple and Some Like It Ancient: Consumer Preferences and WTP for Value-
Added Attributes in Whole Grain Bread.” Food Quality and Preference 52:244-
254.
Train, K.E. 2009. Discrete Choice Methods with Simulation, 2nd. ed. Cambridge
University Press.
Ureña, F., R. Bernabéu, and M. Olmeda. 2008. “Women, Men and Organic Food:
Differences in Their Attitudes and Willingness to Pay. A Spanish Case Study.”
International Journal of consumer Studies 32(1):18-26.
U.S. Department of Agriculture – National Institute of Food and Agriculture (USDA-
NIFA) Organic Agriculture Research and Extension Initiative (OREI). 2014-
2020. Compost Carryover and Cover Crop Effects on Soil Quality, Profitability,
and Cultivar Selection in Organic Dryland Wheat.
Van Doorn, J., and P.C. Verhoef. 2011. “Willingness to Pay for Organic Products:
Differences between Virtue and Vice Foods.” International Journal of Research
in Marketing 28(3):167-180.
111
Yong, A.G., and S. Pearce. 2013. “A Beginner’s Guide to Factor Analysis: Focusing on
Exploratory Factor Analysis.” Tutorials in Quantitative Methods for Psychology
9(2):79-94.
Zepeda, L., H.S. Chang, and C. Leviten-Reid. 2006. “Organic Food Demand: A Focus
Group Study Involving Caucasian and African-American Shoppers.” Agriculture
and Human Values 23(3):385-394.
Zepeda, L., and J. Li. 2007. “Characteristics of Organic Food Shoppers.” Journal of
Agricultural and Applied Economics 39(1):17-28.
Zheng, Y. 2014. “Consumer Preference and Willingness-To-Pay for Locally Produced,
Organic Food: A Stated Choice Approach.” MS thesis, University of Guelph.
112
CHAPTER 3
DO EXTRA LABELS PAY? THE IMPACT OF NON-GMO AND HEALTH
RELATED LABELS ON CONSUMER WILLINGNESS TO PAY FOR ORGANIC
WHEAT PRODUCTS
3.1 Abstract
In this essay, consumer preferences and willingness to pay (WTP) for the organic label
alone and in combination with other labels in the context of wheat products are explored.
The investigated labels are either directly related to the organic label (non-GMO), or
perceived as health promoting (gluten-free, low-carb and sugar-free). The analysis is
performed for two products (bread and cookies) to examine whether results depend on
product type (virtue vs. vice). The data used comes from an online survey conducted in
2017 across 16 U.S. western states. The analysis is completed using multinomial logit and
random parameter logit models. It is found that the combinations of the labels with the
organic label are not widely accepted and they may reduce overall consumer WTP,
however, there are consumers who find these combinations appealing. Also, consumer
knowledge that organic is also non-GMO can increase overall WTP for the combination of
organic and non-GMO labels above WTP for the organic label alone. Finally, it is found
that those with some gluten or wheat intolerance or avoidance represent a profitable market
for organic wheat products.
113
3.2 Introduction
Consumption of organic foods has become one of the most significant trends in the food
industry, and consumer demand for organic food continues to increase each year. In 2017,
the total sales of organic foods were at $45.2 billion in the United States, up 6.4% compared
to the sales in 2016, outpacing the growth in the sales of conventional foods (Organic Trade
Association, 2018). In general, organic foods are more expensive than their conventional
counterparts so the additional costs of producing organics are usually covered. And,
previous studies have found that consumers are willing to pay extra for the organic label
alone in a variety of products, ranging from fresh products such as produce (Jolly, 1991;
Govindasamy and Italia, 1999; Govindasamy, DeCongelio, and Bhuyan, 2006) and meats
(Van Loo et al., 2011) to processed, multi-ingredient food products (Batte et al., 2007).
Like the organic label, the non-GMO (genetically modified organism) label has also
gained importance in the US food market in recent years. According to a report by Nielsen,
sales of non-GMO products jumped from $12.9 billion in 2012 to $21.1 billion in 2016
(USA Today, 2016).
23
The non-GMO label is related to the organic label in the sense that
the absence of GMOs is listed as one of the conditions that each product labelled as organic
must meet, as established by the National Organic Program (NOP) in the United States in
2002. That is, the non-GMO label complements the organic label.
However, findings of a recent study by McFadden and Lusk (2017) suggest that
consumers do not recognize the difference between organic and non-GMO labels, and they
23
At the same time, the number of non-GMO labels in new products introduced to the market increased by
558% between 2012 and 2016 (U.S. Department of Agriculture Economic Research Service – USDA ERS,
2017).
114
confuse the two. Consumers preferring non-GMO over organic labels might be doing so
based on perception rather than actual facts and knowledge, and organic food producers
may be losing sales to non-GMO labeled foods, which represent a cheaper alternative
(Roseboro, 2013). A possible solution to this problem—to provide both labels together—
may not help either, as McFadden and Lusk (2017) find that in the case of products with
both labels, the overall WTP is close to the WTP for organic alone. On the other hand,
Conner and Christy (2004) found that the average WTP for the combination of organic and
non-GMO labels was higher than the WTP for organic alone, but the authors focused on
organic food consumers. Other studies show that when the definitions of the labels are
provided to consumers, they may be willing to pay significantly more for organic labeled
foods relative to non-GMO foods (He and Bernard, 2011), or they may not (Bernard,
Zhang, and Gifford, 2006), but the authors did not examine the effect of organic knowledge
on consumer WTP for the labels alone and together.
Overall, it seems that consumers care greatly about the non-GMO component of
organics and/or they perceive the organic and non-GMO labels as substitutes. To our
knowledge the effect of having both labels together on a product has not been examined
separately for consumers with and without prior knowledge that organic must be non-
GMO
24
, which may be correlated with their interest in these labels and thus affect their
WTP values. This analysis is one of the objectives of this study. Further, we examine
whether consumer overall knowledge of organic products and production systems
24
In this study, we do not consider specifically that consumers may be aware of and concerned about possible
contamination of organic with GMOs. This could lead them to prefer “Non-GMO Project Verified” (NGPV)
label over organic label, since the NGPV label verifies that the GMOs are not present in the final product,
while the organic label examines that GMOs are not used in the production process only.
115
influences their WTP for the labels alone and in combination. Since it has been found
previously that subjective (self-reported) knowledge affects consumer interest in organic
foods more than objective (actual or tested) knowledge (Pieniak, Aertsens, and Verbeke,
2010; Aertsens et al., 2011), we examine the impact of both on consumer preferences for
organic and non-GMO labels. We expect that our findings will provide insights regarding
why some “natural food” retailers experience positive consumer responses to products that
contain both labels (Roseboro, 2015), despite recent research suggesting otherwise.
In addition to organic and non-GMO labels, other labels have received increased
attention recently. Food manufacturers and marketers have been increasing the number of
labels or claims on new food products introduced to the market, recognizing the potential
of labels to attract consumers. For example, in 2009 a new food product had on average
2.1 claims
25
, but in 2016 it was 3.7 claims (USDA ERS, 2017). This indicates that there is
a trend of increased claims on a single product, but the question is whether multiple labels
provide any benefit to consumers and whether they are helpful in deciding to purchase a
product or not, or whether they are instead causing confusion.
In addition to the already mentioned interaction of organic and non-GMO labels,
of which the consumer interest has been driven to a considerable extent by their health
concerns (Gil, Gracia, and Sanchez, 2000; Torjusen et al., 2001; Magnusson et al., 2003;
Baker et al., 2004; Lesch, Anderson, and Wachenheim, 2006; Zepeda, Chang, and Leviten-
Reid, 2006), we examine the interaction of the organic label with other labels perceived as
healthy. We focus on gluten-free, low-carb and sugar-free labels, which appeal to
25
Calculated as the ratio of total number of claims on new products introduced to the market in that year to
number of new products.
116
consumers who seek to avoid ingredients that may cause allergies or lead to a food related
disease, which have been identified as some of the major trends in consumer food demand
in recent years (Asioli et al., 2017). We examine the impact of these labels on the WTP for
organic in the context of two wheat product categories: bread as a staple item, where
organic attribute may appeal to those who by default prefer a gluten-free diet (for other
than health reasons); and cookies as a hedonistic item, where the sugar-free attribute may
appeal to those who by default prefer food labelled as organic.
In summary, we consider these objectives for this study: (1) obtain consumers’
WTP for the organic label alone in two types of organic wheat products—bread and
cookies, (2) examine the impact of combining organic and non-GMO labels on the WTP
for organic products and whether it depends on consumers’ prior knowledge that organic
must be non-GMO, (3) examine the impact of including additional labels perceived by
consumers as healthier (gluten-free, sugar-free or low-carb) on the WTP for organic
products, and (4) evaluate the effect of several factors, including overall tested (objective)
knowledge of organic, self-reported (subjective) familiarity with organic, and wheat/gluten
intolerance/avoidance, on the WTP for the organic label and in combination with other
labels.
3.3 Background and Literature Review
3.3.1 The Complexity of the Organic Label, Consumer Preferences for Components of
Organic and Associations with Other Labels
In the United States, the organic label can be used only on products that are produced
following strict rules developed by the National Organic Program (USDA, 2002). These
117
rules specify the substances, methods and practices that are allowed and prohibited in the
organic farming, including use of genetic engineering, synthetic pesticides and fertilizers,
growth hormones, sewage sludge and irradiation. Thus, the organic label is composed of
multiple characteristics that jointly define organic. The complexity of the organic label has
inspired researchers to examine consumer valuation of the organic label as a whole relative
to its parts, in particular non-GMO.
Some studies find that consumers are not willing to pay significantly more for
organic relative to non-GMO alone (Bernard, Zhang, and Gifford, 2006, using potato chips,
tortilla chips and milk chocolate), and the WTP values for parts of organic are likely not
additive because the sum of WTP for two selected components of organic (non-GMO and
pesticide-free) is found to be significantly not different from WTP value for organic
(Bernard and Bernard, 2010, using potatoes and sweet corn). On the other hand, He and
Bernard (2011) found that products labeled organic elicited significantly higher premiums
on average than products only labeled non-GMO, with the difference in premiums ranging
from 6.80% for tortilla chips to 17.57% for potato chips. In these studies, respondents were
given a neutrally phrased definition of each label, but consumers do not necessarily have
accurate knowledge of labels during a real shopping scenario. In addition, these studies do
not examine the interaction of organic and non-GMO labels, which could be valuable for
those who do not know that by definition organic implies non-GMO.
Conner and Christy (2004) examined the WTP for a bag of organic corn chips that
had a non-GMO label in addition to organic, and found that current consumers of organic
food, who were targeted specifically in this study, are willing to pay on average $0.75 more
118
for a product with both labels relative to the base product with the organic label only, which
cost $1. The respondents were not provided definitions of the labels and only 53% knew
that organic is non-GMO by definition. However, the impact of knowledge on WTP values
has not been discussed in more detail. In a more recent study, McFadden and Lusk (2017)
examined the effect of the interaction of organic and non-GMO labels as well. In contrast
to the findings of the study by Conner and Christy (2004), they found that consumers, not
limited to organic food consumers only, are not willing to pay for non-GMO and organic
labels together significantly more than for each label alone. In fact, they seem to value non-
GMO and organic labels similarly, suggesting the two labels are perceived to be substitutes
rather than complements, but the impact of their prior knowledge of the organic label was
not assessed.
In summary, previous research suggests that consumers care greatly about a few
components of organic and/or they do not have a clear understanding of what organic really
means. As discussed in Stanton and Guion (2015), some segments of the population have
a good understanding of what organic means. But the general population is still confused
about its meaning, which may be exacerbated by companies which, in an attempt to
advertise organic products, include claims that add to the confusion. For example,
consumers in general appear to confuse organic with other labels that are perceived as
sustainable, such as “local” (Risku-Norja and Løes, 2017), or excluding some undesirable
practices during the production process (e.g. no antibiotics or pesticide-free), such as “all-
natural” (Abrams, Meyers, and Irani, 2010) and “natural” (Gifford and Bernard, 2011),
although it is not clear whether they confuse them because they are placed together on a
119
product. However, previous research also suggests that when organic and local labels are
combined together, they may be appealing to specific consumer segments that view the
labels as complements (Hasselbach and Roosen, 2015; Hempel and Hamm, 2016), which
could also be the result of their better understanding of the organic label (Holmes and Yan,
2012). Also, regarding the relationship between organic and natural labels, findings of Lusk
and Briggeman (2009) suggest that “naturalness” is a key value motivating consumer
preferences for organic food. In this study, we examine whether consumers find the
combination of organic and non-GMO labels confusing, or helpful, conditional on their
actual knowledge that organic implies non-GMO.
3.3.2 Organic Food Perceived as Healthier by Consumers
Many studies have looked at consumers’ perceptions of organic foods to provide insights
into why consumers value organics and what motivates their purchase decisions. A
common finding across the studies is that organic food is perceived as healthier and more
nutritious when compared to the conventional alternatives (Magnusson et al., 2001; Lea
and Worsley, 2005; Krystallis, Fotopoulos, and Zotos, 2006; Lee et al., 2013). For example,
Lee et al. (2013) found that consumers evaluated organic versions of processed foods
(cookies, potato chips and yogurt) as more nutritious and having less calories than the
conventional versions, solely based on the organic label, and their WTP for the organic
version was higher. It has been found that people who are concerned about their health tend
to have more positive attitudes toward organic foods (Zepeda, Chang, and Leviten-Reid,
2006), affecting positively their intention to purchase and/or WTP for organic (Gil, Gracia,
120
and Sanchez, 2000; Magnusson et al., 2003; Akgüngör, Miran, and Abay, 2010; Gracia
and de Magistris, 2013; Nasir and Karakaya, 2014).
As shown in the studies above, consumers in general perceive organic products to
be healthier than the conventional alternatives. We aim to explore the impact of including
other labels, perceived by consumers as health promoting, on consumers’ preferences and
WTP for organic, by examining the interactions of the organic label with the other health
claims. The labels we examine are gluten-free, and low-carb in case of bread and sugar-
free in case of cookies. These labels are chosen to represent two different trends driven by
consumers’ health concerns and issues as one of the major tendencies in the food
consumption in the developed countries (Grunert, 2013)—avoidance of some ingredients
due to allergies and intolerances, and due to concerns about or as a result of suffering from
a food related disease (Asioli et al., 2017). We investigate whether these additional labels
in combination with organic are desirable for consumers.
3.3.3 Increasing Importance of Gluten-Free Foods
Gluten-free foods were developed to address needs of people with celiac disease, or
intolerances or allergies to gluten, but over time, more and more consumers became
convinced that gluten-free diet is a healthy lifestyle option and shifted their preferences
towards these types of foods (Potter, Stojceska, and Plunkett, 2014). According to The
Hartman Group’s Health & Wellness report (2015), 26% of consumers in US claim that
they purchase gluten-free foods because they believe they are a “healthier option”, which
was the second most cited reason for purchasing gluten-free foods, following the most
frequently cited “no reason at all” by 35% of respondents, while only 8% stated “I have a
121
gluten sensitivity.” This study also found that in 2015 one in five consumers in the US
claim that they avoid or reduce gluten in their daily diet. The increasing consumer interest
in the foods with gluten-free label is also documented by 36% annual growth in the U.S.
gluten-free foods market between 2010 and 2015 (Berry, 2017).
Given the importance that the gluten-free diet has gained over the past few years,
we aim to examine if there is any connection between preferences for bakery products
labeled as gluten-free and those labeled organic, as both labels are perceived to be healthy,
which to our knowledge has not yet been addressed in the literature. We aim to examine
whether organic wheat products may serve as an alternative to gluten-free bakery products
for consumers who prefer gluten-free products for other than gluten/wheat intolerance
reasons, given that gluten-free bakery products are generally perceived among consumers
to be of lower quality when compared to conventional bakery products (Arendt et al., 2002;
Potter, Stojceska, and Plunkett, 2014).
3.3.4 Increasing Importance of Food with no or Reduced Sugar
Health concerns and issues gave rise to another trend in food consumption, which is
controlling sugar intake and interest in foods with reduced/no sugar labels. Despite mixed
scientific evidence (Stanhope, 2016), many consumers believe that high sugar intake leads
to health issues including diabetes, cardiovascular diseases and obesity. However, as the
number of diabetics in US increased from 25.8 million in 2010 (U.S. Department of Health
and Human Services, Centers for Disease Control and Prevention – HHS CDC, 2011) to
30.3 million in 2015 (HHS CDC, 2017), more and more consumers need to watch closely
their overall carbohydrate intake, including sugars. There is evidence that consumers in the
122
US have decreased their consumption of added sugars (Welsh et al., 2011). And, while
only 30% of US consumers of all ages met the recommendations for daily intake of added
sugars in the period 2007-2010 (HHS and USDA, 2015), this share increased to 42% in the
period 2013-2014 (Bowman et al., 2017), indicating a trend toward decreased consumption
of added sugars. Given the consumers’ perception of organic as health promoting, we are
interested in examining whether consumers value the combination of organic and low-carb
or sugar-free labels, which to our knowledge has not yet been considered in the literature.
3.3.5 Impact of Knowledge of Organic Label on Organic Food Choice
Previous studies have investigated the effect of objective and/or subjective knowledge of
organic food production systems on the consumption of organics.
26
Pieniak, Aertsens, and
Verbeke (2010) and Aertsens et al. (2011) found that subjective knowledge directly affects
the likelihood of consuming organic vegetables. Objective knowledge did not affect the
consumption of organic vegetables directly but contributed significantly to forming
positive attitudes towards organics. Thus, it appears that subjective knowledge has a
stronger impact on the consumption of organic vegetables than objective knowledge.
Similarly, Gracia and de Magistris (2013) found that higher self-reported knowledge of
organic products is associated with an increase in the willingness to purchase organic
products. Further, Gil and Soler (2006) and Mesías Díaz et al. (2012) found positive
relationship between consumers’ tested organic knowledge and their WTP for organic
26
As defined in the discussed studies (Pieniak, Aertsens, and Verbeke, 2010; Aertsens et al., 2011), objective
knowledge is what a consumer actually knows about organic (e.g. it can be evaluated using a test) and
subjective knowledge is what a consumer thinks he/she knows and is influenced by perceptions (e.g. it can
be a self-reported familiarity with organic).
123
products. On the other hand, Li, Zepeda, and Gould (2007) found that the knowledge of
organic production practices does not play a role in organic food purchasing behavior. In
this study, we examine how objective and subjective knowledge of organics affects WTP
for the organic alone and in combination with the other labels. Understanding how
knowledge of organic production systems is related to consumer choice and consumption
is important, since it can provide insights for the marketing of organic food in terms of
whether campaigns aimed at educating consumers about organic can be an effective tool
to increase sales of organic foods or not.
3.3.6 WTP for Organic Wheat Products
In this section, studies examining consumer WTP for products containing organic wheat
are reviewed. For example, average WTP for 1kg whole wheat flour was estimated at 7.5%
over the price of conventional flour in Rodríguez, Lacaze, and Lupín (2007), using a
convenience sample of food shoppers in Argentina. Hempel and Hamm (2016) found that
organic-minded consumers in Germany were willing to pay on average approximately €1
extra for 1kg organic flour, while the non-organic minded consumers were not willing to
pay extra.
Several studies examined WTP for organic bread. Krystallis, Fotopoulos, and Zotos
(2006) extracted a sub-sample of organic food buyers from a sample of primary household
grocery shoppers in Greece and found that their stated WTP for 0.5kg of organic bread was
75.5% above the price of conventional bread. Boxall et al. (2007) used a sample of regular
consumers of wheat bread products in Edmonton, Canada, intercepted at various shopping
outlets and public venues, to elicit WTP a predetermined value for the whole wheat organic
124
bread above the price for conventional bread, either before or after tasting a sample of
organic and conventional bread, while providing information about health or
environmental aspects of organic production, resulting in four treatments. The mean WTP
was 63% above the price of conventional bread, but it differed based on the treatment. They
found that before the tasting occurred, environmental information yielded higher WTP
(86%) than the health information (42%), but after the tasting occurred, health information
yielded higher WTP (79% vs. 43%). This suggests that consumers care about the taste, and
once they confirm it is desirable, health information has a greater effect on WTP values.
Using a sample of primary grocery shoppers in Canada, Zheng (2014) found that
they were willing to pay on average 9.5% above the price of the conventional whole wheat
bread for the certified organic label but were not willing to pay extra for the non-certified
organic label. Teuber, Dolgopolova, and Nordström (2016) used an experimental auction
to elicit students’ and university staff’ bids for 0.5kg whole grain bread with selected
attributes, including organic. The average WTP for organic bread was €0.80 under a blind
tasting scenario, €1.22 when extrinsic quality cues (e.g. info about labels and packaging)
were provided, and €1.19 after both blind tasting and extrinsic quality cues were provided.
Few studies examined the interaction of the organic label combined with other
claims on bread. Bitzios, Fraser, and Haddock-Fraser (2011) examined the interaction of
the organic claim with claims of health benefits and functional ingredients in three
consumer segments, extracted from the stratified sample of UK households. They found
that consumers in all segments were not willing to pay extra for the organic claim alone
and with added claims of health benefit and functional ingredient, only one segment was
125
willing to pay £1 GBP for bread with organic and health benefit claims together.
Hasselbach and Roosen (2015) used a sample of organic food buyers, intercepted at various
shopping outlets in Germany, and found that consumers were not willing to pay extra for
the organic attribute alone in bread, but the estimated premium for organic bread from the
region, considered local, was €4.66.
In summary, the review of studies shows that there is a large variability in consumer
WTP for the organic label on bread. We build on the previous research by examining the
WTP values for the organic attribute alone and when combined with other labels, which
has not been done previously in the US. In contrast to the majority of reviewed studies, we
use a sample of consumers selected from the general population in the U.S. western states,
not limiting the sample to primary household shoppers/organic food buyers/wheat bread
consumers. In addition, the analysis is performed for two different types of wheat products
– bread and cookies, with the goal to examine the impact of a product category on WTP
values (staple vs. hedonistic), since previous studies have found that consumer preferences
for organic attribute depend on the type of product (Ellison et al., 2016).
3.4 Data and Survey Methodology
The data used in the analysis comes from an online survey that was administered in the
U.S. western states in the summer of 2017. The data and survey methodology are described
in more detail in Essay 2.
126
3.5 Model Specification and Methodology
In this essay, the choices of respondents from a set of given alternatives are analyzed to
understand how individual labels and prices contribute to the utility from an alternative and
probability of the choice of an alternative. The choice is a binary variable, equal to 1 when
the alternative is chosen and 0 otherwise, and the analysis is performed using a discrete
choice modeling framework. Within this framework, multinomial logit (MNL) and random
parameter logit (RPL) models are used in this study, as described below.
3.5.1 Multinomial Logit Model
The MNL model is considered a base model within the discrete choice modeling
framework in a situation when the decision maker faces more than two alternatives from
which he/she can choose. The MNL model is useful to identify what factors affect the
decision maker’s level of utility.
The utility of decision maker n from bread/cookies alternative i among j=1,…,J
bread/cookies alternatives, presented in a choice scenario t, 𝑈𝑛𝑖𝑡, is
(3-1) 𝑈𝑛𝑖𝑡 =𝑉𝑛𝑖𝑡 +𝜀𝑛𝑖𝑡 =𝛽𝑋𝑛𝑖𝑡 +𝜀𝑛𝑖𝑡.
Here, 𝑉𝑛𝑖𝑡 is the observed portion of the utility and 𝜀𝑛𝑖𝑡 is unobserved and assumed to be
i.i.d. type 1 extreme value, i.e. uncorrelated across decision makers, alternatives and choice
scenarios. The value of 𝑉𝑛𝑖𝑡 depends on the vector of attributes of the bread/cookies
alternative i presented to decision maker n in a choice scenario t, 𝑋𝑛𝑖𝑡, and the vector of
marginal utilities associated with the attributes, 𝛽, which are assumed to not vary across
respondents within the MNL framework. The goal is to estimate 𝛽 given the observed
attributes and choices made.
127
In this essay, three specifications of the observed utility function are used. In MNL1
model, 𝑋𝑛𝑖𝑡 is a vector of attributes in levels only, which includes dummies for organic,
non-GMO, gluten-free and low-carb (for bread) or sugar-free (for cookies) labels equal to
1 when the label is present and 0 otherwise, alternative specific constant for opt-out
option—none, and price. In MNL2 model, 𝑋𝑛𝑖𝑡 is expanded to include also two-way
interactions of organic label with a) non-GMO, b) gluten-free and c) low-carb (bread) or
sugar-free (cookies) labels equal to 1 when both labels are present and 0 otherwise. The
aim is to examine whether a combination of the organic label with another label has any
impact on the overall level of utility and to gain insights regarding whether the labels are
perceived as complements, substitutes, or independent. If the parameter for the interaction
term of two labels is positive, consumers derive additional utility from both labels that is
beyond the sum of utilities from individual labels, and they are considered complements.
If the interaction term is negative, the labels are considered substitutes and if it is zero, the
labels are independent (McFadden and Lusk, 2017).
In MNL3 model, a three-way interaction of the organic label, the non-GMO label
and knowledge that the organic product should be non-GMO by definition is added to the
vector 𝑋𝑛𝑖𝑡 to examine whether this knowledge has any impact on the utility from the
combination of organic and non-GMO labels.
27
The knowledge is a dummy variable equal
27
This specification omits the main effect of knowledge and two-way interactions of knowledge with non-
GMO and organic labels, respectively. First, the justification for omitting the main effect of knowledge is
that when neither organic nor non-GMO labels are present (i.e. “organic”=0 and “non-GMO”=0), the
knowledge that organic is also non-GMO does not affect the choice. Second, the justification for omitting
the interaction of knowledge with non-GMO label is that when only the non-GMO label is present, the
knowledge that organic is non-GMO does not affect the choice. And third, when only the organic label is
present, the knowledge that organic is non-GMO does not affect the choice. In summary, the knowledge is
assumed to affect the choice only when both organic and non-GMO labels are present.
128
to 1 when the decision maker knows that organic is non-GMO by definition and 0
otherwise. In the context of discrete choice models, only the differences in utility are
identifiable (Train, 2009) and the conventional version of bread/cookies without any labels
(i.e. “organic”=0, “non-GMO”=0, “gluten-free”=0, “low-carb”/”sugar-free”=0, “none”=0)
is set as the reference product. The utility of the reference product is 0, and the utility
associated with any label is interpreted relative to that.
In each choice scenario t, a rational and utility-maximizing decision maker n will
choose the bread/cookie alternative i among j=1,…,J bread/cookie alternatives if and only
if the utility from this alternative, 𝑈𝑛𝑖𝑡, is greater than the utility from all other available
alternatives, 𝑈𝑛𝑗𝑡, for all 𝑗≠𝑖. The algebraic derivation of the logit choice probability of
decision maker n choosing alternative i is described in more detail in Essay 2 and the
resulting formula for the probability is (Train, 2009)
(3-2) 𝑃𝑛𝑖𝑡 =exp(𝑉𝑛𝑖𝑡)
∑exp(𝑉𝑛𝑗𝑡)
𝐽
𝑗=1 =exp(𝛽𝑋𝑛𝑖𝑡)
∑exp(𝛽𝑋𝑛𝑗𝑡)
𝐽
𝑗=1 .
In this essay, each decision maker faces eight choice scenarios within each product
category, making a sequence of choices. Since the unobserved portion of utility is assumed
to be i.i.d. type 1 extreme value, it is uncorrelated over choice scenarios. Thus, the
probability of decision maker n choosing a sequence of alternatives 𝐢 = {𝑖1,…,𝑖𝑇}, where
𝑖𝑡 is an alternative with attributes 𝑋𝑛𝑖𝑡𝑡 chosen in choice scenario t, is a product of logit
choice probabilities over choice scenarios (Train, 2009)
(3-3) 𝑃𝑛𝐢 =∏[ exp(𝑉𝑛𝑖𝑡𝑡)
∑exp(𝑉𝑛𝑗𝑡)
𝐽
𝑗=1 ]=∏[ exp(𝛽𝑋𝑛𝑖𝑡𝑡)
∑exp(𝛽𝑋𝑛𝑗𝑡)
𝐽
𝑗=1 ]
𝑇
𝑡=1
𝑇
𝑡=1 .
129
The vector 𝛽 is estimated by applying maximum likelihood estimation (MLE) procedure
to
(3-4) 𝐿𝐿(𝛽)=∑∑∑𝑦𝑛𝑗𝑡
𝐽
𝑗=1
𝑇
𝑡=1
𝑁
𝑛=1 ln𝑃𝑛𝑗𝑡.
Equation (3-4) represents the log-likelihood function of observing the choices that are
made by 𝑁(=1009) decision makers in the sample, in 𝑇(=8) choice scenarios and from
𝐽(=3) alternatives, where 𝑦𝑛𝑗𝑡 =1 when decision maker n chooses alternative j in choice
scenario t and 𝑦𝑛𝑗𝑡 =0 otherwise. The values of 𝛽 are estimated by maximizing the log-
likelihood function.
The advantage of a MNL model is that it can be easily implemented. However,
there are several assumptions related to this model which limit its application as described
following Train (2009). First, it is required that the independence from irrelevant
alternatives (IIA) property holds, meaning that when a new alternative is introduced to the
choice set, the ratio of the probability of choice of any other two options in the choice set
will not be affected. Second, it is assumed that the preferences of decision makers are
homogeneous, i.e. 𝛽 values are fixed across the population of decision makers. Third, in
case when the decision maker makes a sequence of choices, MNL model assumes that there
is no correlation between the unobserved factors that affect the choices made. The
Hausman and McFadden (1984) test of IIA property is performed to determine whether the
MNL model is valid, or whether the RPL model is more appropriate.
130
3.5.2 Random Parameter Logit Model
If the assumptions of a MNL model are found to be too restrictive, RPL model, also known
as mixed logit model, can be implemented. In contrast to MNL model, RPL model does
not require that the IIA assumption holds, it allows consumer preferences for product
attributes to vary randomly across decision makers, and it can accommodate the correlation
in the unobserved factors over the sequence of choices made by the same decision maker
and over the choice alternatives in a choice scenario (Train, 2009).
In case of a RPL model, the vector of preferences for product attributes, i.e. taste
parameters, is individual-specific and 𝛽𝑛 instead of 𝛽 is used. However, the vector 𝛽𝑛 may
also contain coefficients that are fixed. The distribution 𝑓(𝛽𝑛|𝛩) represents the density of
𝛽𝑛 in the population, with 𝛩 representing the parameters of the density that are to be
estimated. The researcher specifies whether preferences for each attribute are random or
fixed, the type of the distribution for each random parameter, and whether the preferences
for individual attributes are correlated or not, which affects the number of parameters to be
estimated and the overall complexity of the model. Preferences for some attributes can be
specified as fixed either due to practical reasons (e.g. for the purposes of deriving WTP
distribution as described later), or when it is found to be appropriate (e.g. when the
estimated standard deviation is statistically insignificant as described later). The choice of
the distribution for each parameter specified as random depends on the assumption of a
reasonable range of the utility associated with the attribute in the population, as well as
whether it is strictly positive or negative. Normal, triangular, uniform and log-normal
distributions are commonly used (Hensher and Greene, 2003).
131
In the context of a RPL model, the probability that decision maker n makes a
sequence of observed choices 𝐢 = {𝑖1,…,𝑖𝑇} is (Revelt and Train, 1998; Train, 2009)
(3-5) 𝑃𝑛𝐢 =∫∏( exp(𝑉𝑛𝑖𝑡𝑡)
∑exp(𝑉𝑛𝑗𝑡)
𝐽
𝑗=1 )𝑓(𝛽𝑛|𝛩)𝑑𝛽𝑛
𝑇
𝑡=1
=∫∏( exp(𝛽𝑛𝑋𝑛𝑖𝑡𝑡)
∑exp(𝛽𝑛𝑋𝑛𝑗𝑡)
𝐽
𝑗=1 )𝑓(𝛽𝑛|𝛩)𝑑𝛽𝑛
𝑇
𝑡=1 .
Here, 𝛽𝑛 is a vector of both fixed and random parameters, distributed as 𝑓(𝛽𝑛|𝛩), thus the
RPL choice probability is the integral of the MNL choice probability over all possible 𝛽𝑛
values. Here, the main interest is to estimate the set of parameters 𝛩, which describe the
distribution of individual taste parameters 𝛽𝑛 in the population (Revelt and Train, 1998).
The integral in equation (3-5) does not have a closed form solution. Thus, in case
of the RPL model, a simulated maximum likelihood estimation (SMLE) procedure is used
instead of the MLE procedure outlined in equation (3-4) for the MNL model. For each
decision maker, SMLE starts with a draw of many values for each 𝛽𝑛 value from its
assumed distribution and each value is used to evaluate the logit choice probability in
equation (3-3). The average of these probabilities is obtained for each decision maker, and
these simulated averages are then used to construct the simulated log-likelihood function
of the observed choices for all decision makers in the sample (Revelt and Train, 1998;
Train, 2009).
Using the RPL framework, models RPL1, RPL2, and RPL3 are estimated following
the specification of MNL models described previously but allowing for random
132
preferences. However, the decision needs to be made whether each individual parameter
should be random or fixed, and for a random parameter, what distribution it should follow.
Initially, starting with RPL1 model, it is assumed that preferences for each attribute
vary in the population following a normal distribution, except price which is specified to
be fixed, following previous literature (e.g. Revelt and Train, 1998; Lusk and Schroeder,
2004; Van Loo et al., 2011; Janssen and Hamm, 2012; Hasselbach and Roosen, 2015).
28
The normal distribution is chosen under the assumption that decision makers can value
each label either positively or negatively. In the next step, parameters with insignificant
standard deviations are re-specified as fixed and the model is re-estimated (Barreiro‐Hurle,
Gracia, and de Magistris, 2010; Janssen and Hamm, 2012). Ultimately, the Bayesian
Information Criterion (BIC) is used to select the best model within the RPL1 specification.
In the next step, two-way interaction terms are added to this model to obtain the starting
RPL2 model. Again, the preferences for the interactions of labels are assumed to be random
and normally distributed, and different specifications for the parameters with insignificant
standard deviations (random vs. fixed) are tested. Finally, the same procedure is applied to
determine the specification of RPL3 model. In this essay, it is also assumed that the
preferences for attributes are uncorrelated. The set of distribution parameters 𝛩 to be
estimated contains a vector of mean values of taste parameters, 𝛽, and a variance-
covariance matrix, 𝛴. In 𝛴, the diagonal terms represent variance 𝜎2 of each taste parameter
indicating how preferences vary in the population. The off-diagonal terms equal 0,
28
For example, Revelt and Train (1998) state that the specification of fixed price parameter “allows easy
derivation of the distribution of the WTP.”
133
assuming that the preferences for attributes are uncorrelated. During the simulation
process, 2000 Halton sequences are used to draw the 𝛽𝑛 values.
3.5.3 Willingness to Pay
After the estimation of MNL and RPL models, we use the Hausman and McFadden (1984)
test of the IIA property and model selection criteria (Akaike Information Criterion – AIC,
Bayesian Information Criterion – BIC, and log-likelihood) to determine the final model.
The mean coefficients of the selected final model are then used to calculate mean WTP
values. For example, the mean WTP value for the organic label is calculated as the negative
ratio of the parameter estimate for the organic label, 𝛽𝑜𝑟𝑔𝑎𝑛𝑖𝑐, to the price parameter
estimate, 𝛽𝑝𝑟𝑖𝑐𝑒,
(3-6) 𝑊𝑇𝑃𝑀𝑒𝑎𝑛,𝑜𝑟𝑔𝑎𝑛𝑖𝑐 =−𝛽𝑜𝑟𝑔𝑎𝑛𝑖𝑐
𝛽𝑝𝑟𝑖𝑐𝑒 .
To determine the significance of the calculated mean WTP values, we use the Delta method
(Oehlert, 1992).
In the case of MNL models, the preferences are fixed and the value of 𝛽 is the same
for every decision maker. In the case of RPL models, the preferences for labels may vary
over decision makers, following a normal distribution with mean 𝛽 and standard deviation
𝜎 as discussed in the previous section. As a result, the WTP values for these labels vary as
well and the distribution of WTP values can be derived. For example, assuming that the
preferences for the organic label follow a normal distribution with standard deviation
𝜎𝑜𝑟𝑔𝑎𝑛𝑖𝑐 and the price coefficient 𝛽𝑝𝑟𝑖𝑐𝑒 is fixed, the standard deviation of the WTP values
for the organic label is calculated as (Hensher and Greene, 2003)
134
(3-7) 𝑊𝑇𝑃𝑆𝐷.,𝑜𝑟𝑔𝑎𝑛𝑖𝑐 =−𝜎𝑜𝑟𝑔𝑎𝑛𝑖𝑐
𝛽𝑝𝑟𝑖𝑐𝑒 .
This convenience in finding the WTP distribution is the reason why the price coefficient
𝛽𝑝𝑟𝑖𝑐𝑒 is set as fixed.
29
We are also interested in understanding what factors affect consumers’ mean WTP
values for the labels alone and in combination. The examined factors include overall level
of organic knowledge and familiarity, wheat and/or gluten intolerance and/or avoidance,
and socio-demographics. The final model is estimated for each subgroup of decision
makers formed based on the above factors, and mean WTP values are compared across
these subgroups. For example, to examine the effect of gender on mean WTP values, the
model is estimated separately for males and females, and then mean WTP values for each
gender group are compared. The reason for not including these factors in the utility function
directly is to avoid issues associated with the estimation of a large number of parameters.
29
Alternatively, the price coefficient can be set as random following log-normal distribution, but past studies
find that in this case the parameters of WTP distribution can become unrealistically high (Hess, Bierlaire,
and Polak, 2005; Hole and Kolstad, 2012). A relatively new approach is to estimate models in “WTP space”
(Train and Weeks, 2005), where the models are re-specified in a way that WTP values are estimated directly,
allowing the assumptions to be made about the distribution of WTP values and not restricting the price
coefficient to be fixed. It appears that models estimated in the WTP space yield more reasonable WTP
distributions, but on the other hand, estimates obtained in the preference space tend to have a better fit (Train
and Weeks, 2005). In addition, Hole and Kolstad (2012) find that mean WTP values derived from simpler
models in the preference space (that is, MNL and RPL models with uncorrelated coefficients and fixed price)
are similar to the mean WTP values estimated directly in WTP space using more complicated specification
(all coefficients are random and uncorrelated/some correlated). Nevertheless, estimation in WTP space
approach has not been applied widely in the past literature. In this essay, an attempt was made to apply this
approach; however, convergence issues were encountered during estimation.
135
3.6 Results
Results from the estimation of the MNL and RPL models for bread and cookies are reported
in tables 3-1 and 3-2, respectively. Within the group of RPL models, in RPL1 models all
standard deviations for labels alone are statistically significant. Thus, no other
specifications within RPL1 models were tested. In case of the RPL2 and RPL3 models,
some standard deviations of the interaction terms are found to be insignificant, which
indicates fixed preferences. Tables B-1 and B-2 in the Appendix B report all models that
were considered within the RPL2 and RPL3 specifications for bread and cookies,
respectively. BIC was used to select the best models.
Next, results from the estimation of MNL and RPL models in tables 3-1 and 3-2
are compared. The Hausman-McFadden test statistics and values of log-likelihood, AIC,
and BIC across MNL and RPL models within each product category show clearly that RPL
specifications are better. In addition, significant and large standard deviations of some
coefficients imply that some preferences are indeed heterogeneous and not homogeneous,
as assumed by MNL. Also, MNL and RPL models yield different coefficient estimates.
Allowing the preferences to be random results in coefficients that are larger in magnitude,
and in some cases the magnitude is more than double. For some coefficients, the
significance is affected as well. These differences in estimates indicate that in this essay
the model specification matters, and we focus on RPL models next.
136
Table 3-1. MNL and RPL Models, Bread
MNL1
MNL2
MNL3
RPL1
RPL2
RPL3
Means
Price
-0.46***
(0.02)
-0.45***
(0.02)
-0.45***
(0.02)
-1.03***
(0.03)
-1.03***
(0.04)
-1.03***
(0.04)
None
-1.76***
(0.09)
-1.69***
(0.09)
-1.70***
(0.09)
-5.06***
(0.23)
-4.90***
(0.23)
-4.89***
(0.23)
Organic
0.14***
(0.05)
0.21***
(0.07)
0.21***
(0.07)
0.33***
(0.10)
0.55***
(0.14)
0.55***
(0.14)
Gluten-free
-0.01
(0.03)
-0.08*
(0.04)
-0.07
(0.04)
-0.13*
(0.07)
-0.20**
(0.09)
-0.20**
(0.09)
Low-carb
0.04
(0.03)
0.02
(0.04)
0.02
(0.04)
0.09
(0.06)
0.10
(0.08)
0.09
(0.08)
Non-GMO
0.17***
(0.03)
0.31***
(0.04)
0.31***
(0.04)
0.33***
(0.06)
0.58***
(0.08)
0.58***
(0.08)
Gluten-free×Organic
-
0.14***
(0.05)
0.14*
(0.05)
-
0.13
(0.11)
0.13
(0.11)
Low-carb×Organic
-
0.02
(0.05)
0.02
(0.05)
-
-0.05
(0.11)
-0.05
(0.10)
Non-GMO×Organic
-
-0.32***
(0.05)
-0.51***
(0.07)
-
-0.60***
(0.12)
-0.78***
(0.14)
Non-GMO×Organic×
Knowledge1
-
-
0.45***
(0.11)
-
-
0.42**
(0.17)
Standard deviations
None
-
-
-
4.54***
(0.22)
4.54***
(0.22)
4.54***
(0.23)
Organic
-
-
-
2.52***
(0.11)
2.57***
(0.12)
2.55***
(0.12)
Gluten-free
-
-
-
1.37***
(0.10)
1.40***
(0.10)
1.40***
(0.10)
Low-carb
-
-
-
0.69***
(0.11)
0.75***
(0.11)
0.74***
(0.11)
Non-GMO
-
-
-
0.71***
(0.11)
0.66***
(0.13)
0.65***
(0.13)
Non-GMO×Organic
-
-
-
-
0.75***
(0.24)
0.76***
(0.24)
Log-Likelihood
-8,160.8
-8,148.3
-8,129.6
-5,603.0
-5,588.5
-5,585.3
AIC
16,333.6
16,314.6
16,279.3
11,228.0
11,206.9
11,202.6
BIC
16,382.2
16,387.5
16,360.2
11,317.0
11,328.4
11,332.1
IIA test 1 statistic2
23.2***
21.8***
22.3***
-
-
-
IIA test 2 statistic2
48.8***
46.8***
49.2***
-
-
-
Note: *, **, *** denote significance of the coefficients at 10%, 5% and 1% level, respectively. Standard
errors are in the parentheses. The number of observations used in the estimation of each model is 24,216.
1 Knowledge = 1 if respondent knows that organic is also non-GMO and 0 otherwise.
2 Hausman-McFadden test of IIA property: Test 1 statistic is obtained after excluding “organic” alternative.
Test 2 statistic is obtained after excluding “none” alternative. H0: Difference in coefficients not systematic.
137
Table 3-2. MNL and RPL Models, Cookies
MNL1
MNL2
MNL3
RPL1
RPL2
RPL3
Means
Price
-0.44***
(0.02)
-0.44***
(0.02)
-0.44***
(0.02)
-1.06***
(0.04)
-1.08***
(0.04)
-1.08***
(0.04)
None
-2.17***
(0.11)
-2.15***
(0.11)
-2.16***
(0.11)
-6.16***
(0.25)
-6.21***
(0.27)
-6.22***
(0.27)
Organic
-0.01
(0.05)
0.02
(0.07)
0.02
(0.07)
-0.06
(0.10)
0.11
(0.13)
0.10
(0.13)
Gluten-free
-0.08**
(0.03)
-0.08*
(0.04)
-0.08*
(0.04)
-0.28***
(0.08)
-0.25***
(0.10)
-0.25***
(0.10)
Sugar-free
-0.31***
(0.04)
-0.29***
(0.04)
-0.29***
(0.04)
-0.85***
(0.10)
-0.77***
(0.11)
-0.77***
(0.11)
Non-GMO
0.04
(0.03)
0.04
(0.04)
0.04
(0.04)
0.12*
(0.06)
0.18**
(0.08)
0.18**
(0.08)
Gluten-free×Organic
-
-0.01
(0.06)
-0.01
(0.06)
-
-0.12
(0.12)
-0.12
(0.12)
Sugar-free×Organic
-
-0.05
(0.06)
-0.05
(0.06)
-
-0.41***
(0.15)
-0.41***
(0.15)
Non-GMO×Organic
-
0.00
(0.05)
-0.10
(0.07)
-
-0.10
(0.12)
-0.15
(0.14)
Non-GMO×Organic×
Knowledge1
-
-
0.24**
(0.11)
-
-
0.12
(0.17)
Standard deviations
None
-
-
-
4.68***
(0.22)
4.69***
(0.21)
4.69***
(0.21)
Organic
-
-
-
2.27***
(0.11)
2.24***
(0.12)
2.24***
(0.12)
Gluten-free
-
-
-
1.43***
(0.11)
1.49***
(0.11)
1.49***
(0.11)
Sugar-free
-
-
-
1.99***
(0.12)
1.98***
(0.12)
1.98***
(0.12)
Non-GMO
-
-
-
0.68***
(0.12)
0.67***
(0.12)
0.68***
(0.12)
Sugar-free×Organic
-
-
-
-
1.36***
(0.24)
1.37***
(0.24)
Log-Likelihood
-8,097.7
-8,097.5
-8,093.0
-5,489.2
-5,482.3
-5,482.0
AIC
16,207.4
16,212.9
16,205.9
11,000.4
10,994.5
10,996.0
BIC
16,256.0
16,285.8
16,286.9
11,089.5
11,115.9
11,125.6
IIA test 1 statistic2
113.8***
579.1***
586.8***
-
-
-
IIA test 2 statistic2
90.4***
98.0***
97.8***
-
-
-
Note: *, **, *** denote significance of the coefficients at 10%, 5% and 1% level, respectively. Standard
errors are in the parentheses. The number of observations used in the estimation of each model is 24,216.
1 Knowledge = 1 if respondent knows that organic is also non-GMO and 0 otherwise.
2 Hausman-McFadden test of IIA property: Test 1 statistic is obtained after excluding “organic” alternative.
Test 2 statistic is obtained after excluding “none” alternative. H0: Difference in coefficients not systematic.
138
Within RPL models for both bread and cookies, the addition of the two-way
interactions of labels affects the values of some coefficients in RPL2 models relative to
RPL1 models, while the addition of the three-way interaction in RPL3 models affects the
coefficients only marginally relative to RPL2 models. Given our interest to examine how
mean WTP values for labels alone and the interactions of labels vary over consumer
subgroups and considering the marginal changes in coefficient estimates going from RPL2
to RPL3 model, the RPL3 model is chosen as the final model. Although RPL3 model does
not have the lowest BIC, the change in BIC from RPL2 to RPL3 model is marginal and
RPL3 model has the lowest AIC and/or the highest log-likelihood. The discussion of results
from RPL3 models follows next, and it is supplemented with means and standard
deviations of WTP values, which are reported in table 3-3. The mean WTP values were
calculated based on the equation (3-6) and the standard deviations (SD) based on the
equation (3-7).
Table 3-3. WTP Distributions Derived from the Results of RPL3 Models
Labels/Label Interactions
Bread
Cookies
Mean
SD
Mean
SD
Organic
$0.54
$2.49
$0.10a
$2.07
Gluten-free
-$0.19
$1.36
-$0.23
$1.38
Low-carb/Sugar-free
$0.09a
$0.73
-$0.71
$1.83
Non-GMO
$0.56
$0.64
$0.17
$0.63
Gluten-free×Organic
$0.13a
-
-$0.11a
-
Low-carb×Organic/Sugar-free×Organic
-$0.05a
-
-$0.38
$1.26
Non-GMO×Organic
-$0.76
$0.74
-$0.14a
-
Non-GMO×Organic×Knowledge
$0.41
-
$0.11a
-
Note: a Values are not statistically significant at 10% or better.
139
3.6.1 Consumer Preferences for Individual Labels
First, the utility function coefficients and WTP values for individual labels alone are
examined. On average, consumers value the organic label significantly positively in bread,
but for cookies the positive coefficient is insignificant. Estimated mean WTP for the
organic label on bread is $0.54 and $0.10
30
for cookies. This result supports previous
findings of Van Doorn and Verhoef (2011) that consumers in general find the organic label
less attractive in hedonistic food items. In both bread and cookies, the non-GMO label
alone is valued on average positively and gluten-free label negatively. However, these
labels are valued more positively and less negatively for bread. Mean WTP for the non-
GMO label is $0.56 and $0.17 and for the gluten-free label -$0.19 and -$0.23 for bread and
cookies, respectively.
Interestingly, in the case of bread, mean WTP for the non-GMO label is close to
the WTP for the organic label. This suggests that on average, consumers either do not have
a good understanding of what constitutes organic, or they care mostly about the non-GMO
component of organic. The latter may be related to concerns about possible contamination
of organic products with GMOs. In that case, consumers may choose the non-GMO label
over the organic label, as it requires testing for possible contamination, while the organic
label ensures that the GMOs are not used in the production process only. For cookies, it is
even more interesting to find that consumers value the non-GMO attribute positively, yet
they are not willing to pay extra for the organic label, although it covers non-GMO and
more. Again, it might be because they are not aware that organic must be non-GMO, but
30
In cookies, the estimated WTP is not statistically significant.
140
also it might be that they associate organic with a healthy diet, which does not correspond
well with the hedonistic nature of cookies. It appears that consumers like the idea of a safer
product, but perhaps they feel that organic label negatively affects some of the features
they expect from cookies, e.g. enjoyable taste, relative to the non-GMO label.
Furthermore, mean estimate for the low-carb label parameter in bread is
insignificant, while the sugar-free label parameter in cookies is negative and significant.
Mean WTP for low-carb label in bread is $0.09 and insignificant, but consumers need a
discount of $0.71 for a sugar-free label in cookies. There is likely a connection between
generally lower consumer preferences for organic and sugar-free labels in cookies as a
hedonistic product, relative to bread. However, the standard deviations of the parameters
and WTP values for all labels alone are significant and large for both bread and cookies,
meaning that there are some consumers who value each of these labels alone
positively/negatively and may have a positive/negative WTP for each one.
3.6.2 Consumer Preferences for Combinations of Labels
Next, we examine the effect of combining the organic label with the remaining labels on
the overall utility and WTP values. The overall utility/WTP value for a combination of two
labels is calculated as the sum of the marginal utilities/WTP values for the labels alone and
their interaction. First, we find that the interactions of organic and gluten-free labels on
bread and cookies, organic and low-carb labels on bread, and organic and non-GMO labels
on cookies are all insignificant, which means that these label combinations do not increase
nor decrease the overall consumer utility/WTP value beyond the sum of the utilities/WTP
values for these labels alone and the labels are perceived as independent. In addition, in
141
these cases the standard deviations of the interaction coefficients are not different from
zero (not included in the reported final models), which means that consumers have similar
preferences regarding these label interactions.
On the other hand, for bread, the interaction of organic and non-GMO label is
negative on average. Thus, when both organic and non-GMO labels are provided on bread,
the overall utility is lower than the sum of the utilities from the labels alone and these labels
are perceived as substitutes. In terms of WTP, overall mean WTP for the combination of
both labels on bread is $0.34 (= $0.54 + $0.56 - $0.76), which is lower than the mean WTP
for each label alone ($0.54 for organic and $0.56 for non-GMO). This result is in line with
the findings of McFadden and Lusk (2017). However, the standard deviation of the
interaction term is significant and of a similar magnitude as the mean of the interaction
term. Given the properties of normal distribution, this means that there is a segment of
consumers who derive additional positive utility beyond the sum of the individual utilities
from these labels alone and they are willing to pay more than the sum of WTP for the
individual labels. These consumers view organic and non-GMO labels as complements.
Further, we find that the specific knowledge that organic must be non-GMO
reduces the negative impact of combining these two labels on the overall utility and WTP.
Total mean WTP for the combination of organic and non-GMO labels, conditional on the
knowledge that organic is non-GMO, increases to $0.75 (= $0.34 + $0.41) and becomes
greater than the WTP for each label alone, but smaller than the sum of WTP for each label.
The moderating effect of knowledge is consistent across respondents since the standard
deviation is insignificant (not included in RPL3 models). On the other hand, for cookies,
142
the interaction of the organic and non-GMO labels does not have any (negative or positive)
effect on the overall utility and WTP, and the knowledge that organic must be non-GMO
does not play a role either. In summary, consumer preferences for the combination of
organic and non-GMO label depend on the product category, and it may or may not depend
on the knowledge that organic is also non-GMO.
As mentioned previously, consumers on average dislike the sugar-free label on
cookies, and in combination with the organic label the overall utility and WTP decrease
even more. When the two labels are combined, consumers require a discount of $1.09 (=
$0.71 + $0.38). However, the significant and large standard deviation of the interaction
term implies that there is a heterogeneity in preferences for the combination of these two
labels. Again, given the properties of the normal distribution, there exists a segment of
consumers who value this combination beyond the sum of the utilities and WTP values
associated with these labels alone.
3.6.3 Impact of Objective and Subjective Knowledge of Organic Label on WTP
RPL3 models have been estimated for subgroups of consumers, formed based on different
criteria to examine the impact of various factors on the mean WTP values for the labels
alone and combinations of labels. In this section, the impact of tested knowledge of organic
label (objective knowledge) and self-reported familiarity with organic label (subjective
knowledge) are evaluated and compared. To test the knowledge of respondents, they were
asked six questions related to organic food production and products (including whether
organic production allows for the use of GMOs). Respondents with zero to one correct
answer were assigned to the group of “low” knowledge (N=356), respondents with two to
143
three correct answers were assigned to the group of “medium” knowledge (N=490), and
respondents with four correct answers or more were assigned to the group of “high”
knowledge (N=163). Respondents were also asked about their familiarity with organic food
production standards, and they were divided into groups of “familiar” (N=288), “not
familiar” (N=493) and “unsure” (N=228). Overall, despite the mainstream nature of
organic foods, the share of those who have a relatively good understanding of or consider
themselves familiar with organics is relatively low at 16% and 23%, respectively.
Mean WTP values for labels alone and interactions of labels along with the 95%
confidence intervals per each consumer subgroup are plotted in figure 3-1. Consumers with
a medium and high knowledge of organic production standards are willing to pay more for
the organic label on bread, while those with low knowledge are not. For cookies, only those
with medium knowledge are willing to pay more for the organic label. This suggests that
objective knowledge plays a role in consumer interest in the organic label only for specific
products. It is different for subjective knowledge. Those who think that they are familiar
with organic production standards are willing to pay more for the organic label in both
bread and cookies. Thus, it appears that the link between the subjective knowledge of
organic and the interest in organic is stronger than in case of the objective knowledge. This
result supports findings of Pieniak, Aertsens, and Verbeke (2010) and Aertsens et al. (2011)
that subjective knowledge affects the likelihood of consuming organic vegetables more
than objective knowledge. It is possible that those who already purchase organic food
consider themselves familiar with organic and are willing to pay more, but they do not
necessarily have high factual knowledge of organics.
144
a) Objective knowledge
b) Subjective knowledge
Figure 3-1. Mean WTP values and 95% confidence intervals per consumer
subgroup based on a) objective knowledge and b) subjective knowledge
Note: “ORG” = organic, “GF” = gluten-free, “LC/SF” = low-carb (bread) or sugar-free (cookies), “NG” =
non-GMO. Numbers are reported in tables B-3 and B-4 in Appendix B.
145
Finally, those with low knowledge and those who claim that they are unfamiliar with
organics are not willing to pay extra for organic bread nor cookies, indicating that lack of
organic knowledge, either objective or subjective, limits interest in organics.
The WTP for the non-GMO label is positive within all subgroups for bread, but for
cookies only within the group of those with medium knowledge and those who claim they
are familiar with organics. Comparison of mean point estimates of WTP values shows that
only those with low knowledge and those unfamiliar with organics are willing to pay more
for the non-GMO label than the organic label. This finding suggests that both objective and
subjective knowledge affect whether organic and non-GMO labels are confused, as
expected.
Next, the effects of label interactions on the overall mean WTP (i.e. WTP for the
combination of two labels) are examined across groups, with focus on those that are
significantly different from zero. Only those who are not familiar with organics have
positive mean WTP for the interaction of organic and gluten-free labels on bread equal to
$0.33, but their mean WTP for these labels alone is $0 and -$0.31, respectively, resulting
in overall mean WTP for both labels equal to $0.02. The interaction of organic and sugar-
free labels on cookies is negative and significant within the group of consumers with low
knowledge (-$0.54) and those who are unfamiliar (-$0.51) or unsure (-$0.56) about
organic, resulting in overall negative WTP for the combination of these labels, given also
their zero WTP for the organic label and negative WTP for the sugar-free label alone.
Further, in bread only, the coefficient for the interaction of organic and non-GMO
labels, ranging from -$1.06 to -$0.46 within five of the six groups identified based on
146
perceived and tested knowledge, reduces the overall mean WTP for both labels, except for
those with high objective knowledge of organics. Considering these five groups, the
negative interaction coefficient reduces the WTP for the combination of these labels below
the WTP for organic alone, except for those who claim that they are familiar with organics.
They see some value in having both labels together (+$0.30 above their mean WTP for the
organic label alone) regardless of their knowledge that organic is non-GMO. Further, the
negative interaction effect is partially offset within the group of those with medium
knowledge and those who are unsure about organic if they know that organic is non-GMO,
and their overall WTP for both labels exceeds their WTP for the organic label alone by
$0.22 and $0.28, respectively. This discussion illustrates the complexity of the issue of
evaluating consumer WTP for combination of organic and non-GMO labels, and how
knowledge that organic is also non-GMO can affect overall WTP positively.
Finally, the interaction terms for the remaining subgroups for bread and cookies,
which have not been discussed, are not statistically different from zero—the labels are
evaluated independently from each other. However, the mean WTP for all labels alone
(besides the organic label) is either negative or zero for the majority of the subgroups,
which means that the overall WTP for the combination of the organic label with these
labels is either lower than or equal to the WTP for the organic label alone. Only those
familiar with organic and those with medium knowledge of organic are WTP extra for low-
carb on bread and non-GMO on cookies, resulting in overall mean WTP for a product with
both labels greater than WTP for a product with organic label only.
147
3.6.4 Impact of Wheat and/or Gluten Intolerance and/or Avoidance on WTP
In this section, the impact of wheat and/or gluten intolerance and/or avoidance (WGIA) on
mean WTP values for labels alone and their combinations is examined. Respondents were
asked to indicate whether they have any wheat or gluten intolerance and whether they avoid
wheat or gluten for any other reasons. They were split into two groups – one group made
of those with WGIA (N=230) and another group made of those with no WGIA, i.e.
NWGIA (N=779). Mean WTP values for labels alone and their combinations, along with
95% confidence intervals, are plotted in figure 3-2.
Figure 3-2. Mean WTP values and 95% confidence intervals per consumer
subgroup based on wheat/gluten intolerance/avoidance
Note: “ORG” = organic, “GF” = gluten-free, “LC/SF” = low-carb (bread) or sugar-free (cookies), “NG” =
non-GMO. Numbers are reported in table B-5 in Appendix B.
148
Focusing on the individual labels first, mean WTP for organic and non-GMO labels
is positive in WGIA group for both products, but in NWGIA group for bread only. Also,
the WGIA have significantly higher mean WTP for both labels alone than the NWGIA. As
expected, the WGIA are also willing to pay extra for the gluten-free label in both bread and
cookies, while the NWGIA require a discount if the label is present on any of the examined
products. Similarly, the NWGIA dislike the sugar-free label on cookies, while the mean
WTP of the WGIA is not different from zero.
Looking further at the label interactions, the overall mean WTP for organic and
low-carb labels in bread increases when these labels are combined for the WGIA (+$0.74),
while the effect is zero for the NWGIA. For cookies, the combination of organic and sugar-
free labels does not affect overall WTP for the WGIA, but the effect on WTP is negative
for the NWGIA (-$0.47).
Regarding the interaction of organic and non-GMO labels, for WGIA the effect on
overall WTP is negative for bread (-$1.04), but overall mean WTP for the combination of
these labels is higher than the mean WTP for each label alone, and the knowledge that
organic should be non-GMO does not play any role. For NWGIA, the effect of combining
both labels is also negative (-$0.72) but it results in overall mean WTP below their WTP
for each of these labels alone; however, not if they have the knowledge that organic is non-
GMO (+$0.49 increase in mean WTP). In summary, the WGIA are on average willing to
pay more than the NWGIA for the combination of organic and non-GMO on bread ($1.72
vs. $0.09 without knowledge and $0.59 with knowledge).
149
The interactions of organic and non-GMO labels on cookies, and organic and
gluten-free labels on both bread and cookies, are not different from zero for both WGIA
and NWGIA. While for WGIA the overall mean WTP for the combination of these labels
is greater than WTP for the organic label alone (positive WTP for each label alone), for
NWGIA it is either lower or does not change (the WTP values for gluten-free labels are
negative and for non-GMO zero).
The discussed findings illustrate that there are some major differences in the
valuation of individual labels and their combinations between the WGIA and NWGIA. The
WGIA have overall higher WTP for examined labels alone including organic and they
respond differently to the combinations of labels than the NWGIA. It is also found that
64% and 49% of the WGIA purchased organic bread and cookies in the past month,
respectively, while only 33% and 17% of the NWGIA purchased organic versions of these
products. This further shows that organic wheat products appeal particularly to the WGIA
and based on their estimated WTP values they represent a profitable market segment.
3.6.5 Impact of Socio-Demographic Variables
In this section, mean WTP values are compared across groups created based on selected
demographic (gender, age) and socio-economic (income, education, residency in
California) variables. The most relevant findings are highlighted with the focus on the
differences between the groups in terms of WTP for the organic label alone relative to the
non-GMO label, and combinations of labels. Mean WTP values and 95% confidence
intervals for demographic and socioeconomic groups are plotted in figures 3-3 and 3-4,
respectively.
150
a) Gender
b) Age
Figure 3-3. Mean WTP values and 95% confidence intervals per consumer
subgroup based on a) gender and b) age
Note: “ORG” = organic, “GF” = gluten-free, “LC/SF” = low-carb (bread) or sugar-free (cookies), “NG” =
non-GMO. Numbers are reported in tables B-6 and B-7 in Appendix B.
151
a) Income
b) Education
Figure 3-4. Mean WTP values and 95% confidence intervals per consumer
subgroup based on a) income, b) education and c) residency in California
Note: “ORG” = organic, “GF” = gluten-free, “LC/SF” = low-carb (bread) or sugar-free (cookies), “NG” =
non-GMO. Numbers are reported in tables B-8, B-9, and B-10 in Appendix B.
152
c) Residency in California
Figure 3-4 (continued). Mean WTP values and 95% confidence intervals per
consumer subgroup based on a) income, b) education and c) residency in California
Note: “ORG” = organic, “GF” = gluten-free, “LC/SF” = low-carb (bread) or sugar-free (cookies), “NG” =
non-GMO. Numbers are reported in tables B-8, B-9, and B-10 in Appendix B.
In terms of age, consumers in group 25-44 have the highest mean WTP for the
organic label in both bread and cookies, which is consistent with findings in the literature
that WTP for organics tends to decrease with age (Hughner et al., 2007), although others
suggest that age does not play any role (He and Bernard, 2011). Further, it is found that
household income above $90,000 and a bachelor’s degree or higher are associated with
positive mean WTP for the organic label on bread and cookies. This finding is in line with
the findings of previous studies that consumers with higher income and/or education are
willing to pay more for organics (Govindasamy and Italia, 1999; Strzok and Huffman,
2015). Also, residents of California are on average willing to pay more than non-residents
153
for organic bread but not for organic cookies. Thus, California may be an attractive market
due to its size and relatively high consumer WTP for some organic products, but not
necessarily all.
On the other hand, for bread, less than $90,000 income and less than a bachelor’s
degree are associated with positive and higher mean WTP for non-GMO relative to WTP
for organic, which is insignificant. For cookies, in these groups the mean WTP estimate
tends to be higher for non-GMO than organic as well, but the values are not statistically
significant. Those who reside outside of California are willing to pay significantly more
for non-GMO than organic label on both products. In summary, those with lower income
and education and those outside of California tend to prefer non-GMO label over organic.
Among all socio-demographic groups, the interaction of organic and non-GMO
labels in bread decreases the overall WTP for these labels together below the mean WTP
for organic label alone, except for those in age group 25-44, but even for this group the
overall mean WTP is close to the WTP for organic alone. However, if the knowledge that
organic is also non-GMO is accounted for, the mean WTP for the combination of labels on
bread increases above the WTP value for organic alone for those in age group 45-64,
females, California residents, and those with income below $60,000. For cookies, this
knowledge increases mean WTP for the combination of organic and non-GMO labels from
zero to a positive value only in the group with income under $30,000. In summary, the
knowledge that organic should be non-GMO plays a role in overall WTP for these labels
together, but it depends on the socio-demographics.
154
3.7 Conclusions
While many studies have examined consumer WTP for organic labels on a variety of food
products, studies that examine the interaction of organic labels with other labels are scarce.
In this essay, the impact of additional labels, when combined with an organic label, on
overall WTP is investigated. The analysis was conducted using two different wheat product
categories, bread and cookies, to note any differences between a staple and a hedonistic
food item. The labels selected for the analysis are non-GMO, gluten-free, and low-carb in
the case of bread and sugar-free in the case of cookies. Multinomial logit (MNL) and
random parameter logit (RPL) models are used to perform the analysis, where RPL model
accounts for the possibility of heterogeneous consumer preferences. The data were
collected using an online survey, which was administered in the summer of 2017 across 16
U.S. western states. In total, 1,009 valid responses were received.
Results show that, on average, consumers are willing to pay extra for the organic
label on bread but not on cookies. It appears that the healthy image of organic label
interferes with the hedonistic nature of cookies. This finding suggests that consumers prefer
organic labels more on staple food items than on hedonistic food items, which has also
been found previously (Van Doorn and Verhoef, 2011). Further, it is found that consumer
interest in the organic label alone is more consistently associated with self-reported
familiarity with organics rather than tested knowledge, and thus it is in line with the
findings of Pieniak, Aertsens, and Verbeke (2010) and Aertsens et al. (2011).
Second, in comparison to mean WTP for organics, consumers are willing to pay a
similar amount for the non-GMO label on bread, but a higher amount for cookies. This
155
suggests that consumers either do not know that organic is also non-GMO, or they confuse
the two labels, or they care mostly about the non-GMO component of organics. Similar
WTP values for organic and non-GMO labels alone have been observed in previous studies
(Bernard, Zhang, and Gifford, 2006; McFadden and Lusk, 2017). But the gap found
between WTP for the organic and non-GMO labels in cookies relative to bread further
indicates that consumers find the organic label less desirable in a hedonistic food product.
Next, considering the overall knowledge of and familiarity with organics, it is found that
specifically those with low knowledge and those unfamiliar are in fact willing to pay more
for the non-GMO label than the organic, suggesting that both objective and subjective
knowledge affect whether organic and non-GMO labels are confused.
Third, the interaction of organic and non-GMO labels on bread decreases mean
WTP for these labels in the whole sample as found in McFadden and Lusk (2017) as well.
But it depends on the age, WGIA, overall familiarity with and knowledge of organic, and
specific knowledge that organic is also non-GMO, whether the overall WTP for these
labels together is higher or lower than mean WTP for the organic label alone. Similarly,
the interaction of organic and sugar-free labels on cookies decreases overall mean WTP
for these labels in the whole sample. However, even the sugar-free label alone in cookies
is on average valued negatively by consumers, with an exception of consumers with some
WGIA and younger consumers. WGIA is also among factors determining whether the
interaction is negative or not, in addition to residency in California, income group, and
familiarity with and knowledge of organic. This shows that there are many factors that may
influence consumer preferences for the combination of organic with other labels.
