Measuring Political Knowledge in Cross-National Contexts:
Enhancing Comparability between Different Political Information Structures
University of Brasília
University of Brasília
Quite a lot has been written about the extent to which ordinary citizens know about and
understand politics, almost all of it focused on the American case. As a result, little has been
written about how to accurately compare citizens´ knowledge across different political contexts.
We analyze the effects of political institutions (the nature and competitiveness of the parties, the
presence or absence of compulsory voting requirements, etc.) and the information environment
(the roles of public versus private media, the nature of the laws governing freedom of the press,
etc.), which vary mostly cross-nationally, on citizens´ political knowledge with the goal of
creating an index that can be used as an explanatory variable to explain other political
phenomena. Estimating multi-level models on CSES data from a wide range of countries, this
paper examines both the sources of the country-level variation political sophistication and the
ways in which institutional arrangements and the information environment can help or handicap
citizens learn about politics.
Despite some still-flickering dissent, it has long been widely accepted that most people
know very little about most aspects of politics, and the usually implicit presumption is that this is
true, in somewhat varying degree, always and everywhere. But this variation in degree,
insufficient though it may be to alter descriptions of every public as knowing relatively little,
may nonetheless be consequential. The timbre of a country’s politics may well depend, in at least
some measure, on the political knowledgeability of its citizens.
Where the public is more
knowledgeable, the distributions of policy attitudes and votes should track those of values and
interests more closely; voting and other forms of participation should be more common; and the
media’s treatment of politics may be less frivolous.
From a sizable number of single-country surveys, we do know that the distribution of
political knowledge is routinely right-skewed, with relatively few people knowing very much.
We also know, at the individual level, that knowledge tightens the connections between interests
and values, on the one hand, and policy attitudes and votes, on the other (Bartels 1996, Della
Carpini and Keeter 1997, Althaus 1998), and that it increases the probability of voting (Della
Carpini and Keeter 1997).
What we do not know is how knowledge’s distribution, determinants, and effects may
vary across countries. Scattered single-country surveys, with relatively few shared items
(generally not including knowledge items) permit only rough and casual comparisons at best. At
worst, though, the comparisons are actively misleading.
For example, knowing the identity of the President of the United States is not an
equivalent task to knowing the identity of the President of Israel. The President of the U.S. is the
preeminent figure in American politics, whereas the President of Israel occupies a far less
important position in day-to-day politics. The President of Israel attracts far less media coverage,
but just as importantly, exists embedded within a different institutional configuration, with
different numbers of parties, different rules for apportioning legislative seats, and different cues
for voters. Measurement equivalence, in other words, is a constant worry when making cross-
Knowledge is one of a close-knit family of variables including “sophistication,” “expertise,”
“awareness,” “cognitive complexity,” and “information” (in the sense of information held).
Some of these terms are synonymous, others subtly different, but knowledge is the most
straightforwardly measured and highly correlated with the rest, and the literature consequently
seems to be converging on it (compare, e.g., Luskin, 2002, 2003 with Luskin, 1987).
Although this last factor could, of course, be a cause as well as an effect of a more knowledgeable public.
national comparisons of political knowledge (see Elkins and Sides 2008; Elff 2009; Stegmueller
2011; Matsubayashi and Turgeon 2012).
In this paper, we specify a preliminary model for measuring and explaining political
knowledge in cross-national contexts. We then use our results to provide the initial specifications
for an index that will succinctly capture the complexity of countries’ political information
structures, which would be available for use as an independent variable for models seeking to
explain political behavior.
Measuring Political Sophistication in a Comparative Context
Some work has indeed been done on measurement equivalence and the institutional
determinants of political knowledge. Gordon and Segura (1997), for instance, use Eurobarometer
data to provide one account of the determinants of knowledge in twelve Western European
countries. Clark (2013), meanwhile, finds that the range of parties available, as measured by the
effective number of electoral parties, increases voters’ political knowledge. Banducci, Giebler,
and Kritzinger (2015) argue that citizens’ knowledge of party positions is affected by the
strength of cues provided by the news media. These studies, however, are the exceptions.
Comparing anything across countries requires at least roughly comparable measures
(Elkins and Sides 2008). This is a greater challenge for political knowledge than for some other
variables because the information it might matter to know varies so widely across countries.
What are comparable pieces of information to have (or lack) in the U.S. versus France versus
Japan versus Peru? Institutions, actors, and issues vary, both across countries and (especially in
the case of actors and issues) over time, leading formally parallel facts to vary in both salience
and importance. An Israeli’s being able to identify the president of Israel shows more knowledge
than a Brazilian’s being able to identify the president of Brazil, a more salient office.
Political scientists have made advances in comparing other aspects of political systems
cross-nationally while taking into account the contextual elements of each country. Laakso and
Taagepera's (1979) Effective Number of Parties (ENP) index, for example, systematically takes
into account the actual influence of political parties to allow for one to make more accurate
cross-national comparisons of party systems. It has been cited 3288 times as of the writing of this
A similar sort of index could be made for political knowledge.
According to Google Scholar on 17 April 2017.
The most straightforward way of gauging political knowledge is via open- or closed-
ended questions that ask for factual information about public figures, current events, the
background to policy decisions, or the workings of political institutions. Their defining
characteristic, and their great strength, is that they have unambiguously correct answers. Their
weakness, however, is that it is difficult to know what facts are comparable across national
contexts. A Canadian’s knowing which party controls the Canadian House of Commons, for
instance, matters more (assuming a majority government in Canada) than an American’s
knowing which party controls the U.S. House of Representatives, a less powerful body. The
CSES affords three factual items per country, varying enormously from country to country in
what they ask about.
An alternative approach, developed by Luskin (1987) and Zaller (1992), constructs
knowledge items from placing parties or candidates on policy or ideological scales.
placement, say, of the British Labour Party on the left side of a left-right scale, or of the British
Labour Party to the left of the British Conservative party, is treated as correct, and all other
placements and don’t-know (DK) responses as treated as incorrect.
Following Luskin and
Bullock (2004), we term the first version of this approach as gauging “absolute correctness” and
the second as gauging “relative correctness.”
The biggest advantage, in the comparative context, of this approach is that the ideological
or policy locations of the most prominent political parties are probably more comparable pieces
of information to have or lack. The number of parties, and therefore, the salience of the most
salient parties (generally the biggest) may vary across countries, but placing, for example, the
three biggest parties of a given country on a left-right continuum is more clearly the same task in
different countries than is answering superficially parallel factual questions about political
figures, institutions, or issues. Its disadvantage, however, is it is less clear what should count as
correct. As a result, indices composed of such placement-based items correlate somewhat less
strongly with criterion variables related to knowledge than indices based on factual items
(Luskin and Bullock 2004).
A variant, adopted by Gordon and Segura (1997), measures the distance between the
respondent’s placement and the party’s “actual” location on the left-right continuum. This
The poles of the ideological dimension space are known as “left” and “right” in Europe, as
“liberal” and “conservative” in the U.S. and Japan.
method offers the advantage of more graduated measurement, but at the cost of adding a further
layer of debatability because parties’ “exact” locations are impossible to know with any
confidence. They can, however, be estimated with mean placements, whether by the whole
sample (as in Gordon and Segura 1997), by the most knowledgeable respondents, or by experts.
In our view, the latter two are superior options, but the results in Luskin and Bullock (2004), as
well as our own explorations, suggest that no distance-based measure works particularly well.
For present purposes, we adopt an absolute placement-based measure based on the
respondent’s placements of the major political parties (numbering between two and six, with a
mean of 4.43 and a standard deviation of 1.31). This means that placing the Labour Party on the
left side of the left-right scale would be measured as correct (and received a value of 1),
regardless of where the Conservative party was placed or whether the Labour Party was placed
to the left of the Conservatives. DKs and midpoint responses (which resemble DKs) are
measured as incorrect (and received a value of 0). The results in Luskin and Bullock (2004), as
well as our own work, suggest that this is the optimal way of converting placements into
knowledge items, largely because the between-country variance is larger than that of other
measures, while within-country variance is smaller, as one would expect.
This paper starts by describing and trying to explain cross-national variation in political
knowledge. Using data from the Comparative Study of Electoral Systems, we examine both the
individual-level determinants within given countries and the ways in which their effects depend
on political institutions and the information environment, which are constant at any given
moment within countries, but vary across them. As a preliminary step, we also consider how to
measure political knowledge in cross-national contexts, and how to use this political information
context as an explanatory variable.
The recent advent of coordinated cross-national surveys, notably including the World
Values Survey, the Pew Global Attitudes Survey, and, at the regional level, LAPOP, the
Eurobarometer, Latinobarometer, and Afrobarometer, opens new possibilities for systematic
This is at odds with Mondak (2001)but consistent with Luskin and Bullock (2004)
We have also tried measures based on relative placements and on factual items but do not present those results
here. For more on what measures of political knowledge are best for cross-national comparison, see Turgeon (2015)
comparative analysis of political knowledge (among other variables).
Here we employ data
from the Comparative Study of Electoral Systems (CSES), one of the richest and widest-reaching
of the lot.
The CSES is a collaborative program of cross-national research conducted in many
different parts of the world. The main purpose of the CSES is to allow researchers to examine
cross-national variation in political behaviors and attitudes. The data collected by the CSES are
of three kinds: individual or micro-level measures of political behaviors and attitudes (e.g., vote
choice, party and candidate evaluations, and the like); district-level measures about elections
(e.g., electoral returns, number of candidates and parties, and the like); and national or system-
level measures about the elections and political system at the time of the election (electoral rules,
regime characteristics, and the like). The individual-level data include measures of political
knowledge and a number of variables plausibly affecting it. So far, the CSES has completed four
modules, providing information about 159 elections that have taken place between 1996 and
2016 in over 50 countries. The sampling method, sample size, method of interviewing, and other
design details vary—widely—with the election study.
At the individual level, the factors affecting political knowledge can be grouped under the
headings of the opportunity, the ability, and the motivation to learn about politics (Luskin 1990).
Opportunity includes exposure to political information in the media, occupation, and education;
ability includes intelligence; and motivation includes political interest and (to a lesser degree)
education. In Luskin's (1990) U.S. results, motivation and ability appear to account for a great
deal of variation, and opportunity for very little. Interest has by far the biggest effect, intelligence
some, but education (controlling for interest and intelligence) virtually none.
But a good many more macro-level factors having to do the with the nature of
the political system, the circumstances of the day, the nature and behavior of the mass media,
and the political culture may also affect political knowledge. These have been much less
examined, though Gordon and Segura (1997) have made a nice start. Here, we offer a multilevel
(HLM) model to explain variation across individuals, countries, and elections.
Accompanied, to be sure, by new challenges, for both measurement (Elkins and Sides 2008) and analysis (Kedar
and Shively 2005).
Our individual-level modeling is constrained by the CSES’s selection of variables, and
not everything that theoretically should affect sophistication is available in all CSES studies. As
a result, we excluded intelligence and political interest, the prime movers in Luskin’s (1990)
results, for want of sufficient data. Our hope is that some of the other regressors— education,
most notably—may appropriate their effects in their absence.
At the individual level, our regressors are:
Education. Findings about education’s effect on political knowledge have been starkly
mixed, with some studies suggesting a very large effect, and others essentially none. These
differences may largely rest on other variables it has to contend with. With interest, intelligence,
and occupation controlled for, any remaining effect of education should stem from its educative
value (a matter of opportunity rather than ability or motivation). Here, without controlling for
interest or intelligence, we may hope that education proxies them. Operationally, education here
is measured on an 8-point scale where higher values indicate more formal education. A value of
1 indicates no formal education at all while 8 indicates the respondent has completed a university
Income. Higher incomes generally mean more time to devote to politics. Income may
also proxy some of what Luskin (1990) calls “political impingement,” or the exposure to
political information and the incentives to process it as part of one’s job. Higher-income
occupations doubtless tend to be more politically impinged, although the correlation is doubtless
far from perfect. We measure income on a 5-point scale, placing respondents into the appropriate
quintile. Higher values denote higher incomes.
Gender (Male). This is a dummy variable with a value of 1 denoting men. In many
places, men appear to know more politics than women. This is likely because politics has
For further information, see http://www.umich.edu/~cses/.
historically shut women out of the profession or deprived them of the vote, and still creates
difficulties for women to engage in politics in many places (see Fraile and Gomez 2015;
Lizotte and Sidman 2009).
Age. This is simply measured in years. In established democracies, age is well known to
affect knowledge. Older people are less distracted by life’s start-up costs, and even those who
remain relatively uninterested in politics have had a longer period in which to pick up incidental
knowledge as they grow older. It is plausible to see this life-cyclical effect as non-monotonic,
with the largest gains coming between youth and middle age (as is true of turnout) and then
declining from there. As a result, we modeled this variable as an exponential term.
Marital Status (Married). This is a dummy variable, with a value of 1 denoting people
who are married or living in marriage-like relationships. Being married increases one’s
probability of voting, although the reason is unclear. It may be that being married makes people
happier and likelier to engage in all manners of pro-social behavior, including voting. It may also
be that it is harder to hide a failure to vote from a spouse than to do so from friends and
coworkers. At any rate, it is at least plausible that some of this effect on voting may carry over to
Employment Status (Unemployed). This is another dummy variable, with the value
being 1 for those who are unemployed. We think that it is plausible that it has an effect on
political knowledge because it has a well-established parallel effect on turnout. In this case,
however, there is a clearer extension of the same logic to political knowledge. Being unemployed
is demoralizing, and the search for employment distracting. The unemployed therefore have both
diminished motivation and diminished opportunity to inform themselves about politics.
Occupation (Blue-Collar). This is an admittedly pale version of occupation qua political
impingement, a dummy variable with a value of 1 for those who hold blue-collar jobs. Blue-
collar jobs only rarely involve much political information or incentives to pay it heed, although
not all white-collar jobs or forms of self-employment do either. Some jobs considered here as
blue collar jobs include occupations related to agriculture and fishery, clerk positions in the
service industry, and all jobs requiring workers to operate machinery.
Residence (Rural). This is another dummy variable, with a value of 1 if the respondent
lived in a rural area. We use another rationale borrowed from the literature on turnout to justify
its presence. Even controlling for education, income, and occupation, people in urban areas tend
to have higher voter turnout and, it is reasonable to suspect, higher knowledge levels as well.
Public-sector jobs. This is another dummy variable, with a value of 1 if the respondent
had a job in the public sector and 0 if she did not. Working in the government could make one
more attentive to politics, particularly given that changes in the political fortunes of certain
groups can easily result in the loss of one’s job in certain contexts.
At the system/election level, we consider variables having to do with political
institutions, the party system, and the mass media (a number of them familiar from Gordon and
Segura 1997). Some vary over time as well as across countries—operationally, given the timing
of the observations, they vary by election. Others vary, at least, in practice, over a span of a just a
few decades, only with the country. To save words, we shall make references to the “election
level,” although a good part of the variance at that level is in fact by country. We do not
necessarily, of course, think the present roster of variables is exhaustive.
Effective Number of Parties. Contrary to Gordon and Segura (1997), who argue that
more numerous parties should decrease the costs of acquiring political information, up to a point,
we believe that more numerous parties should monotonically increase them. It is harder to keep
track of three parties than two, of four parties than three, etc. We use the Laakso and Taagepera
(1979) measure of the effective number of parties, which gives less weight to marginal parties.
The measure is where si is the proportion of seats of the ith party. The measure is based on the
CSES’s information about the six largest parties.
District Magnitude. This is a variable that measures how many officials are elected in
each district. It is a four-category measure, rather inaptly known as “electoral competitiveness”
or “district magnitude.” This measure ranges from single-member districts to three broad
varieties of PR. On the one hand, the more candidates to be selected in one district, the more
complex the institutional environment might become, thereby decreasing knowledge (Lloyd
2016). On the other hand, however, proportional representation systems might also induce voters
to seek out more information by encouraging them to express their first preference more readily
at the ballot box (Arnold 2007).
Disproportionality of Representation. Distributions of seats that visibly and
chronically fail to reflect the distribution of votes are an irritant to many voters, especially those
on the losing side of the mistranslation, which in turn stimulate greater interest and learning.
Following Lijphart (1984), we measure this as the mean difference between the vote and seat
shares of the two largest parties.
Compulsory Voting. Given that compulsory voting induces citizens to vote, often under
threat of punishment, we expected a positive effect on political knowledge. Knowing that voting
was required could give citizens added motivation to seek out information on politics (Gordon
and Segura 1997).
Unicameral Legislature. Bicameral legislatures figure to make the relationship between
votes and policy outcomes murkier, thereby increasing information costs and decrease political
knowledge. Again following Lijphart (1984), we use a dummy variable distinguishing countries
coded in the CSES dataset as having national legislatures consisting of a single chamber.
Public TV Viewership. This is the percentage of the aggregate market share of the five
most-viewed TV stations that belongs to state-owned stations, as measured by Djankov et al.
(2003). Given the tendency of public TV stations to be richer in political information content
(Fraile and Iyengar 2014), people living in countries with larger public TV viewership should
have greater exposure to political information.
Age of Regime. In some countries regimes and parties rapidly appear and vanish, in
others they last for decades or longer. The longer the same regime stays in place, the easier it
should be for people to discern parties’ and candidates’ left-right locations. Operationally,
therefore, this variable is the number of years since the last time that the country’s Polity score
changed by at least three points from the previous year. This data comes from the Electoral
Integrity Project (Norris et al. 2016).
Media (Non-)Freedom. This is the degree to which each country’s press is free,
according to Freedom House’s Freedom of the Press study. Higher scores mean less freedom: a
score between 0 and 30 signified that the country in question had a free press. A country scoring
between 31 and 60 was classified as partly free. Lastly, a score between 60 and 100 meant that
the country was considered not free (Freedom House 2016). Following Schoonvelde (2013), we
argue that a free media is essential for ensuring an informed electorate, and that government-
controlled press organs will make it more difficult for voters to learn about politics.
Extremity of the Parties. Parties nearer the center are harder to locate. More people
will place a party whose real location is 6 (slightly right of center) at 0-5 (dead center or left-of-
center) than one whose real location is 8, as the results in Luskin, Cautres, and Lowrance (2003)
confirm. For present purposes, we use the parties’ mean distance from the midpoint (5), a
variable that thus ranges from 0 (when all the parties are exactly at the midpoint) to 5 (when they
are all at 0 or 10). Again, we take the “real” locations from the CSES expert ratings.
Public Purchasing Power (log) and Gini coefficient. These variables follow the same
logic as the income variable at the individual level, but from a macro standpoint, with the Gini
coefficient included to also measure inequality within each country.
Because these variables are multilevel, with individual-level observations embedded
within countries, we express the hypotheses that they all affect individual-level political
knowledge in a multilevel model (Bryk and Raudenbush 1992; Steenbergen and Jones 2002).
This avoids the likely violation of classical assumptions in the “naïve pooling” of all individual-
level observations, regardless of country (as in Gordon and Segura 1997). The problem is that
the disturbances associated with given observations are unlikely to be “spherical”—i.e., to be
independent or have the same variance across elections (Burton, Gurrin, and Sly 1998). Naïve
pooling tends to bias the estimated standard errors downward and thus to produce falsely
“significant” results (Barcikowski 1981).
More precisely, we propose the following two-level linear multilevel model:
Individual−level:PKij =β0j ∑βpj xpij +εij ,
System−level:β 0j=γ 00 ∑γ0q zqj+δ0j
PKij=γ 00 ∑γ0q zqj +δ 0j ∑βpjxpij+ε ij
where PKij is the political knowledge of the ith individual in the jth election study, xpij is the ith
individual observation in the jth election study on the pth individual-level regressor, xqj is the jth
country-level observation on the qth country-level regressor, β0j and βpj are the jth election’s
intercept and the pth regressor’s slope in the individual-level equation for political knowledge,
and γ00 and γ0q are the intercept and the qth regressor’s slope in the election-level equation for
the jth election’s individual-level intercept β0j, and are the disturbances of individual and
election-level equations (1) and (2) respectively (both assumed to be multivariate normal and
assumed to be independent of each other).
In substantive terms, the individual-level equation (1) expresses political knowledge as a
linear function of education, income, age, gender, blue collar worker, rural residence, being
unemployed, and being married. The election-level equation (2) expresses the individual-level
equation’s intercept β0j and thus political knowledge as a linear function of the age of the
parties, the electoral system’s district magnitude, the national legislature’s being unicameral, the
disproportionality of representation, the effective number of parties, public TV’s percentage of
TV viewership, and the party mean extremity.
Note that “linear” here means linear and additive in the parameters. There is actually
one nonlinearity (strictly speaking, non-additivity) in the variables. As a crude approximation of
the argument above, we include the product of the respondent’s age and the age of the parties,
thereby allowing age to be less of an advantage in younger democratic systems (which tend to
have younger parties).
We estimate equation (3) by maximum likelihood. The results, in Table 1, show all eight
individual-level regressors as having statistically significant effects in the expected direction.
Table 1. ANOVA
Number of observations
Number of groups (elections)
Note: Table entries are maximum likelihood estimates with estimated standards errors in parentheses. *<.05
To derive something like a pseudo-R2 for this individual-level equation, we estimate the
linear model expressing Kij as a function simply of a full set of election dummies (Q – 1, where
Q is the number of elections. This can also be seen as an ANOVA model. The relevant results
are the election-level and residual variances, the latter attributed to the individual-level.
Table 2. Determinants of Political Knowledge
Union membership was also included at the individual level, but with no significant effect.
Effective number of parties
% public TV
Note: Table entries are maximum likelihood estimates with estimated standard errors in parentheses. *<.05
One can see from these results that, as expected, virtually all the individual-level
variables had a significant effect on political knowledge. Income, education, and age had positive
effects, as did being married, male, or having a public-sector job. Being unemployed, a blue-
collar worker, or living in rural areas, however, have negative effects.
At the systemic level, party extremity has positive effects on political knowledge and a
lack of media freedom had negative effects, as predicted. However, other systemic variables
diverged from our expectations. Firstly, PPP and the Gini coefficients, along with
unicameralism, ENP, district magnitude, and disproportionality all had no effect on political
knowledge. Furthermore, regime age, compulsory voting, and percent public TV had effects
opposite those we predicted. These results deserve further research in order to better understand
why they diverged so far from our theoretical expectations.
Another important result to take into account is the effect of our model on the variance
components. The unexplained variance at the election-level for the initial ANOVA model in
Table 1 is more than halved by our two multilevel models (conversely, the individual-level
unexplained variance hardly budges). This indicates that our model does an effective job at
explaining election-level variance in political knowledge and indicates that our index will indeed
This paper, far from representing the end of our project, instead represents the beginning.
Firstly, we plan to update our results with the most recent modules from CSES to determine if
our results still hold. Secondly, and more importantly, we will take the system-level variables
that have an effect on political knowledge—namely, regime age, party extremity, media non-
freedom, compulsory voting, and percent public TV. We will use item response theory to
determine the precise combination for these variables (see Elff 2009; Stegmueller 2011 for
examples of this technique).
We hope that, much like Laakso and Taagepera's (1979) index for the effective number
of parties in a political system, this index can be used as independent variable in other analyses.
For example, Lloyd (2016) uses the effective number of parties as a proxy variable for the
complexity of political information structures when trying to explain why increases in income
lead to decreases in clientelist voting in some countries but not others. This index will be a more
accurate measurement of this important concept, thereby allowing for an indicator that is more
accurate and causally proximate to the concept of political information structures.
Our hope, consequently, is that the results of this analysis and its successors may
ultimately be helpful for future researchers looking to understand country effects of political
knowledge. We also hope that it will prove useful for both reformers of existent political systems
and designers of nascent democracies to the extent that more knowledgeable publics, as we
firmly believe, make for better—more representative—democracy.
Althaus, Scott L. 1998. “Information Effects in Collective Preferences.” American Political
Science Review 92: 545–81.
Arnold, Jason Ross. 2007. Contextualizing Political Knowledge: A Cross-National, Multilevel
Banducci, Susan, Heiko Giebler, and Sylvia Kritzinger. 2015. “Knowing More from Less: How
the Information Environment Increases Knowledge of Party Positions.” British Journal of
Political Science: 1–18. http://www.journals.cambridge.org/abstract_S0007123415000204.
Barcikowski, Robert S. 1981. “Statistical Power with Group Mean as the Unit of Analysis.”
Journal of Educational Statistics 6(1967): 267–85.
Bartels, Larry M. 1996. “Uninformed Votes: Information Effects in Presidential Elections.”
American Journal of Political Science 40: 194–230.
Bryk, Anthony, and Stephen W Raudenbush. 1992. Hierarchical Linear Models: Applications
and Data Methods. Newbury Park, CA: Sage Publications.
Burton, A, L Gurrin, and P Sly. 1998. “Extending the Simple Linear Regression Model to
Account for Correlated Responses: An Introduction to Generalized Estimating Equations
and Multi-Level Mixed Modeling.” Statistics in Medicine 17(March): 1261–91.
Carpini, Michael Delli, and Scott Keeter. 1997. What Americans Know about Politics and Why It
Matters. New Haven: Yale University Press.
Clark, Nicholas. 2013. Contextual Dynamics and Political Knowledge : The Role of Institutional
Quality in an Informed Citizenry.
Djankov, Simeon, Caralee McLiesh, Tatiana Nenova, and Andrei Shleifer. 2003. “Who Owns
the Media.” Journal of Law and Economics XLVI(October): 341–81.
Elff, Martin. 2009. “Political Knowledge in Comparative Perspective: The Problem of Cross-
National Equivalence of Measurement.” In Midwest Political Science Association, , 1–27.
Elkins, Zachary, and John Sides. 2008. 1997 The Vodka Is Potent, but the Meat Is Rotten:
Evaluating Equivalence in Cross-National Survey Data.
Florida, J Mondak. 2015. “Developing Valid Knowledge Scales.” 45(1): 224–38.
Fraile, Marta, and Raul Gomez. 2015. “Why Does Alejandro Know More about Politics than
Catalina? Explaining the Latin American Gender Gap in Political Knowledge.” British
Journal of Political Science: 1–22. https://www.cambridge.org/core/article/why-does-
Fraile, Marta, and Shanto Iyengar. 2014. “Not All News Sources Are Equally Informative: A
Cross-National Analysis of Political Knowledge in Europe.” The International Journal of
Press/Politics 19(3): 275–94.
Gordon, Stacy B., and Gary M. Segura. 1997. “Cross–National Variation in the Political
Sophistication of Individuals: Capability or Choice?” The Journal of Politics 59(1): 126–47.
Kedar, Orit, and W. Phillips Shively. 2005. “Introduction to the Special Issue.” Political Analysis
Laakso, Markku, and Rein Taagepera. 1979. “‘Effective’ Number of Parties: A Measure with
Application to West Europe.” Comparative Political Studies 12(1): 3–27.
Lijphart, Arend. 1984. Patterns of Democracy: Government Forms and Performance in Thirty-
Lizotte, Mary-Kate, and Andrew H. Sidman. 2009. “Explaining the Gender Gap in Political
Knowledge.” Politics & Gender 5(2): 127.
Lloyd, Ryan. 2016. “Explaining Resilience in Clientelist Voting.”
Luskin, Robert C. 1987. “Measuring Political Sophistication.” American Journal of Political
Science 31(4): 856–99.
———. 1990. “Explaining Political Sophistication.” Political Behavior 12: 331–61.
———. 2002. “From Denial to Extenuation (and Finally Beyond): Political Sophistication and
Citizen Performance.” In Thinking About Political Psychology, ed. James H. Kuklinski.
New York: Cambridge University Press.
———. 2003. “The Heavenly Public: What Would a Fully Informed Citizenry Be Like?” In
Electoral Democracy, eds. Michael MacKeun and George Rabinowitz. Ann Arbor:
University of Michigan Press, 238–61.
What+Would+a+Fully+Informed+Citizenry+Be+Like?#0 (July 29, 2014).
Luskin, Robert C, and John Bullock. 2004. “Re(: )Measuring Political Sophistication.” In
Midwest Political Science Association,.
Luskin, Robert C, Bruno Cautres, and Sherry Lowrance. 2003. Measuring Political
Sophistication in France.
Matsubayashi, Tetsuya, and Mathieu Turgeon. 2012. Citizen Competence and the Institutional
Norris, Pippa, Ferran Martinez i Coma, Alessandro Nai, and Max Groemping. 2016. Perceptions
of Electoral Integrity, August 2016 Release (PEI_4.5 Codebook).
Schoonvelde, Martijn. 2013. “Media Freedom and the Institutional Underpinnings of Political
Knowledge.” Political Science Research and Methods (October): 1–16.
Steenbergen, Marco R, and Bradford S Jones. 2002. “Modeling Multilevel Data Structures.”
American Journal of Political Science 46(1): 218–37.
Stegmueller, Daniel. 2011. “Apples and Oranges? The Problem of Equivalence in Comparative
Research.” Political Analysis 19(4): 471–87.
Turgeon, Mathieu. 2015. Measuring and Explaining Political Sophistication in a Comparative
Zaller, John. 1992. The Nature and Origins of Mass Opinions. New York: Cambridge University