Need help with some writing assignments

AJ122001

Media2.pdf

Home >Government homework help >Need help with some writing assignments

OXFORD BULLETIN OF ECONOMICS AND STATISTICS, 80, 6 (2018) 0305–9049 doi: 10.1111/obes.12252

Does the Media Help the General Public in Understanding Inflation?*

David-Jan Jansen† and Matthias Neuenkirch‡,§

†De Nederlandsche Bank, Financial Stability Division, 1000 AB Amsterdam, The Netherlands (e-mail: [email protected]) ‡Faculty IV – Economics, University of Trier, 54296 Trier, Germany §CESifo, Munich, Germany

Abstract

This paper studies whether media information helps the general public in understanding inflation. We combine detailed Dutch household survey data on media usage, inflation perceptions, and inflation expectations. We find no evidence that more-often informed members of the general public do better in understanding inflation. In fact, more frequent readership of popular newspapers is associated with slightly less accurate inflation per- ceptions. There is also no evidence that usage of non-print media leads to more accurate of views on inflation. One implication of these results is that central banks might need to consider more direct ways of engaging with the general public.

I. Introduction

This paper asks whether media usage helps members of the general public in understanding inflation. Since the seminal work by Jonung (1981), it is well known that inflation percep- tions and expectations vary across sociodemographic clusters. One of his findings was that expected rates declined with age, which could be explained by different inflation experi- ences of cohorts over their lifetime. Recently, similar evidence is provided by Malmendier and Nagel (2016), who analyse over 50 years of data from the Reuters/Michigan Survey of Consumers. The list of relevant background factors is extensive and includes, in addition to age, variables such as gender, income, and education. The relevance of sociodemographic backgrounds for views on inflation has been affirmed in many studies, including Bryan and Venkatu (2001), Souleles (2004), Bruine de Bruin et al. (2010), and Ehrmann, Pfajfar and Santoro (2017).

JEL Classification numbers: D1, D8, E3, E5. *We thank Mark Boukes, Wändi Bruine de Bruin, Carin van der Cruijsen, Lena Dräger, Michael Ehrmann, Jakob

de Haan, Bernd Hayo, Richhild Moessner, Edith Neuenkirch, Florian Neumeier, Sanne Peeters, Ricardo Reis, Kalle Rinne, Maarten van Rooij, Stefania Rossi, Francesco Zanetti (the Editor), three anonymous referees and participants of the 2016 European Public Choice Conference, the 2016 World Finance Conference, the Workshop Household Surveys in Macroeconomics in Hamburg, the 2017 ESCB Research Conference at the Banco de España, and the 2018 Luxembourg Workshop on Household Finance and Consumption for useful comments and suggestions.

1186 Bulletin

The manner in which the media reports on economic developments could be one factor that explains heterogeneity in views on inflation, as media consumption also differs across sociodemographic backgrounds. Papers that study the potential role of economic reporting often take a perspective that differs from a traditional rational-expectations approach. In particular, these papers incorporate the idea that economic agents are not necessarily always fully attuned to the relevant news on the economy. In a general setting, (2003) builds on insight from information theory. He uses the idea of rational inattention, which follows from the notion that agents have only a finite capacity to process information. One of the implications is that the way in which people react to news depends on how the media presents it. Carroll (2003) builds a theoretical model, using insights from epidemiology. He assumes that individuals obtain information on the macroeconomy from the media, but only probabilistically. As a consequence, there is a delay before relevant macroeconomic news reaches every member of the public. Given that his model generates slow-moving changes in aggregate expectations, it provides a microfoundation for sticky-information type models (Mankiw and Reis, 2003). In line with the predictions of his model, Carroll finds evidence that people are better informed when there is more news coverage.

Although the idea that media coverage affects people’s views on inflation is appealing, the empirical evidence remains somewhat mixed. Some papers detect evidence supportive of news effects. Lamla and Lein (2015) find that media reporting has had meaningful impact on inflation perceptions and contributed to their sharp rise in the aftermath of the euro cash changeover. Dräger and Lamla (2017) reveal that when people have heard news on inflation, they are more likely to adjust their forecasts. Lamla and Maag (2012) document that media coverage affects disagreement of consumers. In particular, the level of disagreement increases with heterogeneity of media coverage and declines in the amount of reports. On the other side of the spectrum, Pfajfar and Santoro (2013) find only weak evidence in support of media effects. They test several predictions of Carroll’s theory using microdata from the Michigan Survey. Their conclusion is that hearing news on prices does not necessarily produce better forecasts, although it does increase the likelihood that people revise their expectations. Using survey data for Dutch households, Van der Cruijsen, Jansen and De Haan (2015) show that more intensive use of media information improves respondents’ understanding of the ECB. However, they find no empirical evidence that media consumption has a direct effect on the likelihood of formulating realistic inflation expectations. There are also papers where news effects only occur under certain conditions. For instance, Lamla and Lein (2015) do find that intensive news reporting improves the accuracy of inflation expectations. However, this effect only occurs if the news is not framed negatively. Dräger (2015) finds evidence that in Sweden, media effects have occurred, but then mostly during times of increasing inflation.1

Given the mixed evidence on news effects, this paper wants to look further at if and how media consumption matters for views on inflation. In line with this aim, a first distinctive aspect of this paper is the use of granular data on media usage. Related papers on news effects either rely on a national measure of media reporting (Carroll, 2003; Lamla and Lein, 2015), or use a general question from surveys on whether people have heard news about

1 There is also a related literature that studies knowledge of monetary policy and the role of central bank commu-

nication. See, e.g. Carvalho and Nechio (2014), Dräger, Lamla and Pfajfar (2016), Hayo and Neuenkirch (2018).

Role of Media in Understanding Inflation 1187

inflation (Pfajfar and Santoro, 2013; Dräger and Lamla, 2017). Our paper innovates by using detailed information on the usage of individual media sources. In particular, we use information on readership of eight national Dutch newspapers over a period of four years. An important benefit is that we can look into the quality of individual media outlets. We collect this information via the CentERpanel, an Internet-based panel which has been used to collect information regarding household finances of Dutch households since the early 1990s. Our primary focus is on print media. However, the 2017 wave of the survey also contains information of other media outlets: television, radio, social media and internet.

A second distinctive aspect of this paper is the fact that we take selection of media sources into account. We do so by including a question on participants’ political prefer- ences. Here, our paper follows the idea that people select those media sources that cater to their opinions. In Mullainathan and Shleifer (2005), a key assumption is that readers hold beliefs that they like to see confirmed. In an analysis of U.S. newspapers during the early 20th century, Gentzkow, Shapiro and Sinkinson (2014) find empirical confirmation that households prefer reading like-minded news. In a related contribution, Gentzkow and Shapiro (2006) show that firms will tend to distort information to make it conform to consumers’ prior beliefs. Also, Gentzkow and Shapiro (2011) find that, although ideolog- ical segregation of online news consumption is low, it is higher than segregation for most offline news consumption. There is also broader evidence that the ideologies that people subscribe to are relevant for views on the economy. Blinder and Krueger (2004) find that in the United States, ideology is the most important determinant of people’s opinions on major policy issues. Van der Cruijsen et al. (2015) find that people who have thought about their ideological positions are more likely to have realistic inflation expectations.

We find no real support for the idea that more-often informed members of the general public do better in understanding inflation. In fact, more frequent readership of some types of newspapers is associated with slightly less accurate inflation perceptions. We base these conclusions on a range of random-effects panel regressions that have absolute errors in perceptions or expectations as dependent variables. These conclusions remain unchanged in a number of robustness checks, such as adding lagged errors or instrumenting percep- tion errors. Also, a set of cross-sectional regressions using data collected in 2017 finds no evidence that non-print media outlets, such as television or Internet, help in understanding inflation. Overall, our paper casts further doubt on the idea that media usage contributes much to knowledge on economic developments. In doing so, our paper presents evidence that is in line with earlier findings by Pfajfar and Santoro (2013) for the U.S. and Van der Cruijsen et al. (2015) for the Netherlands. Compared to the first paper, we extend the literature by focusing on a different country than the U.S. Compared to the second paper, we extend the evidence by using longitudinal data rather than a single cross- section. Overall, this paper innovates by using detailed data on the actual media information that individuals use.

In a practical sense, one implication of the sobering results of this paper could be that monetary policy authorities might need to consider more direct ways of engaging with the general public. The effects of monetary policy operate, to a large extent, through expectations (Woodford, 2001). Blinder et al. (2008) suggest that the relevant actors are not necessarily only financial market participants but also households. Likewise, Bernanke (2010) argues that a better understanding of the macroeconomy may help the public in

1188 Bulletin

making more-informed decision. In a similar vein, Haldane (2017) argues that a lack of understanding may have significant societal costs, for instance in the form of excessive indebtedness or higher likelihoods of financial panics. Our results suggest then that reaching out to the general public may not be a straightforward exercise for central banks. The recent experience of some central banks in engaging directly with the public will be interesting to monitor, especially since first evidence on the effectiveness of these efforts is positive (Haldane and McMahon, 2018).

II. Research design and data

This paper uses detailed data on views on inflation, media usage, and political preferences. We gathered data on these three types of variables from four waves of a customized survey among Dutch households. This data collection took place between 2014 and 2017, always in the month of December. We collected the data by submitting short questionnaires (shown at the end of the paper) of roughly 2,500 members of the CentERpanel. This panel is the basis for an often-used survey among the Dutch population, the DNB Household Survey (DHS). The DHS has information on both economic and psychological aspects of household decision-making for a representative sample of Dutch citizens of 16 years and older. The DHS history extends back to the early 1990s, and the DHS forms the basis for a wide range of papers on household finance. In this paper, we also use a standard set of sociodemographic variables from the DHS in the regressions.2

To construct measures of inflation perceptions and expectations, we gave panel members two questions on consumer prices. The first question asked participants to indicate with what percentage consumer prices had approximately changed during the current calendar year. The question on perceptions gave respondents the option to select one out of eleven values, while there was also the possibility to say ‘I do not know’. A second question asked for people’s expectations regarding consumer price changes for the next calendar year. The question on expectations had eleven ranges to choose from, while adding an option ‘This is difficult to estimate’. In both questions, the respondents could also choose negative values or ranges, as the surveys took place at a time when the possibility of negative inflation was widely discussed.3 In the empirical analyses, we work with variables that transform the original answer categories into numeric values. The main purpose is to facilitate the interpretation. The transformation is implemented by making two assumptions. First, we restrict the top and bottom categories to −4.5%/+4.5% (for perceptions) and −5%/+5% (for expectations). Second, in case of inflation expectations, we use the midpoint of the intervals.

As dependent variables, we use absolute errors in inflation perceptions and expectations. Given that the surveys always took place in December, we compare the reported figures for perceptions against the annual change in consumer price inflation in that particular calendar year. We compare the reported numbers for expectations against realized inflation in year t + 1. The realizations are the year-on-year growth rates for the consumer price

2 See also: https://www.centerdata.nl/en/projects-by-centerdata/dnb- household-survey-dhs.

3 The DHS does not include information on inflation perceptions, and its standard question on inflation expectations

does not allow respondents to report negative numbers. Table D1 shows the empirical results generally hold when using the standard DHS variable on expectations (pr0).

Role of Media in Understanding Inflation 1189

(a)

(b)

Figure 1. Errors in inflation perceptions and expectations. Notes: Boxplots for the errors that Dutch households’ make in formulating inflation perceptions (panel A) and expectations (panel B). Boxplots are based on four surveys among Dutch households via the CentERpanel that took place in December of 2014, 2015, 2016, and 2017. Perceptions were asked for current year inflation, while expectations referred to next-year inflation. The boxes cover the region between the 25th and the 75th percentile. The realization for 2018 was not yet available at time of drafting.

index as reported by Statistics Netherlands. The realizations are currently available for the calendar years 2014–2017. Figure 1 show boxplots for the dependent variables per year.4 On average, both the absolute errors for perceptions and expectations are around 1 percentage point. However, there is a large heterogeneity, and the distribution is strongly skewed to the right, which means that many people make relatively large errors in pinpointing the levels of inflation.

The third question in our survey asked respondents to report how frequently they were informed about consumer prices via various print media outlets. Participants had the op- tion of selecting one or more of the major Dutch newspapers (Algemeen Dagblad, NRC Handelsblad, De Telegraaf, Trouw, De Volkskrant), one magazine (Elsevier), free news- papers (such as Metro), and local or regional print media. An important benefit of focusing on print media is the likely difference in outlook and nature of reporting. In terms of outlook, it is interesting to note that the NRC Handelsblad uses the motto ‘Lux et Liber- tas’, while the AD (founded in 1946) was originally closely affiliated to the NRC. In the case

4 Figure B1 reports the distributions for the perceptions and expectations themselves.

1190 Bulletin

of Elsevier, its editorial approach was historically close to both liberal and catholic political ideas, although the magazine characterizes its current approach as ‘Edge without ideology’. Trouw was founded during World War II by a group of orthodox-protestants, while De Volkskrant (founded in 1919) has its roots in the Catholic labour movement. De Telegraaf was founded in the late 19th century. In the 1960s and 1970s, it often took strong positions against socialist and progressive views. It has always been one of the most-read newspapers in The Netherlands. Currently, there are two main free newspapers (Metro and Spits), which are most readily available to users of public transport in urban areas.

Respondents reported the frequency of receiving information on prices via these eight outlets on a scale from 1 (never) to 5 (very frequently). Based on these replies, we construct indicators of the intensity of newspaper usage, where we also take the type of newspaper into account. For this latter point, we follow Boukes and Vliegenthart (2017) in categorizing the eight print outlets into two subcategories: (i) quality newspapers (NRC Handelsblad, Trouw, De Volkskrant, Elsevier) and (ii) popular newspapers (Algemeen Dagblad, free newspapers, De Telegraaf, local or regional newspapers). Then, for each of these two subcategories, we construct a variable for intensity of usage. Here, we follow Blinder and Krueger (2004) in constructing a variable labelled QH that measures the intensity of usage by dividing the number of outlets which are used either ‘frequently’ or ‘very frequently’ by the total number of outlets used. By construction, QH is bounded between zero and one, and a higher value indicates that a larger fraction of the available outlets are used intensively.

Most of the respondents only have a low intensity of media usage, a result that is in line with the findings of Blinder and Krueger (2004) for the U.S. The four panels in Figure 2 report the distributions of QH for the four survey waves. Each panel reports the distribution of QH for quality newspapers (solid line) and popular newspapers (dashed line). QH is plotted on the horizontal axis, while the vertical axis lists the percentage of respondents. Most of the distribution is massed at values between 0 and 0.25, showing that, at best, only one outlet is used at least frequently. This holds for quality newspapers as well as popular newspapers.

As noted above, for the 2017 survey wave we included an additional question on the usage of non-print media. We asked respondents to indicate how often they received in- formation on consumer prices via four sources: television, radio, Internet, or social media. Figure 3 shows the distribution of replies. The grey bars represent the percentage of respondents that use a particular media outlet (almost) never, while the white bars rep- resent those who use a particular outlet sometimes. The black bars denote those who use a particular media outlet (very) regularly. The media outlets are ranked in increasing order of usage on the horizontal axis. As can be seen, television was the most important source of information on prices in 2017, while newspapers were, in fact, least often used. Compared to the results by Van der Cruijsen et al. (2015), who described survey data collected in 2009, we find that the relative importance of newspapers for obtaining information on the economy has strongly decreased.

With a fourth and final question, we collected background information on the respon- dents’ political preferences. We asked participants to indicate whether they would use one or more of five terms to describe their political preferences. These five terms were as follows: liberal, socialist, christian-democratic, conservative and progressive. Our aim

Role of Media in Understanding Inflation 1191

(a) (b)

Figure 2. Intensity of newspaper usage. Notes: Distribution of survey replies regarding the intensity of newspaper usage (QH) for quality papers and popular (including local or regional) papers. The measure QH is computed as in Blinder and Krueger (2004) and denotes the fraction of newspapers that are read either frequently or very frequently.

Figure 3. Media usage in 2017. Notes: Histograms based on replies by survey respondents in 2017. Height of bars represents percentage of respondents that use a particular media type (almost) never, sometimes, or (very) regularly to be informed about consumer prices. For newspapers, the figure shows data for the newspaper that is least-often read and the newspaper that is most-often read. In addition, data are shown for usage of social media, Internet, radio and television.

1192 Bulletin

TABLE 1

Descriptive statistics for sociodemographic variables

(1) (2) (3) (4) 2014 2015 2016 2017

Male 0.59 0.55 0.55 0.55 Age: 16–34 0.06 0.08 0.12 0.11 Age: 35–49 0.17 0.22 0.22 0.21 Age: 50–64 0.33 0.31 0.29 0.27 Age: 65+ 0.43 0.39 0.37 0.40 University degree 0.17 0.14 0.15 0.15 Has children 0.25 0.31 0.32 0.32 Lives with partner 0.77 0.77 0.77 0.75 Urban: (Very) low 0.40 0.40 0.39 0.40 Urban: Moderate 0.19 0.22 0.22 0.21 Urban: (Very) high 0.41 0.37 0.39 0.39 Homeowner 0.79 0.78 0.78 0.79 Income: 0–30K 0.37 0.39 0.35 0.34 Income: 30–50K 0.26 0.26 0.26 0.23 Income: over 50K 0.16 0.13 0.15 0.16 Income: NA 0.21 0.22 0.24 0.27 Worker 0.41 0.44 0.47 0.45 Self-employed 0.05 0.05 0.05 0.05 Retired 0.38 0.34 0.32 0.34 Other occupation 0.16 0.17 0.16 0.16

Observations 914 1,760 1,871 1,714

Notes: Summary statistics for a range of sociodemographic variables from four waves of a survey among Dutch households. Numbers represent fractions of the sample per category. Variables include sex, age, education, a dummy for children, a dummy for living with a partner, degree of urbanization, home- ownership, income, and main occupation. Other occupations denotes people looking for a job, students, housemakers, and volunteer workers.

was presenting participants with a list of generally acceptable political labels. However, we did not want to suggest any kind of ranking to participants. Therefore, the questionnaire presented the five political labels as neutrally as possible, without attempting to suggest a particular order or interrelationship. As with the media variables, we also compute a mea- sure of broadness of political views. This variable (labelled QHp) measures the fraction of political labels that a person would use to describe his or her political preferences. In our samples, it turns out that respondents would usually use between one and three labels to describe their political views.

Turning to control variables, taken from the official DHS data sets, Table 1 gives summary statistics for a range of sociodemographic factors that we use in the regressions. The table presents sample means for several variables in each of the four years. The covariates are: gender, age (divided in four categories), education (a dummy for having obtained a university master degree), a dummy for the presence of children in the household, a dummy denoting whether the respondent lives with a partner, a mea- sure for the degree of urbanization around the main residence (in three categories), a dummy for homeownership, a variable measuring gross income (in four categories), and a

Role of Media in Understanding Inflation 1193

T A

B L

E 2

C or

re la

ti on

m at

ri x

of so

ci od

em og

ra ph

ic va

ri ab

le s

(1 )

(2 )

(3 )

(4 )

(5 )

(6 )

(7 )

(8 )

(9 )

(1 0)

(1 1)

(1 2)

(1 3)

(1 4)

(1 5)

(1 6)

(1 7)

(1 8)

(1 9)

(2 0)

(1 )

M al

e 1.

00 (2

) A

ge :

16 –3

4 −0

.1 0

1. 00

(3 )

A ge

: 35

–4 9

−0 .0

8 −0

.2 0

1. 00

(4 )

A ge

: 50

–6 4

−0 .2

3 −0

.3 6

1. 00

(5 )

A ge

: 65

+ 0.

12 −0

.2 7

−0 .4

2 −0

.4 8

1. 00

(6 )

U ni

ve rs

it y

de g

re e

0. 06

0. 08

−0 .0

5 −0

.0 8

1. 00

(7 )

H as

ch il

dr en

−0 .0

7 0.

17 0.

42 −0

.4 7

1. 00

(8 )

L iv

es w

it h

pa rt

ne r

0. 12

0. 05

−0 .0

3 −0

.0 5

0. 23

1. 00

(9 )

U rb

an :

(V er

y) lo

w −0

.0 3

−0 .0

7 0.

07 0.

12 1.

00 (1

0) U

rb an

: M

od er

at e

−0 .0

3 0.

03 −0

.0 5

−0 .0

2 0.

04 −0

.4 2

1. 00

(1 1)

U rb

an :

(V er

y) hi

gh 0.

05 0.

11 −0

.0 5

−0 .1

5 −0

.6 6

−0 .4

1 1.

00 (1

2) H

om eo

w ne

r 0.

06 −0

.0 4

0. 07

−0 .0

4 0.

04 0.

13 0.

28 0.

15 0.

03 −0

.1 7

1. 00

(1 3)

In co

m e:

0 -

30 K

−0 .2

4 0.

04 −0

.1 0

−0 .0

5 0.

12 −0

.1 5

−0 .0

4 −0

.1 6

1. 00

(1 4)

In co

m e:

30 –5

0K 0.

15 0.

03 −0

.0 3

0. 08

−0 .4

3 1.

00 (1

5) In

co m

e: ov

er 50

K 0.

24 −0

.0 7

0. 11

0. 04

−0 .0

9 0.

25 0.

04 0.

03 0.

16 −0

.3 0

−0 .2

2 1.

00 (1

6) In

co m

e: N

A −0

.0 7

−0 .0

2 0.

03 0.

03 −0

.0 3

−0 .0

4 0.

03 −0

.0 3

−0 .0

2 −0

.4 4

−0 .3

3 −0

.2 3

1. 00

(1 7)

W or

ke r

0. 14

0. 35

0. 23

−0 .6

1 0.

09 0.

34 0.

04 0.

09 −0

.2 4

0. 15

0. 18

−0 .0

3 1.

00 (1

8) S

el f-

em pl

oy ed

0. 06

0. 03

−0 .0

8 0.

08 0.

03 −0

.0 3

−0 .0

3 0.

06 −0

.0 5

0. 07

−0 .2

1 1.

00 (1

9) R

et ir

ed 0.

18 −0

.2 3

−0 .3

7 −0

.3 6

0. 82

−0 .0

6 −0

.4 3

0. 03

0. 05

0. 02

−0 .0

5 −0

.0 4

−0 .6

2 −0

.1 5

1. 00

(2 0)

O th

er oc

cu pa

ti on

−0 .2

5 0.

11 −0

.0 4

0. 12

−0 .1

6 −0

.0 9

0. 06

−0 .0

7 −0

.1 4

0. 24

−0 .1

9 −0

.1 6

0. 05

−0 .4

4 −0

.1 1

−0 .3

2 1.

N ot

es :

C or

re la

ti on

co ef

fi ci

en ts

fo r

a ra

ng e

of co

nt ro

l va

ri ab

le s

fr om

fo ur

w av

es of

a su

rv ey

am on

g D

ut ch

ho us

eh ol

ds .T

ab le

di sp

la ys

co rr

el at

io ns

si gn

ifi ca

nt at

th e

fi ve

pe rc

en t

le ve

1194 Bulletin

variable measuring a respondent’s main occupation (in four categories).5 In addition, Table 2 presents a correlation matrix. There are no indications that multicollinearity is an issue.6

A comparison between the means for our samples and those of the official DHS data sets – which would by construction be representative for the Dutch population – shows that our analysis uses, in particular, a relatively high proportion of males, from older age groups, who are well-educated. For instance, the percentage of respondents with a master degree lies around 15%, compared to a figure of 11% for the Dutch population. If anything, that would most likely affect our results in a positive sense, as these would be the individuals with better ability to assess inflationary developments.7

Turning to estimations, our empirical results are based on random-effects panel regres- sions. The analyses rely on a complete-cases sample, where, in addition, answers to all questions are available for at least two out of the four survey waves. This enables us to account for unobserved heterogeneity. To model the absolute error in inflation perceptions, we estimate:

|�pit − �t | = x′it � + �t + � + ui + �it (1) where �p denotes inflation perceptions, �t is realized consumer price inflation according to Statistics Netherlands, i indexes individuals, t denotes time in calendar years, and �t denotes time fixed-effects. The vector x has the sociodemographic covariates listed in Table 1, the QH variables that capture intensity of newspaper usage, and the QHp variable that measures the broadness of political preferences. To model the absolute error in inflation expectations, we start by estimate a similar model to Equation (1). In an additional estimation, we also add the absolute error in perceptions as a control variable. Furthermore, we shows that results based on Equation (1) are robust to a number of changes, such as using lagged dependent variables for perceptions and expectations.

To avoid misinterpretation of our empirical analysis, we have to emphasize that we cannot exclude the possibility of endogeneity. Inasmuch as the regressors are indeed en- dogenous, the estimated coefficients reflect conditional correlations rather than causal effects.

III. Baseline results for effects of media usage

Table 3 reports coefficients and standard errors for three random-effects panel regressions, where the dependent variable is absolute perceptions errors. Column 1 focuses on quality newspapers, while column 2 focuses on popular newspapers. Column 3 reports results for a combined model. The table also reports coefficients for the broadness of political views (QHp) as well as a number of covariates.

We only find a weak indication that more frequent media usage is associated with better knowledge of current inflation levels as the coefficient for quality newspapers is only

5 Detailed descriptions of these variables are in Table A1 of the Appendix. Table A2 in Appendix A lists summary

statistics for the official DHS waves for 2014–2017. 6 As noted by one of the referees, the actual prices observed during shopping would also impact inflation perceptions.

We are not able to control for this channel in the regressions, although to some extent variables such as income and education proxy for heterogeneity in consumption patterns.

7 Appendix E has further details on the age structure of our samples and the relationship with media consumption.

Role of Media in Understanding Inflation 1195

TABLE 3

Accuracy of inflation perceptions

(1) (2) (3) Quality Popular Combined

Political broadness −0.15** −0.17** −0.16** (0.07) (0.07) (0.07)

Quality newspapers −0.07 −0.15* (0.08) (0.08)

Popular newspapers 0.28*** 0.32*** (0.06) (0.07)

Male −0.03 −0.04 −0.04 (0.03) (0.03) (0.03)

Age: 16–34 0.08 0.07 0.08 (0.06) (0.06) (0.06)

Age: 50–64 −0.00 −0.01 −0.01 (0.04) (0.04) (0.04)

Age: 65+ 0.08 0.03 0.05 (0.06) (0.06) (0.06)

University degree −0.22*** −0.20*** −0.20*** (0.04) (0.04) (0.04)

Income: 0 – 30K 0.02 0.01 0.02 (0.04) (0.04) (0.04)

Income: over 50K −0.00 −0.01 −0.01 (0.04) (0.04) (0.04)

Income: NA 0.08* 0.07* 0.07* (0.04) (0.04) (0.04)

Homeowner −0.19*** −0.20*** −0.20*** (0.05) (0.05) (0.05)

Has children −0.01 −0.01 −0.01 (0.04) (0.04) (0.04)

R2 0.05 0.05 0.05 No. clusters 1,788 1,797 1,783 Observations 5,857 6,049 5,815

Notes: Coefficients and standard errors (in parentheses, clustered per household) for random-effects panel regressions. The dependent variable measures the absolute difference between inflation perceptions and realizations for the years 2014−2017. Columns 1 and 2 use an explanatory variable that measures the fraction of quality or popular newspapers that are read ‘frequently’ or ‘very frequently’. Column 3 combines both measures in a nested regression. All columns use a variable that measures the fraction of political preferences an individual would use to describe his or her views. Only selected coefficients are shown; regressions include all covariates listed in Table 1 as well as dummies for years, provinces, and religious positions. ‘Age: 35–49’ and ‘Income: 30K–50K’ serve as base categories. */**/*** denotes significance at the 10%/5%/1% level.

significant at the 10% level in the combined specification (column 3). In contrast, we find that more frequent readers of popular newspapers make larger errors in estimating current inflation as the coefficients of around 0.30 in columns 2 and 3 are significantly different from zero at the 1% level. This means that, compared to an individual who reads none of the popular papers, an individual who reads one of these outlets frequently has perceptions errors that are roughly 8 basis points higher.

1196 Bulletin

TABLE 4

Accuracy of inflation expectations

(1) (2) (3) Quality Popular Combined

Political broadness −0.11 −0.12* −0.12* (0.07) (0.07) (0.07)

Quality newspapers −0.02 −0.03 (0.07) (0.08)

Popular newspapers 0.15** 0.22*** (0.06) (0.07)

Male 0.01 −0.00 −0.00 (0.03) (0.03) (0.03)

Age: 16–34 0.06 0.04 0.06 (0.06) (0.05) (0.06)

Age: 50–64 0.05 0.05 0.04 (0.04) (0.04) (0.04)

Age: 65+ 0.00 −0.01 −0.01 (0.06) (0.06) (0.06)

University degree −0.15*** −0.13*** −0.13*** (0.04) (0.04) (0.04)

Income: 0 – 30K −0.02 −0.03 −0.03 (0.04) (0.04) (0.04)

Income: over 50K −0.06 −0.07* −0.06 (0.04) (0.04) (0.04)

Income: NA 0.04 0.03 0.04 (0.04) (0.04) (0.04)

Homeowner −0.16*** −0.16*** −0.17*** (0.05) (0.05) (0.05)

Has children −0.06 −0.06 −0.06 (0.04) (0.04) (0.04)

R2 0.06 0.06 0.06 No. clusters 1,713 1,724 1,708 Observations 4,225 4,371 4,192

Notes: Coefficients and standard errors (in parentheses, clustered per house- hold) for random-effects panel regressions. The dependent variable measures the absolute difference between inflation expectations and realizations for the years 2015−2017. Columns 1 and 2 use an explanatory variable that measures the fraction of quality or popular newspapers that are read ‘frequently’ or ‘very frequently’. Column 3 combines both measures in a nested regression. All columns use a variable that measures the fraction of political preferences an individual would use to describe his or her views. Only selected coefficients are shown; regressions include all covariates listed in Table 1 as well as dummies for years, provinces, and religious positions. ‘Age: 35–49’ and ‘Income: 30K–50K’ serve as base categories. */**/*** denotes significance at the 10%/5%/1% level.

Next, the estimates confirm the relevance of socio-demographic factors in explaining the heterogeneity in inflation perceptions as education and homeownership, and to some extent income, are important factors. For instance, individuals who obtained a university master degree have around 20 basis points lower perception errors. At the same time, it is important to stress that the amount of variation explained by our models is still quite low.

Role of Media in Understanding Inflation 1197

TABLE 5

Accuracy of expectations: Controlling for perceptions

(1) (2) (3) Quality Popular Combined

Accuracy perceptions 0.42*** 0.42*** 0.42*** (0.02) (0.02) (0.02)

Political broadness −0.05 −0.07 −0.06 (0.06) (0.06) (0.06)

Quality newspapers −0.02 −0.01 (0.07) (0.07)

Popular newspapers 0.06 0.12* (0.06) (0.06)

Male 0.03 0.02 0.02 (0.03) (0.03) (0.03)

Age: 16–34 0.01 −0.00 0.01 (0.05) (0.05) (0.05)

Age: 50–64 0.05 0.04 0.04 (0.04) (0.04) (0.04)

Age: 65+ −0.02 −0.02 −0.03 (0.06) (0.06) (0.06)

University degree −0.07** −0.05 −0.05 (0.03) (0.03) (0.03)

Income: 0 – 30K −0.04 −0.05 −0.05 (0.03) (0.03) (0.03)

Income: over 50K −0.05 −0.06* −0.05 (0.04) (0.04) (0.04)

Income: NA −0.01 −0.01 −0.01 (0.04) (0.04) (0.04)

Homeowner −0.07** −0.07* −0.07** (0.04) (0.04) (0.04)

Has children −0.05 −0.06* −0.05 (0.03) (0.03) (0.03)

R2 0.27 0.27 0.27 No. clusters 1694 1706 1689 Observations 4150 4295 4117

Notes: Coefficients and standard errors (in parentheses, clustered per household) for random- effects panel regressions. The dependent variable measures the absolute difference between inflation expectations and realizations for the years 2015−2017. Columns 1 and 2 use an explanatory variable that measures the fraction of quality or popular newspapers that are read ‘frequently’ or ‘very frequently’. Column 3 combines both measures in a nested regression. All columns use a variable that measures the fraction of political preferences an individual would use to describe his or her views as well as the measure for the accuracy of perceptions as defined in Table 3. Only selected coefficients are shown; regressions include all covariates listed in Table 1 as well as dummies for years, provinces, and religious positions. ‘Age: 35–49’ and ‘Income: 30K–50K’ serve as base categories. */**/*** denotes significance at the 10%/5%/1% level.

However, this seems not to be an uncommon finding in the literature. For instance, Ehrmann et al. (2017) report R2s of around 5%, while Dräger and Lamla (2017) report pseudo-R2s of less than 1%. In fact, Dräger and Lamla (2017) p. 958) explicitly mention that ‘the low explanatory power of the regressions clearly reveals more has to be done.’ From that perspective, it is interesting that the measure for the broadness of political views is also

1198 Bulletin

strongly associated with the extent to which an individual understands current inflation. All estimated coefficients for QHp in Table 3 are significantly different from zero at the 5% level. Though the economic significance is not very large, with coefficients of around −0.10, it is still non-negligible. The importance of political factors is, to the best of our knowledge, new in the literature on household expectation formation. Exploring the role of political views further may, therefore, provide an interesting avenue for future research.

Turning to how people formulate views on future inflation, Table 4 reports regression results for models where absolute expectation errors are the dependent variables. The results are broadly comparable to those reported in Table 3. Most importantly, we again find that more frequent readership of popular newspapers is associated with significantly higher absolute errors in inflation expectations. The coefficients for readership of quality newspapers are, again, negative but not significantly different from zero.

One apparent question is whether the differences these two types of newspapers are driven by the newspapers themselves or the readers. As indicated by the analysis in Appendix C, readers of quality and popular newspapers are indeed different. However, we control for these socio-demographic factors in all estimations, implying that we can rule out observable differences in the readership as driver of the results. Nevertheless, we cannot preclude that unobservable differences across readers might affect the accuracy of perceptions and expectations beyond media influence.

Since Jonung (1981), we know that perceptions are highly relevant for formulating inflation expectations. Therefore, we also estimated models based on Equation (1) where we added the accuracy of perceptions as an additional explanatory variable. Indeed, we find that the accuracy of perceptions captures most of the effects of the socio-demographic factors, as its inclusion leaves only a few of the coefficients for the other variables significant (Table 5). Interestingly though, we still find evidence that readership of popular newspapers is associated with a lower ability to formulate accurate inflation expectations (column 3).8

IV. Robustness and extensions

We perform four additional analyses to assess the robustness of our findings based on the model in Equation (1). In all cases, the conclusion remains that there is no real contribution of media consumption to understanding inflation.9 Table 6 summarizes the results. The first two columns of Table 6 show results when we introduce lagged dependent variables into the panel regressions. Given the persistence in actual inflation, current inflation is likely to be affected by its own lags. When adding a lagged dependent variable to the baseline model, the results with respect to popular newspapers carry over in case of inflation perceptions (column 1). We also find that the persistence in perceptions and expectations is not very pronounced (0.16 and 0.06 respectively). Indeed, in case of absolute expectation errors, current perceptions errors are more important than lagged expectation errors (column 2).

8 We checked whether the difference between columns 2 and 3 in Table 5 is driven (i) by the different sample sizes

or (ii) by collinearity between the indicators for readership of quality and popular newspapers. Restricting the sample in column 2 to the smaller sample of the nested specification yields an estimate of 0.12 for popular newspapers that is also significant at the ten-percent level.

9 We thank the referees for suggesting most of these robustness checks.

Role of Media in Understanding Inflation 1199

T A

B L

E 6

A cc

ur ac

y of

in fla

ti on

pe rc

ep ti

on s

an d

ex pe

ct at

io ns

: R

ob us

tn es

s te

st s

(1 )

(2 )

(3 )

(4 )

(5 )

P er

ce pt

io ns

E xp

ec ta

ti on

s E

xp ec

ta ti

on s

E xp

ec ta

ti on

s E

xp ec

ta ti

on s

L ag

.D ep

.V ar

. L

ag .D

ep .V

ar .

IV In

te ra

ct io

ns C

ou nt

er fa

ct ua

L (A

cc .p

er ce

pt io

ns )

0. 16

** *

(0 .0

2) L

(A cc

.e xp

ec ta

ti on

s) 0.

06 **

* (0

.0 2)

A cc

.p er

ce pt

io ns

0. 40

** *

(0 .0

2) 0.

67 **

* (0

.0 6)

0. 43

** *

(0 .0

4) 0.

39 **

* (0

.0 2)

… P

er ce

pt io

n er

ro r

< 1

−0 .1

5* **

(0 .0

3) P

ol it

ic al

br oa

dn es

s −0

.2 3*

** (0

.0 8)

−0 .0

2 (0

.0 8)

0. 06

(0 .0

8) −0

.0 4

(0 .0

9) −0

.0 6

(0 .0

6) …

A cc

.p er

ce pt

io ns

−0 .0

2 (0

.0 7)

Q ua

li ty

ne w

sp ap

er s

−0 .0

4 (0

.1 0)

−0 .0

5 (0

.0 8)

−0 .0

5 (0

.0 9)

0. 17

(0 .1

2) −0

.0 2

(0 .0

7) …

A cc

.p er

ce pt

io ns

−0 .1

6 (0

.1 2)

P op

ul ar

ne w

sp ap

er s

0. 26

** *

(0 .0

8) 0.

06 (0

.0 8)

0. 01

(0 .0

8) 0.

08 (0

.0 9)

0. 12

* (0

.0 6)

… A

cc .p

er ce

pt io

ns 0.

03 (0

.0 7)

R 2

0. 14

0. 28

0. 25

0. 27

0. 28

N o.

cl us

te rs

1, 59

4 1,

30 8

1, 96

4 1,

68 9

1, 68

9 O

bs er

va ti

on s

3, 92

9 2,

42 3

2, 41

7 4,

11 7

4, 11

N ot

es :

C oe

ffi ci

en ts

an d

st an

da rd

er ro

rs (i

n pa

re nt

he se

s, cl

us te

re d

pe r

ho us

eh ol

d) fo

r ra

nd om

-e ff

ec ts

pa ne

l re

g re

ss io

ns .

T he

de pe

nd en

t va

ri ab

le m

ea su

re th

e ab

so lu

te di

ff er

en ce

be tw

ee n

in fl

at io

n pe

rc ep

ti on

s (e

xp ec

ta ti

on s)

an d

re al

iz at

io ns

fo r

th e

ye ar

s 20

14 −2

01 7

(2 01

5− 20

17 ).

C ol

um ns

1 an

d 2

in cl

ud e

th e

re sp

ec ti

ve la

gg ed

de pe

nd en

t va

ri ab

le .

C ol

um n

3 us

es la

gg ed

pe rc

ep ti

on er

ro rs

as an

in st

ru m

en t

fo r

cu rr

en t

pe rc

ep ti

on er

ro rs

(F -s

ta ti

st ic

of fi

rs t

st ag

e: F

(1 ,1

29 8)

= 15

9. 41

** *)

. C

ol um

n 4

in cl

ud es

in te

ra ct

io ns

of pe

rc ep

ti on

er ro

rs w

it h

bo th

ne w

sp ap

er in

te ns

it y

va ri

ab le

s as

w el

la s

th e

in di

ca to

r fo

r po

li ti

ca lb

ro ad

ne ss

.C ol

um n

5 in

cl ud

es an

in te

ra ct

io n

of th

e pe

rc ep

ti on

er ro

r w

it h

an in

di ca

to r

va ri

ab le

fo r

pe rc

ep ti

on er

ro rs

of on

e pe

rc en

ta ge

po in

t or

lo w

er .

O nl

y se

le ct

ed co

ef fi

ci en

ts ar

e sh

ow n;

re g

re ss

io ns

in cl

ud e

al l

co va

ri at

es li

st ed

in T

ab le

1 as

w el

l as

du m

m ie

s fo

r ye

ar s,

pr ov

in ce

s, an

d re

li gi

ou s

po si

ti on

s. */

** /*

** de

no te

s si

gn ifi

ca nc

e at

th e

10 %

/5 %

/1 %

le ve

1200 Bulletin

A second robustness check addresses the issue that respondents filled in current-year and next-year inflation in two subsequent questions in our survey. It could be the case that people were unsure about inflation developments in the first place, and decided to fill in the exact same numbers for these two questions. Thus, we compared the answers to both questions and indeed find that quite a few – though by no means all – of the answers are within an interval of ±0.5% (the percentage is 76%, see Table B1 in the Appendix for details). To assess the possible effects, we implemented an instrumental-variables (IV) approach, where we use lagged perception errors to instrument current perception errors in a model that has expectation errors as dependent variable. The outcomes of this regression can be compared against those in column 3 of Table 5. We find that the coefficient on perception errors in the IV approach (column 3 of Table 6) is significantly larger than in the baseline results. A further analysis (available upon request) reveals this increase is not due to the decrease in the number of observations in the IV model. More importantly, we still find no evidence that more frequent readership of various types of newspapers leads to a better understanding of inflation.

A third robustness analysis studies how perceptions interact with media usage in formulating thoughts about future inflation. To assess this, column 4 of Table 6 shows results for a panel regression where we interact perception errors with the two newspaper

TABLE 7

The contribution of non-print media to accurate perceptions

(1) (2) (3) (4) (5) Soc. Med. Internet Radio TV Combined

Political broadness −0.23* −0.24* −0.22* −0.23* −0.24* (0.13) (0.13) (0.13) (0.13) (0.13)

Quality newspapers 0.00 0.04 −0.10 −0.10 0.00 (0.20) (0.20) (0.20) (0.20) (0.21)

Popular newspapers 0.32** 0.39*** 0.35** 0.34** 0.30** (0.15) (0.15) (0.14) (0.14) (0.15)

Info via Social Media 0.04 0.03 (0.02) (0.03)

Info via Internet 0.01 −0.03 (0.02) (0.03)

Info via Radio 0.04* 0.02 (0.02) (0.03)

Info via Television 0.05* 0.04 (0.02) (0.03)

R2 0.07 0.07 0.07 0.07 0.07 No. clusters 1,322 1,324 1,331 1,346 1,298 Observations 1,564 1,572 1,577 1,599 1,538

Notes: Coefficients and standard errors (in parentheses, clustered per household) for cross- section least squares regressions where the dependent variable measures the accuracy of inflation perceptions in 2017. Estimations include five-point scale variables for the usage of other media sources (social media, Internet, radio and television) that range from 1 (never) to 5 (very frequently). Only selected coefficients are shown; regressions include all covariates listed in Table 1 as well as dummies for provinces and religious positions. */**/*** denotes significance at the 10%/5%/1% level.

Role of Media in Understanding Inflation 1201

TABLE 8

Probability of over-, accurate, and underestimation of perceptions and expectations

Coefficients Prob(Under) Prob(Correct) Prob(Over)

Panel A: Perceptions Political broadness −0.15 0.03 0.02 −0.05

(0.10) (0.02) (0.01) (0.03) Quality newspapers −0.19 0.04 0.02 −0.06

(0.15) (0.03) (0.02) (0.04) Popular newspapers 0.33*** −0.06*** −0.04*** 0.10***

(0.11) (0.02) (0.01) (0.03) Age: 16–34 −0.06 0.01 0.01 −0.02

(0.08) (0.02) (0.01) (0.03) Age: 50–64 −0.09 0.02 0.01 −0.03

(0.07) (0.01) (0.01) (0.02) Age: 65+ −0.22** 0.04** 0.03** −0.07**

(0.10) (0.02) (0.01) (0.03)

No. clusters 1,783 Observations 5,815

Panel B: Expectations Political broadness 0.06 −0.01 −0.00 0.02

(0.12) (0.02) (0.01) (0.03) Quality newspapers 0.09 −0.02 −0.01 0.02

(0.17) (0.03) (0.01) (0.05) Popular newspapers 0.43*** −0.09*** −0.03*** 0.12***

(0.13) (0.03) (0.01) (0.04) Age:16–34 0.21** −0.04** −0.02** 0.06**

(0.10) (0.02) (0.01) (0.03) Age: 50–64 0.05 −0.01 −0.00 0.01

(0.08) (0.02) (0.01) (0.02) Age: 65+ −0.05 0.01 0.00 −0.01

(0.12) (0.02) (0.01) (0.03)

No. clusters 1,708 Observations 4,192

Notes: Coefficients and average marginal effects alongside their standard errors (in paren- theses, clustered per household) for random-effects panel ordered probit regressions. The dependent variables are ternary indicators measuring whether perceptions or expectations were overestimated (category 1), correctly estimated (category 0), or underestimated (cate- gory -1). Only selected coefficients and marginal effects are shown; regressions include all covariates listed in Table 1 as well as dummies for years, provinces, and religious positions. ‘Age: 35–49’ serves as base category. */**/*** denotes significance at the 10%/5%/1% level.

intensity variables as well as the indicator for political broadness. All interactions, however, are individually and jointly (�2(3) = 1.91; P = 0.59) insignificant.

The final robustness analysis looks further at the relevance of accurate perceptions for formulating views on next-year inflation. Here, we distinguish between ‘good’ and ‘bad’ perceptions and see if there is any asymmetry in the effect on expectations. The median error for perceptions is roughly one percentage point in our sample. Hence, we utilize this value as the threshold for ‘good’ versus ‘bad’ perception errors. We then interact the

1202 Bulletin

perception error with a dummy for good perceivers. Column 5 in Table 6 indicates that the effect of perception errors on expectation errors is weaker (0.24) for good perceivers compared to bad perceivers (0.39). This finding implies that good perceivers formulate their expectations more independently from perceptions than bad perceivers.

In a first extension, we study the effects of usage of other media outlets. Here, we utilize the data that we collected from the additional question in the 2017 wave of the survey. We include variables that measure the usage of television, radio, Internet and social media in cross-section least-squares regressions that use perception errors as dependent variables. The regressions also include the measures for political broadness and readership of two types of newspapers. Table 7 shows coefficients and standard errors. We find no evidence that more frequent usage of the other media outlets helps in formulating inflation perceptions. In contrast, there is the suggestion that those who receive more information via radio or television have a few basis points higher perception errors (columns 3 and 4).

In a second extension, we study what drives the probability of misestimating inflation. This extension is motivated by the fact that respondents generally overestimate the levels of inflation, as is evident from the distributions shown in Appendix B. To further look into this issue, we ran two random-effects panel ordered probit regressions. Table 8 has coefficients and average marginal effects. Panel A of this table focuses on perceptions, while Panel B has results for expectations. The dependent variables are ternary indicators that take the value 1 if inflation was overestimated, 0 if inflation was correctly estimated, and −1 if inflation was underestimated.10 Overall, we find that more frequent usage of popular newspapers increases (decreases) the likelihood of overestimating (underestimating) inflation. This finding confirms the main findings for absolute errors in Tables 3–5.

V. Conclusions

This paper finds no real support for the hypothesis that more-often informed members of the general public have a better understanding of inflation. In contrast, the estimates indicate that a higher frequency of receiving information via popular newspapers goes hand in hand with slightly larger errors in inflation perceptions. Overall, the estimation results suggest, if anything, only a modest economic significance. These essentially negative conclusions on media effects are broadly in line with Pfajfar and Santoro (2013) and Van der Cruijsen et al. (2015). Further investigating why news may not necessarily be helpful is important in future research. Exploring the role of quality of the information that is provided would be one way to do so.

The sobering nature of this paper’s findings offers food for thought for monetary policymakers. Nowadays, central banks increasingly focus on the role of expectations management via communication and transparency (Woodford, 2001; Blinder et al., 2008). Indeed, central banks have been using communication more actively in recent years. A recent survey of central bank heads finds that communication might become even more important to central banks in the near future (Blinder et al., 2017). Our findings suggest that reaching out to the general public via traditional media channels may not be a

10 In case of perceptions, we consider those values that are closest to the actual realization as correctly estimated. For

instance, if the realization is 0.6% then 1% is assigned to the category ‘correctly estimated.’ In case of expectations, we utilize the intervals provided in the questionnaire to create the category ‘correctly estimated’.

Role of Media in Understanding Inflation 1203

straightforward exercise for central banks. It is, therefore, interesting to see recent efforts by central banks to further engage with the public. For instance, in late-2017, the Bank of England started augmenting the information in its Inflation Report with content aimed explicitly at a less-specialist audience. A first analysis of this new approach, based on a controlled experiment, suggest that more straightforward communication indeed improves the likelihood that people’s beliefs align with the forecast of the central bank (Haldane and McMahon, 2018). Building on these initial findings, it would be interesting to explore fur- ther to what extent direct engagement with the public might be a complement or substitute for relying on the more traditional media channels.

Survey questions

We report the four relevant questions from the four questionnaires, which were submitted to the roughly 2,500 members of the CentERpanel in the month December of the four years 2014–2017. In 2014, these four questions were part of a longer questionnaire (available upon request). The response rates to the four surveys were, respectively, 70.2%, 89.1%, 92.5%, and 88.0%.

With what percentage have Dutch consumer prices approximately changed in 2014 (2015/2016/2017) compared to 2013 (2014/2015/2016)?

• Prices declined by more than 4%. • Prices declined by 4%. • Prices declined by 3%. • Prices declined by 2%. • Prices declined by 1%. • Prices remained unchanged. • Prices increased by 1%. • Prices increased by 2%. • Prices increased by 3%. • Prices increased by 4%. • Prices increased by more than 4%. • Do not know

With what percentage will Dutch consumer prices approximately change in 2015 (2016/2017/2018) compared to 2014 (2015/2016/2017)?

• Prices will decline by more than 4%. • Prices will decline by between 3 and 4%. • Prices will decline by between 2 and 3%. • Prices will decline by between 1 and 2%. • Prices will decline by between 0 and 1%. • Prices will remain more or less unchanged. • Prices will increase by between 0 and 1%. • Prices will increase by between 1 and 2%. • Prices will increase by between 2 and 3%. • Prices will increase by between 3 and 4%.

1204 Bulletin

• Prices will increase by more than 4%. • This is difficult to estimate

How often did you read about Dutch consumer prices in these newspapers or magazines in 2014 (2015/2016/2017)?

• Algemeen Dagblad • Elsevier • Free newspaper (e.g. Metro) • NRC Handelsblad • Regional or local newspaper • De Telegraaf • Trouw • De Volkskrant 1. Never 2. Almost never 3. Sometimes 4. Frequently 5. Very frequently 6. Do not know

Would you use one of the following terms to describe your political preferences? (it is possible to give more than one answer)

• Liberal • Socialist • Christian-Democratic • Conservative • Progressive 1. No 2. Yes 3. Do not know

Additional question for 2017 wave: How often did you hear or read about Dutch consumer prices via these media outlets in 2017?

• Internet • Social media • Radio • Television

1. Never 2. Almost never 3. Sometimes 4. Frequently 5. Very frequently 6. Do not know

Role of Media in Understanding Inflation 1205

Appendix A Variable description & summary statistics for official DHS data

Table A1 gives a full overview of all variables used in the regressions, as well as a short description on their construction.

TABLE A1

Variable description

Variable Description

Absolute perception and expectation errors Accuracy perceptions Absolute difference between inflation perceptions and realizations Accuracy expectations Absolute difference between inflation expectations and realizations

Political preferences and intensity of newspaper usage Political broadness Fraction of political preferences used to describe views Quality newspapers Fraction of quality newspapers that are read ‘(very) frequently’ Popular newspapers Fraction of popular newspapers that are read ‘(very) frequently’

Media usage in 2017 (five-point scale variables from 1 = never to 5 = very frequently) Social media Usage of social media Internet Usage of the Internet Radio Usage of radio Television Usage of television

Socio-demographic control variables (binary dummies) Male Male respondents Age: 16–34 Age 16–34 Age: 35–49 Age 35–49 Age: 50–64 Age 50–64 Age: 65+ Age 65 and older University degree University Master degree Has children Children in household Lives with partner Partner in household Urban: (Very) low Area with more than 1,500 addresses per km2

Urban: Moderate Area with 1,000–1,500 addresses per km2

Urban: (Very) high Area with less than 1,000 addresses per km2

Homeowner Homeowner Income: 0–30K Gross household income lower than 30,000 euros Income: 30–50K Gross household income between 30,000 and 50,000 euros Income: over 50K Gross household income more than 50,000 euros Income: NA Gross household income not reported Worker Workers Self-employed Self-employees Retired Retired Other occupation Looking for a job, students, housemakers, volunteer workers No religion No religious denomination (Roman-)Catholic (Roman-)Catholic Protestant Protestant Other religion Other religious denomination

1206 Bulletin

TABLE A2

Descriptive statistics for official DHS data sets

(1) (2) (3) (4) 2014 2015 2016 2017

Male 0.49 0.49 0.49 0.49 Age: 16–34 0.18 0.15 0.20 0.20 Age: 35–49 0.24 0.22 0.20 0.22 Age: 50–64 0.18 0.22 0.19 0.20 Age: 65+ 0.18 0.23 0.22 0.23 University degree 0.11 0.09 0.11 0.10 Has children 0.58 0.55 0.54 0.48 Lives with partner 0.86 0.85 0.85 0.80 Urban: (Very) low 0.41 0.42 0.41 0.41 Urban:Moderate 0.20 0.22 0.21 0.21 Urban: (Very) high 0.39 0.36 0.38 0.38 Homeowner 0.81 0.81 0.79 0.78 Income: 0–30K 0.18 0.21 0.19 0.20 Income: 30–50K 0.13 0.13 0.13 0.13 Income: over 50K 0.07 0.07 0.07 0.07 Income: NA 0.62 0.59 0.61 0.60 Worker 0.41 0.38 0.41 0.42 Self-employed 0.04 0.04 0.04 0.04 Retired 0.15 0.19 0.18 0.18 Other occupation 0.28 0.29 0.26 0.27

Notes: Based on 2014−2017 waves of DNB Household Survey. Other occupations denotes people looking for a job, students, housemakers, and volunteer workers.

Table A2 presents summary statistics for sociodemographic variables in the DNB Household Survey waves between 2014 and 2017. These means for the official DHS data set – by construction representative for the Dutch population – can be compared against the means for the samples used in this paper, which are listed in Table 1. Such a comparison shows that the samples in this paper have, in particular, an overrepresentation of males, from older age categories, who are well educated.

B Distribution of inflation perceptions and expectations

Figure B1 shows boxplots for perceptions (top panel) and expectations (bottom panel). In addition, the blue diamonds denote the realizations for inflation in respective calendar years.

As shown in the top panel of Figure B1, between 2014 and 2016 inflation perceptions showed a downward trend, in line with realized inflation. Also, in each of the four years, the means (denoted by the grey circles) as well as the medians (denoted by the plus-signs) of inflation perceptions are in line with realized inflation. However, for each year, but in particular for 2016, there was a large degree of variation in perceptions. Also, inflation per- ceptions were generally higher than the actual levels of inflation, which is an important mo-

Role of Media in Understanding Inflation 1207

(a)

(b)

Figure B1. Inflation perceptions and expectations. Notes: Boxplots for Dutch households’ perceptions (panel A) and expectations (panel B) of consumer price changes. The blue diamonds denote the realization of consumer price inflation according to Statistics Nether- lands. Boxplots are based on four surveys among Dutch households via the CentERpanel that took place in December of 2014, 2015, 2016, and 2017. Perceptions were asked for current year inflation, while expectations referred to next-year inflation. The boxes cover the region between the 25th and the 75th percentile.

tivation for the analysis described in Table 8 of the main text. The bottom panel of Figure B1 suggests similar conclusions for inflation expectations.

Next, we compare the responses to the questions on perceptions and expectations to check if respondents tend to answer both questions similarly or differently. The answers are categorized as approximately equal if the difference is no larger than ±0.5%. Table B1 provides an overview over the four waves of the survey. Indeed, roughly 75% percent of the respondents provide similar answers to both questions. Focusing on those respondents with different answers we find that, in 2014, perceptions were higher than expectations, whereas we find the opposite for 2016 and 2017. These findings coincide with the decrease in actual inflation from 2014 to 2015 (from 1.0% to 0.6%) and the increase from 2016 to 2017 (from 0.3% to 1.4%).

1208 Bulletin

TABLE B1

‘Transition matrix’ of perceptions and expectations

Wave �p >> �e �p ≈ �e �p << �e 2014 239 (17.3%) 1,021 (74.1%) 118 (8.6%) 2015 161 (9.0%) 1,442 (80.9%) 180 (10.1%) 2016 182 (9.1%) 1,512 (75.2%) 316 (15.7%) 2017 191 (10.4%) 1,351 (73.5%) 297 (16.1%)

Notes: Table shows the relationship between a respondent’s answers to the questions on inflation perceptions and expectations. The answers are categorized as approximately equal ( �p ≈ �e ) if the difference is no larger than ±0.5%.

C Newspaper readership

This section studies intensity of newspaper usage. Table C1 shows coefficients for two random-effects panel regressions, where the dependent variables measure the intensity of newspaper readership using the QH variables. As explanatory variables, we employ the political broadness measures and a set of sociodemographic factors. We find evidence for

TABLE C1

Intensity of newspaper usage

(1) (2) Quality Popular

Political broadness 0.05*** (0.01) 0.04*** (0.01) Male 0.01 (0.00) 0.02*** (0.01) Age: 16–34 0.00 (0.01) −0.03** (0.01) Age: 50–64 0.02*** (0.01) 0.04*** (0.01) Age: 65+ 0.04*** (0.01) 0.07*** (0.01) University degree 0.06*** (0.01) −0.06*** (0.01) Income: 0 – 30K −0.01 (0.01) 0.01 (0.01) Income: over 50K 0.02*** (0.01) 0.02* (0.01) Income: NA −0.00 (0.01) 0.01* (0.01) Homeowner 0.02*** (0.01) 0.01 (0.01) Has children −0.01 (0.01) −0.01 (0.01) (Roman-) Catholic −0.01 (0.01) 0.04*** (0.01) Protestant −0.01 (0.01) 0.01 (0.01) Other 0.00 (0.01) −0.03*** (0.01) R2 0.10 0.09 No. clusters 1,865 1,872 Observations 6,428 6,629

Notes: Coefficients and standard errors (in parentheses, clustered at the household level) for random-effects linear panel regressions. The dependent variables measure the fraction of quality and popular news- papers that are read ‘frequently’ or ‘very frequently’. All columns use an explanatory variable that measures the fraction of political prefer- ences an individual would use to describe his or her views. Only se- lected coefficients are shown; regressions include all covariates listed in Table 1 as well as dummies for years and provinces. ‘Age: 35– 49’, ‘Income: 30K–50K’, and ‘No religion’ serve as base categories. */**/*** denotes significance at the 10%/5%/1% level.

Role of Media in Understanding Inflation 1209

a positive relationship between QHp and QH. In addition, factors such as age, education, and religion are relevant for differences in readership.

D Results using standard expectations variable from DHS

The official DNB Household Survey data does not include information on inflation percep- tions, and the standard DHS question on inflation expectations does not allow respondents to report negative numbers. For these two reasons, this paper decided to work with a customized questionnaire. Nevertheless, it is interesting to compare the baseline results reported in the paper with regressions that rely on the standard DHS variable on inflation expectations. This variable (pr0) measures on a ten-point scale that ranges from 1% to 10% the replies to the question: ‘What is the most likely (consumer) prices increase over the next twelve months, do you think?’

Table D1 presents three random-effects panel regression models. The dependent vari- able measures the accuracy of the pr0 variable for the calendar years 2015 till 2017. The findings from the baseline regressions in Table 4 are broadly confirmed. The coefficients for quality newspapers are negative, but insignificant. Readership of popular newspapers is associated with higher expectations error, though the coefficients are also insignificant.

TABLE D1

Accuracy of expectations using pr0 variable from DNB Household Survey

(1) (2) (3) Quality Popular Combined

Political broadness −0.21** −0.25*** −0.22** (0.09) (0.09) (0.09)

Quality newspapers −0.13 −0.16 (0.11) (0.12)

Popular newspapers 0.07 0.11 (0.09) (0.10)

R2 0.20 0.21 0.21 No. clusters 1,617 1,628 1,614 Observations 4,026 4,167 3,996

Notes: Selected coefficients and standard errors (in parentheses, clus- tered at the household level) for random-effects panel regressions. The dependent variable measures the accuracy of inflation expectations us- ing the standard question from the DHS. This variable (pr0) measures on a ten-point scale that ranges from 1% to 10% the replies to the ques- tion: ‘What is the most likely (consumer) prices increase over the next twelve months, do you think?’ Columns 1 and 2 use an explanatory variable that measures the fraction of quality and popular newspapers that are read ‘frequently’ or ‘very frequently’. Column 3 combines both measures in a nested regression. All columns use a variable that mea- sures the fraction of political preferences an individual would use to describe his or her views. Only selected coefficients are shown; re- gressions include all covariates listed in Table 1 as well as dummies for years, provinces, and religious positions. */**/*** denotes significance at the 10%/5%/1% level.

1210 Bulletin

E Role of age structure

This section focuses on the role of age structure in our sample. First, we compare the age structure of our sample against the age structure of the Dutch population (aged 16 or older). We take our data from Statistics Netherlands. Table E1 compares the percentage of respondents versus the percentage within the same age group in the Dutch population. It turns out we have oversampled the elderly. However, if anything, that would most likely affect our results in a positive sense, as these would be the individuals with better ability to assess inflationary developments (see also Table 8).

We now turn to the question of whether media usage depends on age. Table E2 presents selected coefficient from regressions where the dependent variables measures the frequency of usage of various media sources. To ensure comparability across models, we present cross- section regressions based on the data collected for 2017. We find that older age groups are more likely to obtain information via television and radio. Young respondents are more likely to be informed via social media.

TABLE E1

Age structure of sample compared to Dutch population

2014 2015 2016 2017

Sample NL Sample NL Sample NL Sample NL

16–34 0.06 0.28 0.08 0.29 0.12 0.29 0.11 0.29 35–49 0.17 0.26 0.22 0.25 0.22 0.25 0.21 0.25 50–64 0.33 0.25 0.31 0.25 0.29 0.25 0.27 0.26 65 – 0.43 0.21 0.39 0.22 0.37 0.22 0.40 0.23

Notes: Numbers present percentages of either the samples or Dutch population (age 16 over over) per year. Data for NL is taken from Statistics Netherlands.

TABLE E2

Media usage by age group in 2017

(1) (2) (3) (4) (5) (6) Quality Popular Soc. Med. Internet Radio TV

Age: 16–34 0.01 −0.02 0.27*** 0.14 −0.05 −0.14 (0.01) (0.01) (0.10) (0.10) (0.09) (0.10)

Age: 50–64 0.02*** 0.05*** −0.11 −0.02 0.12 0.27*** (0.01) (0.01) (0.08) (0.08) (0.08) (0.07)

Age: 65+ 0.04*** 0.09*** −0.09 −0.09 0.25** 0.49*** (0.01) (0.02) (0.11) (0.12) (0.12) (0.10)

R2 0.08 0.11 0.05 0.03 0.04 0.12 No. clusters 1,572 1,610 1,566 1,579 1,581 1,606 Observations 1,952 2,003 1,934 1,955 1,960 1,999

Notes: Selected coefficients and standard errors (in parentheses, clustered at the household level) for least-squares regressions. The dependent variable measures the frequency of usage of various media outlets. The regressions also contain the standard set of covariates listed in Table 1. ‘Age: 35–49’ serves as base category. */**/*** denotes significance at the 10%/5%/1% level.

Final Manuscript Received: June 2018

Role of Media in Understanding Inflation 1211

References

Bernanke, B. S. (2010). ‘Monetary policy objectives and tools in a low-inflation environment’ Speech at the Federal Reserve Bank of Boston, 15 October 2010.

Blinder, A. S. and Krueger, A. (2004). ‘What does the public know about economic policy, and how does it know it?’ Brookings Papers on Economic Activity, Vol. 1, pp. 327–387.

Blinder, A. S., Ehrmann, M., Fratzscher, M., de Haan, J. and Jansen, D. (2008). ‘Central bank communication and monetary policy: A survey of theory and evidence’, Journal of Economic Literature, Vol. 46, pp. 910– 945.

Blinder, A. S., Ehrmann, M., de Haan, J. and Jansen, D. (2017). ‘Necessity as the mother of invention: Monetary policy after the crisis’, Economic Policy, Vol. 92, pp. 707–755.

Boukes, M. and Vliegenthart, R. (2017). ‘A general pattern in the construction of economic newsworthiness? Analyzing news factors in popular, quality, regional and financial newspapers’, Forthcoming at Journalism, https://doi.org/10.1177/1464884917725989.

Bruine de Bruin, W., Vanderklaauw, W., Downs, J., Fischhoff, B., Topa, G. and Armantier, O. (2010). ‘Expecta- tions of inflation: The role of demographic variables, expectation formation, and financial literacy’ Journal of Consumer Affairs, Vol. 44, pp. 381–402.

Bryan, M. and Venkatu, G. (2001). ‘The demographics of inflation opinion surveys’ Federal Reserve Bank of Cleveland Economic Commentary, October 2001.

Carroll, C. D. (2003). ‘Macroeconomic expectations of households and professional forecasters’, Quarterly Journal of Economics, Vol. 118, pp. 269–298.

Carvalho, C. and Nechio, F. (2014). ‘Do people understand monetary policy?’, Journal of Monetary Economics, Vol. 66, pp. 108–123.

Dräger, L. (2015). ‘Inflation perceptions and expectations in Sweden – Are media reports the missing link?’, Oxford Bulletin of Economics and Statistics, Vol. 77, pp. 681–700.

Dräger, L., Lamla, M. J. and Pfajfar, D. (2016). ‘Are survey expectations theory-consistent? The role of central bank communication and news’, European Economic Review, Vol. 85, pp. 84–111.

Dräger, L. and Lamla, M. J. (2017). ‘Imperfect information and consumer inflation expectations: Evidence from microdata’, Oxford Bulletin of Economics and Statistics, Vol. 79, pp. 933–968.

Ehrmann, M., Pfajfar, D. and Santoro, E. (2017). ‘Consumers’ attitudes and their inflation expectations’, International Journal of Central Banking, Vol. 13, pp. 225–259.

Gentzkow, M. and Shapiro, J. (2006). ‘Media bias and reputation’, Journal of Political Economy, Vol. 114, pp. 280–316.

Gentzkow, M. and Shapiro, J. (2011). ‘Ideological segregation online and offline’, Quarterly Journal of Eco- nomics, Vol. 126, pp. 1799–1839.

Gentzkow, M., Shapiro, J. and Sinkinson, M. (2014). ‘Competition and ideological diversity: Historical evi- dence from US newspapers’, American Economic Review, Vol. 104, pp. 3073–3114.

Haldane, A. G. (2017). ‘Everyday Economics’, Speech at Nishkam High School, Birmingham, 27 November 2017.

Haldane, A. G. and McMahon, M. (2018). ‘Central bank communication and the general public’, Forthcoming at the American Economic Association Papers & Proceedings, Vol. 108, pp. 578–583.

Hayo, B. and Neuenkirch, E. (2018). ‘The influence of media use on layperson monetary policy knowledge in Germany’, Scottish Journal of Political Economy, Vol. 65, pp. 1–26.

Jonung, L. (1981). ‘Perceived and expected rates of inflation in Sweden’, American Economic Review, Vol. 71, pp. 961–968.

Lamla, M. J. and Lein, S. (2015). ‘Information rigidities, inflation perceptions, and the media: Lessons from the euro cash changeover’, Economic Inquiry, Vol. 53, pp. 9–22.

Lamla, M. J. and Maag, T. (2012). ‘The role of media for inflation forecast disagreement of households and professional forecasters’, Journal of Money, Credit and Banking, Vol. 44, pp. 1325–1350.

Malmendier, U. and Nagel, S. (2016). ‘Learning from inflation experiences’, Quarterly Journal of Economics, Vol. 131, pp. 53–87.

Mankiw, N. G. and Reis, R. (2003). ‘Sticky information: A model of monetary non-neutrality and structural slumps’, in Aghion, P., Frydman, R., Stiglitz, J. S. and Woodford, M. (eds.), Knowledge, Information and

1212 Bulletin

Expectations in Modern Macroeconomics: In Honor of Edmund S. Phelps, Princeton, NJ: Princeton University Press.

Mullainathan, S. and Shleifer, A. (2005). ‘The market for news’ American Economic Review, Vol. 95, pp. 1031–1053.

Pfajfar, D. and Santoro, E. (2013). ‘News on inflation and the epidemiology of inflation expectations’, Journal of Money, Credit and Banking, Vol. 45, pp. 1045–1067.

Souleles, N. S. (2004). ‘Expectations, heterogeneous forecast errors, and consumption: Micro evidence from the Michigan consumer sentiment surveys’, Journal of Money, Credit and Banking, Vol. 36, pp. 39–72.

Van der Cruijsen, C. A. B., Jansen, D. and De Haan, J. (2015). ‘How much does the general public know about the ECB’s monetary policy? Evidence from a survey of Dutch households’, International Journal of Central Banking, Vol. 11, pp. 169–218.

Woodford, M. (2001). ‘Monetary Policy in the Information Economy’, Economic Policy for the Information Economy, Kansas City: Federal Reserve Bank of Kansas City, pp. 297–370.

Copyright of Oxford Bulletin of Economics & Statistics 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.