Environmental History
hamb
ELECTRICANDPLUG-INHYBRIDVEHICLEDEMANDLESSONSFORANEMERGINGMARKET.pdf
ELECTRIC AND PLUG-IN HYBRID VEHICLE DEMAND: LESSONS FOR AN EMERGING MARKET
TAMARA L. SHELDON, J. R. DESHAZO and RICHARD T. CARSON
Understanding demand in the new plug-in hybrid electric vehicle (PHEV) market is critical to designing more effective adoption policies. We use stated preference data from an innovative choice experiment to estimate demand for PHEVs relative to battery electric vehicles (BEVs) and to explore heterogeneity in demand for these vehicles. We find the gap between willingness to pay for PHEVs and their price premium over conventional vehicles is on the order of current subsidies, while that of BEVs is an order of magnitude larger. We use a latent class model to show PHEVs draw a different consumer segment into the market. (JEL Q5, R41)
I. INTRODUCTION
Policymakers have sought to spur demand for plug-in electric vehicles (PEVs) through a vari- ety of policy incentives. The economic rationale for the design of these policy incentives has been based on the presence and size of environmen- tal and knowledge-spillover externalities. The desired effect of policies targeting these exter- nalities is to adjust consumers’ ex post demand for these vehicles in ways that enhance overall social welfare. However, understanding of con- sumer demand for these vehicles and associated interactions with policy incentives is incomplete because automakers have recently differentiated their PEV product mix.
Automakers have added plug-in hybrid elec- tric vehicles (PHEVs), which may be fueled by either electricity or gasoline, to the early mix of battery electric vehicles (BEVs), which are fueled only by electricity. By adding PHEVs, automakers sought to eliminate consumers’ “range anxiety” associated with the limited travel range of smaller-battery electric vehicles. PHEVs also represented a vehicle design innovation that enabled many automakers to adapt pre-existing vehicle designs to plug-in electric refueling, thus
Sheldon: Assistant Professor, Department of Economics, University of South Carolina, Columbia, SC 29204. Phone (803) 777-6828, Fax (803) 777-6876, E-mail [email protected]
DeShazo: Professor and Director, Luskin School of Public Affairs, University of California, Los Angeles, CA 90095. Phone (310) 593-1198, Fax (310) 267-5443, E-mail [email protected]
Carson: Professor, Department of Economics, University of California, San Diego, CA 90093. Phone (858) 822-2262, Fax (858) 534-7040, E-mail [email protected]
eliminating their need to design entirely new models. For instance, there are now PEV ver- sions of the Ford Fusion and Honda Accord, as well as the Mitsubishi Outlander and the Porsche Panamera. The attractiveness of the PHEVs rel- ative to BEVs to automakers has been revealed by the decision to introduce a substantial number of PHEVs to the market (see Table 1) with plans for many more in the relatively near future as rumored in the trade press (e.g., the Audi A3 e-tron and the Hyundai Sonata Plug-in). Con- sumers have thus far exhibited a preference for PHEVs relative to BEVs by purchasing relatively more of them, as shown in Figure 1.
Within the literature, researchers have under- taken innovative studies of consumer demand for BEVs (Brownstone, Bunch, and Train 2000; Bunch et al. 1993; Hidrue et al. 2011); however, research on PHEV demand remains limited. Most existing research studies were implemented before PHEVs were commercially available and they focused on design priorities for vehicle attributes (Axsen and Kurani 2009; Kurani, Heffner, and Turrentine 2008) as well as qualitative market trial studies (Caperello and Kurani 2012; Graham-Rowe et al. 2012).
ABBREVIATIONS
ASC: Alternative Specific Constant BEV: Battery Electric Vehicles HOV: High Occupancy Vehicle ICE: Internal Combustion Engine IIA: Independence of Irrelevant Alternatives PEV: Plug-in Electric Vehicle PHEV: Plug-in Hybrid Electric Vehicle WTP: Willingness to Pay
695
Economic Inquiry (ISSN 0095-2583) Vol. 55, No. 2, April 2017, 695 – 713
doi:10.1111/ecin.12416 Online Early publication November 10, 2016 © 2016 Western Economic Association International
696 ECONOMIC INQUIRY
TABLE 1 PEV Model Introductions
2012 – 2013 2014 – 2015
Model Make PEV Type Model Make PEV Type
Model S Variations Tesla BEV i3 BMW BEV 2012 Smart Fortwo ed. Daimler BEV E-Golf VW BEV e6 BYD BEV i8 BMW PHEV Chevy Spark GM BEV Cayenne S E-Hybrid Porsche PHEV Scion iQ Toyota BEV 918 Spyder Porsche PHEV RAV4 EV Toyota BEV Soul EV Kia BEV C-Max Energi Ford PHEV B-Class Electric Mercedes-Benz BEV Fusion Energy Ford PHEV A3 e-tron Audi PHEV Fit EV Honda BEV Infinity LE Nissan BEV GCE Amp BEV Model X Tesla BEV MLe Amp BEV A3 e-tron Audi PHEV Accord PHV Honda PHEV Golf twinDRIVE VW PHEV F3DM BYD PHEV Sonata Plug-in Hybrid Hyundai PHEV F6DM BYD PHEV Outlander Sport PHV Mitsubishi PHEV 500 Elettrica Chrysler-Fiat BEV A4 e-quattro Audi PHEV Cadillac ELR GM PHEV V60 Plug-in Hybrid Volvo PHEV Prius Plug-in Hybrid Toyota PHEV Panamera Porsche PHEV Focus Electric Ford BEV
Closest to our work here is Axsen and Kurani (2013), who survey a sample of recent new car buyers in San Diego who are asked to play a design game where they get to assemble vehicles by allocating points to different attribute options. They find that PEVs are preferred to regular hybrids, which in turn are preferred to regular vehicles. PHEVs dominate BEVs. An important finding from this study is that PHEVs with shorter ranges may be more commercially viable than more expensive longer-ranged PHEVs.1
A. Understanding Demand to Guide Policy Design
Several important questions relevant to under- standing the need for, and design of, public policies remain unanswered. A critical empiri- cal question is how large are the differences in consumer demand for BEVs, PHEVs, and inter- nal combustion engines (ICEs), ceteris paribus? Answering this question helps us to understand the magnitude of importance of the PHEV as a vehicle innovation in the growth of the PEV mar- ket. This relative preference information is also critical in determining whether vehicle purchase
1. This study in some ways can be seen as the inverse of ours. We focus on prospective new car buyers in California at a time when a substantial number of PHEVs and BEVs have already been introduced and look at choices between compet- ing vehicles that are described by attributes rather than having recent buyers assemble preferred vehicle configurations from sets of attributes.
incentives will even be needed to encourage PHEV purchases, and if so, how effective they are likely to be in compensating for utility dif- ferentials across types of vehicles. Lastly, under- standing utility differentials enables economists to evaluate the size of “free rider” losses associ- ated with vehicle purchase incentives for BEVs versus PHEVs, as well as the aggregate public revenues needed to support these rebate policies.2
Beyond vehicle purchase incentives, there are also important questions about how differences in consumer demand for BEVs and PHEVs interact with other public policy incentives. For example, some researchers have suggested that demand for BEVs, relative to PHEVs, may be more sensitive to the presence of residential and publicly accessible recharging infrastruc- ture since BEVs cannot operate using gasoline (Egbue and Long 2012; Khan and Kockelman 2012). If true, this might explain how the policy provision for charging infrastructure and PEV- friendly buildings will affect the relative rates of purchase of BEVs and PHEVs. In addition, many states allow BEVs and PHEVs to use
2. DeShazo, Sheldon, and Carson (2016) find that rebates are more cost-effective not only when they target consumer segments with more marginal consumers, but also when they target segments with fewer infra-marginal consumers. For example, they find that it is optimal to allocate higher rebates to BEV purchases than to PHEV purchases since there are more infra-marginal PHEV purchasers who receive the rebate and who would have purchased the PHEV even in the absence of the rebate.
SHELDON, DESHAZO & CARSON: DEMAND FOR PLUG-IN HYBRIDS 697
FIGURE 1 PEV Registrations in California by Month
high occupancy vehicle (HOV) lanes. When predicting PEV market growth impacts, it may be useful to policymakers to better understand whether there are differences in how HOV access induces demand for BEVs versus PHEVs.
Better understanding consumer valuation of PHEVs and their attributes can also inform us of how this new market is likely to evolve as newer vehicle models come to market. For example, estimating consumer preferences for PHEV range can help in understanding how consumer demand will likely respond to second generation, extended-range PHEVs that are expected to be available in the next several years.
B. Demand Modeling Strategy
Using stated preference data from a survey of California new car buyers, we estimate discrete choice models that allow us to compare demand
for BEVs, PHEVs, and conventional ICE vehi- cles. Not only is this one of the first studies to investigate relative demand for different PEV technologies, but our analysis also utilizes inno- vative experimental design techniques, including a Bayesian D-efficient design that enables a more efficient estimation, as well as a pivoting on cur- rent preference and prices for non-PEV vehicles in order to make the choices faced by survey respondents more realistic.
We estimate three models that allow us to explore heterogeneity of preferences for PEVs from several angles. First, we estimate a mixed logit model that allows for the esti- mated preference parameters to randomly vary. Second, we estimate an alternative specific con- stant logit, which provides insight into what consumer characteristics tend to be associated with different aspects of the preference parameter distributions. Finally, we estimate a latent class
698 ECONOMIC INQUIRY
model, which allows us to uncover customer profiles of market segmentation.
Stated preferences approaches have long been used to look at new consumer products (e.g., Louviere and Hensher 1983), including an influential early study on electric car attributes (Beggs, Cardell, and Hausman 1981). There is a strong motivation to rely on stated preference data for this purpose because there is no market data on which to estimate the preference param- eters of interest. Typically, reasonable results that are useful for decision making are obtained (Louviere, Hensher, and Swait 2000), although this is not always the case. Carson and Groves (2007) show that when a choice task for a private good is viewed as influencing a decision as to whether to offer a new good for sale, there is an incentive to over-estimate the propensity to buy in order to expand the future choice set, but when the choice task is viewed as influencing the pricing of an existing good then the incentive is for the respondent to appear to be more price sensitive than would be true in the parallel mar- ket transaction. This stands in stark contrast to the public goods case (Carson, Groves, and List 2014) where truthful preference revelation is the dominant strategy for a single take-it-or-leave-it binary choice with a coercive payment vehicle (e.g., tax or utility bill) which is perceived as con- sequential by the survey respondent in the sense of having a non-zero probability of influencing the government’s decision. Carson and Groves (2007) argue that a multinomial choice question, which we use in this study, under strong, but plausible conditions (i.e., respondent does not want to induce the company to offer products with attributes perceived as undesirable), has desirable incentive properties with respect to the revelation of marginal willingness for a change in one of a good’s attributes but not necessarily for total willingness to pay for an individual good. This conjecture is directly supported by experimental evidence (e.g., Lusk and Schroeder, 2004). It is also indirectly supported by a large number of papers using the approach of Swait and Louviere (1993) to combine revealed and stated preference data which finds that the pref- erence parameters from the two types of data are typically statistically indistinguishable once all of the preference parameters from one of two data types is allowed to vary by a single scale factor. This finding is consistent with marginal willingness to pay (WTP) estimates being similar, but is silent on whether there is a difference in total WTP estimates. A divergence
between total WTP estimates is often reflected in the purchase/no purchase decision. This can be seen in the recent Kesternich et al. (2013) paper on Medicare Part D health insurance where the stated preference data produce an estimate of a higher propensity to buy the insurance than the revealed preference data (although as the paper argues there may be good reasons for this divergence), but statistically indistinguish- able estimates for the attributes of insurance policies conditional on buying. We bypass the purchase/no purchase decision here by screening for respondents who indicate they will be in the market for a new car in the relatively near future.3 The policy issues this paper focuses on rely on understanding how changes in attribute levels influence the distribution of vehicles that would be purchased.
II. SURVEY DESIGN AND DATA
We administered an online survey to a rep- resentative sample of Californian new car buy- ers and obtained a sample of 1,261 completed surveys.4 The survey first gathered household, vehicle, and demographic data. Next, the survey elicited body and brand preferences. Respondents were asked to choose the top two vehicle body types (out of 12 options) they were most likely to select for their next new vehicle purchase, as shown in Figure 2. Then respondents were asked to select the top three brands (out of the 20 most popular brands by sales volume in California in 2012) they were most likely to select for their next new vehicle purchase, as shown in Figure 3.
Next, respondents were shown four sets of five vehicles, as shown in Figure 4, and in each set were asked to choose which of the five vehicles they were most likely to select for their next new vehicle purchase. The total set of 20 vehicles respondents chose from included all conventional vehicles (including ICE vehicles, hybrid electric vehicles, and diesel-fueled vehicles) on the new vehicle market as of the fall of 2013 that are
3. There will no doubt be some individuals who buy new vehicles in the relatively near future that we exclude from our sample. For instance, someone’s vehicle may unexpect- edly breakdown or the individual may receive an unexpected increase in income or wealth. Likewise, some of our respon- dents will decide to keep their current vehicle after suffering adverse income/wealth shocks or because their vehicle turns out not to need expected future repairs.
4. Of the respondents who completed an initial screener, approximately 42% both qualified as potential new car buyers and completed the survey.
SHELDON, DESHAZO & CARSON: DEMAND FOR PLUG-IN HYBRIDS 699
FIGURE 2 New Car Buyer Survey: Body Choice
700 ECONOMIC INQUIRY
FIGURE 3 New Car Buyer Survey: Brand Choice
of both the top brand and top body selected by respondents. The remainder of the 20 included a random draw of vehicles that are of the top body choice and second or third brand choice, or of the second body choice and top brand choice. In cases where the set of vehicles that meets these criteria is less than 20, the remainder of the vehicles were a random selection of vehicles that are of either of the top body selections or of the top brand selections. Finally, respondents were asked to choose which one of the four vehicles chosen as top picks out of the 20 vehicles in the previous four questions they would be most likely to select for their next new vehicle purchase, as shown in Figure 5. This “top” vehicle and its characteristics are carried through to subsequent questions in the survey.
Respondents were provided with information on BEV and PHEV technologies and introduced to PEV attributes, including refuel price, electric range, and HOV lane access. Finally, respondents were asked to choose between the conventional version, two BEV versions, and two PHEV ver- sions of the vehicle they previously indicated as their top choice. In each choice set the first
column displayed the conventional vehicle, and we randomized whether the two BEVs or PHEVs appeared in the subsequent columns. Attribute levels vary for each vehicle version as shown in Table 2, with price pivoting off the price of the existing conventional vehicle. An example choice set is shown in Figure 6. By choosing between five versions of the top vehicle, respon- dents are encouraged to assume that everything else (e.g., trim and performance) except the listed attributes are identical. This allows us to focus on how respondents make tradeoffs between vehicle technology, price, refuel cost, electric range, and HOV lane access.
We use NGENE software to design the choice experiment. We sought an experimental design to minimize the variance of the estimated coef- ficients of the specified utility function that underlies the logit models. The efficiency of an experimental design can be greatly improved if we know the approximate magnitude or even just the sign of the true parameters (Scarpa and Rose 2008). For example, by assuming that the coefficient on price is negative, or that consumer utility for an alternative is reduced
SHELDON, DESHAZO & CARSON: DEMAND FOR PLUG-IN HYBRIDS 701
FIGURE 4 New Car Buyer Survey: Top Vehicle Choice
as that alternative gets more expensive, we no longer need an experimental design that can distinguish between a negative or positive coef- ficient, but can instead more precisely estimate a negative coefficient.
Specifically, we use an algorithm in NGENE that allows us to maximize the amount of information we are able to extract from our choice experiment by minimizing the variance – covariance estimator of the vector of utility function coefficients. The algorithm searches through potential experimental designs with different combinations and levels of attributes. We select the experimental design with the smallest determinant of the asymptotic variance – covariance matrix, also known as the D-error.5 To further increase the efficiency of the design, we specify Bayesian priors. That is, for each coefficient that we seek to estimate, we specify an assumed a priori distribution. We base these assumptions on parameter estimates from earlier studies looking at PEV attributes (Acht- nicht, Bühler, and Hermeling 2012; Brownstone, Bunch, and Train 2000; Bunch et al. 1993; Ewing and Sarigöllü 2000; Golob et al. 1993; Hidrue et al. 2011; Qian and Soopramanien 2011).
5. For more details see Scarpa and Rose (2008).
To make the choice experiment more realis- tic for respondents, we employ a pivot design. Price levels are designed to be percentages of a reference value. The price of the top conven- tional vehicle chosen by a respondent becomes her reference price, and the different price lev- els she sees are the percentage levels as speci- fied by the experimental design multiplied by the reference price. For example, a respondent who selects a conventional model that costs $30,000 would see BEV and PHEV versions of that model that cost $31,500, $34,500, $37,500, or $45,000. On the other hand, a respondent who is consid- ering the luxury end of the market and selects a conventional model that costs $60,000 would see BEV and PHEV versions of that model that cost $63,000, $69,000, $75,000, or $90,000. Our pivot design anchors the alternatives in reality, avoiding unrealistic vehicles. For example, the maximum value for the price attribute is 150% of actual market price. In 2013, a new Camry cost approximately $22,000. Therefore, the maximum price for a Toyota Camry PEV in a choice set would be $33,000.
To incorporate the pivoting price attribute levels in the experimental design, NGENE’s algorithm uses relative attribute levels rather than absolute attribute levels for price. However, in calculating the efficiency of the design, the
702 ECONOMIC INQUIRY
FIGURE 5 New Car Buyer Survey: Top Vehicle Choice
algorithm must assume some reference level. Therefore, we assume four different segments: (1) economy and compact cars, (2) midsize and large cars, (3) SUVs, trucks, and minivans, and (4) luxury vehicles. For each segment we assume the price is the average of that vehicle type from the new vehicle universe. The algorithm utilizes a model averaging approach according to the actual market shares of the four segments.
Table A1 in the Appendix gives definitions of all the variables used in our analysis. Most of these variables were collected in the survey. We obtained average gasoline prices in December 2013 by Census Tract from Gas Buddy Organiza- tion Inc. From the U.S. Department of Energy’s Alternative Fuels Data Center we obtained a mea- sure of publicly available PEV charger density, which we define as the number of level 2 chargers within a 5-mile radius of the population centroid of a Census Tract as of December 2013.
III. MODEL SPECIFICATION
The standard multinomial logit can model the probability of selecting a vehicle over other alternatives. In this model, a respondent selects the vehicle that gives her greater utility than
any other available alternative. The utility of each alternative is a function of its attributes. The estimated coefficients tell us how a change in each attribute (e.g., an increase in range) impacts utility.
Individual n receives utility Uni from choosing alternative i:
(1) Uni = Vni + εni.
The probability of individual n selecting alter- native i is the probability her utility from i is greater than her utility from choosing any other available alternative:
(2) πni = Prob ( Vni + εni ≥ Vnj + εnj
) ; ∀j = i.
If we assume εni’s are independently dis- tributed Type-I extreme value errors and a linear utility function, such that Vni = x
′ i 𝛃, where xi is
a vector of attributes of i and 𝛃 is a vector of parameters, then we can model the probability of individual n choosing alternative i as:
(3) πni = exp
( μnx
′ i 𝛃 )
J∑ j=1
exp ( μnx
′ j 𝛃 ) ,
SHELDON, DESHAZO & CARSON: DEMAND FOR PLUG-IN HYBRIDS 703
TABLE 2 Attribute Levels
Purchase pricea (% of conventional) Gasoline 100% BEV 105%, 115%, 125%, 150% PHEV 105%, 115%, 125%, 150%
Gasoline refuel cost ($ per gal) Gasolineb $4.00, $4.40, $4.80, $5.60 BEV n/a PHEVc $2.00, $2.20, $2.40, $2.80
Electric refuel costd ($ per gal equivalent) Gasoline n/a BEV $0.90, $1.10, $1.50, $2.50 PHEV $0.90, $1.10, $1.50, $2.50
Gasoline range (miles) Gasoline 300 BEV 300 PHEV 0
Electric range (miles) Gasoline n/a BEV 50, 75, 100, 200 PHEV 10, 20, 40, 60
HOV access Gasoline No BEV No, yes PHEV No, yes
aThe respondent sees price in dollars. For example, a respondent who selected a conventional model that costs $30,000 would see BEV and PHEV versions of that model that cost $31,500, $34,500, $37,500, or $45,000.
bAt the time the survey was administered, average gaso- line cost in California was approximately $4 per gallon.
cThe average gasoline fuel economy of PHEVs as of December 2013 was 41 mpg, which is roughly double the fuel economy of our gasoline vehicle universe of 20 mpg. Therefore we choose a baseline gasoline refueling cost for PHEVs that is half that of gasoline vehicles.
dAt the time the survey was administered, the average overnight electricity rate in California was roughly 16 cents per kWh and the average vehicle economy of electric vehicles was 3.5 miles per kWh, suggesting an average cost per electric mile of $0.046. The average cost per mile of gasoline vehicles
in our vehicle universe is $4∕gal
20 mi∕gal = $0.20 per mile. Thus
on average, refueling cost for electric miles is 23% of the $4 per gallon refueling cost for gasoline miles, or $0.92/gal. Therefore we choose a baseline electric refueling cost of $0.90 per gallon equivalent.
where μn is a scale parameter commonly assumed to equal 1.
In this model, the coefficients are fixed, effec- tively assuming that all respondents have the same preferences (e.g., all respondents have the same value for a BEV, all else being equal). The logit model exhibits the independence of irrele- vant alternatives (IIA), meaning that the odds of choosing vehicle j over vehicle k are indepen- dent of the choice set for all pairs j, k, which may imply unrealistic substitution patterns. The stan- dard logit model does not allow for heterogeneity of preferences.
The first model we estimate that relaxes this assumption is a mixed logit. In the mixed logit model, developed by Train (1998), the coefficients of the utility function are random parameters for which we can specify a distri- bution. For example, if we assume a coefficient is normally distributed, we estimate both the mean and standard deviation of that coefficient. This model allows for heterogeneous preferences across respondents and does not necessarily exhibit the IIA property, thereby allowing for more flexible substitution patterns. Structurally, the mixed logit model is similar to the standard logit except the parameters of the utility function are assumed to be random, not fixed, and the probability of individual n selecting alternative i becomes:
(4) πni = ∫ exp
( μnx
′ i 𝛃 )
J∑ j=1
exp ( μnx
′ j 𝛃 ) f (𝛃|𝛉) ∂𝛃,
where f (𝛃|𝛉) is the density function of 𝛃. A drawback of the mixed logit model is that
it does not tell us where different respondents are in the estimated distribution of preferences.6 In other words, it does not tell us which respondents have which preferences.
The alternative specific constant (ASC) logit and the latent class logit offer two different meth- ods of further exploring heterogeneity. The ASC logit, developed by McFadden (1974), is a con- stant parameter logit where explanatory variables in the utility function include not only alterna- tive attributes but also respondent characteristics. The ASC logit estimation therefore tells us how respondent characteristics impact their odds of selecting a BEV or PHEV relative to the gaso- line version. The ASC logit is similar to the standard logit except the utility function includes consumer characteristics:
(5) Vni = x ′ i𝛃 + z
′ n𝛄,
where zn is a vector of characteristics of individ- ual n and 𝛄 is a vector of parameters.
The latent class model is similar to the ASC logit model in that preferences are heteroge- neous across respondents characteristics. The latent class model segments the population
6. Technically, it is possible to make the mean or variance of a mixed logit parameter a function of observed covariates, but in practice this is rarely done to problems because such models tend to be numerically unstable and frequently do not converge to a well-defined maximum value.
704 ECONOMIC INQUIRY
FIGURE 6 New Car Buyer Survey: PEV vs. Conventional Vehicle Choice Module
into different classes, where preferences for each class are estimated separately, and class membership of respondents is determined by their characteristics.
Assume existence of S segments in a popu- lation. The probability of consumer n choosing alternative i conditional on membership in seg- ment s, where s =1, .. , S, is:
(6) πni|s = exp
( x′
i 𝛃s )
J∑ j=1
exp (
x′ j 𝛃s ) .
Allowing latent membership for segmentation to be:
(7) M∗ns = y ′ n𝛌s + ζns,
where M∗ns is membership likelihood function for individual n to be in segment s, yn is vector
of both psychometric constructs and socioeco- nomic characteristics, 𝛌s is vectors of parameters, ζns is independently distributed Type-I extreme value errors.
We can model the probability of consumer n belonging to segment s as:
(8) πns = exp
( y′n𝛌s
) S∑
s=1 exp
( y′n𝛌s
) .
The probability of consumer n choosing alter- native i is the sum across segments of the prob- ability of her selecting alternative i conditional on segment membership times her probability of segment membership:
(9) πni = S∑
s=1 πnsπni|s
SHELDON, DESHAZO & CARSON: DEMAND FOR PLUG-IN HYBRIDS 705
TABLE 3 Mixed Logit Results
Price Normally Distributed Price Log Normally Distributed
Mean SD Mean SD
Price ($1,000) −0.226*** 0.194** −2.520*** 0.397 (0.028) (0.089) (0.257) (0.320)
BEV −1.301** 4.007*** −1.605*** 4.348*** (0.656) (0.950) (0.460) (0.817)
PHEV 1.738** 2.745*** 1.921*** 2.423***
(0.772) (0.461) (0.407) (0.428) Range 0.014*** 0.004 0.017*** 0.007***
(0.002) (0.003) (0.002) (0.002) Refuel −0.158** 0.057 −0.128 0.005
(0.072) (1.095) (0.096) (0.240) HOV 0.311** 0.302 0.261*** 0.400**
(0.128) (0.753) (0.087) (0.159) Observations 24,940 24,940 Log pseudolikelihood −5,959 −5,931
Notes: Weighted to represent population of California new car buyers. Robust standard errors in parentheses, clustered by respondent.
*p < .1; **p < 0.05; ***p < .01.
(10)
πni = S∑
s=1
exp ( y′n𝛌s
) S∑
s=1 exp
( y′n𝛌s
) exp
( μsx
′ i 𝛃s )
J∑ j=1
exp ( μsx
′ j 𝛃s ) .
IV. RESULTS
A. Mixed Logit Model
Table 3 shows the results of the mixed logit estimation. The first two columns are estimated assuming that the price coefficient is normally distributed. The second two columns assume the price coefficient is log normally distributed.7
Specifications with log normally distributed price coefficients have a better model fit. This is unsur- prising since the log normal distribution allows for the mean to be greater than the median, which might be the case if some respondents are very price sensitive. Table 3 shows that on average (and all else being equal), respondents have a negative preference for BEVs relative to conventional gasoline vehicles (the omitted cat- egory), a positive preference for PHEVs, a pos- itive preference for increased range and HOV
7. A log-normal distribution assumption for a parameter implies the coefficient should be positive. Therefore, we trans- form price, multiplying it by − 1 for the estimation, and trans- form the resulting positive coefficient back post-estimation, multiplying by − 1. Therefore, the price coefficient for the log-normal specification shown in Table 3 is negative.
access, and a negative preference for higher refu- eling costs.
Figure 7 shows kernel density plots of individ- ual respondents’ estimated coefficients, using a sampling method from Revelt and Train (2000). The distribution of the (negative) price coeffi- cient appears to be log normal, as shown in Figure 7a. The median price coefficient is around 0.3 and the mean is substantially higher, sug- gesting a sizable fraction of respondents are very price sensitive.
Figure 7b shows that the distribution of coef- ficients for BEVs is bi- or perhaps even trimodal. While most respondents have a negative coeffi- cient for BEVs of around −2, a small portion of the population has a positive preference for BEVs, and a significant portion of the population has an even stronger dislike of BEVs. Similarly, Figure 7c shows that the distribution of coeffi- cients for PHEVs is bimodal, with a minority of respondents having a coefficient around −2, but a majority of respondents having a strong positive preference for PHEVs with a coefficient closer to 4.
While range has a positive coefficient for all respondents, the distribution of the range coefficient as shown in Figure 7b also exhibits bimodality, with some respondents caring signif- icantly more than others, perhaps due to different commute distances.
Figure 7e shows that a minority of respon- dents does not seem to care about refueling costs with a coefficient of zero, but that a majority of
706 ECONOMIC INQUIRY
FIGURE 7 Mixed Logit Coefficient Distributions
respondents do care about refueling costs with a coefficient around −2. Similarly, Figure 7f shows that a large majority of respondents values HOV lane access, but a minority does not, which may reflect a lack of local HOV lane access.
Table 4 shows the mean estimates of WTP for vehicle attributes obtained using the Hensher and
Greene approach (Hensher and Greene 2003).8
We find that the average WTP for a BEV is about −$4,900. Out of current BEVs on the market as
8. To calculate the mean WTP for each attribute, we took the mean of 10,000 random draws from the distribution of the attribute’s coefficient divided by the exponential of a random draw from the distribution of the price coefficient.
SHELDON, DESHAZO & CARSON: DEMAND FOR PLUG-IN HYBRIDS 707
TABLE 4 Willingness to Pay
WTP (Price Normally
Distributed)
WTP (Price Log Normally
Distributed)
BEV −$18,693 −$4,906 PHEV $12,873 $6,783 Additional mile of
electric range $81 $57
Additional $ per gallon refuel cost
−$874 −$430
HOV access $1,555 $903
of early 2014 that have a comparable ICE model, the BEVs are priced at an average premium of $18,411 (see Table 5 for details). We find that the average WTP for a PHEV is nearly $6,800. Out of PHEVs on the market as of early 2014 that have a comparable ICE model, the PHEVs are priced at an average premium of $11,024 (see Table 5 for details). This suggests that the gap between WTP and the price premium for BEVs is very high, on the order of $23,000, while the gap between WTP and the price premium for PHEVs is much smaller, on the order of $4,000. State level incentives are typically a few thousand dol- lars, and the federal income tax incentive is up to $7,500. This suggests that current financial incen- tives will stimulate fewer BEV purchases, but could stimulate more PHEV purchases. This is consistent with DeShazo, Sheldon, and Carson’s (2016) finding that California’s PEV rebate pol- icy induces more marginal PHEV purchases than marginal BEV purchases.
The average survey respondent would pay approximately $589 per year on refueling costs per $1 increase in $/gallon equivalent. This is based on the assumption that the respondent re- fuels once every week and a half, and that the respondent’s fuel tank capacity is 17 gallons. These are the average values based on the sur- vey responses. Thus, the WTP for refuel savings of $1 per gallon of $430 implies a high discount rate, with an expected payback period of just under 1 year.9
9. This payback period is calculated using mean WTP for refuel savings across the whole sample, including consumers like those in Segment 2 of the Latent Class model who have no intention of purchasing a PEV. Using the results from the latent class model, we find that Segment 1 has a payback period of 2 years and Segment 3 has a payback period of 8.7 years, which is close to full valuation. Segment 2 has a payback period of less than 6 months (though the WTP estimate is not statistically different from 0).
TABLE 5 Price Comparison of Internal Combustion
Engine (ICE) Vehicles and PEVs of the Same Model
ICE MSRP BEV MSRP Premium
Smart Fortwo $13,270 $25,000 $11,730 Chevrolet Spark $12,170 $26,685 $14,515 Ford Focus $16,810 $35,170 $18,360 Toyota RAV4 $23,550 $49,800 $26,250 Honda Fit $15,425 $36,625 $21,200 Average premium $18,411
ICE MSRP PHEV MSRP Premium
Ford C-Max $25,170 $32,920 $7,750 Ford Fusion $21,970 $34,700 $12,730 Honda Accord $21,955 $39,780 $17,825 Toyota Prius Plug-in $24,200 $29,990 $5,790 Average premium $11,024
Notes: MSRPs are taken from automakers’ websites and http://www.edmunds.com. MSRPs as of March 2014.
We find that the average respondent is willing to pay about $900 for free single-occupant HOV lane access. The survey does not state that the HOV benefit would expire, which implies the HOV benefit would be good for the lifetime of the PEV.10 Bento et al. (2014) estimate the average annual rent of a hybrid HOV sticker in southern California to be $743, with a net present value of $4,800. Shewmake and Jarvis (2014) estimate an average premium of $3,200 for a hybrid with an HOV sticker, which translates into a yearly value of $625. In comparison, our estimated WTP of $900 is low. However, Bento et al. (2014) derive their estimate by dividing the value of travel time savings from using HOV lanes by the number of hybrids on the road. Shewmake and Jarvis (2014) use data on sales of used hybrids. Unlike these two studies, our estimate of WTP for HOV lanes is the average across the new car buying population, including consumers who indicated they would not purchase a PEV. As shown in Section IV.C, our WTP estimate is higher for the consumer population who would consider purchasing a PEV.
The mixed logit results show that there is considerable heterogeneity in preferences across BEVs and PHEVs, as well as across consumers. Sections IV.B and IV.C attempt to better understand the underlying sources of this heterogeneity.
10. In California, PEV drivers currently have this benefit through 2019.
708 ECONOMIC INQUIRY
TABLE 6 Alternative-Specific Constant Logit, Main
Results
Price ($1,000s) −0.062*** (0.009)
BEV −9.701**** (3.604)
PHEV −8.936*** (2.701)
Range 0.033***
(0.003) Range squared −0.0001***
(.00001) Refuel −0.086**
(0.045) HOV 0.239***
(0.057) Observations 24,620 Log pseudolikelihood −6,732
Notes: Weighted to represent population of California new car buyers. Robust standard errors in parentheses, clustered by respondent.
*p < .1; **p < .05; ***p < .01.
B. Alternative-Specific Constant Logit Model
Tables 6 and 7 show the results of the ASC logit estimation. The coefficient on price in Table 6, −.06, is smaller in absolute value than the −2.5 estimated by the preferred specification in Table 3. The former estimate assumes the coefficient is fixed, while the latter estimate assumes the coefficient follows a log normal distribution and allows for the mean to be greater than the median, which might be the case due to a small fraction of respondents being very price sensitive. The coefficients on refueling costs and HOV access are similar between Tables 3 and 6. The BEV and PHEV coefficients are not directly comparable, as those in Table 6 must be adjusted by respondent characteristics as shown in Table 7. For example, the coefficient on Gas Price in Table 7 is approximately 1.5, and the gas price in most Census Tracts during December of 2013 was greater than $3, such that at least 3 * 1.5 = 4.5 must be added to both the BEV and PHEV coefficients in Table 6.
Due to the complexity of the model, we are unable to achieve convergence in the maximum likelihood estimation of the mixed logit when we include a quadratic range term in the speci- fication. We are able to achieve convergence in the ASC logit estimation when a quadratic range term is included. When we include this term, we get more precision on the refueling cost coef- ficient and we find that consumers’ utility for
TABLE 7 Alternative-Specific Constant Logit, ASC
Results
BEV PHEV
Small body −0.126 −0.014 (0.210) (0.196)
Household vehicles 0.091 0.196*
(0.122) (0.112) Outlet 0.367 0.394*
(0.237) (0.214) Parking at work 1.967*** 0.809
(0.627) (0.566) Commute under 20 miles −0.803** −0.681***
(0.316) (0.263) Use gas mode daily −1.302*** −1.345***
(0.364) (0.283) HOV access 0.123 0.456***
(0.161) (0.135) Pro environment 0.886*** 0.427**
(0.215) (0.195) Early adopter 0.207*** 0.130***
(0.055) (0.050) Charging station density 0.004 0.010
(0.020) (0.020) Gas price 1.598 1.795**
(0.979) (0.714) Low income (<$30k) −0.228 0.148
(0.354) (0.315) High income (>$100k) −0.415* −0.070
(0.233) (0.206) Observations 24,620 24,620 Log pseudolikelihood −6,732 −6,732
Notes: Weighted to represent population of California new car buyers. Robust standard errors in parentheses, clustered by respondent.
*p < .1; **p < .05; ***p < .01.
range exhibits decreasing returns. This is consis- tent with the literature (Brownstone, Bunch, and Train 2000; Bunch et al. 1993). The linear and quadratic range coefficients suggest an optimal electric range of 165 miles.
Table 6 shows that all else being equal, con- sumers prefer PHEVs to BEVs. Table 7 shows that having pro environment preferences and self- identifying as an early adopter increase a respon- dent’s WTP for both BEVs and PHEVs, although relatively more for BEVs.
Respondents with round-trip commutes under 20 miles are less likely to select PEVs. This may be because a shorter commute would accrue less refueling cost savings, making it more difficult for the consumer to justify the higher upfront cost of a PEV.
The environmental benefits associated with driving a PHEV depend on the relative num- ber of miles driven in electric versus gasoline mode. While the California Air Resources Board
SHELDON, DESHAZO & CARSON: DEMAND FOR PLUG-IN HYBRIDS 709
currently assigns higher rebates to BEVs in the belief they are associated with greater environ- mental benefits than PHEVs, it is sometimes argued that PHEVs may result in close to the same environmental benefits if daily commut- ing can be done in all-electric mode (California Environmental Protection Agency 2007). PHEVs do not invoke range anxiety or impair the abil- ity to take longer occasional trips. The results in Table 7 support this assertion. Respondents who anticipate needing to utilize gasoline mode on a daily basis if they owned a PHEV are much less likely to purchase either a BEV or a PHEV. This effect is similar for BEVs and PHEVs, suggesting prospective PHEV drivers are equally as moti- vated to commute primarily in all-electric mode, even though they do not face the same total range constraints as BEVs.
The positive coefficients on outlet access in Table 7 suggest that respondents who have an electrical outlet near their home parking spot are more likely to purchase a PEV. This is con- sistent with earlier studies (Axsen and Kurani 2009; Hidrue et al. 2011). Notably, outlet access appears just as important for PHEVs as BEVs, even though PHEVs do not require the electric battery be charged in order to drive the vehi- cle in gasoline mode. However, when we replace the outlet variable with an indicator variable for whether the respondent lives in a single-family house, this coefficient is positive and statisti- cally significant at the 10% level for BEVs but smaller and not statistically different from zero for PHEVs.11 This may suggest that BEV owners are more comfortable plugging into an outlet at their single-family residence while PHEV owners living in multifamily housing are also comfort- able plugging into a less private or less exclusive outlet near their residential parking spot.
The coefficient on the indicator for whether a respondent parks in a garage while at work is positive and highly statistically significant for BEVs but smaller and not significant for PHEVs. Respondents with access to a parking garage at work may anticipate a higher likelihood of charg- ing access while at work, which would increase their utility for PEVs. These coefficients sug- gest that workplace charging is a more important issue for BEV adoption than PHEV adoption. The
11. If we substitute the Outlet variable with Single House, the BEV coefficient on Single House is 0.427* (0.234) and the PHEV coefficient on Single House is 0.151 (0.207), with other coefficients not significantly different. We do not include Outlet and Single House in the same specification due to concerns about collinearity.
coefficients on public charging station density are positive but not statistically different from zero.
The coefficients on HOV lane access are positive, but that for BEVs is smaller than that for PHEVs and not statistically significant. This suggests that new car buyers who have access to HOV lanes are more likely to purchase PHEVs, and that government policies allowing free single-occupant HOV lane access increase consumer probability of purchasing PHEVs. Sheldon and DeShazo (2015) find that Califor- nia’s HOV lane policy had a positive impact on both BEV and PHEV adoption, with relatively more impact on the PHEV market.
The coefficient on number of household vehi- cles is positive for both vehicle types, although only statistically significantly greater than zero for PHEVs. This lends support to the “Hybrid Household” hypothesis that households with larger vehicle fleets are more likely to diversify their vehicle holdings with alternative vehicles (Kurani, Turrentine, and Sperling 1996).
The coefficients on small body type are not statistically different from zero, implying that respondents who are likely to purchase a new vehicle that is a hatchback or small sedan are nei- ther more nor less likely than other respondents to select a PEV. Although the majority of PEVs on the market have historically been smaller vehi- cles, this result is unsurprising because in our choice experiment, respondents were allowed to choose PEV versions of any body type.
C. Latent Class Model
Tables 8 and 9 show the results of a latent class estimation assuming three segments, using a variety of sociodemographic variables and atti- tudes to determine segment membership. Note that the latent class groups are helpful in explain- ing the kernel density estimate of coefficients. For example, Figure 7b shows that there are three peaks in the BEV coefficient distribution: one at a large negative number, the biggest at a small neg- ative number, and the third and smallest peak at a near-zero positive number. These three peaks are consistent with the three BEV preferences of the different segments.
Table 8 shows consumer Segment 3 has a pos- itive WTP for PHEVs and a WTP for BEVs that is approximately zero. This class is by far the most receptive to BEVs. Table 9 shows that self-identified environmentalists and early adopters are more likely to be in Segment 3. Con- sumers who reside in single-family houses and
710 ECONOMIC INQUIRY
TABLE 8 Latent Class Model: Segment Preferences
Segment 1 Segment 2 Segment 3
Price ($1,000s) −0.193*** −0.387*** −0.024*** (0.016) (0.052) (0.007)
BEV −3.752*** −3.031*** −0.197 (0.382) (0.485) (0.300)
PHEV 0.643** −1.531*** 0.511** (0.298) (0.403) (0.251)
Range 0.051*** 0.013** 0.018***
(0.003) (0.006) (0.003) Range squared −0.0002*** −.00003 −.00003***
(0.00002) (0.00002) (0.00001) Refuel −0.219*** −0.088 −0.123**
(0.073) (0.105) (0.052) HOV 0.382*** −0.073 0.232***
(0.089) (0.156) (0.064) Class share 42.4% 26.1% 31.5% Observations 24,940 24,940 24,940
Note: Standard errors in parentheses. *p < .1; **p < .05; ***p < .01.
younger consumers are also more likely to be in Segment 3. These findings support the notion that demand for BEVs is driven by strong environ- mental preferences and eagerness to adopt new technologies. These findings also confirm ear- lier results that households with home charging infrastructure are relatively more likely to pur- chase PEVs.
Table 8 shows consumer Segment 2 has a neg- ative WTP for both BEVs and PHEVs. This is also the most price sensitive segment. Segment 2 has less strong preferences for range and is indifferent toward refueling cost and HOV lane access, perhaps as a result of their low likelihood of selecting a PEV. The results in Table 9 show that consumers who are less educated, more con- servative, less concerned about the environment, and tend not to be early adopters are more likely to belong to this segment.
Consumer Segments 2 and 3 are consistent with prevalent beliefs about the PEV market, in which there is a class of consumers that is enthusiastic about PEVs and another class that will have nothing to do with PEVs. Consumer Segment 1 is the most interesting, because this segment has more nuanced preferences and also represents the largest of the three seg- ments. Table 8 shows consumer Segment 1 has a negative WTP for BEVs but a positive WTP for PHEVs. They are more price sensitive than Segment 3.
Consumers who have HOV lane access, who do not live in single-family houses, and who are more liberal are more likely to belong to
TABLE 9 Latent Class Model: Segment Membership
Segment 1 Segment 2 Segment 3a
Household size −0.049 −0.238*** 0.000 (0.070) (0.077)
Household vehicles 0.231** 0.172 0.000 (0.106) (0.113)
Age under 35 −0.648*** −0.305 0.000 (0.205) (0.217)
Age over 60 0.548** 0.504* 0.000 (0.255) (0.258)
Low income (<$30k) 0.322 0.108 0.000 (0.262) (0.267)
High income (>$100k) 0.349* 0.074 0.000 (0.211) (0.225)
College education 0.056 −0.290 0.000 (0.187) (0.197)
Use gas mode daily 0.005 0.793** 0.000 (0.382) (0.357)
Single house −0.398** −0.313 0.000 (0.194) (0.204)
HOV access 0.050 −0.436*** 0.000 (0.121) (0.133)
Pro environment −0.641*** −1.088*** 0.000 (0.175) (0.191)
Early adopter −0.077* −0.219*** 0.000 (0.046) (0.049)
Liberal 0.332* −0.017 0.000 (0.189) (0.212)
Constant 0.277 1.439*** 0.000 (0.405) (0.407)
Class share 42.4% 26.1% 31.5% Observations 24,940 24,940 24,940
Note: Standard errors in parentheses. aSegment 3 is the baseline segment that the other seg-
ments are compared to. *p < .1; ** p < .05; ***p < .01.
Segment 1, as shown in Table 9. Respondents fit- ting this profile tend to live in urban areas. Addi- tionally, consumers who are older, have higher incomes, and are more educated are more likely to belong to Segment 1. This segment’s positive preference for PHEVs appears to stem not from environmental or early adopter preferences but rather from more pragmatic reasons such as refu- eling cost savings and HOV lane access. This seg- ment’s negative preference for BEVs may be in part driven by less access to home charging.
The latent class results show that the BEV market may be constrained since less than a third of the new car buying population seems willing to consider purchasing a BEV, all else being equal. A much larger fraction of the pop- ulation, and one that breaks out of the early adopter/environmentalist niche, seems willing to consider purchasing a PHEV.
The $900 estimate for WTP for HOV lane access in Section IV.A is the mean WTP across
SHELDON, DESHAZO & CARSON: DEMAND FOR PLUG-IN HYBRIDS 711
the across the whole sample, including con- sumers like those in Segment 2 who have no intention of purchasing a PEV. Using the results from Table 8, we find that Segment 1 has a WTP of $1,979 and Segment 3 has a WTP of $9,667. Segment 2 has a WTP that is not statistically dif- ferent from zero. Thus, the population-weighted average WTP for Segments 1 and 3 (consumers who would actually consider purchasing a PEV) is $5,256, which is in line with previous stud- ies (see Section IV.A). Assuming the consumer keeps the PEV for 6 years12 and that the HOV sticker retains its benefit for these 6 years, our $5,256 estimate equates to an annual value of $876.
V. IMPLICATIONS FOR POLICY AND THE EMERGING MARKET
Results from the mixed logit model suggest that the gap between WTP and the price premium for BEVs is very high, on the order of $23,000, while the gap between WTP and the price pre- mium for PHEVs is much smaller, on the order of $4,000. This suggests that financial incentives of a few thousand dollars, similar to current sub- sidy levels, will stimulate fewer BEV purchases, but could stimulate more PHEV purchases.
In the ASC logit model we find that con- sumers’ utility for range exhibits decreasing returns. The linear and quadratic range coeffi- cients suggest an optimal electric range of 165 miles. A similar calculation for the latent class model suggests optimal ranges for Segments 1, 2 and 3 of 127.5, 216.7, and 300 miles, respec- tively. Segment 3 is the most likely to choose a BEV and is the least price sensitive, so it makes sense this segment is willing to pay for a longer range. Segment 1 is more likely to purchase a PHEV, such that a more cost-effective, shorter range vehicle may be sufficient. In the ASC logit model we also find evidence that prospective PHEV drivers are equally as motivated to com- mute primarily in all-electric mode, even though they do not face the same total range constraints as BEVs.
In the mixed logit model, we find that the aver- age respondent is willing to pay about $900 for free single-occupant HOV lane access. However, when restricting our analysis to the latent classes of consumers who would consider purchasing a
12. According to http://articles.latimes.com/2012/feb/ 21/business/la-fi-mo-holding-cars-longer-20120221, new car owners hold on to their vehicles for an average of 6 years.
PEV, we find an annual WTP of $876. In the ASC logit model, the coefficients on HOV lane access are positive, but that for BEVs is smaller than that for PHEVs and not statistically signifi- cant. This suggests that new car buyers who have access to HOV lanes are more likely to purchase PHEVs, and that government policies allowing free single-occupant HOV lane access increases consumer probability of purchasing PHEVs.
In the ASC logit model we find that charg- ing close to home access appears just as impor- tant for PHEVs as BEVs, even though PHEVs do not require the electric battery to be charged in order to drive the vehicle in gasoline mode. These results suggest that home charging is just as important to consumers considering a PHEV purchase. However, we also find evidence that BEV owners are more comfortable plugging into an outlet at their single-family residence while PHEV owners living in multifamily housing are also comfortable plugging into a less private or less exclusive outlet near their residential parking spot. Our latent class model similarly suggests that consumers living in a single-family house- hold are more likely to purchase BEVs. In the ASC logit model we also find evidence that the ability to charge at work is more important for BEV adoption than PHEV adoption.
The latent class model reveals three distinct consumer segments. About a quarter of the new car buyer population seems to be less urban, more conservative, and have strong negative prefer- ences for all PEVs. A third of the population has pro-environmental preferences and a tendency for early adoption. This is the only segment that does not have a strong negative preference for BEVs. The last segment, Segment 1, tends to be more urban, older, higher income, and more edu- cated. These consumers have a strong negative preference for BEVs but a strong positive pref- erence for PHEVs. This positive preference for PHEVs appears not to stem from environmental or early adopter preferences. This segment’s neg- ative preference for BEVs may be in part driven by less access to home charging.
The latent class results show that the BEV market may be constrained since less than a third of the new car buying population seems will- ing to consider purchasing a BEV, all else being equal. On the other hand, a much larger and more general population seems willing to con- sider purchasing a PHEV and even has a positive willingness to pay for this technology relative to a conventional gasoline vehicle. This suggests that the addition of PHEVs to the market may
712 ECONOMIC INQUIRY
stimulate PEV demand in consumer segments who would otherwise be unlikely to purchase a BEV. These findings also imply that many PHEV purchasers would not purchase a BEV, and such sales would represent growth in the over- all PEV market rather than cannibalization of the BEV market. We speculate that due to the strong negative preferences for BEVs in most of the population and cost differentials that are large relative to subsidy levels being considered by policy makers, much of the future growth of the PEV market will be driven by demand for PHEVs from Segment 1. Finally, our results
should be understood as a snapshot of preference early in the market’s development. As consumers learn more about these vehicles, and the choice set of vehicles expands on the supply side, their preferences will change in the future. A com- mon implicit assumption by many policymakers is that consumer preferences will change in a way that increases demand for PEVs as more of them hit the road. This assumption could be formally tested by administering our survey instrument several years from now to a similarly defined pop- ulation of prospective car buyers.
APPENDIX
TABLE A1 Definition of Variables
Variable Name Description
BEV Indicator for whether the chosen vehicle is a BEV PHEV Indicator for whether the chosen vehicle is a PHEV Range Electric range of chosen vehicle (miles) Refuel Refueling cost of chosen vehicle ($ per gallon equivalent) HOV Indicator for whether the chosen vehicle is granted free single-occupant access to high
occupancy vehicle lanes Small body Binary variable for if the respondent indicated that the vehicle she is most likely to select for
her next new vehicle purchase is a compact car, midsize car, or hatchback Household size Number of members of household, including respondent Household vehicles Number of vehicles in respondent’s household Age under 35 Binary variable for if respondent is less than 35 years old Age over 60 Binary variable for if respondent is more than 60 years old Outlet Binary variable that equals 1 if the respondent indicated an electrical outlet located within 100
feet of her home parking spot Single house Binary variable for if respondent lives in a one-family house detached from any other house
or a one-family house or condo attached to one or more houses Parking at work Binary variable for if the respondent indicated she parks her vehicle in a commercial lot or
garage while at work Commute under 20 mi Binary variable for if the respondent indicated that the shortest electric range she would need
for daily commute is under 20 miles Use gas mode daily Binary variable for if the respondent purchased a PHEV, she anticipates using gasoline mode
almost daily HOV access Binary variable that equals 1 if the respondent indicated she could use HOV lanes for her
daily commute or weekend travel Pro environment Binary variable for if the respondent indicates that environmental issues are very or extremely
important to her personally Early adopter Early adopter scorea
Liberal Binary variable for if the respondent identifies her political ideology as liberal (versus conservative or moderate)
Charging station density Publicly available level 2 charging stations within a 5 mile radius of population centroid of the Census Tract in which the respondent (in tens) lives as of December 2013
Gas price Average price per gallon of gasoline of the Census Tract in which the respondent lives in December 2013
High income (>$100k) Binary variable that equals 1 if the respondent’s household income is greater than $100,000 Low income (<$30k) Binary variable that equals 1 if the respondent’s household income is less than $30,000 College education Binary variable for if respondent has a Bachelor’s degree or higher education
aEarly adopter score is between 0 and 5. For each of the five following statements, one point is allocated toward the early adopter score if the respondent agrees or strongly agrees with the statement: (1) I usually try new products before other people do, (2) I often try new brands because I like variety and get bored with the same, (3) When I shop I look for what is new, (4) I like to be the first among my family and friends to try something new, and (5) I like to tell others about new brands or technology.
SHELDON, DESHAZO & CARSON: DEMAND FOR PLUG-IN HYBRIDS 713
REFERENCES
Achtnicht, M., G. Bühler, and C. Hermeling. “The Impact of Fuel Availability on Demand for Alternative-Fuel Vehicles.” Transportation Research Part D, 17, 2012, 262 – 9.
Axsen, J., and K. S. Kurani. “Early US Market for Plug- in Hybrid Electric Vehicles.” Transportation Research Record: Journal of the Transportation Research Board, 2139(1), 2009, 64 – 72.
. “Hybrid, Plug-in Hybrid, or Electric — What Do Car Buyers Want?” Energy Policy, 61, 2013, 532 – 43.
Beggs, S., S. Cardell, and J. Hausman. “Assessing the Poten- tial Demand for Electric Cars.” Journal of Economet- rics, 17, 1981, 1 – 19.
Bento, A., D. Kaffine, K. Roth, and M. Zaragoza-Watkins. “The Effects of Regulation in the Presence of Multiple Unpriced Externalities: Evidence from the Transporta- tion Sector.” American Economic Journal: Economic Policy, 6(3), 2014, 1 – 29.
Brownstone, D., D. S. Bunch, and K. Train. “Joint Mixed Logit Models of Stated and Revealed Preferences for Alternative-Fuel Vehicles.” Transportation Research Part B, 34, 2000, 315 – 38.
Bunch, D. S., M. Bradley, T. F. Golob, R. Kitamura, and G. P. Occhiuzzo. “Demand for Clean-Fuel Vehicles in California: A Discrete-Choice Stated Preference Pilot Project.” Transportation Research Part A, 27A(3), 1993, 237 – 53.
California Environmental Protection Agency. “Status Report on the California Air Resources Board’s Zero Emission Vehicle Program.” California Environmental Protection Agency, Air Resources Board, 2007.
Caperello, N. D., and K. S. Kurani. “Households’ Stories of Their Encounters with a Plug-in Hybrid Electric Vehi- cle.” Environment and Behavior, 44(4), 2012, 493 – 508.
Carson, R. T., and T. Groves. “Incentive and Informational Properties of Preference Questions.” Environmental and Resource Economics, 37, 2007, 181 – 210.
Carson, R. T., T. Groves, and J. A. List. “Consequentiality: A Theoretical and Experimental Exploration of a Single Binary Choice.” Journal of the Association of Environ- mental and Resource Economists, 1, 2014, 171 – 207.
DeShazo, J. R., T. L. Sheldon, and R. T. Carson. Forthcom- ing. “Designing Policy Incentives for Cleaner Technolo- gies: Lessons from California’s Plug-in Electric Vehicle Rebate Program.” Journal of Environmental Economics and Management, 2016.
Egbue, O., and S. Long. “Barriers to Widespread Adoption of Electric Vehicles: An Analysis of Consumer Attitudes and Perceptions.” Energy Policy, 48, 2012, 717 – 29.
Ewing, G., and E. Sarigöllü. “Assessing Consumer Pref- erences for Clean-Fuel Vehicles: A Discrete Choice Experiment.” Journal of Public Policy and Marketing, 19(1), 2000, 106 – 18.
Golob, T. F., R. Kitamura, M. Bradley, and D. S. Bunch. “Pre- dicting the Market Penetration of Electric and Clean- Fuel Vehicles.” The Science of the Total Environment, 134, 1993, 371 – 81.
Graham-Rowe, E., B. Gardner, C. Abraham, S. Skippon, H. Dittmar, R. Hutchins, and J. Stannard. “Mainstream Consumers Driving Plug-in Battery-Electric and Plug- in Hybrid Electric Cars: A Qualitative Analysis of Responses and Evaluations.” Transportation Research Part A, 46, 2012, 140 – 53.
Hensher, D. A., and W. H. Greene. “The Mixed Logit Model: The State of Practice.” Transportation, 30, 2003, 133 – 76.
Hidrue, M. K., G. R. Parsons, W. Kempton, and M. P. Gardner. “Willingness to Pay for Electric Vehicles and Their Attributes.” Resource and Energy Economics, 33, 2011, 686 – 705.
Kesternich, I., F. Heiss, D. McFadden, and J. Winter. “Suit the Action to the Word, the Word to the Action: Hypothetical Choices and Real Decisions in Medicare Part D.” Journal of Health Economics, 32(6), 2013, 1313 – 24.
Khan, M., and K. M. Kockelman. “Predicting the Market Potential of Plug-in Electric Vehicles Using Multiday GPS Data.” Energy Policy, 46, 2012, 225 – 33.
Kurani, K. S., T. Turrentine, and D. Sperling. “Testing Elec- tric Vehicle Demand in ‘Hybrid Households’ Using a Reflexive Survey.” Transportation Research Part D, 1(2), 1996, 131 – 50.
Kurani, K. S., R. R. Heffner, and T. Turrentine. “Driving Plug- in Hybrid Electric Vehicles: Reports from US Drivers of HEVs Converted to PHEVs, Circa 2006 – 07,” Insti- tute of Transportation Studies, University of California, Davis, 2008.
Louviere, J. J., and D. A. Hensher. “Using Discrete Choice Models with Experimental Design Data to Forecast Consumer Demand for a Unique Cultural Event.” Jour- nal of Consumer Research, 10, 1983, 348 – 61.
Louviere, J. J., D. A. Hensher, and J. D. Swait. Stated Choice Methods: Analysis and Applications. New York: Cam- bridge University Press, 2000.
Lusk, J. L., and T. C. Schroeder. “Are Choice Experiments Incentive Compatible? A Test with Quality Differen- tiated Beef Steaks.” American Journal of Agricultural Economics, 86, 2004, 467 – 82.
McFadden, D. “Conditional Logit Analysis of Qualitative Choice Behaviour,” in Frontiers in Econometrics, edited by P. Zarembka. New York: Academic Press, 1974, 105 – 42.
Qian, L., and D. Soopramanien. “Heterogeneous Consumer Preferences for Alternative Fuel Cars in China.” Trans- portation Research Part D, 16, 2011, 607 – 13.
Revelt, D., and K. Train. “Customer-Specific Taste Parame- ters and Mixed Logit: Households’ Choice of Electricity Supplier.” Working Paper No. E00-274, Department of Economics, University of California, Berkeley, 2000.
Scarpa, R., and J. M. Rose. “Design Efficiency for Non- Market Valuation with Choice Modelling: How to Mea- sure It, What to Report and Why.” Australian Journal of Agricultural and Resource Economics, 52(3), 2008, 253 – 82.
Sheldon, T. L., and J. R. DeShazo. “How Does the Presence of HOV Lanes Affect Plug-in Electric Vehicle Adoption in California? A Generalized Propensity Score Approach.” Working Paper, Department of Economics, University of South Carolina, 2015.
Shewmake, S., and L. Jarvis. “Hybrid Cars and HOV Lanes.” Transportation Research Part A, 67, 2014, 304 – 19.
Swait, J., and J. J. Louviere. “The Role of the Scale Parameter in the Estimation and Comparison of Multinomial Logit Models.” Journal of Marketing Research, 30, 1993, 305 – 14.
Train, K. E. “Recreation Demand Models with Taste Dif- ferences over People.” Land Economics, 74(2), 1998, 230 – 39.
Copyright of Economic Inquiry is the property of Wiley-Blackwell and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
