Quarterly Journal of Political Science, 2010, 5: 1–25
Political Information Acquisition for
Social Exchange∗
Gani Aldashev
Department of Economics and CRED, University of Namur (FUNDP), 8 Rempart de la
Vierge, 5000 Namur, Belgium. E-mail: [email protected].
ABSTRACT
Why do citizens get politically informed in a democracy? On one hand, being
informed allows a citizen to participate in political discussions within her social
network. On the other hand, having an informed opinion can help her to extend
her social network. This paper builds a simple model on these insights and finds
that effort in political information acquisition has inverted-U shape in the size of
social network. The data from the 2000 American National Election Study and
the 2002–2006 European Social Surveys confirm this theory: political information
acquisition, political knowledge, and interest in politics increase with the size of
social network, at a decreasing rate. The effect of social network is much weaker
for the political efficacy measures for the United States, but not for Europe.
“One way of acquiring opinions in …the personality-enriching manner is to give
them definite shape only after they have passed through intense confrontation
with other views …” (Hirschman, 1989)
∗
The author thanks the editors of the Journal, two anonymous referees, Alberto Alesina, David
Austen-Smith, Jean-Marie Baland, Oriana Bandiera, Marc Bellemare, Tim Besley, Alberto Bisin,
Gabrielle Demange, Eliana La Ferrara, Bruno Frey, Markus Goldstein, John Matsusaka, Matthias
Messner, Jean-Philippe Platteau, Guido Tabellini, and Hylke Vandenbussche for their useful
comments.
Replication Data available from:
http://dx.doi.org/10.1561/100.00009009_supp
MS submitted 2 February 2009; final version received 7 February 2010
ISSN 1554-0626; DOI 10.1561/100.00009009
© 2010 G. Aldashev
2
Aldashev
“Most people acquire political information so that they can participate in the
conversation at parties, not in order to decide how to vote” (Tullock, 1967)
Why do citizens get politically informed in a democracy? Understanding political
information acquisition in large elections poses a challenge: if getting informed takes
time and the instrumental payoffs are negligible (because the probability of being pivotal
quickly goes to zero when the size of electorate increases), one cannot easily explain
why so many citizens spend time getting informed. This is the well-known “rational
ignorance” result first stated by Downs (1957) and Tullock (1967) in the decisiontheoretic framework (see Martinelli, 2006 for the statement of the result in the gametheoretic framework). Tullock summarizes this result as follows: “It would almost never
be rational to engage in much study in order to cast a “well-informed” vote” (Tullock,
1967: 114) (however, the opening quote by Tullock shows that he was well aware of
other potential avenues for understanding information acquisition). A simple look at the
data shows that this model does not correspond to reality. The U.S. National Election
Study (ANES) of 2000 shows that about 30% of respondents followed in full at least one
TV debate between Gore and Bush. The 2002–2006 European Social Surveys (ESS)
indicate that about one-sixth of respondents spend 30 minutes or more per day reading
newspaper articles about politics. The instrumental-motivation rational choice models
cannot explain these basic facts, and such failure calls for a better model of voters’
motivation to acquire political information.1
One strand of literature chooses to remain agnostic regarding voter motivation and
tries instead to explain variation in political informedness by looking at the cost of
information (Besley and Burgess, 2002; Stromberg, 2004, Prat and Stromberg, 2006;
Larcinese, 2007). However, understanding cost variation alone does not suffice to explain
real-life patterns in political informedness of citizens. Boeri and Tabellini (2007) show,
using data from an Italian survey, that voters’ information matters for the viability
of key welfare state reforms. Their findings suggest that informing citizens about the
long-run payoffs from reforms might help to overcome broad opposition and create
political support for reforming the welfare state. However, they find that better supply
of information does not imply more information acquisition: citizens living in regions
where media cover the reforms extensively are not informed better than those living in
regions with less media coverage.
Everyday observation suggests that political campaign periods are also the periods of
heated political discussions among citizens, and having an informed opinion serves as a
“ticket” for entering into such discussions. Benz and Stutzer (2004) state that “in the
weeks preceding the vote [at the Swiss referendum on joining the EU in 1992], it was
almost impossible not to get involved in the fierce discussions on the subject, and consequently, the incentives to be informed were high”. This intuition presents a promising
alternative route to understanding information acquisition: citizens may spend time
1
Feddersen and Sandroni (2006) show that including ethical concerns in citizens’ utility function
theoreticallty substantially increases the fraction of citizens becoming informed, even when the cost
of getting informed is positive. Measuring the extent to which ethical concerns drive information
acquisition remains an open empirical question.
Political Information Acquisition for Social Exchange
3
acquiring political information to form opinions that serve them in discussions and
social interactions with other fellow citizens. Ohr and Schrott (2001), using the data
from state-level elections in Germany, find that in deciding to get politically informed,
citizens place most weight to the social expectations about them being informed.
These findings indicate that the role of social environment and social network of a
citizen is fundamental for her political information acquisition. Enjoying participation in
political debates with close friends should be a strong motivating force to get informed.
Moreover, information serves not only for enjoying one’s time within one’s social network, but also plausibly to enlarge one’s network. Becoming a friend with someone met
at a dinner party is more likely if one can entertain an informed political discussion
during the party, especially when important elections are forthcoming or some key political event has recently occurred. In other words, social capital, like any other form of
capital, is an asset in the accumulation of which individuals invest rationally (as argued
by Glaeser et al., 2002), and political conversations may well serve this purpose.2
This paper builds a simple model of political information acquisition inspired by
these ideas. A citizen acquires political information that serves her to engage in political discussions within the existing social network, but also to make new friends. This
combination of consumption and investment incentives implies — under standard concave technology assumptions — that the overall incentive to acquire information has an
inverted-U-shaped relationship with the size of social network.
I confront this theoretical prediction with data from the 2000 ANES and the 2002–
2006 ESS. I find that the effort to acquire political information, the level of political
knowledge, and political interest and participation increase with the size of the social
network, at a decreasing rate. The effect of social network is much weaker for the political
efficacy measures for the United States, but not for Europe.
These findings open several interesting directions for future research. I discuss them
in the concluding section of this paper.
MODEL
Consider a citizen living for two periods, t = 1, 2. In each period, she decides the effort
that she puts into political information acquisition, et . This effort translates into the
amount of political information, qt :
qt = q(et ).
Let’s assume that this information is period-specific (e.g., it can be about the issues specific to elections in each period: the characteristics of candidates, their policy positions,
etc.) and it does not accumulate across periods. Exerting effort implies a cost, c(et ), with
standard concavity assumptions: c > 0 and c > 0.
2
Skaperdas (2003) also studies a model built around the consumption motivation for political information acquisition. However, his paper gives no role for the investment motive and does not
empirically test the model.
4
Aldashev
The payoffs of political information are of two types: private and social exchange.
Private payoffs, v(qt ), with v < 0 < v , are the payoffs that the citizen gets from
making better private decisions using the acquired information. The social exchange
payoffs come from using political information in debates with other citizens in one’s
social network. I denote the quantity of social exchange activity with s. The size of the
citizen’s social network is n. The social exchange is produced using a production function
with political information and the size of social network as inputs:
s = s(q(e), n).
(1)
The nature of these payoffs is clearly psychological and can be interpreted in terms of the
expressive returns studied by Brennan and Lomasky (1993) for voting: a citizen enjoys
utility from expressing her informed opinion, and the larger is the group to which she
can express her opinion, the larger is her utility.
Moreover, one’s political information serves not only to enjoy utility from social
exchange within the current network, but also to increment the social network for the next
period. La Due Lake and Huckfeldt (1998) find that politically relevant social capital
(i.e., the part of social capital that enhances participation) is built up in social networks
as a by-product of information discussions. Let’s assume then
nt+1 = f (nt , qt ),
(2)
where f (.) is a known function.
The problem of the citizen is to maximize her lifetime payoff (net of effort costs) by
choosing the effort levels devoted to political information acquisition:
max v(q(e1 )) + s(q(e1 ), n1 ) − c(e1 ) + β[v(q(e2 )) + s(q(e2 ), n2 ) − c(e2 )],
e1 ,e2
subject to
n2 = f (n1 , q(e1 )).
(3)
(4)
Here, β denotes the intertemporal discount factor. When solving Problem (3) – (4),
the citizen takes as given the initial social network size, n1 .
The first-order conditions for e1 and e2 are:
∂v dq
∂s dq
∂s ∂f dq
dc
∂V
=
+
+β
−
=0
(5)
∂q1 de1
∂q1 n1 de1
∂n2 ∂q1 n1 de1
de1
∂e1
∂v dq
∂s dq
dc
∂V
=
+
−
= 0.
(6)
∂e2
∂q2 de2
∂q2 n2 de2
de2
The interpretation of these first-order conditions is as follows. Consider first Equation (5). It has four additive terms (three positive and one negative). The first term is the
private marginal benefit. The second term is the first-period marginal social exchange
benefit of effort to get informed. It is made of two parts: the second part describes
how much the citizen’s information increases as a result of incremental effort, which
Political Information Acquisition for Social Exchange
5
is then multiplied by the first part that describes how much one more unit of political
information increases the social exchange activity of the citizen (at the given size of her
social network). Let’s call this term the consumption benefits of political information. The
third term is the intertemporal marginal benefit of effort spent for political information
acquisition. An additional unit of effort translates into more information dedq1 , and
each extra unit
implies a larger social network of the citizen in the sec information
of
∂f ond period ∂q1 . This, in turn, translates into higher social exchange in the second
n1
period, appropriately discounted. Let’s call this term investment benefits of political information. Finally, the fourth term describes the marginal cost of effort. A rational citizen
spends the first-period effort up to the point where total marginal benefit equals the
marginal cost.
Next, consider Equation (6). Since there is no further accumulation of social network,
this condition is made of only three terms: the private marginal benefit, the benefit
from second-period social exchange, and the marginal cost of effort spent for political
information acquisition in Period 2.
Consider again Equation (5). Let’s apply the standard diminishing marginal returns
assumption for consumption and investment benefits. Intuitively, it is likely that when
the citizen’s social network is small and increases marginally, the consumption benefits
(which are initially low) increase rapidly. Subsequent expansion of social network will,
however, bring in lower increments in consumption benefits. The opposite is true for
investment benefits: when social network is relatively small, making new friends by
investing into political information is relatively easy. However, as the social network
size keeps expanding, making additional friends via such investment becomes relatively
more difficult. These two mechanisms jointly imply that the overall incentive to acquire
political information has an inverted-U-shaped relationship with the social network size.
To further elucidate these mechanisms, below I construct a simple theoretical example
that illustrates this result.
A Simple Example
Let’s suppose that the totality of social exchange of the citizen is composed of many
individual interactions. In each individual interaction, the social exchange activity is
produced using two inputs: the amount of political information that the citizen brings in
(which I assume is simply equal to q) and the time spent in the individual interaction, τ.
This single-interaction social exchange, which I denote with s, is produced using a
production function of the Cobb-Douglas form:
s = q α τ 1−α ,
with 0 < α < 1. In other words, time spent discussing with a partner and the political
information brought in are (imperfectly) substitutable.
The citizen has total time resource for social exchange, T , that she divides equally
between all n members of her social network. Thus, she enters into n interactions and
6
Aldashev
spends Tn of time in each interaction. Her overall social exchange production function
(1) in a given period can now be written as
T 1−α
(7)
= T 1−α q α nα .
s = n
s = n qα
n
Note then that the marginal benefit (in terms of producing social exchange) of political
information is increasing and concave in the network size, n:
∂s
= αT 1−α q α−1 nα ,
∂q
(8)
given that nα is increasing and concave in n.
Intuitively, the current-period (consumption) marginal benefits from political information acquisition are increasing and concave in the number of people with whom the
citizen interacts. A larger social network has two effects, going in the opposite direction.
On one hand, larger network implies more interactions. On the other hand, the citizen
can devote less time to each interaction, and thus each interaction delivers lower payoffs.
When n is low, an increase in network size brings in high incremental social activity: this
is because the second effect is still very small (given that for low n, inside each individual
interaction time is relatively abundant while information is relatively scarce). As n gets
bigger, however, the second effect becomes more important (given that inside each individual interaction time now turns into the relatively scarce input). At the limit, when n
is very high, the first and the second effects are of almost equal size and opposite sign —
thus, the total effect is almost zero. In other words, for a very extensive social network,
any further increase in the network size does not bring in any incremental consumption
benefits from political information.
Let’s now turn to the process of accumulation of social network. I assume that the pool
of all agents with whom the citizen can interact is L and her potential acquaintances are
all agents which are not in her current social network. The creation of new members of
the social network is governed by a Cobb–Douglas production function, with political
information (q) and the pool of potential acquaintances (L − n) as inputs. In other words,
political information and the pool of potential partners are imperfectly substitutable for
creating new acquaintances. When the pool of potential partners is large, one can make
new friends even with a relatively low knowledge of politics; however, when the pool of
potential acquaintances becomes smaller, to create the same number of new friends, one
needs a deeper knowledge of politics (for instance, more arguments to discuss).
Furthermore, let’s suppose that a share of acquaintances disappear between Periods 1
and 2 (for example, some citizens can move out, which can imply severing the links with
them). Then, the evolution function for social capital of the citizen (4) can be written as:
γ
n2 = f (n1 , q1 ) = (1 − δ)n1 + q1 (L − (1 − δ)n1 )1−γ .
(9)
Note that the investment benefits from political information acquisition are decreasing
and concave in the size of the current social network:
∂f
= γq γ−1 (L − (1 − δ)n1 )1−γ .
(10)
∂q1
Political Information Acquisition for Social Exchange
7
The intuition for this is as follows. When the initial social network is small, the number
of potential acquaintances is relatively abundant. In this case, a small increase in the
social network size does not reduce much the marginal payoff (in terms of new friends)
of investing into more political information. As the network size becomes bigger, any
further increase makes the number of potential acquaintances scarcer (given the level
of q). Thus, the negative effect of a larger network on the marginal payoff (in terms of
new friends) of having more information gets bigger. At the limit, when the network size
is very close to L, any further increase in the network size basically eliminates all the
investment incentives to acquire political information.
Given the considerations above, now I can characterize the shape of the optimal firstperiod information acquisition. Consider again Equation (5). The marginal cost of effort
is independent of the network size, and so is the marginal private benefit of effort. The
benefit from social exchange, composed of the consumption and investment incentives
to acquire political information, evolves non-monotonically in the initial network size.
Given Equation (8), the consumption incentives to acquire political information are
increasing and concave in n1 . Given Equation (10), the investment incentives to acquire
political information are decreasing and concave in n1 . Then, given the concavity of both
relations, the total incentives (i.e., the marginal benefit) to acquire information have an
inverted-U-shape in the initial network size.
Proposition 1 A citizen’s effort to acquire political information is concave in the size of her
social network.
EMPIRICAL STRATEGY
In the rest of this paper, I verify whether this theoretical proposition is in line with
individual-level data from the United States and Europe.
Let’s assume that the benefit from information acquisition for a citizen i can be
modeled as:
(11)
Ui = µ0 Mi + µ1 Mi2 + λ Xi + εi ,
where Xi is the vector of individual characteristics, Mi is the size of i’s social network,
and εi is the error term distributed normally with mean 0.
All of the variables measuring political information acquisition in my empirical analysis
are of ordered response type. This implies that one does not observe the latent variable
1
2
J
Ui . Let U i < U i < · · · < U i be unknown cut points, and define:
1
Pi = 0 if Ui ≤ U i
1
Ui
Pi = 1 if
< Ui ≤
...
J
Pi = J if Ui > U i .
(12)
2
Ui
I can thus estimate the coefficients µ0 and µ1 using maximum-likelihood ordered
probit procedure.
8
Aldashev
There are two potential problems with this specification. First, some third individuallevel political characteristic, such as intrinsic political motivation, may drive both the
size of social network and the political information acquisition. For instance, politically
motivated citizens may also be socially active, as well as acquire more political information, than less politically motivated citizens. To control for this problem, I add citizens’
past political behavior (measured by having voted in the previous presidential elections)
into the vector of controls Xi .
Second and similarly, some third individual-level social-skill characteristic, such as
intrinsic skills in accumulating social capital, may drive both the social network size and
the political information acquisition. Citizens more skillful in building social capital
may tend both to naturally have a larger social network and to acquire more political
information. To account for this problem, I add individual social-activity characteristics
(trust in other people, readiness to serve on jury, membership in an organization, and
attendance of religious services) into the vector of controls Xi .
DATA AND VARIABLES
The first source of data is the 2000 American National Election Study (ANES). It was
conducted around the November 2000 presidential elections (a set of questions was asked
before the elections, while the remaining part was asked after the elections). The sample
contains 1,807 individuals; however, as most of my variables of interest were measured in
the post-election part, my analysis mainly concentrates on a subsample of around 1,550
observations. The main reason for choosing this particular data set is that it contains
explicit measures of an individual’s political discussions within her social network.3
The battery of dependent variables that I use can be classified into four categories:
1. Political information acquisition/seeking. These variables include: intensity of
watching the presidential debates on TV between Gore and Bush, intensity of
listening to radio programs about the campaign, and whether the respondent has
seen any information about the campaign on the Internet;
2. Political interest/participation. These measures include: attention paid to presidential campaign news, attention paid to congressional campaign news, the degree
to which the respondent follows government and public affairs, and turnout in
2000 elections;
3. Political knowledge. The measures in this category are: the knowledge score on
party control (in the Senate and the House), public figure recognition score,
the knowledge score on presidential and vice-presidential candidates (their state
and religion), and the interviewer’s evaluation of the respondent’s political
knowledge; and
4. Political efficacy. These variables include: feeling about having a good understanding of politics, feeling that the respondent’s vote matters, feeling that public
3
Unfortunately, these measures are not available for other rounds of the ANES.
Political Information Acquisition for Social Exchange
9
officials care what people think, and feeling that people have say about what government does.
My main explanatory variable is constructed as follows. An interviewer asked each
respondent whether the respondent has a person with whom she discusses politics. In
case of an affirmative answer, the interviewer asked the respondent whether she has
another discussant. This procedure was repeated twice more, which allows to measure
the social network in which the individual discusses politics up to four people. I thus
have constructed the index of size of political-discussion social network of a respondent,
which varies from 0 to 4.
The size of social network is truncated at a relatively low value (many citizens might
have more than four people with whom they discuss politics). Because of this shortcoming
of the data, in most regressions I use the set of binary variables for each level of social
network size (instead of using the continuous measure of network size and its square).
This allows to see at which sample levels of network size the dependent variable measures
reach their sample maxima.
The battery of control variables comprises individual demographic and socioeconomic
characteristics (age, gender, marital status, race, income, education, home-ownership,
and length of residence in the community) and variables aimed at capturing unobservable
political motivation and social-skill heterogeneity (turnout in the 1996 presidential elections, trust in other people, readiness to serve in a jury, membership in an organization,
and the regular attendance of religious services).
Table 1a presents the summary statistics for the key variables. On average, citizens
report to have close to two people with whom they discuss politics regularly. However,
there is a lot of heterogeneity: one-quarter of respondents do not discuss politics with
anyone, and more than one-fifth discuss politics with (at least) four people. Similarly, on
all measures of political information seeking, interest, knowledge, and efficacy, one can
see that there is substantial variation in the data (standard deviations are relatively high).
To check whether my findings apply in a different context, I have chosen as the second
source of data the 2002–2006 Cumulative European Social Survey (ESS). This biannual
survey covers over 30 countries in Europe and includes a number of political information
and interest variables. More importantly, it contains also a measure of individual social
interactions, which I use as a proxy for the size of social network. Notably, the data set
is much bigger than the ANES 2000 round: my typical regression for ESS is based on
over 50,000 observations.4
The battery of dependent variables in this data set belongs to three categories:
1. Political information acquisition/seeking. These variables include: intensity of
watching TV news about politics and current affairs, intensity of listening to radio
programs about politics, and intensity of reading newspaper articles about politics;
2. Political interest/participation. These measures include: how much the respondent is interested in politics and turnout in the latest national legislative elections;
and
4
The overall sample is over 120,000 observations, but because of missing information on several
control variables, my subsample reduces down to about 50,000 observations.
10
Aldashev
Table 1a. ANES — Summary statistics for some key variables.
Variable
Political discussion/social networks
Size of political-discussion network
No one with whom disc. politics
Discuss politics with 1 person
Discuss politics with 2 people
Discuss politics with 3 people
Discuss politics with 4 people
Political information acquisition/seeking
Intensity of watching presidential TV
debates
Intensity of listening to radio programs
about the campaign
Has seen information on Internet about
the campaign
Political interest/participation
Attention paid to these presidential
campaign news
Attention paid to these congressional
campaign news
Follow government and public affairs
Voted in 2000 elections
Political knowledge
Party control knowledge score
Public figure recognition score
President and VP knowledge score
Interviewer’s evaluation of respondent’s
political knowledge
Political efficacy
Feel to have good understanding of
politics
Feel that my vote matters
Feel that public officials care what
people think
Feel that people have say about what
government does
No. of obs.
Mean
SD
Min
Max
1,550
1,550
1,550
1,550
1,550
1,550
1.864
0.257
0.187
0.201
0.144
0.211
1.479
0.437
0.390
0.401
0.351
0.408
0
0
0
0
0
0
4
1
1
1
1
1
1,549
0.984
0.777
0
2
1,551
0.729
1.012
0
3
973
0.477
0.500
0
1
1,551
2.454
1.046
0
4
1,550
1.439
0.988
0
4
1,543
1,554
1.671
0.761
0.967
0.427
0
0
3
1
1,555
1,547
1,554
1,550
1.045
1.090
3.190
2.162
0.900
1.114
1.852
1.094
0
0
0
0
2
4
8
4
1,542
2.558
1.153
0
4
1,442
1,547
0.649
1.674
1.138
1.239
0
0
4
4
1,548
2.125
1.327
0
4
Political Information Acquisition for Social Exchange
11
3. Political efficacy. These variables include: disagreeing that politics is too complicated to understand, the perception of difficulty of making up one’s mind about
political issues, trust in national parliament, general trust in politicians, trust in
the European Parliament, and the degree of satisfaction with democracy in one’s
country.5
My main explanatory variable in this data set is the frequency of socialization, measured on 0–6 scale. The value of 0 is given if the respondent states that she does not
socialize at all, 1 if she socializes less than once a month, 2 if she socializes about once
a month, etc. I also use, in robustness analysis, the variable that measures taking part in
social activities as compared to others of the same age. This relative measure varies from
0 to 4, with 0 standing for taking part in social activities much less than most people of
the same age, 1 – less than most, 2 – about the same, etc.
Both of these measures are imperfect: the relative distances between different values
are not the same. Thus, in my regressions I employ the set of binary variables for each level
of frequency of socialization (instead of using the continuous measure and its square).
The battery of control variables comprises individual demographic and socioeconomic
characteristics (age, gender, marital status, income categories), years of education completed, place of residence (city, suburb, town, village, or countryside), and variables aimed
at capturing unobservable social-skill heterogeneity (trust in other people, membership
in an organization, and the regular attendance of religious services). Given substantial variation in political systems across European countries, I also add country-level
dummies in all the regressions run on the ESS data.
Table 1b presents the summary statistics for the key variables in the ESS data set. On
average, frequency of socialization is relatively high: the value of 3.96 is very close to the
score of 4, which stands for socialization once a week. In fact, looking at the distribution
across different categories, one can see that individuals that do not socialize or socialize
less than once a month together comprise less than 10% of the sample, while those that
socialize several times a week or every day comprise almost half of the sample. Similar
to the 2000 ANES data set, there is substantial heterogeneity in all measures of political
information acquisition, interest, and efficacy.
EVIDENCE
ANES 2000
If the data are in line with my theoretical model, one should see that the size of social
network has an increasing and concave relationship with measures of political information acquisition, knowledge, and interest. Before running the full-fledged econometric
analysis, though, let’s look at the simple correlations between size of social network and
political information acquisition.
5
The ESS does not contain measures of political knowledge of the respondent.
12
Aldashev
Table 1b. ESS — Summary statistics for some key variables.
Variable
Social networks
Frequency of socialization
Do not socialize
Socialize less than once a month
Socialize once a month
Socialize several times a month
Socialize once a week
Socialize several times a week
Socialize every day
Taking part in social activities as compared
to others of same age
Taking part in social activities = much less
than most
Taking part in social activities = less than
most
Taking part in social activities = about the
same
Taking part in social activities = more than
most
Taking part in social activities = much more
than most
Political information acquisition/seeking
Intensity of watching TV news about
politics/current affairs on average
weekday
Intensity of listening to radio news about
politics/current affairs on average
weekday
Intensity of reading newspaper news about
politics/current affairs on average
weekday
Political interest/participation
How much interested in politics
Voted in last national legislative elections
Political efficacy
Disagree that politics is too complicated to
understand
Difficulty in making up my mind about
political issues
Trust in national parliament
Trust in politicians
Trust in the European Parliament
Satisfaction with democracy in my country
No. of obs.
Mean
SD
Min
Max
124,274
124,274
124,274
124,274
124,274
124,274
124,274
124,274
122,166
3.964
0.022
0.074
0.086
0.181
0.180
0.285
0.172
1.722
1.587
0.146
0.262
0.281
0.385
0.384
0.451
0.377
0.944
0
0
0
0
0
0
0
0
0
6
1
1
1
1
1
1
1
4
122,166
0.113
0.316
0
1
122,166
0.255
0.436
0
1
122,166
0.461
0.498
0
1
122,166
0.140
0.347
0
1
122,166
0.031
0.173
0
1
120,419
1.980
1.287
0
7
99,173
1.625
1.507
0
7
93,590
1.210
0.894
0
7
124,294
114,579
1.390
0.786
0.896
0.410
0
0
3
1
122,397
1.854
1.145
0
4
121,744
1.962
1.050
0
4
120,441
121,725
108,354
118,856
4.658
3.733
4.632
5.393
2.464
2.327
2.387
2.420
0
0
0
0
10
10
10
10
13
0
1
2
3
Political Information Acquisition for Social Exchange
0
1
2
Intensity of watching TV debates
Attention to presidential campaign news
3
4
Public figure recogn.
Figure 1. Average political information acquisition, by size of social network — ANES
2000.
Figure 1 plots the means of three political information acquisition variables (intensity
of watching TV debates between Gore and Bush, public figure recognition score, and
attention paid to 2000 presidential campaign news), for each level of social network
size. The pattern is clear: for all the three variables, citizens with larger network size
tend to report higher values of political information acquisition. The concavity of the
relationship is less clear. This might be because the measure of network size is truncated
at a relatively low value, or because the investment incentives described in my theoretical
model are empirically less important.
The graphical evidence, while suggestive, cannot be considered as conclusive, because
it does not allow to control for numerous individual characteristics that might be driving
both the social network size and political information acquisition. Let’s then proceed
to the econometric analysis, which permits to control for these characteristics and thus
to separate the effect of network size. Table 2 shows my regression results with measures of political information acquisition as dependent variables.6 Even after controlling
for a large battery of individual characteristics, I find that the citizens with a larger
social network size are more likely to declare higher intensity of watching presidential
TV debates, more frequent listening to radio programs about the campaign, and more
intensive search for information about the campaign on the Internet. The regression
coefficients on network-size dummies (the omitted category is “Discuss politics with
4 people”) are negative and highly statistically significant and the absolute value of the
6
In Columns 1 and 2, the dependent variables are ordered response, thus I use ordered probit,
whereas in Column 3, the dependent variable is binary, thus I use simple probit regression. I report
the standard errors, clustered at the state level, in parentheses.
14
Aldashev
Table 2. ANES — Network size and political information acquisition.
No one with whom disc. politics
Discuss politics with 1 person
Discuss politics with 2 people
Discuss politics with 3 people
Age
Age, squared
Gender = female
Marital status
Race = African-American
Race = Latino
Race = Other
Household income (category)
Home-ownership
Education (category)
Length of residence in community, years
Attend religious services (frequency)
Would serve on jury duty
Trust in other people
Member of an organization
Turnout in 1996 Pres. elections
Observations
Estimation procedure
(1)
Intensity of watching
presidential TV
debates
(2)
Intensity of listening
to radio programs
about the campaign
(3)
Has seen information
on Internet about the
campaign
−0.622
(0.097)
−0.200
(0.090)
−0.138
(0.089)
−0.209
(0.089)
−0.019
(0.013)
0.000
(0.000)
−0.192
(0.063)
0.081
(0.083)
−0.092
(0.119)
0.118
(0.149)
0.099
(0.123)
0.030
(0.012)
−0.023
(0.081)
0.031
(0.022)
−0.001
(0.002)
0.073
(0.021)
0.024
(0.084)
0.056
(0.054)
−0.074
(0.060)
0.373
(0.046)
1258
Ordered probit
−0.741
(0.111)
−0.370
(0.092)
−0.243
(0.099)
−0.089
(0.088)
0.023
(0.013)
−0.000
(0.000)
−0.378
(0.065)
0.090
(0.072)
−0.137
(0.089)
−0.004
(0.174)
−0.173
(0.178)
0.012
(0.013)
−0.068
(0.092)
0.095
(0.023)
−0.002
(0.002)
0.026
(0.025)
0.095
(0.059)
−0.045
(0.062)
0.053
(0.058)
0.244
(0.091)
1258
Ordered probit
−0.256
(0.047)
−0.136
(0.052)
−0.042
(0.067)
−0.096
(0.045)
0.017
(0.008)
−0.000
(0.000)
−0.126
(0.039)
−0.111
(0.037)
0.046
(0.070)
0.084
(0.054)
−0.044
(0.088)
0.001
(0.005)
0.019
(0.035)
0.053
(0.011)
−0.002
(0.001)
0.004
(0.014)
0.065
(0.039)
0.055
(0.042)
−0.011
(0.042)
−0.047
(0.050)
785
Probit
Note: Benchmark category for network size is “Discuss politics with 4 people”
Political Information Acquisition for Social Exchange
15
coefficient is monotonically declining in network size, almost everywhere. Moreover,
there is (a somewhat weaker) evidence that the strength of the effect of a larger social
network decreases with its size: the differences in the absolute values of coefficients on
the network-size dummies tend to decrease with the size of network, which indicates a
concave relationship. Statistically, for all the three regressions, the equality-of-coefficient
tests (not reported here, available upon request from the author) reject the equality
of coefficients for the first two categories (“No one with whom discuss politics” and
“Discuss politics with 1 person”) and cannot reject the equality of coefficients of the
subsequent pairs of categories.
Note that these results indirectly disconfirm the findings of the line of research that
argues that social embeddedness helps to get political information for free or at a very
low cost (Campbell et al., 1960; Fiorina, 1981; Huckfeldt, 1983, 2001; Zaller, 1989;
McClurg, 2003, 2006). The predictions of those models are difficult to reconcile with
the evidence that citizens with a larger social network tend to actively invest more in
political information acquisition.
These results confirm my theoretical model. However, there are several potential
caveats that have to be addressed. First, the findings might be driven by the particularities
of the U.S. presidential elections; it might be interesting to see whether the results are
robust when one looks at other elections. Second, since political information is not
restricted to elections, one may wonder about the relevance of my theoretical model to
political-information behavior of citizens beyond the elections. Third, given the wellestablished link between political informedness and participation, it is interesting to
check whether this social exchange motivation has any relevance for turnout. Fourth, the
self-reported measures of information acquisition might be unreliable: if a respondent
feels any negative connotation of not being politically informed, she might over-report
the extent to which she invests into getting informed.7 Moreover, this over-reporting
might be related to social network size, because respondents with a larger social network
might feel the negative connotation stronger than those with a smaller network. It is
then necessary to check whether the model explains not only the variation in selfreported information acquisition, but also the variation in objective measures of political
knowledge.
To address these concerns, I repeat my analysis using several alternative measures of
political information acquisition, and find that my basic findings are robust. Tables 3a
and 3b report the regression results with political interest/participation and political
knowledge measures as dependent variables.8
7
8
Burden (2000) discusses in detail the problem of over-reporting of turnout in NES data. Even
though my analysis mainly concerns political information, the underlying causes of over-reporting
might be similar.
To economize on space, Tables 3a–3c do not report the coefficients on control variables (I use the
same individual controls as in the regressions of Table 2). In Table 3a, for Columns 1–3 ordered
probit is used. In Column 4, the dependent variable is binary, thus I use simple probit. In Table 3b,
for Columns 1–4 ordered probit is used. Column 5 is a robustness check and reports the results of
a simple OLS regression. In Table 3c, I use ordered probit in all the regressions.
16
Aldashev
Table 3a. ANES — Network size and political interest/participation.
(1)
Attention paid
to these
presidential
campaign news
No one with whom
−0.820
disc. politics
(0.102)
Discuss politics
−0.347
with 1 person
(0.101)
Discuss politics
−0.287
with 2 people
(0.081)
Discuss politics
−0.253
with 3 people
(0.088)
Observations
1258
Estimation
Ordered probit
procedure
(2)
Attention paid
to these
congressional
campaign news
(3)
Follow
government
and public
affairs
−0.516
−0.762
(0.101)
(0.089)
−0.067
−0.347
(0.095)
(0.104)
−0.252
−0.260
(0.085)
(0.084)
0.039
−0.064
(0.092)
(0.096)
1257
1253
Ordered probit Ordered probit
(4)
Voted in 2000
elections
−0.164
(0.034)
−0.120
(0.054)
−0.028
(0.047)
−0.003
(0.046)
1257
Probit
Note: Benchmark category for network size is “Discuss politics with 4 people”
In Table 3a, the dependent variables are measures of political interest and participation.
In Columns 1 and 2, the dependent variables are the attention paid to presidential
and to congressional campaign news, respectively. Comparing the results of these two
regressions allows me to check whether the findings are driven by the peculiarity of the
presidential elections. In Column 3, the dependent variable is the extent to which the
respondent follows government and public affairs. This measure has the advantage of
not being directly related to elections and campaigns, and thus this regression checks the
relevance of my model to citizens’ political-information behavior beyond the elections. In
Column 4, the dependent variable is turnout at the 2000 elections. This regression checks
whether political discussions in social networks have consequences on participatory
behavior of citizens.
In Table 3b, I employ four different objective measures of citizens’ political knowledge:
the score on the knowledge about which party controls the Senate and the House,
the score on recognizing several key public figures, the score on the knowledge about
presidential and vice-presidential candidates, and the interviewer’s evaluation of the
respondents’ political knowledge. These regressions allow me to address the concern
about the possible over-reporting by citizens of their political information acquisition
efforts.
Looking at the regression results, one sees that except for two coefficients — on
“Discuss politics with 1 person” in Column 2 of Table 3a, and on “Discuss politics
with 2 people” in Column 1 of Table 3b, the findings are strikingly similar to those
−0.688
(0.104)
−0.271
(0.097)
−0.112
(0.064)
−0.023
(0.128)
1251
Ordered probit
−0.583
(0.120)
−0.270
(0.105)
0.080
(0.124)
−0.088
(0.124)
1259
Ordered probit
Note: Benchmark category for network size is “Discuss politics with 4 people”
Observations
R2
Estimation procedure
Discuss politics with 3 people
Discuss politics with 2 people
Discuss politics with 1 person
No one with whom disc. politics
(2)
(3)
Ordered probit
−0.424
(0.126)
−0.118
(0.107)
−0.048
(0.119)
0.020
(0.115)
1259
Party control
Public figure President and VP
knowledge score recognition score knowledge score
(1)
Ordered probit
−0.762
(0.104)
−0.322
(0.080)
−0.119
(0.100)
−0.013
(0.095)
1258
(4)
Interviewer’s
evaluation of
respondent’s
political
knowledge
Table 3b. ANES — Network size and political knowledge.
−0.602
(0.181)
−0.179
(0.163)
−0.071
(0.182)
0.018
(0.186)
1257
0.30
OLS
President and VP
knowledge score
(5)
Political Information Acquisition for Social Exchange
17
18
Aldashev
Table 3c. ANES — Network size and political efficacy.
(1)
(2)
Feel to have
good underst.
of politics
Feel that my
vote matters
(3)
(4)
Feel that public
Feel that
officials care people have say
what people
about what
think
govt does
No one with whom
−0.671
0.567
0.051
0.021
disc. politics
(0.090)
(0.112)
(0.095)
(0.083)
Discuss politics
−0.197
0.321
0.106
0.128
with 1 person
(0.089)
(0.159)
(0.109)
(0.115)
Discuss politics
−0.314
0.241
0.001
−0.051
with 2 people
(0.085)
(0.120)
(0.097)
(0.078)
Discuss politics
−0.160
0.017
0.146
0.186
with 3 people
(0.102)
(0.121)
(0.093)
(0.084)
Observations
1252
1160
1253
1254
Estimation
Ordered probit Ordered probit Ordered probit Ordered probit
procedure
Note: Benchmark category for network size is “Discuss politics with 4 people”.
reported above: the regression coefficients on network-size dummies are negative, highly
statistically significant in most cases, are monotonically decreasing in absolute value, and
the differences in absolute size get smaller as the network size increases. These findings
strengthen the evidence in favor of my theoretical model.
A further interesting question is what effect the network size has on the citizens’
perception of political efficacy. To answer this question, I repeat my econometric analysis
with four measures of political efficacy as dependent variables: the extent to which
the respondent agrees with the statements “I feel that I have a good understanding of
politics”, “I feel that my vote matters”, “I feel that public officials care about what people
think”, and “I feel that people have a say about what government does”.
Table 3c reports the findings of these regressions. In Column 1, the dependent variable
is the extent to which the respondent feels to have a good understanding of politics. The
results resemble those found in earlier regressions: citizens with a larger social network
tend to feel that they have a better understanding of politics than those with a smaller
network. In Column 2, the dependent variable is the extent to which the respondent feels
that her vote matters. Here, the results are strikingly different from those above: all the
regression coefficients on network-size dummies are positive, the first three coefficients
are highly statistically significant, and there is a negative monotonic relationship between
network size and feeling that the respondent’s vote matters. In other words, citizens
that discuss politics with more people tend to be more pessimistic about the political
importance of their individual votes. This might be because of the unique feature of the
19
0
.5
1
1.5
2
Political Information Acquisition for Social Exchange
0
1
2
3
4
5
6
Reading politics in newspapers
Interest in politics
Disagree that politics is too complicated to understand
Figure 2. Average political information acquisition, by frequency of socialization —
ESS 2002–2006.
2000 presidential election, in that no president-elect was nominated immediately after
the elections because of vote re-counting problems. On the other hand, the results in
Columns 3 and 4 indicate that there is no robust statistical relationship between network
size and the feeling that public officials care about what people think or the feeling that
people have a say in what government does. Thus, the pessimistic feelings of citizens
with larger political-discussion networks concern only votes and elections, and not the
political accountability in general. Jointly, these findings suggest that better political
informedness of citizens related to larger social networks might imply a weaker feeling
of personal political efficacy, if — as a consequence of being better informed — citizens
become aware of problems in the functioning of their democratic system.
ESS 2002–2006
To test my theoretical model in a different context, I use the data from the 2002–2006
Cumulative European Social Survey (ESS). Given that the data set does not contain
measures of social network size, I use as the main explanatory variable the frequency
of socialization. In other words, I have to adopt an implicit assumption that larger
social network size is correlated with higher frequency of socialization. Even though this
assumption may not apply always (for instance, some citizens with a small network might
interact frequently within their network), it is likely to apply for the majority of citizens.
Figure 2 plots the means of three political information acquisition variables, by frequency of socialization. There is a clear inverted-U-shaped pattern for all the three
variables: citizens with intermediate frequency of socialization read more about politics
in newspapers, declare a stronger interest in politics, and are more likely to disagree that
20
Aldashev
politics is too difficult to understand, than citizens with very low or very high frequency
of socialization.
Turning now to the econometric analysis, Table 4 shows my regression results with
political information acquisition measures as dependent variables.9 In Columns 1, 2,
and 3, the dependent variables are, respectively, the intensity of political information
acquisition on TV, radio, and newspapers. In all the three specifications, there is a concave
relationship between the frequency of socialization and information acquisition. In other
words, citizens that socialize more frequently exert higher effort to get informed via TV,
radio, and newspapers, but the incremental effort gets smaller (and beyond some point
turns negative) as the frequency of socialization increases.
Table 5a presents regression results with political interest and participation as dependent variables.10 In Column 1, the dependent variable is the interest in politics, whereas
in Column 2, the dependent variable is turnout in the latest national legislative elections.
In both specifications the dependent variable has a clear concave relationship with the
frequency of socialization. Citizens that socialize more often declare to be more interested in politics and are more likely to vote. However, the incremental effect gets ever
weaker as the frequency of socialization increases.
Table 5b presents regression results with measures of political efficacy as dependent
variables. Contrarily to the results found for the United States in 2000, in all the specifications, the concave relationship between political efficacy and frequency of socialization
is still present. There are three possible explanations for this difference in results. First,
the explanatory variable is not exactly the same across two sets of regressions. In other
words, one might have found results similar to those for the United States, if one were to
use the size of political-discussion network as the main explanatory variable for the ESS
regressions. Second, the dependent variables are not exactly the same across two sets
of regressions. For instance, the ESS does not have a measure whether the respondent
feels that her vote matters. It is possible that if I were to use this measure for the ESS
regression, I could have found the pattern similar to the one in the United States. Finally,
the relationship between political efficacy and social network size might be inherently
different in the United States and in Europe. For example, Europeans with relatively
larger social networks might be more satisfied with the functioning of democracy in
their countries (as a consequence of being more informed about politics), while their
American counterparts might be satisfied less, perhaps because of the particular outcome
of the 2000 presidential elections. Understanding which of these different possibilities
explains the pattern that we find in the data is an interesting question that we leave for
future work.
For both the ANES and the ESS data, I perform further robustness checks.11 For
the ANES data, first, given that political culture might vary across the U.S. states, I
9
10
11
In all specifications (except Column 2 in Table 5a) I use ordered probit. Column 2 in Table 5a
reports the results where the dependent variable is binary, thus I use simple probit.
To economize on space, Tables 5a and 5b do not report the coefficients on control variables (I use
the same individual controls as in the regressions of Table 4).
I do not report regression tables here, but they are available upon request.
Political Information Acquisition for Social Exchange
21
Table 4. ESS — Network size and political information acquisition/seeking.
Freq. of socialization =
none
Freq. of socialization =
less than once a month
Freq. of socialization =
once a month
Freq. of socialization =
several times a month
Freq. of socialization =
once a week
Freq. of socialization =
several times a week
Age
Age, squared
Gender = female
Marital status
Years of full-time
education completed
Place of residence =
suburb
Place of residence =
town
Place of residence =
village
Place of residence =
countryside
Trust others
Has worked in a non-political
organization last year
Attend religious services
(frequency)
Observations
Estimation procedure
(1)
TV watching,
news/politics/
current affairs on
average weekday
(2)
Radio listening,
news/politics/
current affairs on
average weekday
(3)
Newspaper
reading, politics/
current affairs on
average weekday
−0.165
(0.034)
−0.074
(0.021)
−0.037
(0.020)
−0.009
(0.016)
−0.046
(0.016)
0.001
(0.014)
0.038
(0.002)
−0.000
(0.000)
−0.150
(0.009)
0.041
(0.010)
0.013
(0.001)
0.006
(0.017)
−0.015
(0.014)
−0.079
(0.014)
−0.118
(0.022)
−0.004
(0.002)
0.059
(0.013)
−0.021
(0.003)
54750
Ordered probit
−0.176
(0.043)
−0.086
(0.025)
−0.023
(0.023)
−0.006
(0.018)
−0.036
(0.018)
0.001
(0.016)
0.034
(0.002)
−0.000
(0.000)
−0.071
(0.010)
−0.027
(0.012)
0.008
(0.002)
0.028
(0.018)
0.030
(0.016)
0.054
(0.016)
0.155
(0.024)
0.001
(0.002)
0.117
(0.014)
0.009
(0.004)
45290
Ordered probit
−0.189
(0.050)
−0.129
(0.027)
−0.041
(0.024)
0.020
(0.019)
−0.034
(0.019)
0.023
(0.017)
0.015
(0.002)
0.000
(0.000)
−0.212
(0.011)
−0.005
(0.013)
0.054
(0.002)
−0.010
(0.020)
−0.092
(0.017)
−0.142
(0.017)
−0.137
(0.025)
0.019
(0.003)
0.178
(0.015)
0.001
(0.004)
43084
Ordered probit
Note: Benchmark category for network size is “Frequency of socialization = every day”. Coefficients
on income categories and country dummies are not shown.
22
Aldashev
Table 5a. ESS — Network size and political interest/participation.
Freq. of socialization = none
Freq. of socialization = less than once a month
Freq. of socialization = once a month
Freq. of socialization = several times a month
Freq. of socialization = once a week
Freq. of socialization = several times a week
Observations
Estimation procedure
(1)
How much
interested in
politics
(2)
Voted in the last
national elections
−0.339
(0.035)
−0.136
(0.022)
−0.033
(0.020)
0.014
(0.017)
0.004
(0.016)
0.027
(0.015)
56247
Ordered probit
−0.107
(0.015)
−0.031
(0.008)
−0.010
(0.007)
−0.003
(0.006)
−0.000
(0.006)
0.010
(0.005)
52877
Probit
Note: Benchmark category for network size is “Frequency of socialization = every day”.
repeat the regressions adding state dummies, and find that the effect of social network
size is very similar to that found in the regressions without state dummies. Second, the
battery of individual controls might be still insufficient to claim that the variation in
the size of social network is exogenous. To address this endogeneity problem, I repeat
the regressions with the length of residence (and its square) as the main explanatory
variable.12 It is likely that the reasons for residential mobility are not directly related
to political information acquisition behavior. If so, I can consider the variation in the
length of residence as exogenous. The results of these regressions are similar to those
with network size as the main explanatory variable: political information acquisition
has a highly statistically significant inverted-U-shaped relationship with the length of
residence.
For the ESS data, I perform a robustness analysis using as the main explanatory
variable the relative frequency of socialization (as compared to others of the same age),
instead of the absolute frequency of socialization. I still find the concave relationship
between the relative frequency of socialization and political information acquisition.
12
I cannot employ the instrumental-variables technique (instrumenting the size of social network
with the length of residence), because the length of residence might have an effect on political
information acquisition not only via the network size.
−0.198
(0.034)
−0.073
(0.021)
−0.021
(0.020)
0.002
(0.016)
0.020
(0.016)
0.010
(0.014)
55658
Ordered
probit
−0.094
(0.034)
−0.049
(0.021)
−0.000
(0.019)
0.019
(0.016)
0.028
(0.016)
0.014
(0.014)
54959
Ordered
probit
Trust in
country’s
parliament
Ease of making
mind up about
political issues
−0.438
(0.034)
−0.202
(0.021)
−0.098
(0.020)
−0.051
(0.016)
−0.042
(0.016)
−0.014
(0.014)
55438
Ordered
probit
(3)
(2)
Note: Benchmark category for network size is “Frequency of socialization = every day.”
Freq. of socialization =
none
Freq. of socialization =
less than once a month
Freq. of socialization =
once a month
Freq. of socialization =
several times a month
Freq. of socialization =
once a week
Freq. of socialization =
several times a week
Observations
Estimation procedure
(1)
Disagree that
politics is too
complicated to
understand
−0.152
(0.034)
−0.006
(0.021)
0.035
(0.019)
0.067
(0.016)
0.054
(0.016)
0.036
(0.014)
55498
Ordered
probit
Trust in
politicians
(4)
Table 5b. ESS — Network size and political efficacy.
−0.190
(0.036)
−0.066
(0.022)
−0.003
(0.020)
−0.002
(0.016)
0.008
(0.016)
0.004
(0.015)
50001
Ordered
probit
Trust in the
European
Parliament
(5)
−0.178
(0.034)
−0.068
(0.021)
−0.059
(0.019)
0.001
(0.016)
0.027
(0.016)
0.010
(0.014)
54546
Ordered
probit
(6)
How satisfied
with the way
democracy works
in country
Political Information Acquisition for Social Exchange
23
24
Aldashev
CONCLUSION
If citizens are influenced by their social environment and benefit from their social network, getting politically informed gives two kinds of social payoffs. On one hand, being
informed allows the citizen to participate in political debates within her network and
enjoy the utility of sharing her informed opinion with her friends. On the other hand,
being able to entertain an informed discussion about political topics may allow her to
make new friends and extend her social network. This paper formalizes these mechanisms in a simple model and shows that the data from the 2000 American National
Election Study and the 2002–2006 European Social Surveys are in line with the model’s
predictions.
I acknowledge that my model has some limitations. Most importantly, the theoretical
model is built in a decision-theoretic framework: the citizen does not take into account
the behavior of her potential discussion partners. In reality, there are many potential
topics of discussion (within and beyond politics) and being informed on one topic while
others are uninformed about it probably does not pay off. This implies that citizens coordinate on most salient topics (some of which during the campaign period are inevitably
related to the elections). This model says nothing about how citizens decide in this
framework and how such coordination occurs. It is likely that mass media play a key role
in coordinating citizens’ information acquisition patterns, and investigating this idea
(maintaining the motivational foundations of the model) seems to be a fruitful direction
for further research.
Two other extensions seem natural for this work. First, given that political informedness is one of the determinants of voter turnout, one can build a model of voter participation based on informational mechanisms described in this paper. In such a model, a
citizen’s decision to vote would be based in part on her political informedness, which is,
in turn, linked to her social incentives to acquire information.
Second, the model can be extended to analyze the effects of introducing a new medium
on voter information and turnout. While a new medium generally implies lower costs of
acquiring political information, it may also affect the demand side of information. The
existing literature finds contrasting results: Stromberg (2004) shows that the introduction of radio had increased turnout, while Gentzkow (2006) finds that the expansion of
television had a negative effect on turnout. This paradox can be explained in the context
of my model: television is known to destroy social capital (Putnam, 2000; Olken, 2006),
and thus the effect of TV on the demand side of information (both lower consumption
and investment incentives to acquire information) is perhaps stronger than the effect
on the supply side. Conversely, radio is unlikely to have had a strong negative effect on
social capital, and this explains the positive effect of radio on turnout. In today’s era of
Internet, a question naturally arises: was the effect of the massive expansion of Internet
on the information demand side (i.e., the negative effect) stronger than the one on the
supply side (i.e., the positive effect)? An extended model based on the one presented in
this paper and an empirical test should be able to provide an answer to this fascinating
question.
Political Information Acquisition for Social Exchange
25
REFERENCES
Benz, M., and A. Stutzer. 2004. “Are Voters Better Informed When They Have a Larger Say in Politics?
Evidence from the European Union and Switzerland.” Public Choice 119: 31–59.
Besley, T., and R. Burgess. 2002. “The Political Economy of Government Responsiveness: Theory and
Evidence from India.” Quarterly Journal of Economics 117: 1415–1452.
Boeri, T., and G. Tabellini. 2007. “Does Information Increase Political Support for Pension Reform?”
Working paper, Bocconi University.
Brennan, G., and L. Lomasky. 1993. Democracy and Decision: The Pure Theory of Electoral Preference.
New York: Cambridge University Press.
Burden, B. 2000. “Voter Turnout and the National Election Studies.” Political Analysis 8: 389–398.
Campbell, A., P. Converse, W. Miller, and D. Stokes. 1960. The American Voter. New York: Wiley.
Downs, A. 1957. An Economic Theory of Democracy. New York: Harper and Row.
Feddersen, T., and A. Sandroni. 2006. “Ethical Voters and Costly Information Acquisition.” Quarterly
Journal of Political Science 1: 287–311.
Fiorina, M. 1981. Retrospective Voting in American National Elections. New Haven: Yale University
Press.
Gentzkow, M. 2006. “Television and Voter Turnout.” Quarterly Journal of Economics 121: 931–372.
Glaeser, E., D. Laibson, and B. Sacerdote. 2002. “An Economic Approach to Social Capital.” Economic
Journal 112: F437–F458.
Hirschman, A. 1989. “Having Opinions — One of the Elements of Well-being?" American Economic
Review 79: 75–79.
Huckfeldt, R. 1983. “Social Contexts, Social Networks, and Urban Neighborhoods: Environmental
Constraints on Friendship Choice.” American Journal of Sociology 89: 651–669.
Huckfeldt, R. 2001. “The Social Communication of Political Expertise.” American Journal of Political
Science 45: 425–438.
La Due Lake, R., and R. Huckfeldt. 1998. “Social Capital, Social Networks, and Political Participation.”
Political Psychology 19: 567–584.
Larcinese, V. 2007. “The Instrumental Voter Goes to the News-Agent: Information Acquisition,
Marginality and the Media.” Journal of Theoretical Politics 19: 249–276.
Martinelli, C. 2006. “Would Rational Voters Acquire Costly Information?” Journal of Economic Theory
129: 225–251.
McClurg, S. 2003. “Social Networks and Political Participation: The Role of Social Interaction in
Explaining Political Participation.” Political Research Quarterly 56: 448–464.
McClurg, S. 2006. “The Electoral Relevance of Political Talk: Examining the Effect of Disagreement
and Expertise in Social Networks on Political Participation.” American Journal of Political Science
50: 737–754.
Olken, B. 2006. “Do Television and Radio Destroy Social Capital? Evidence from Indonesian Villages.”
NBER Working Paper No. 12561.
Ohr, D., and P. Schrott. 2001. “Campaigns and Information Seeking: Evidence from a German State
Election.” European Journal of Communication 16: 419–449.
Prat, A., and D. Stromberg. 2006. “Commercial Television and Voter Information.” Working Paper,
London School of Economics.
Putnam, R. 2000. Bowling Alone: The Collapse and Revival of American Community. New York: Simon
and Schuster.
Skaperdas, S. 2003. “Turning ‘Citizens’ into ‘Consumers’: Economic Growth and the Level of Public Discourse,” in Rational Foundations of Democratic Politics (ed. A. Breton et al.), New York:
Cambridge University Press.
Stromberg, D. 2004. “Radio Impact on Public Spending.” Quarterly Journal of Economics 119: 189–191.
Tullock, G. 1967. Toward a Mathematics of Politics. Ann Arbor: University of Michigan Press.
Zaller, J. 1989. “Bringing Converse Back In: Modeling Information Flow in Political Campaigns.”
Political Analysis 1: 181–234.