THE GENDER GAP AND POLITICAL KNOWLEDGE

Document Sample
THE GENDER GAP AND POLITICAL KNOWLEDGE Powered By Docstoc
					THE GENDER GAP AND POLITICAL KNOWLEDGE: MEN AND WOMEN IN NATIONAL AND STATE POLITICS

James C. Garand Emogine Pliner Distinguished Professor Department of Political Science Louisiana State University Baton Rouge, Louisiana 70803-5433 Office: (225) 578-2548 Email: pogara@lsu.edu Emily Guynan M.A. Candidate Department of Political Science Louisiana State University Baton Rouge, Louisiana 70803-5433 Email: eguyna1@lsu.edu Monique Fournet B.A. Candidate Department of Political Science Louisiana State University Baton Rouge, Louisiana 70803-5433 Email: mfourn1@lsu.edu

Paper presented at the 2004 annual meeting of the Southern Political Science Association, New Orleans, Louisiana, January 8-10, 2005.

2

THE GENDER GAP AND POLITICAL KNOWLEDGE: MEN AND WOMEN IN NATIONAL AND STATE POLITICS Abstract Previous research has found that men have higher levels of political knowledge than women, both in general and after the effects of various control variables are taken into account. In this paper we explore the contours and determinants of this gender gap in political knowledge. Using data from the 2000 American National Election Study (ANES) and the 2002 Louisiana Survey, we develop a series of models in which we depict political knowledge as a function of gender, socioeconomic and demographic attributes, political attitudes and engagement, media exposure, and political life circumstances. We find that gender effects in political knowledge persist, even in the face of statistical controls. Men and women differ on their mean values for a number of control variables, so the inclusion of a wide range of independent variables does result in a moderate reduction in the magnitude of gender coefficients. We also find that the gender gap appears to be somewhat stronger for national-level political knowledge rather than state-level political knowledge. Moreover, we consider the possibility that men and women differ in their relative propensities to give incorrect and don’t know responses to knowledge questions, and our results from bivariate and multivariate analyses suggest that women are more likely to give both incorrect and don’t know responses. Finally, we test a stereotype threat model of gender differences in political knowledge, and our results from these models are somewhat inconclusive. Overall, our findings suggest that gender matters for political knowledge, with women exhibiting consistently lower levels of political knowledge across a wide range of model specifications. The persistence of gender differences is somewhat perplexing, insofar as we account for several explanations for why men have higher levels of political knowledge then women

1

The study of political knowledge has drawn considerable scholarly attention in recent years. Scholars have described levels of political knowledge in the American public, generally finding that American citizens have, on average, relatively low levels of knowledge about politics (Bennett, 1989; Delli Carpini and Keeter, 1996). Other researchers have focused attention on the measurement of political knowledge, debating the types of items to be included in scales of political knowledge as well as how to interpret incorrect answers and nonresponses (Luskin, 1987; Delli Carpini and Keeter, 1993; Nadeau and Niemi, 1995; Mondak, 1999, 2001; Mondak and Creel, 2001). Still other scholars have developed models to explain individuals’ levels of political knowledge (Delli Carpini and Keeter, 1996; Mondak, 1999; Lambert, Curtis, Kay, and Brown, 1988; Davis and Silver, 2003; Gidengil, Goodyear-Grant, Nevitte, Blais, and Nadeau, 2003). Finally, there is a debate among scholars concerning the implications of low levels of political knowledge in the mass public on the workings of democratic political systems (Lupia and McCubbins, 1998; Popkin, 1991; Bartels, 1996; Garand and Lichtl, 2000). One of the most interesting findings in this body of research is that there is considerable gender gap in political knowledge. Simply, scholars have found that men are much more knowledgeable about politics than women (Delli Carpini and Keeter, 1996; Kenski and Jamieson, 2001; Kenski and Jamieson, 2001; Verba, Burns, and Schlozman, 1997; Mondak, 1999; Gidengil et al., 2003). For instance, Delli Carpini and Keeter (1996) develop a full model of political knowledge, using data from both the 1988 American National Election Study (ANES) and a survey conducted in Virginia in 1989; they find that the coefficient for gender is negative and significant, even controlling for the effects of other variables that are related to political knowledge. Mondak (1999) also finds a negative coefficient for gender in his multivariate model of political knowledge in the 1992 ANES; more specifically, he finds that women are more likely then men to respond both incorrectly and with a “don’t know” response to knowledge questions. Moreover, Davis and Silver (2003) find that men have significantly higher levels of political knowledge. The finding of a gender gap in political knowledge also extends both cross-nationally and through the life span. In a study of political knowledge and political engagement across 19 nations, Claibourne and Sapiro find that women have significantly lower levels of political knowledge in all nations, controlling for basic demographic variables. Moreover, using data from the 2000 Canadian Election Study, Gidengil et al. (2003) find strong and consistent gender differences in political knowledge. Finally, the gender gap in political knowledge extends to pre-adults and adolescents (Hess and Torney, 1967; Jennings and Niemi, 1981; Niemi and Junn, 1999). The observation of gender differences in political knowledge is somewhat perplexing. Women and men have similar IQ levels, so observed differences in political knowledge for men and women cannot be due to differences in intelligence (Delli Carpini and Keeter, 1996: 195). Academic preparation also would not appear to be a good candidate for explaining gender differences in political knowledge, particular since girls and women perform better in school than do boys and men; for instance, Bowen and Bok (1998) find that female students in their College and Beyond Study are more likely to graduate and have a higher class rank than similarly-situated male students, and Vars and Bowen (1998) use the same data set to find a significant positive effect of female gender on cumulative grade point average. Since intellectual ability and academic preparation can be eliminated as possible explanations, scholars are forced to consider other possible alternatives. Some observers note that women and men differ in terms of important demographic characteristics (e.g., education), and that these differences matter for levels of political knowledge. Others contend that women have political predispositions and attitudes that discourage political participation; for instance, Verba, Burns, and Schlozman (1997) note that women exhibit systematically lower levels of political engagement (i.e.,

2

political efficacy, information about and interest in politics), and these characteristics might discourage the acquisition of political information by women relative to that of men. Some researchers also suggest that the life circumstances of women differ from those of men, particularly in terms of the relative responsibility for raising children and juggling family and work responsibilities (Delli Carpini and Keeter, 1996; Verba et al., 1997). These disproportionate responsibilities are thought to have a disengagement effect that disadvantages women in terms of their political knowledge relative to that of men. In this paper we build on previous research to explore the contours and determinants of the gender gap in political knowledge. We begin by describing general patterns of political knowledge for men and women using data from the 2000 American National Election Study (ANES) and the 2002 Louisiana Survey. Second, we develop and test a general model of political knowledge, focusing on how the magnitude of gender effects decreases as one controls for the effects of independent variables that differentiate men and women and that are thought to account for the effects of gender on political knowledge. Third, we consider the possibility that men and women differ in their relative propensities to give incorrect and “don’t know” responses. Rather than use general knowledge scales that do not differentiate between incorrect and “don’t know” responses, we estimate a series of multinomial logit models of political knowledge using specific knowledge items recoded to include correct, incorrect, and “don’t know” responses. Finally, we test a stereotype threat model of gender differences in political knowledge, suggesting that gender differences in political knowledge may be at least partially explained by gender-of-interview effects. PREVIOUS RESEARCH The study of gender differences in political knowledge has drawn some specific attention in the scholarly literature. While scholars have examined the gender gap in political knowledge, there is still no satisfactory answer as to why men know more about politics than women. Even when one controls for variables representing the various explanations, the gender gap in political knowledge remains intact, albeit somewhat diminished. Differences in Propensities to Hazard a Guess One of the most intriguing ideas about why men know more about politics than women relates to hypothesized differences in how men and women respond to survey questions. Gidengil et al (2003: 1) raise questions about “the assumption that the sex difference in opinion expression reflects women’s socialized reticence about politics.” Gidengil et al. argue that perhaps the appropriate question is not “Why do women not know as much as men?”, but rather “Why are men so willing to express opinions?” By taking male behavior as the norm, women’s behavior is automatically viewed as deviant. Gidengil et al. argue that what should be examined is the quality of the opinions given, suggesting that women may be more likely to answer “don’t know” to survey questions, but when women do give answers they are more likely to be correct and their opinions are more likely to be well informed. They go on to argue that men may be masking a lack of political opinions by being so willing to express opinion while women, though more likely to answer “don’t know”, do not feel pressure to answer unless they are ready to express well informed opinions. Mondak (1999, 2001) reinforces this view in his research on political knowledge. He examines how political knowledge is measured and suggests that incorrect and “don’t know” answers should not be grouped together because they imply very different processes. Individuals can be informed,

3

misinformed, uninformed, and partially informed. A “don’t know” reply by a respondent can also indicate that the respondent has a low confidence level. For these reasons Mondak suggests that incorrect and “don’t know” responses should not be combined. Partially informed and misinformed individuals are getting information, although it is incomplete in some cases and misinterpreted in others, while uninformed individuals are not obtaining or retaining any information at all. These differences in degree are important, and valuable information on individuals’ levels of political knowledge is lost when uninformed individuals are lumped together along with individuals who might know the answer but for whatever reason are not confident enough to respond positively. The idea that knowledge scales might tap into something other than knowledge is also important to realize (Mondak, 1999: 61). Confidence levels, propensity to take risks, and social desirability can all influence whether individuals will respond to knowledge questions. They are particularly significant when comparing the political knowledge of the sexes. Men are usually more confident in expressing their opinions, particularly on political matters that have been traditionally assigned to them as their domain, while women might be more hesitant to voice their opinion. Although, Mondak is argues that incorrect answers and “don’t know” answers should not be collapsed, it is not clear how these responses should be handled. One option is to word survey questions so as not to give survey respondents the opportunity to provide a “don’t know” response. Alternatively, one can include all responses in the dependent variable and treat them as ordered or discrete alternatives (Mondak, 1999: 80). The questions of how to measure knowledge and the realization that these measures can measure attributes other than political knowledge is important to realize because of the potential affects they can have on empirical results. Possible Interviewer Effects Another factor that affects how willing men and women are to provide answers to knowledge questions is the gender of interviewer effect. The gender of the interviewer of a survey may seem like an insignificant variable to consider, particularly in telephone surveys where the respondent does not directly interact with the interviewer in person; however, interviewer gender can have important effects on survey response, especially to political knowledge questions. Banducci (2003) examines this effect and hypothesizes that men may be more likely to overestimate interest in politics and political knowledge when interviewed by a woman, while women may be more likely to do so with male interviewers. She also argues that, because there are generally more female than male interviewers, differences based on gender-of-interviewer effects may be greatly exaggerated. Banducci presents two explanations for why the gender of the interviewer matters: (1) social desirability--i.e., the need to conform to desirable behavior and attitudinal expectations; and (2) disconfirming negative stereotypes--i.e., the need to disconfirm negative stereotypes that individuals expect the interview to hold (Banducci, 2003: 3). Social desirability could explain why men are so willing to offer responses to survey questions. The general argument is that politics is perceived to be a man’s world, and so men perceive that it is their duty to know about politics and offer responses to political questions, even if they do not know anything about the issue or are poorly informed. This would especially be the case when the interviewer is a woman, as men would feel the need to live up to society’s expectations. Women, on the other hand, might be expected to engage in disconfirming negative stereotypes about themselves. Because politics is perceived as a man’s world, women may overestimate their interest and/or knowledge in order to show that women too can follow, understand, and participate intellectually in politics. This would happen with a male interviewer because women might be afraid to display ignorance to a male interviewer. Banducci (2003: 5) suggests considering surveys as gendered conversations, contending that “men will tend to answer survey questions in a way that assures their status and power while

4

women in trying to reach consensus will avoid conflict or answers that are threatening to the interviewer.” She finds that there are consistent gender differences in engagement, but only for certain types of questions, such as those involving political interest and newspaper readership. Ultimately, she finds that both men and women show conforming behavior with male interviewers. She finds only one case where men exhibit conforming behavior with a female interviewer. Instead she found that male interviewers elicited more conforming behavior with both men and women, and this results in an overestimate of political knowledge levels. Campaign Context and Ease of Information Acquisition Kenski and Jamieson (2001) examine the gender gap in political knowledge in relation to the 2000 presidential election campaign. They hypothesize that “gender will be a significant predictor of political knowledge with females knowing less political information about the presidential candidates than male counterparts” (2001: 2). Political campaigns are important sources of information, but the authors suggest that this information is not beneficial to all. More highly educated individuals will have easier access to campaign information and they will have the ability to process more complicated and in-depth information than less-educated individuals. As a result, even a highly informative campaign can increase the knowledge gap between highly educated and less educated individuals. Kenski and Jamieson find that as the 2000 election campaign progressed the gap between highly educated and less educated individuals increased, while the gap between men and women decreased. They suggest that as elections draw near women may gain knowledge relative to men because more of the information that they see as relevant is presented later in the campaign, while earlier in the campaign other less relevant information is presented. Kenski and Jamieson conclude by suggesting that women use online processing (i.e., processing information and then forgetting it), while men record information and remember it; moreover, women use more elaborate cognitive measures as elections draw near and can hence close the gap in political knowledge more quickly. Social Capital and Social Interaction Another important factor that affects political knowledge is social capital. Robert Putnam (2000) in his book Bowling Alone assumes that “social capital allows political information to spread (Putnam 2000: 343). Gidengil et al. (2003) examine the connection between gender, knowledge, and social capital. The main question they study is, “Do women’s associational involvements and informal social networks bring them into contact with people who like to talk about politics, and if they do, does this stimulate greater political discussion on women’s part?” Social networks are important because it is through contact and discourse with others that information is passed along. As such, they are or can be significant sources of political information. Gidengil et al. find that just belonging to a social group is not enough. The type of association to which one belongs has a significant impact on the type of information to which members are exposed. They find that men and women belong to different types of social groups and are exposed to different types of information. Women tend to belong to more homogeneous, community-based groups, while men tend to belong to more heterogeneous, instrumental groups. This difference is important because women are less exposed to individuals with different views and ideas while men are exposed to a diverse group of people through work and their groups. Community-based groups thus tend to isolate those who belong to the group from the broader world of information while instrumental groups help broaden horizons because there are more diverse members. The idea of information quality over information quantity is confirmed in their finding that women who belonged to four or more groups have barely as much knowledge as men who

5

belonged to no groups. Through their work, men are exposed to people in different occupations, with different views, and with different values. This exposure provides the opportunity to learn and, most importantly, discuss political events. The authors come to the conclusion that until women become more involved in job-related, instrumental organizations that provide many opportunities for political discourse, they will have a hard time catching up to men in the area of political knowledge. Interest and Political Engagement Verba, Burns, and Schlozman (1997) examine the effects of gender on political participation and engagement. They find that women are “less politically interested, informed, and efficacious than men and this has consequences for political participation” (1997: 1051). Because men are more interested in politics they acquire the resources that are necessary for political participation. Women, on the other hand, do not acquire these resources as readily because of lack of interest and/or lack of opportunity. However, Verba et al. agree with previous authors that the acquisition of political information seems to be domain specific (1997: 1054). For instance, women may know a great deal about school politics because that is an area that has been traditionally assigned to women; arguably, women have a vested interest in school politics because they are the primary caregivers of their children and school directly affects them and their children. Resources are factors that greatly determine whether and how much individuals participate in politics, as well as how much they know about politics. Resources include education, income, and civic skills. These resources have been traditionally more available to men than to women; however, Verba et al. find that even these differences do not fully account for the gender gap in participation. They find, for instance, that women who live in a state with a female U.S. senator showed an increase in political knowledge (1997: 1066). This helps to demonstrate that political knowledge is selective. Women become more interested in politics when there are female politicians involved. This supports Kenski and Jamieson’s (2001) finding that women become more interested in politics when what they perceive as more relevant information is presented. Verba et al. conclude that, “[g]ender differences in political interest and information seem to reflect a genuine difference in the taste for politics” (1997: 1070). It may be that men and women are interested in different things, with men being more interested in politics than women. The authors suggest that future study should focus on how these preferences are formed in boys and girls. Political interest, knowledge, and participation have serious consequences for representation. If one segment of society makes its voice heard through various channels of participation then that group will have a better chance of representation than a group that does not participate. ********** In sum, the literature on political knowledge and engagement points to a number of possible explanations for the gender gap in political knowledge. Some scholars see gender-based knowledge differences as at least partly an artifact of methodology, particularly in terms of how survey instruments measure political knowledge, as well as how incorrect and “don’t know” responses are to be interpreted. Others see gender differences as “real,” insofar as they exist and are explainable by differences in individuals’ characteristics (e.g., political attitudes, demographic attributes, cognitive abilities) or the contexts within which they reside. There appears to be some validity to these approaches to studying gender differences in political knowledge, though additional research is needed to disentangle the level of empirical support for these two approaches.

6

MODELLING GENDER DIFFERENCES IN POLITICAL KNOWLEDGE What explains gender differences in political knowledge? Why are men more knowledgeable about politics than women? In this section we lay out the general contours of a model of political knowledge, focusing on the effects of gender as well as on the effects of independent variables that are related to political knowledge and that possibly differentiate men and women. In particular, we consider the effects of variables that represent individuals’ capacity to learn about politics, their exposure to political information, their general engagement in politics, and characteristics of their personal contexts that promote political information acquisition. We should note from the outset that we rely on two data sets for our analysis. The first, the 2000 American National Election Study survey, is a standard survey conducted before and after the 2000 national elections. Most readers will be well aware of the ANES series. On the other hand, the 2002 Louisiana Survey will be less familiar to most readers. The Louisiana Survey was conducted in June 2002, with 1103 completed interviews. While the focus of the survey was the spending and taxation preferences of Louisiana citizens (Garand and Blais, 2003), we also included a wide range of items on respondents’ knowledge of various national and state political figures, as well as data on a variety of independent variables that should be related to political knowledge. The inclusion of national and state political knowledge items differentiates the 2002 Louisiana Survey from the 2000 ANES and other ANES surveys. A descriptive summary of the variables used in this study can be found in Appendix 1. Dependent Variable: Political Knowledge As a starting point, it is important to discuss our measure of political knowledge, which serves as the dependent variable in each of our models. The measurement of political knowledge has drawn considerable attention in the literature (Delli Carpini and Teeter, 1996; Mondak, 1999, 2001). There is a wide range of political knowledge domains, including knowledge of political institutions and processes, people and players, domestic politics, foreign affairs, national politics, and state and local politics (Delli Carpini and Teeter, 1996; Niemi and Junn, 1999). Perhaps the most widely-used domain is what Delli Carpini and Teeter refer to as the “people and players” domain, which involves asking survey respondents if they can identify specific political figures. These questions are a regular feature of the ANES surveys, and knowledge scales based on these items have high reliability and validity. Typically, the wording of these questions is as follows: Now we have a set of questions concerning various public figures. We want to see how much information about them gets out to the public from television, newspapers, and the like. To help us do that, I’d like to ask you some questions about your knowledge of politics. Most people will not know the answers to many of these; if you don’t know, don’t worry about it, just tell me and we’ll move on to the next one. The first is Dick Cheney. Do you happen to know what job or political office he now holds? Respondents are typically asked to identify a series of political figures, with three coding options available: (1) correct answer; (2) incorrect answer; and (3) don’t know response. For the purposes of creating general knowledge scales, we recoded the answers to each item so that it is coded as a dichotomous variable, with correct answers coded 1, and other responses coded 0. Knowledge scales are created by summing the individual dichotomous variables, and these scales typically have high levels of reliability and validity.

7

For our data from the 2000 ANES, we base our knowledge scales on six political knowledge items. Each of these items involves asking whether or not respondents can correctly identify political figures or information about the presidential candidates: (1) Trent Lott, U.S. Senate majority leader; (2) William Rehnquist, Chief Justice of the U.S. Supreme Court; (3) Tony Blair, Prime Minister of Great Britain; (4) Janet Reno, Attorney General of the United States; (5) the home state for George W. Bush (Texas); and (6) the home state for Al Gore (Tennessee). Each of these items is coded 1 for correct responses, and 0 otherwise.1 We sum these six items together, creating a knowledge scale ranging from 0 (low knowledge) to 6 (high knowledge). The scale has a reasonable level of reliability (α = 0.76). For the 2002 Louisiana Survey, we base our knowledge scale on a somewhat broader set of questions, some of which reflect knowledge of national political figures and some of which reflect knowledge of state political figures. National knowledge questions include items in which respondents are asked if they can identify the following political figures: (1) Dick Cheney, VicePresident of the United States; (2) William Rehnquist, Chief Justice of the U.S. Supreme Court; (3) Tom Daschle, U.S. Senate minority leader; and (4) Dennis Hastert, Speaker of the U.S. House of Representatives. The reliability for an additive scale based on these four items is at acceptable levels (α = 0.69). The state knowledge scale is based on items on whether or not respondents can identify the following state political figures: (1) Mary Landrieu, junior member of the U.S. Senate from Louisiana; (2) John Breaux, senior U.S. senator from Louisiana; (3) Mike Foster, Governor of Louisiana; (4) Fox McKeithan, Louisiana Secretary of State; and (5) Charles DeWitt, Speaker of the Louisiana House of Representatives. A scale based on these five items has borderline reliability (α = 0.66). However, a global political knowledge scale based on all nine items has stronger reliability (α = 0.79). We focus our attention on the global political knowledge scale, though parts of our analysis are directed at the narrower national and state scales. Moreover, in order to ascertain the effect of gender on correct, incorrect, and “don’t know” responses, we also estimate separate models for each of the specific knowledge items. Independent Variable: Gender Obviously, the key independent variable in our analysis is gender, which we measure as a dichotomous variable, coded 1 for women and 0 for men. If there is a gender difference in political knowledge, we would expect the coefficient for the gender variable to be negative, indicating that women have lower levels of political knowledge than men. One of our key concerns is what happens to the gender coefficient as we include additional independent variables to our model that are designed to account for those gender differences. In a simple bivariate model (i.e., without control variables), the coefficient for gender is expected to be negative. The intercept in this model represents the mean level of political knowledge for men (i.e., when gender = 0), while the coefficient for the gender variable represents the difference in mean levels of political knowledge for men and women. Of course, the mean level of political knowledge for women can be obtained by adding the intercept and the b coefficient. However, when one begins to include theoretically-relevant control variables (e.g., education, strength of partisanship, etc.), the coefficient for the gender variable should be diminished to the extent that the additional independent variables differentiate men and women. If we include independent variables that account for the gender gap in political knowledge, the coefficient for gender should be reduced to a value of 0. Analytically speaking, our task is to model knowledge so that the coefficient for gender approaches a value of 0 and is rendered nonsignificant. If this outcome occurs, it means that we

8

have accounted for the explanations of gender differences in the additional independent variables included in our model. Independent Variables: Socioeconomic and Demographic Characteristics The effect of socioeconomic and demographic characteristics on political knowledge has been well documented in the literature (Delli Carpini and Keeter, 1996; Gidengil et al., 2003; Mondak, 1999). We include several socioeconomic and demographic variables in our models. First, in previous research most scholars have found that age is positively related to political knowledge, suggesting that older individuals are more knowledgeable about politics than younger citizens. Presumably, learning about politics is a life-long process, and individuals who have experienced political life and have been exposed to the political world over the course of the life span are more likely than younger individuals with less political experience and exposure to have accumulated high levels of political knowledge. Given this, we measure respondents’ age as the chronological number of years since birth. We hypothesize that the coefficient for age will be positive, indicating that older individuals have higher levels of political knowledge. Second, educational attainment is one of the strongest predictors of political knowledge, with education effects observed for all types of political information (Delli Carpini and Keeter, 1996). Formal education has an effect on political knowledge in at least three ways. First, education provides individuals with direct exposure to information about politics, primarily through courses in civics, history, and other social sciences (Niemi and Junn, 1999). Second, education provides individuals with learning skills, whereby they are more likely to internalize and organize information about politics in their daily lives. Third, individuals with high levels of education are likely to begin life with higher levels of cognitive capability, which is translated into a greater propensity to learn about the political world. We measure education in both the 2000 ANES and the 2002 Louisiana Survey using the same seven-point scale of formal educational attainment, ranging from 0 (less than 9th grade completed) to 6 (advanced degree). We hypothesize that education will be positively (and, we expect, strongly) related to political knowledge. Third, the effect of income on political knowledge has drawn some scholarly attention. Specifically, Delli Carpini and Keeter (1996) find that individuals with high incomes have higher levels of political knowledge, raising questions about the class-based nature of political information and the record of political mobilizations institutions targeted toward the working class in the United States. One might speculate that the self-interest rationale for seeking information would be invariant across income groups; arguably, those at the lower income levels have as much at stake in learning about politics as those with high income. However, previous work—most notably, Delli Carpini and Keeter (1996)—has called this into question. In order to capture these possible effects, we include a measure of family income in each of our models. This variable is measured in the 2000 ANES using a 22-point scale of family income, ranging from 0 (less than $4,999 per year) to 21 (greater than $200,000 per year). For the 2002 Louisiana Survey, we utilize an eight-point scale ranging from 0 (under $10,000 per year) to 7 (over $70,000 per year). If family income has the hypothesized effect on political knowledge that is compatible with previous research, we would expect the coefficient for family income to be positive. Fourth, we hypothesize that levels of political knowledge will vary as a function of race and ethnicity. Delli Carpini and Keeter (1996) and Davis and Silver (2003) find that blacks have lower levels of political knowledge than whites; for Delli Carpini and Keeter, this is true even in the face of statistical controls that might be thought to explain the black-white knowledge gap. We speculate that a similar pattern will be exhibited for Hispanic respondents, who along with language differences face many of the knowledge constraints faced by black citizens. In both of our data sets

9

we create a black respondent variable, coded 1 for black respondents, and 0 otherwise. We also create an Hispanic respondent variable, coded 1 for Hispanic respondents and 0 otherwise. Because of the small number of Hispanics in the Louisiana sample, we measure this Hispanic variable only for the 2000 ANES survey. We hypothesize that the coefficients for these two variables will be negative, indicating lower levels of knowledge than similarly situated white respondents. Finally, we consider the self-interest effect of home ownership on political knowledge. Home owners have made a substantial investment of resources to purchase their homes, and the are likely to be acutely aware of anything that affects their investment. They obviously have a special interest in government activity, particularly as it relates to taxes, housing values, interest rates, and the quality of life in their communities. Arguably, this has the incentive for home owners to seek out and retain political information. We measure home ownership as a dichotomous variable, coded 1 for home owners and 0 otherwise. We hypothesize that the coefficient for home ownership will be positive, indicating that home owners have higher levels of political knowledge than other individuals. Independent Variables: Political Attitudes and Engagement We also consider the possibility that individuals’ political attitudes and level of political engagement are related to political knowledge. In particular, we contend that the intensity and direction of political attitudes, attentiveness to and interest in politics, and feelings about the responsiveness and trustworthiness of the political system will all affect individuals’ level of political knowledge. Intensity and direction of political attitudes. First, we suggest that the intensity and direction of partisan identification will be related to levels of political knowledge. Simply, we hypothesize that strong partisans will be significantly more knowledgeable, since they are likely to hold more intense feelings about politics and will be more likely to seek out information about politics. We also consider the possibility that there are partisan differences in political knowledge, though the direction of the coefficient is not clear on theoretical grounds. In order to capture these effects, we measure partisan identification in both surveys as a seven-point scale, ranging from 0 (strong Democrat) to 6 (strong Republican). We then create a measure of folded partisanship, denoted strength of partisanship and measured as a four-point scale ranging from 0 (pure independent) to 3 (strong partisan). We hypothesize that strength of partisanship will be positively related to political knowledge. The expected direction for the coefficient for the raw partisan identification variable is less clear, though we include it in the model to capture potential partisan direction effects. Second, another important political attitude is ideological identification. As is the case with partisanship, we speculate that intensity of ideological identification will be positively related to political knowledge. Simply, strong liberals and strong conservatives should have higher levels of political knowledge than moderates, in large part because of the intensity of viewpoint held by strong ideologues. We also consider the possibility that there are knowledge differences across different ideological groups, with liberals possibly being more or less knowledgeable than conservatives. In both surveys we measure ideological identification using a seven-point scale, ranging from 0 (strong liberal) to 6 (strong conservative). As is the case with partisanship, we create a folded measure of ideology, measured as a four-point scale and ranging from 0 (moderate) to 3 (strong ideologue). We speculate that strong ideologues will have higher levels of political knowledge than moderates, and hence the coefficient for folded ideology should be positive. As is the case for partisanship, the directional effect for the raw ideological identification measure is less clear, though we include it in the model to capture potential directional effects of ideology.

10

Political engagement. We suggest that individuals who are engaged in politics—i.e., those who pay attention to and are interested in politics—will be more likely to seek out information and develop knowledge about politics. We include several political engagement variables. First, in both surveys we use items designed to measure respondents’ interest in politics. For the 2000 ANES we rely on an item asking individuals about how interested they are and how much they pay attention to political campaigns; this variable is measured as a three-point scale, ranging from 0 (not much interested) to 2 (very much interested). For the 2002 Louisiana Survey, the focus of the question is slightly different, insofar as we asked respondents about how much attention they pay to “government and politics.” Here again, this variable is measured as a three-point scale, ranging from 0 (not much interested) to 2 (very much interested). Moreover, we include two additional items in the 2002 Louisiana survey, one asking about how much attention respondents pay to national news about “government affairs and politics,” the other asking how much attention respondents pay to “local news on government affairs and politics in the state of Louisiana.” Both of these items are coded as five-point scales, ranging from 0 (none) to 4 (a great deal). It should be noted that these more specialized items are only included in the models relating to national and state political knowledge. For both data sets, we hypothesize that interest in and attentiveness to politics will be positively related to political knowledge. Second, for the analysis of data from the 2000 ANES survey only, we include two additional political engagement variables as independent variables in our model. We measure the degree to which individuals follow “government and public affairs” using a four-point scale, ranging from 0 (hardly at all) to 3 (most of the time). We expect the coefficient for this variable to be positive., indicating that individuals who follow politics will have higher levels of political knowledge than other individuals. Moreover, we include an item from the 2000 ANES that measures the degree to which the respondent considers him- or herself more opinionated than others. This variable is measured as a five-point scale, ranging from 0 (a lot fewer than average) to 4 (a lot more than average). Here again, we expect this variable to be positively related to political knowledge, controlling for the effects of other variables.2 Attitudes toward government and politics. We also suggest that how citizens view the government and the American political system will have a strong effect on political knowledge. First, external political efficacy is the degree to which individuals perceive government as being responsive to them. We suggest that individuals who exhibit high levels of external efficacy will see a payoff in the acquisition of political knowledge, insofar as they perceive that they can use that knowledge as currency to influence a political system that pays attention to their preferences. On the other hand, individuals who see government as being unresponsive to them will be less likely to see a positive result from the acquisition of knowledge, and so these individuals will have relatively low levels of political knowledge. We measure external political efficacy as factor scores obtained through principle components analysis of the following two items: (1) degree of agreement with statement that "I don't think public officials care much what people like me think;" and (2) degree of agreement with statement that “People like me don't have any say about what the government does. For the 2000 ANES, the eigenvalue is 1.41, with 70.5% of the variance explained. For the 2002 Louisiana Survey a full range of efficacy items are not available, so we measure efficacy based on responses to the following item: “Over the years, how much attention do you feel that Louisiana government pays to what people think when it decides what to do—a good deal, some, or not much?” This variable is coded as a three-point scale, ranging from 0 for low efficacy, and 2 for high efficacy. We hypothesize that external political efficacy will be positively related to political knowledge. Second, we consider the effects of political trust on levels of political knowledge. We suggest that individuals who have trust in the political system will also see a payoff in gaining knowledge

11

about politics, while those who distrust government will consider the acquisition of knowledge about government an politics a relative waste of time. To measure political trust, we generate factor scores through a principle components analysis of the following four items: (1) the degree to which respondents think that government wastes money paid in taxes; (2) whether respondents would say that government is run by a few big interests; (3) how much of the time respondents would say that they can trust the government in Washington to do what is right; and (4) the extent to which respondents think that the people running the government are crooked. For the 2000 ANES the eigenvalue for this analyisis is 1.93, with 48.4% of the variance explained. For the 2002 Louisiana Survey, we include in our political trust scale the first, third, and fourth items listed above, with the questions reconfigured to make Louisiana government and politics the target of the items. The eigenvalue for the analysis generating the trust factor scores is 1.80, with 60% of the variance explained. Third, there is considerable tension in American society between those who advocate greater provision of goods and services by the public sector and those who advocate the allocation of goods and services through free market mechanisms or based on individual initiative. We suggest that individuals who are opposed to government provision of goods and services are likely to be highly motivated and actively engaged in that opposition, and hence these individuals will have higher levels of political knowledge than other citizens, controlling for the effects of other variables. In both the 2000 ANES and 2002 Louisiana Survey, we use three items that represent a forced choice between pro-government and anti-government approaches to solving policy problems. The questions used to measure this scale are as follows: Next, I am going to ask you to choose which of two statements I read comes closer to your own opinion. You might agree to some extent with both, but we want to know which one is closer to your views. One, the less government, the better; or two, there are more things that government should be doing. 1. The less government the better 2. More things government should be doing One, we need a strong government to handle today's complex economic problems; or two, the free market can handle these problems without government being involved. 1. Need a strong government to handle complex economic problems 2. Free market can handle without government involvement One, the main reason government has become bigger over the years is because it has gotten involved in things that people should do for themselves; or two, government has become bigger because the problems we face have become bigger. 1. Government is bigger because involved in things people should handle themselves 2. Government has become bigger because problems are bigger Using these items, we have created a scale, denoted the antigovernment scale, based on a principal components analysis of these three items. For the 2000 ANES, the eigenvalue is equal to 1.98, with 66% of the variance explained. For the 2002 Louisiana Survey, the eigenvalue is equal to 1.85, with 62% of the variance explained. High values on this scale represents hostility to government provision of goods and services, while a low score represents support for government provision. We

12

expect that those who advocate provision of goods and services through the free market or through individual initiative will be motivated by their opposition and hence more knowledgeable about politics; hence, we hypothesize that the coefficient for this variable will be positive. Fourth, citizens differ in terms of the degree to which they perceive policy differences between the two political parties. Some citizens see the Democrats and Republicans as the proverbial “Tweedledee and Tweedledum,” insofar as they perceive the Democrats and Republicans as staking out similar policy positions. For these citizens, there may be little reason to invest substantial time and effort in acquiring political knowledge, since the electoral choices confronting them do not matter very much at all. On the other hand, other individuals perceive substantial differences between the two parties, usually seeing the Democrats as the relatively liberal party and the Republicans as the relatively conservative party. For these citizens, the stakes in politics are substantial, insofar as these policy differences mean that it really matters which party controls government. In order to measure polarized party placement, it is necessary to have respondents’ placement of the Democratic and Republican parties. These data are available for the 2000 ANES, but not for the 2002 Louisiana Survey. Using data from the 2000 ANES only, we measure polarized party placement as the absolute value of the distance between respondents’ placements of the Republican and Democratic parties. We expect voters who perceive polarized parties to be more knowledgeable than voters who perceive the political parties as being similar, so the coefficient for this variable should be positive. Independent Variables: Media Exposure Knowledge about politics does not materialize out of thin air. Rather, individuals must be exposed to the information from family, close friends, or the media. Arguably, one of the most important predictors of political knowledge is exposure to various news media. Individuals who are regularly exposed to the news media are more likely to be exposed to information about politics, which is a prerequisite for developing knowledge about politics. In our analyses of both the 2000 ANES and 2002 Louisiana Survey data, we include several media variables as predictors of political knowledge. These media variables represent different kinds of media, which appear to differ in the degree to which they effectively transmit information to individuals. First, we create a variable to represent individuals’ exposure to national television news. For the 2000 ANES, this variable is measured as the number of days in which the respondent watches national television news; hence this variable ranges from 0 to 7. For the 2002 Louisiana Survey, we adopt an alternative measurement strategy, relying on individuals’ responses to the question of whether or not they had watched national television news in the past 24 hours. In both cases we expect the coefficient for this variable to be positive, indicating that those who watch national television news on a regular basis will have higher levels of political knowledge. Second, we consider the effects of local television news on political knowledge. For the 2000 ANES, we rely on two items relating to local news viewership. The first deals with how many days respondents watch early evening local television news, while the second involves how many days respondents watch late night local television news. We create a simple additive scale that combines the values of these two variables; hence this variable ranges from 0 (i.e., respondent does not ever watch local television news) to 14 (i.e., respondent watches both the early evening and late night local television news every day). For the 2002 Louisiana Survey, we measure this variable as a simple dichotomy, coded 1 for individuals who have watched local television news in the past 24 hours, and 0 otherwise. We hypothesize that the coefficient for these variables will be positive.

13

Third, newspaper news is generally thought of as a superior form of distributing information about politics. For the 2000 ANES we measure this variable as the number of days in which respondents read a daily newspaper; this variable ranges from 0 to 7. For the 2002 Louisiana Survey, we code this variable 1 for respondents who have read a daily newspaper in the past 24 hours, and 0 otherwise. Here again, we expect newspaper readers to have higher levels of political knowledge than other similarly situated citizens, so the coefficient for this variable is expected to be positive. Independent Variables: Personal Life Circumstances One set of variables that might be related to political knowledge involves personal life circumstances. These personal life circumstances are often thought to explain the gap in levels of political knowledge for men and women. In particular, many observers contend that women have joint responsibilities for work and family that many men do not have, and the result, so the argument goes, is that women are more likely to be so overburdened by work and family that they do not have adequate time to acquire detailed information about politics. We include several personal life circumstances variables in our models. First, for the 2000 ANES survey we consider respondents’ marital status; this variable is coded 1 for married respondents, and 0 otherwise. We hypothesize that married individuals will have higher levels of political knowledge, so the coefficient for this variable should be positive.3 Second, we consider the effects of children on political knowledge. As any parent will state, raising children is a time-consuming activity, and this diminishes the amount of time that parents can spend on learning about politics. For the 2000 ANES, we utilize a measure to capture the effects of having school-age children; this variable is coded 1 for respondents who report having a child under the age of 18, and 0 otherwise. We hypothesize that the coefficient for this variable will be negative, indicating that parents of children are expected to have lower levels of political knowledge. For the 2002 Louisiana Survey, we use a question designed to ascertain if respondents have children of school age. We recode this variable as 1 for individuals who have children of school age, and 0 otherwise. We also are able to determine if respondents have children older than 18 years; this variable is coded 1 for respondents with adult children, and 0 otherwise. Third, we combine information from the gender, marital status, and child variables to create a variable for single motherhood. This variable is coded 1 for single mothers—i.e., women who are single and have children—and 0 otherwise. Being a single parent is often very stressful, if for no other reason that single parents usually have sole responsibility for raising their children. Given the burdens of raising children in a single-parent family, we hypothesize that single mothers will have lower levels of political knowledge.4 Fourth, we explore the effects of employment on political knowledge. The direction of the effect of employment is unclear. On one hand, employment gives individuals exposure to political information through their interaction with coworkers; hence one would expect employment to have a positive effect on political knowledge, both for men and women. On the other hand, employment is a time-consuming activity that can make it difficult for individuals to take sufficient time to learn about politics; hence one might expect that employment and political knowledge will be negatively related. We include in our model an employment variable, coded 1 for employed respondents and 0 otherwise. Given the ambiguous expected direction for this coefficient, we test our hypothesis of employment effects using a two-tailed test of statistical significance.

14

It is also the case that employment does not act in a vacuum. Rather, employment may combine with gender and parenthood to affect individuals’ level of political knowledge. By itself, employment may or may not increase or decrease political knowledge. But the combination of employment and parenthood might be expected to diminish political knowledge, particular if, as many commentators believe, women are more likely than men to take on family responsibilities in addition to their work obligations. Hence, it is possible that the gender gap in political knowledge is really due to which women are relatively overburdened compared to men; if this is the case, it may be the case that men have more time and energy to seek out and acquire knowledge about politics. Moreover, if this is the case, then women who are not both employed and a parent should have similar levels of political knowledge as men, controlling for the effects of other independent variables. We include a series of separate interaction variables for (1) gender and employment, (2) gender and parenthood, and (3) gender, employment, and parenthood. If employment and parenthood each have a depressing effect on political knowledge for women, but less so for men, the coefficient for the gender-employment and gender-parenthood interactions should be negative. But more importantly, if parenthood and employment combined have a stronger effect on reducing political knowledge for women than men, the coefficient for the three-way interaction of gender, employment, and parenthood should be negative. Independent Variables: Interviewer Effects We also account for the possibility that interviewer characteristics have an impact on individuals’ level of political knowledge. One process that might generate a relationship between interviewer characteristics and political knowledge involves the stereotype threat hypothesis, which has had substantial currency in recent years. This thesis is most closely associated with the work of Steele (1997), who suggests that individuals who are the target of a widely-held negative stereotype about their competence are likely to under-perform when confronted with a task—usually a test— that relates to that stereotype. For instance, Steele and Aronson (1995) find that describing an exam as “a test of intellectual ability” generates a stereotype threat for African Americans, who do less well on the test than other African Americans who had the test introduced to them in a mannuer that does not activate the stereotype threat. The idea is that African Americans are aware of the stereotype that they have lower levels of intellectual ability, and when the stereotype is activated they respond with relatively weak performance. Similarly, Davis and Silver (2003) find that African Americans exhibit lower levels of political knowledge than whites when they have a white interviewer (i.e., high threat) than when they have a black interviewer. Consistent with the stereotype threat thesis, the levels of political knowledge for whites are unaffected by the race of interviewer. We consider the possibility that gender differences in political knowledge are at least partly a function of the gender of interviewer. In particular, we suggest that there is a widely-held stereotype that women are less knowledgeable about politics than men. If so, women confronted with political knowledge questions posed by male interviewers should have lower levels of knowledge than women answering political knowledge questions posed by female interviewers, since those negative stereotypes are likely to be engaged for women with male interviewers. Male respondents should exhibit similar levels of political knowledge, regardless of the gender of the interviewer. Unfortunately, data on the gender of interviewer are not available for the 2002 Louisiana Survey, but there is an interviewer gender variable available for the 2000 ANES. Hence for this data set we create an interviewer gender variable, coded 1 for female interviewers and 0 for male interviewers. Moreover, to estimate the effect of stereotype threat, we create an interaction variable for respondent gender and interviewer gender; this variable is coded 1 for female respondents interviewed by female interviewers, and 0 otherwise. If the stereotype threat process is at work, one

15

would expect the coefficient for the simple interviewer gender variable to be equal to 0, but the coefficient for the interaction variable should be positive. Such a pattern in the coefficients indicates that female respondents whose stereotype threat is diminished by exposure to a female interviewer will exhibit higher levels of political knowledge than female respondents facing high stereotype threat implied by exposure to a male interviewer. EMPIRICAL RESULTS Patterns of Political Knowledge for Men and Women We begin with a simple descriptive summary of the gender gap in political knowledge for respondents in the 2000 ANES and 2002 Louisiana Survey. In Table 1 we present the distribution of responses to political knowledge items from the 2000 ANES, broken down by gender, while in Table 2 we do the same thing for political knowledge items from the 2002 Louisiana Survey. Respondents are classified into three categories on each item: (1) answered correctly, meaning that the respondent’s answer to the question matches the correct answer; (2) answered incorrectly, meaning that the respondent did attempt to offer an answer but that the answer was not correct; and (3) answered “don’t know” (hereafter denoted DK), meaning that the respondent was unable to offer an answer. Three things stand out for the results reported in Tables 1 and 2. First, men clearly have a greater propensity than women to provide correct answers to the questions posed in these two surveys. In Table 1 men are more likely to be able to identify Trent Lott (14.0% to 4.7%, a difference of 9.3%), William Rehnquist (18.0% to 4.9%, 13.1%), Tony Blair (41.3% to 29.3%, 12.0%), Janet Reno (66.5% to 46.4%, 20.1%), George Bush’s home state (94.2% to 86.0%, 8.2%), and Al Gore’s home state (75.8% to 61.5%, 14.3%). In Table 2 men are more likely to identify Dick Cheney (78.2% to 69.2%, 9.0%), Tom Daschle (44.7% to 27.0%, 17.7%), William Rehnquist (41.8% to 25.7%, 16.1%), Dennis Hastert (23.4% to 6.9%, 16.5%), Mary Landrieu (73.5% to 70.1%, 2.4%), Charles Dewitt (17.1% to 10.6%, 6.5%), John Breaux (73.5% to 63.6%, 9.9%), Fox McKeithan (41.3% to 31.2%, 10.1%), and Mike Foster (94.0% to 88.0%, 6%). The gender differences in correct answers range from 8.2% to 20.1% for 2000 ANES data and from 2.4% to 17.7% for 2002 Louisiana Survey data. In no case are women more likely than men to answer a given knowledge question correctly. Given these distributions, it would appear that men have higher levels of political knowledge than women. Second, if women are less likely to answer political knowledge questions correctly, how are they answering? A close examination of the results in Table 1 and 2 reveals that women are consistently more likely than men to respond to political knowledge questions with DK responses. For all of the items in Tables 1 and 2, the percentage of DK responses for women exceeds the percentage of DK responses for men. In some cases, the difference is substantial; for instance, 74.5% of women give DK responses when asked to identify Trent Lott, compared to 50.0% of men, for a difference of 24.5%. On the other hand, very few men (3.1%) or women (4.8%) give DK responses to the question about identifying George Bush’s home state, and the gap is only 1.7%. Overall, though, women are more likely than men to give DK responses. Third, men and women differ across specific items in their propensities to give incorrect answers, but overall men and women are about equally likely to give incorrect answers. The gender gap in incorrect answers lacks a consistent pattern. On some items, men are more likely than women to give incorrect answers, but on other items women are more likely to give incorrect answers. On several items men and women are relatively indistinguishable in terms of their

16

propensities to give incorrect answers. All in all, the differences in behavior between men and women are focused on the greater likelihood of men to give correct answers and the greater likelihood of women to give DK responses. Few scholars studying political knowledge will be surprised that women are less likely to give correct responses and more likely to give DK responses. How should these patterns be interpreted? On one hand, Mondak (1999) (and echoed by Gidengil et al., 2003) suggests that DK responses do not necessarily reflect a lack of knowledge. Rather, some individuals might have partial knowledge, and the DK response represents an unwillingness to risk an answer that might be incorrect. Some individuals with partial knowledge may be more willing than others to offer an answer than others, and perhaps men differ from women in ways that lead them to offer answers with only partial knowledge. If men with partial knowledge are more likely to answer knowledge questions than women with the same level of partial knowledge, then men will at least occasionally provide correct answers and hence overall appear to be more knowledgeable than women. Moreover, Mondak suggests that individuals may have psychological attributes (e.g., self-confidence, risk taking) that makes it more or less likely to offer answers to political knowledge questions when they have only partial knowledge. If men have greater levels of self-confidence and a greater propensity to take risks, they may be more likely than women to answer knowledge questions with only partial knowledge. While there is no doubt a great deal of validity to Mondak’s assertions, we are skeptical that these processes provide a full explanation for the gender gap in political knowledge. For one thing, men are clearly able to provide more correct answers than women, and the gender gap in correct answers is large enough to make it unlikely that it is a function of the greater propensity of men to guess when they have partial knowledge. Women are also roughly equally likely to provide incorrect answers, so women would have to be much more likely than men to have partial knowledge and funnel their answers into the DK category rather than make an effort to guess. Men with partial knowledge would also have to be more likely to guess and offer a correct answer. There would also have to be a great deal of correct guessing by men to result in the observed gender gap. All of this is possible, but given the open-ended nature of the questions that requires individuals to recall the correct answer rather than just recognize the correct answer from a set of alternatives, we think that it is unlikely to account for the overall gap in political knowledge levels for men and women. As noted, we also create political knowledge scales for the 2000 ANES and 2002 Louisiana Survey data. These scales are created by summing the number of correct answers across the component political knowledge items, and they serve as the dependent variables in much of our analysis. In Table 3 we report the means and distribution of these scales, broken down by gender. Here again, we find that men have consistently higher levels of political knowledge than women. For the general knowledge scale from the 2000 ANES, the knowledge scores for women are skewed toward the low end of the distribution; approximately 62% of women correctly answer two questions or less, compared to approximately 46% of men. Moreover, men correctly answer an average of 2.64 items, compared to 2.02 items for women; this difference in mean knowledge levels for men and women is highly significant (t = -8.00). In this national sample neither men nor women have overwhelmingly high levels of political knowledge, but these data suggest that women have even lower levels of knowledge than men. For the 2002 Louisiana Survey, a similar pattern is observed. For the 10-point general political knowledge scale, men have a mean number of correct answers of 4.88, compared to a mean for women of only 3.92; on average, men answer almost one more question correctly than women, out of a total of nine questions. Only 25% of women answer more than five questions correctly, while

17

40% of men are able to answer more than five questions correctly. This pattern extends to both the national and state political knowledge scales. Men answer 1.88 knowledge questions about national political figures correctly, compared to only 1.29 for women. The gender gap in political knowledge is narrower (but still discernible) for state political figures, with men answering 3.00 questions correctly compared to 2.63 correct answers for women. All of these differences in means are highly significant. All in all, men exhibit higher levels of political knowledge than women. This is true for a general knowledge scale measured in a national sample, as well as for a general knowledge scale and specific national and state knowledge scales measured in a Louisiana sample. The knowledge levels for women tend toward the low end of the knowledge distribution, while the knowledge levels for men are typically balanced around the middle of the knowledge distribution. The key question left unanswered is: what explains the gender gap in political knowledge? Why are men more knowledgeable about politics than women? Estimating Gender Effects for Gender Knowledge Scales In order to gain some leverage over these questions, we estimate a full model of political knowledge, with the general knowledge scales depicted as a function of a variety of socioeconomic and demographic variables, political attitudes and engagement, media exposure, and personal life circumstances. Our strategy is to estimate a full model that includes a range of independent variables that might be expected to differentiate men and women and that, therefore, might account for the significant effect of gender on political knowledge. If we include in our models control variables that account for gender differences, then the coefficient for gender should shrink to nonsignificance. 2000 ANES Survey. In Table 4 we report the OLS regression estimates for our full model of political knowledge, using data from the 2000 ANES. Model (1) includes socioeconomic, demographic, political attitudes, and media exposure independent variables, while in Model (2) we include the same set of independent variables as well as the personal life circumstances variables. As a starting point, we should note that in a bivariate model the coefficient for gender suggests a difference in political knowledge for men and women of about six-tenths of a point (b = -0.625, t = -8.00). The question is whether this gender gap in political knowledge is diminished by the inclusion of independent variables that should account for differences in political knowledge for men and women. In Model (1) of Table 4, we find that the coefficient for gender is negative and highly significant (b = -0.426, t = -5.60), though it is noteworthy that the size of the coefficient is reduced by about one-third by the inclusion of control variables. Taking into account the effects of the other independent variables in the model, men and women still differ by about four-tenths of a point in their levels of political knowledge. This suggests that we have not accounted for the full range of independent variables that explain gender differences in political knowledge. Several independent variables are found to have a strong effect on individuals’ level of political knowledge. Among the socioeconomic and demographic variables, the effects of education stand out; as education increases, so do levels of political knowledge (b = 0.235, t = 8.64). Race is also an important variable, with blacks exhibiting political knowledge levels that are almost seven-tenths of a point lower than similarly situated individuals in other racial groups (b = -0.682, t = -5.12). Some of the political attitudes and engagement variables are also important predictors of political knowledge. Attentiveness to politics, following politics closely, antigovernment attitudes, polarized party placement, and political efficacy all have significant positive effects on political knowledge.

18

Individuals who pay attention to politics, follow politics closely, are hostile to government solutions rather than private or free market solutions to policy problems, perceive the political parties as being different on policy grounds, and have high levels of political efficacy are much more knowledgeable about politics than other individuals. Finally, we find that media exposure is related to political knowledge. Watching national television news (b = 0.037, t = 2.19) and reading a daily newspaper (b = 0.051, t = 3.65) increase political knowledge, though it is interesting to note that getting news about politics from local television news has a null (or, if anything, a negative) effect on political knowledge. It would appear that national television news and newspapers are good sources of information about politics. In Table 4, Model (2) we add personal life circumstances to our model, and the result is a further diminution of the gender coefficient. Some observers have speculated that differences in political knowledge for women reflect differences in how women balance work and family responsibilities compared to men. Controlling for several variables representing these personal life circumstances, as well as the control variables included in Model (1), we find that the gender effect is reduced by about one-half (b = -0.325, t = -2.46). Part of this is due to the inclusion of interaction variables representing various combinations of personal life circumstances that one might expect to diminish political knowledge, especially among women. But it is important to note that the gender effect remains, even after estimating the effects of various personal life circumstances. It is noteworthy that only one of the independent variables reflecting personal life circumstances has an effect on political knowledge. Single mothers have significantly lower levels of political knowledge, controlling for the effects of other independent variables included in the model (b = -0.430, t = -2.11). This finding suggests that the well-documented stresses and timeconsuming nature of single motherhood diminishes the ability to acquire political information. Other personal life circumstances appear to have little effect on political knowledge. The coefficients for marital status, parenthood, and employment variables are all small and nonsignificant. Employment does not appear to have a different effect on knowledge for women and men (b = -0.040, t = -0.24), nor does employment with children (b = -0.075, t = -0.92). All in all, it appears that women continue to have lower levels of political knowledge than men, but that single parenthood accentuates gender differences in political knowledge. Among these independent variables, what accounts for knowledge differences between men and women? The coefficients in our model provide tantalizing evidence about the particularly characteristics of individuals who are most likely to have high levels of political knowledge. It is possible that these characteristics differ for men and women, and these differences could account for the fact that men have higher levels of political knowledge than women in a bivariate model but that this effect is moderately attenuated when other variables are included as predictors of political knowledge. Our next step is to ascertain which of our significant predictor variables actually differentiate men and women. The identification of these variables will begin to pinpoint the particularly characteristics that make men and women different in their levels of political knowledge. In Table 5 we report mean values on our independent variables, calculated separately for women and men. As one can readily see from these results, women and men are significantly different from each other in a number of ways. First, among the socioeconomic and demographic variables in the ANES sample, women have significantly lower levels of education and family income. Since education is so strongly (and positively) related to political knowledge, and since women respondents have, on average, significantly lower levels of education than men, it is likely that the gender gap in education at least partly accounts for the gender gap in political knowledge.

19

Family income is, on the other hand, unrelated to political knowledge, so the gender gap in income is unlikely to contribute much to the gender gap in political knowledge. Second, men and women differ considerably on several of the variables in the political attitudes and engagement category. Men and women differ significantly in terms of partisan identification, ideology, folded ideology, attentiveness to politics, follows politics, degree to which they are opinionated, political efficacy, and antigovernment attitudes. Among these variables, attentiveness to politics, follows politics, political efficacy, and antigovernment attitudes are positively related to political knowledge. Women are less attentive to politics, follow politics less frequently, have lower levels of political efficacy, and much lower levels of antigovernment attitudes; these attitudes and behaviors undoubtedly diminish political knowledge levels for women relative to knowledge levels of men. On the other hand, ideological conservatism is negatively related to political knowledge, so the fact that women are, on average, more liberal than men means that this contributes toward shrinking the gender gap in political knowledge. Taken as a whole, the political attitudes and engagement variables contribute to a widening of the gap in political knowledge for men and women. Third, media exposure also contributes to the gender gap in political knowledge. Women are both more likely to use local television news and less likely to use newspapers as sources of political information. Newspaper exposure has a strong positive effect on political knowledge, so the greater propensity of men to obtain political information from newspapers increases their level of political knowledge vis-à-vis that of women. Moreover, the greater likelihood that women use local television news might also have the effect of widening the gender knowledge gap, since local television news exposure depresses political knowledge, though this effect is only at the border of conventional levels of statistical significance. While we cannot be entirely sure of the local television news effect, we can be confident that the gender gap in political knowledge is enhanced by the greater likelihood that men rely on newspaper news. Finally, among the personal life circumstances variables, none both (1) have a significant effect (Table 4) and (2) differentiate men and women (Table 5). Hence it is unlikely that these variables contribute to the knowledge gap in political knowledge. 2002 Louisiana Survey. In Table 6 we also estimate a variation on our political knowledge model using data from the 2002 Louisiana Survey. We begin by noting that in a bivariate model the coefficient for gender suggests a difference in political knowledge for men and women of almost a full point (b = -0.952, t = -6.69) on a ten-point scale. Clearly women have lower levels of political knowledge then men in the Louisiana sample. As one can readily see from Model (1), gender continues to have a strong effect on political knowledge, even controlling for the effects of other independent variables (b = -0.729, t = -4.51). The magnitude of the coefficient is reduced somewhat in the full model, though controlling for other variables reduces the gender effect only about 25%. Obviously, gender has an independent effect on political knowledge above and beyond the effects of other variables. Many of the control variables in our Louisiana model are significantly related to political knowledge. All of the socioeconomic and demographic variables have significant effects on the dependent variable, with age, education, family income, and home ownership having positive effects on political knowledge. Black citizens have levels of political knowledge that are approximately four-tenths of a point lower than those of other citizens. Among the political attitudes and engagement variables, we find that the coefficients for partisan identification, folded partisanship, attentiveness to politics, and antigovernment attitudes are all positive and statistically significant; Republicans, strong partisans, those who are attentive to politics, and those who prefer

20

private sector or market-based responses to policy problems are more knowledgeable about politics. We also find that national television news and newspaper news exposure have positive effects on political knowledge, but that local television news exposure does not; this is in keeping with the findings from the 2000 ANES (reported above) that national television and newspaper news sources produce stronger political knowledge effects. In Model (2) we add personal life circumstances variables to the variables included in Model (1). Interestingly, the magnitude of the coefficient for gender increases (b = -0.890, t = -3.08) compared with the coefficient in Model (1), but this is no doubt due to the inclusion of various interaction variables. Regardless, the significant gender effects on political knowledge remain in place, and the coefficients for most of the other independent variables are of similar magnitude, direction, and significance. We also find that children have a depressing effect on political knowledge, with individuals with children having political knowledge levels that are about one-half of a point lower than those without children (b = -0.532, t = -2.48). Interestingly, none of the interaction effects are statistically significant, indicating that women are no more or less sensitive to employment, having children, or both. Now that we estimates of the effects of various independent variables on political knowledge, do any of these independent variables account for gender differences in political knowledge? As we did for the 2000 ANES data, we calculate mean differences for men and women on each of the independent variables, and these differences are reported in Table 7. If a given independent variable has a significant effect on political knowledge, and if men and women differ significantly in their values on that variable, then it is highly likely that that independent variable helps to account for some of the gender gap in political knowledge. As one can see, men and women differ significantly in their values on several key independent variables. First, there is a significant difference between men and women respondents in terms of family income, race, and home ownership, with women respondents differing from men respondents in terms of having significantly lower family incomes, a higher likelihood of being black, and a lower likelihood of being a homeowner. Since family income and home ownership are positively and significantly related to political knowledge, and since being black lowers political knowledge (see Table 6), the difference between men and women on these three variables very likely contributes to increasing the gender gap in political knowledge. Simply, women have attributes on three important predictors of political knowledge that are associated with lower knowledge. Second, in terms of political attitudes and engagement, women have significantly lower values on partisan identification, attentiveness to politics, political trust, and antigovernment attitudes, along with higher mean values on folded partisanship. Partisans identification (Republican), attentiveness to politics, and antigovernment attitudes all have very strong positive effects on political knowledge, yet women have significantly lower mean values on these variables; hence some of the gender gap in political knowledge can be explained by gender differences on these variables. Third, the effect of media exposure is a wash. Women have significantly greater local television news exposure, but this is the one media variable that is unrelated to political knowledge. Moreover, national television news and newspaper news exposure have positive effects on political knowledge, but neither of these two variables differentiate men and women significantly. Finally, none of the personal life circumstances variables appear to contribute to the gender gap in political knowledge. In sum, our analysis of the general political knowledge scales from the 2000 ANES and the 2002 Louisiana Survey reveals a consistent, strong gender effect. Regardless of model specification, the coefficients for gender are negative and significant, indicating that women have lower levels of political knowledge than men even in the face of a wide range of statistical controls.

21

Moreover, we find that many of the independent variables included in our models account for part of the gender gap in political knowledge. In simple bivariate models of political knowledge, the coefficients for gender are sizeable and statistically significant, but in the multivariate models the gender coefficients are decreased in magnitude. This suggests that the independent variables included in our models account for some of the linkage between gender and political knowledge. We find that several of our independent variables both differentiate men and women and have a significant effect on political knowledge. In particular, men and women differ in their values on several socioeconomic and demographic variables (i.e., education, income), political attitudes and engagement variables (i.e., attentiveness to politics, follow politics, antigovernment attitudes, polarized party placement), and media exposure. Several of these variables undoubtedly contribute to a reduction in the overall size of the gender coefficient. Nonetheless, even as we are able to account for some of the gender gap in political knowledge, there is a substantial amount of this gender gap that remains and is left unexplained by our model. Differences in Gender Effects for National and State Political Knowledge One possibility that is worth considering is that the effect of gender differs for national and state political knowledge domains. Specifically, some scholars speculate that men are generally more interested in (and hence more knowledgeable about) national politics, and that women are more interested in (and hence more knowledgeable about) more localized politics (Delli Carpini and Keeter, 1996; Darcy, Welch, and Clark, 1987). We do not have data on local political knowledge, but in the 2002 Louisiana Survey we asked respondents political knowledge questions that cover both the national and state political domains. To be sure, Delli Carpini and Keeter (1996) find that gender has a significant effect on national and state political knowledge, but not for local political knowledge. On the other hand, if women have a localized focus to their political interest, we would expect women to be more knowledgeable about politics as it gets closer to having a direct impact on their lives. Hence we expect the gender gap to be stronger for the national political knowledge scale than for the state political knowledge scale; for the latter, the gender gap in political knowledge should be diminished or close to 0. In Table 8 we report the OLS estimates for our separate models of national and state political knowledge. For national political knowledge, gender has a strong negative effect on the dependent variable (b = -0.451, t = -4.76). This coefficient suggests that, controlling for the effects of other independent variables in the model, men score about six-tenths of a point higher on the five-point national political knowledge scale than women. This coefficient is approximately 25% smaller in magnitude than the effect of gender on national political knowledge in a simple bivariate model (b = -0.592, t = -7.80), so controlling for the effect of other variables has a moderate effect on the magnitude of the gender coefficient in this instance. On the other hand, based on the estimates for the model of state political knowledge found in the second column of Table 8, we find that the effect of gender is substantially lower than in the national political knowledge model (b = -0.227, t = -2.23). Simply, gender has a weaker effect on political knowledge in the state politics domain, with women exhibiting levels of state political knowledge that are only about one-quarter of a point (on a six-point scale) lower than that of men. In a simple bivariate model for state political knowledge, the coefficient for the gender variable (b = -0.360, t = -4.27) is also smaller that the comparable coefficient for the national political knowledge scale. All of this is in keeping with speculations that the political interests of women are more likely to be manifested in state politics rather than national politics. For the state political knowledge

22

model, the gender effect represents only about 4.5% of the range of the six-point state political knowledge scale, while for the national political knowledge model the gender effect represents about 11.3% of the range of the five-point national politics scale. While the effect of gender is significant in both political knowledge domains, the gender gap is clearly smaller for state political knowledge than for national political knowledge. Gender Effects and Individual Knowledge Items This far we have explored the determinants of overall patterns of political knowledge by focusing attention on general political knowledge scales comprised of several specific political knowledge items. However, are there gender effects across a wide range of these knowledge items? In this section we explore the effects of gender on the ability of respondents to identify various political figures. We move to this disaggregated level of analysis for two reasons. First, it is possible that there is some variability in gender effects across specific knowledge items. The only way to explore these separate effects is to estimate separate models for each knowledge item. Second, and more importantly, using the individual knowledge items permits us to estimate the effect of gender on both correct responses and DK responses. In the first part of our analysis of individual items, we code our knowledge items as binary variables, coded 1 for correct answers and 0 for other responses. However, in the second part of our analysis of individual items, we code the knowledge items to include correct, incorrect, and DK responses, and this permits us to estimate a multinomial logit model to assess the degree to which gender is related to each of these responses. We begin by examining the degree to which gender is related to the likelihood of giving correct responses to each of our knowledge items. In Table 9 we report binary logit coefficients for the gender variable in our models for each of the knowledge items.5 Clearly, gender is strongly related to political knowledge, even controlling for the effects of other independent variables. For the knowledge items from the 2000 ANES, gender has a significant negative effect on the propensity to give a correct answer in five of the six cases. This suggest that women are explicitly less likely to be able to identify correctly Trent Lott, William Rehnquist, Janet Reno, and the home state of each of the presidential candidates. The only exception is the Tony Blair item; in this case the gender coefficient is negative, but it fails to achieve conventional levels of statistical significance. For the knowledge items from the 2002 Louisiana Survey, the gender effects on political knowledge are a bit more ambiguous, but these results still point to a significant gender effect. The coefficients for gender are negative for all nine models, though in some cases the gender coefficient is not statistically significant. For five of the nine knowledge items (i.e., Tom Daschle, William Rehnquist, Dennis Hastert, John Breaux, and Mike Foster), the gender coefficient is both negative and statistically significant, and for a sixth item (i.e., Charles Dewitt) the coefficient is negative and significant at a more relaxed .10 level. The gender coefficient is nonsignificant for Dick Cheney, Mary Landrieu, and Fox McKeithan. But overall, for most of the knowledge items the gender coefficient is negative, indicating that women are less likely to respond with a correct answer. Among those unable to give correct answers, is there a difference between men and women in their relative propensities to give incorrect and DK responses? In Table 10 we report multinomial logit coefficients for the gender variable from a series of models estimated using the trichotomous political knowledge items. We estimate the effect of gender on incorrect and DK responses, with correct responses the excluded comparison group. A positive coefficient indicates that women are more likely to give incorrect and DK responses, respectively. As one can readily see, the gender coefficients are positive in 13 of 15 cases for the incorrect-correct pairings; of these six are significant at the .05 level and another two are significant at the .10 level. These coefficients indicate that women are more likely than men to give incorrect responses in most cases, and in no instance are men more likely to give incorrect responses. Moreover, as suggested by Banducci

23

(2003) and Mondak (1999), our results confirm that women are significantly more likely than men to give DK responses. In all 15 models the gender coefficient is positive, and in 9 of 15 of these the coefficient is statistically significant at the .05 level; another two coefficients are significant at the .10 level. These coefficients suggest strongly that women are relatively more likely than men to funnel their lack of knowledge—or, at least, their indecision—into DK responses rather than incorrect responses. How should this pattern of gender coefficients from our binary and multinomial logit models be interpreted? Women are clearly less likely than men to provide correct answers to knowledge questions, and women are more likely than men to respond to knowledge questions with both incorrect answers and DK responses. What is interesting is the clear propensity of women to respond to knowledge questions that they don’t know the answer, without making an effort to hazard a guess. One possibility is that women actually have lower levels of knowledge than men, but that they are less likely to make a guess when they have no knowledge or only partial knowledge. If this is the case, this would reflect merely a different response strategy by lowknowledge women in comparison to low-knowledge men. If low-knowledge women demur from answering, but low-knowledge men offer a guess (no matter how bad the guess is), then DK and incorrect answers have similar meaning. On the other hand, it is possible that DK answers have meaning that is relevant for measuring individuals’ levels of political knowledge. If partial-knowledge men respond to a given political knowledge question by hazarding a guess, while partial-knowledge women respond with a DK response, then the gender gap in political knowledge may be overestimated since partial knowledge will result in a correct answer at least part of the time. We are a bit skeptical of this, since it is difficult for individuals to answer open-ended questions correctly unless they have a pretty good idea of the answer. But it remains a possibility that should be considered more fully in future research. The Stereotype Threat Hypothesis Our data permit us to provide a partial test of the stereotype threat hypothesis, which suggests that women exhibit lower levels of political knowledge to the degree that stereotypes about women’s lack of knowledge are engaged in the survey context. In an ideal research world, we would like to have control over how negative stereotypes are engaged, perhaps by randomly assigning respondents to groups based on the instructions that they receive or characteristics of the interviewer. Fortunately, while the 2000 ANES survey does not randomly assign respondents in a way designed to measure stereotype threat, this survey does include an interviewer gender variable that can be used as a rough surrogate measure.6 We use interviewer gender in a manner similar to how Davis and Silver (2003) use interviewer race to estimate the effect of racial stereotype threat on political knowledge. If women respond to negative stereotypes about their lack of knowledge, we would expect that those stereotypes will be engaged for women being interviewed by men; hence, women being interviewed by men should have lower levels of political knowledge than women interviewed by women. We capture this effect by creating an interaction variable, coded 1 for women respondents who are interviewed by female interviewers. Our working hypothesis is that the coefficient for the interaction variable should be positive, indicating that women being interviewed by female interviewers will have higher levels of political knowledge. In Table 11 we present part of our test of the stereotype threat thesis. In Model (1) we depict our general political knowledge scale as a function of respondent gender, the interaction for

24

respondent gender and interviewer gender, and the remaining independent variables from our previously discussed knowledge models. The simple gender coefficient represents the effect of gender on political knowledge for individuals interviewed by male interviewers. As one can see, the gender coefficient is negative and statistically significant (b = -0.564, t = -3.07), indicating that women have significantly lower levels of political knowledge then men among respondents interviewed by men. Most importantly, and consistent with the stereotype threat hypothesis, the interaction for respondent gender and interviewer gender is positive and statistically significant (b = 0.279, t = 1.86). For respondents interviewed by women, the relationship between gender and political knowledge is weaker (i.e., less negative), suggesting that there is less of a gender gap in political knowledge with women interviewers. In fact, with women interviewers the effect of gender on political knowledge is only -0.285 (i.e., -0.564 + 0.279), compared to -0.564 for respondents with men interviewers. This pattern of coefficients is wholly consistent with what one would expect if a process similar to that suggested by the stereotype threat thesis were in place. One possibility is that interviewer gender has an effect on all respondents, including men, and that the finding of a significant interaction effect in Model (1) is really a function of a global effect of interviewer gender, rather than a stereotype threat process. In Model (2) we add an additional variable for simple interviewer gender; if there is an overall interviewer gender effect, the coefficient for this variable should be positive. As one can see from Model (2), the coefficient for interviewer gender is positive, but it fails to achieve conventional levels of statistical significance (b = 0.174, t = 1.11). Given this, one cannot infer a global gender-of-interviewer effect. However, it is noteworthy that the coefficient for the interaction variable for respondent gender and interviewer gender is rendered nonsignificant by the inclusion of the interviewer gender variable (b = 0.105, t = 0.48). Men interviewed by women increase their knowledge (compared to other men) by 0.174, while women interviewed by women increase their knowledge (compared to other women) by 0.279 (i.e., 0.174 + 0.105). However, taken as a whole these results suggest that women being interviewed by women do not significantly increase their knowledge levels more than men being similarly interviewed by women. The results for Model (2) would seem to call into question the level of empirical support for the stereotype threat thesis. In Table 12 we examine this question somewhat differently by estimating our model separately for women and men. We include all of the independent variables from our previous models, as well as an interviewer gender variable to capture possible stereotype threat effects. The results from these models are not expected to reflect exactly those from Table 11, since we permit the effects of other independent variables in the model to vary for men and women. But if stereotype threat processes are in place, we would expect the coefficient for interviewer gender to be positive and significant in the model estimated with women respondents only, while the coefficient for interviewer gender should be effectively 0 in the model estimated for men respondents only. The results in Table 12 appear to provide modest support the stereotype threat interpretation. In the model estimated for women only, the coefficient for interviewer gender is positive and significant (b = 0.331, t = 2.25). This coefficient suggests that women being interviewed by women increase their political knowledge by about one-third of a point on the 0-6 political knowledge scale. On the other hand, for men the coefficient for interviewer gender is positive but nonsignificant (b = 0.173, t = 1.04). While the magnitude of the difference in interviewer gender coefficients for men and women is not great, the fact remains that for women the effect of interviewer gender is statistically significant, while for men the effect is nonsignificant. Simply, we can infer that interviewer gender has an effect on political knowledge for women respondents, but we cannot make the same inference for men respondents. This would seem to provide solid support for the stereotype threat hypothesis. On the other hand, the difference in how

25

men and women respond to interviewer gender is not stark. Based on the results from Table 11, we cannot infer that the difference is statistically significant. All in all, we are left with inconclusive findings but with enough evidence to warrant further study. Clearly, more work on the subject of stereotype threat as a possible explanation for the gender gap in political knowledge is needed. CONCLUSION In this paper we have explored the gender gap in political knowledge. Previous research has demonstrated that women typically have lower levels of political knowledge then men, and our study is no different. Women are consistently less likely to be able to identify correctly a wide range of national and state political figures, and when we sum these items into political knowledge scales we find that women have lower mean levels of knowledge than men. What is most interesting is that the relationship between gender and political knowledge persists even when we estimate a full model that includes a wide range of independent variables that are expected on theoretical grounds to be related to political knowledge. Many of these independent variables differentiate men and women as well, so one might expect to see a substantial reduction in the magnitude of the gender effect once one accounts for the effects of these other variables. Despite our best efforts, the significant gender coefficient remains, indicating that we have not accounted fully for the explanations of the gender gap in political knowledge. Do demographic and socioeconomic differences between men and women account for the significant effect of gender on political knowledge? What about political attitudes? Political engagement? Media exposure? Personal life circumstances? In each case, the inclusion of independent variables representing these sets of explanations reduces the size of the gender effect, but when all is said and done gender continues to have a strong, significant effect on political knowledge. We also consider the possibility that the gender effect is a methodological artifact based on the greater propensity of women to give DK responses to political knowledge questions; in fact, women are more likely to give DK responses, but they are typically more often to give incorrect responses as well. What about the geographic focus of political knowledge questions? Does the gender gap in political knowledge disappear once we look at political knowledge questions relating to state politics? Here again, even after considering these explanations we find that the significant effect of gender on political knowledge remains. Clearly, we have not accounted fully for the gender gap in political knowledge, and more research is needed to uncover additional explanations for these gender differences. All of this begs the question of whether or not gender differences in political knowledge have major implications for the working of the American political system. Do gender differences in political knowledge matter? There has been a substantial debate in the scholarly literature on the effects of political knowledge on the working of the American political system. Some scholars, most notably Popkin (1991) and Lupia and McCubbins (1998), suggest that a properly working democracy need not have a citizenry with high levels of political knowledge. Individuals rely on simplifying decision rules and decisional shortcuts (e.g., cues from political elites) that permit them to make good decisions without having a high level of political knowledge. Other scholars are less sanguine about the low level of political knowledge among citizens, suggesting that there are real differences in how decision processes work for high- and low-information citizens (Bartels, 1996; Garand and Lichtl, 2000). If high levels of political knowledge are related to quality decision making, and if women have significantly lower levels of political information than men, it is possible that there is also a gender gap in the “quality” of political decision-making, though this is certainly subject to debate and further study. There is a hint of this in the study by Holbrook and Garand (1996), who explore the

26

determinants of accuracy of information about the state of the economy during the 1992 presidential election campaign. Holbrook and Garand find that there is a gender gap in the accuracy of information about the economy, with women in 1992 being much more likely than men to overstate the level of unemployment and inflation. Individuals who misperceived the state of the economy and who exaggerated levels of unemployment and inflation were most likely to vote for Bill Clinton for president. Since women were over-represented among those misperceiving the state of the economy, it is likely that women were more likely to have voted for Bill Clinton than would have been suggested based on actual economic conditions. At least some of the gender gap in support for Bill Clinton may have been the result of the greater likelihood that women had less inaccurate information about the state of the economy in 1992. Findings such as these raise questions about the degree to which political knowledge has an effect on political decision-making in the American democracy. Where do we go from here? Our study is clearly not the final word on the subject of gender differences in political knowledge. First, we suggest that it is important to extend this research program to other contexts, particularly other election years and other political systems. Does the gender gap in political knowledge extend to other election years? Obviously, the American National Election Study series permits an estimation of gender effects on political knowledge for numerous election years. What about other states and other countries? We have used data from the 2002 Louisiana Survey, and Delli Carpini and Keeter (1996) explored the determinants of political knowledge using data from the 1989 Virginia Survey. Is there variation in the effect of gender on political knowledge across states? It is also important to extend this research cross-nationally. Does the gender gap in political knowledge extend to other countries? Is there variation across countries in the magnitude of the gender gap? Second, it is important to focus attention on the effects of context on both political knowledge and, more specifically, the gender gap in political knowledge. Are there features in individuals’ immediate contexts that promote or inhibit the development of political knowledge? For instance, the role of family in shaping political knowledge levels for men and women has been understudied. Part of the reason for this is the lack of good data on politics and families; the ANES surveys and most other surveys do not include interview data with respondent spouses, so it is difficult to study the linkage between political knowledge levels for spouses. Moreover, there are possible contextual effects on political knowledge that could affect the relative levels of knowledge for men and women. For instance, do women candidates increase the level of political knowledge of women visà-vis those of men? Third, the question of whether women with partial knowledge are more likely to give DK responses than men with partial knowledge needs to be resolved. Perhaps some resolution to this debate can be obtained by testing different question formats (e.g., open- vs. closed-ended questions) and differ question wordings. There may be ways of using experimental designs to manipulate information and question wordings in order to ascertain the degree to which there is a gender gap in responses to partial information. Finally, our study is not the last word on the applicability of the stereotype threat thesis to the study of the gender gap in political knowledge. In our view, it is important to replicate Davis and Silver’s (2003) approach for studying the stereotype threat hypothesis, using gender as the subject of study rather than race. This would involve the random assignment of respondents to experimental groups defined by the gender of interviewer, as well as the use of a larger number of male interviewers than in previous studies.

27

ENDNOTES 1. It should be noted that each of the knowledge items in the 2000 ANES and 2002 Louisiana Survey is coded using three possible responses: (1) correct answers; (2) incorrect answers; and (3) don’t know responses. In creating the knowledge scales, we code each of the component variables by combining incorrect and don’t know responses; correct answers are coded 1, and incorrect and don’t know responses are coded 0. However, for parts of our analysis it is crucial to maintain the distinction among correct, incorrect, and don’t know responses, so we retain these three categories for each of the component political knowledge variables. 2. In our analysis of individual knowledge items, we suggest that opinionization will have a particularly strong effect on decreasing the likelihood that respondents give “don’t know” responses. 3. Unfortunately, the marital status variable is not included in the 2002 Louisiana Survey. A marital status variable was included on the original survey protocol, but this variable was inadvertently left out of the survey during the telephone interview stage. Regrettably, we are unable to estimate the effect of marital status in our models estimated using 2002 Louisiana Survey data. 4. We also consider the effects of single parenthood for men, but the results (not shown) suggest that single fatherhood has a null (or perhaps even a slightly positive) effect on political knowledge. 5. For the sake of brevity, we report only the coefficients for the gender variable from each binary logit model. The full set of coefficients for these models can be obtained from the senior author. 6. Unfortunately, the distribution on the interviewer gender variable is heavily skewed, with women representing the lion’s share of interviewers. Among the 1807 respondents, 1581 (87.5%) were interviewed by women, and the remaining 226 (12.5%) were interviewed by men. We would have liked to see a roughly equal number of men and women interviewers, and ideally we would have liked to see the interviewers randomly assigned. An examination of the cross-tabulation for respondent gender and interviewer gender reveals that men and women respondents were about equally likely to have been assigned a men or women interviewer.

28 Table 1. Distribution of correct, wrong, and don’t know responses to political knowledge questions, by gender, 2000 American National Election Study Percentages of Responses ________________________________________________________ Don’t Correct Wrong Know __________________________________________________________________________________ Trent Lott Men (N = 674) Women (N = 881) William Rehnquist Men (N = 674) Women (N = 881) Tony Blair Men (N = 674) Women (N = 881) Janet Reno Men (N = 674) Women (N = 881) George Bush home state Men (N = 674) Women (N = 881) Al Gore home state Men (N = 674) Women (N = 881) 75.8% (511) 61.5% (542) 11.6% (78) 22.0% (194) 12.6% (85) 16.5% (145) 94.2% (635) 86.0% (758) 3.7% (25) 9.2% (81) 3.1% (14) 4.8% (42) 66.5% (448) 46.4% (409) 15.9% (107) 18.8% (166) 17.7% (119) 34.7% (306) 41.3% (278) 29.3% (258) 8.9% (60) 9.3% (82) 49.9% (336) 61.4% (541) 18.0% (121) 4.9% (43) 34.9% (235) 28.5% (251) 47.2% (318) 66.6% (587) 14.0% (94) 4.7% (41) 36.1% (243) 20.9% (184) 50.0% (337) 74.5% (656)

29 Table 2. Distribution of correct, wrong, and don’t know responses to political knowledge questions, by gender, 2002 Louisiana Survey Percentages of Responses ________________________________________________________ Don’t Correct Wrong Know __________________________________________________________________________________ Dick Cheney Men (N = 385) Women (N = 725) Tom Daschle Men (N = 385) Women (N = 725) William Rehnquist Men (N = 385) Women (N = 725) Dennis Hastert Men (N = 385) Women (N = 725) 23.4% (90) 6.9% (50) 16.4% (63) 12.3% (89) 60.3% (232) 80.8% (586) 41.8% (161) 25.7% (186) 14.3% (55) 12.3% (89) 43.9% (169) 62.1% (450) 44.7% (172) 27.0% (196) 14.3% (55) 16.3% (118) 41.0% (158) 56.7% (411) 78.2% (301) 69.2% (725) 8.5% (31) 6.6% (48) 13.8% (53) 24.1% (175)

____________________________________________________________________________

30 Table 2 (continued) Percentages of Responses ________________________________________________________ Don’t Correct Wrong Know __________________________________________________________________________________ Mary Landrieu Men (N = 385) Women (N = 725) Charles Dewitt Men (N = 385) Women (N = 725) John Breaux Men (N = 385) Women (N = 725) Fox McKeithan Men (N = 385) Women (N = 725) Mike Foster Men (N = 385) Women (N = 725) 94.0% (362) 88.0% (638) 2.6% (107) 3.2% (23) 3.4% (119) 8.8% (64) 41.3% (159) 31.2% (226) 21.3% (82) 15.0% (109) 37.4% (144) 53.8% (390) 73.5% (283) 63.6% (461) 6.0% (23) 8.6% (62) 20.5% (79) 27.9% (541) 17.1% (66) 10.6% (77) 20.5% (79) 12.1% (88) 62.3% (240) 77.2% (560) 73.5% (283) 70.1% (508) 9.9% (38) 10.9% (79) 16.6% (64) 19.0% (138)

_______________________________________________________________________

31 Table 3: Distribution of political knowledge scales, by gender, using data from the 2000 ANES and 2002 Louisiana Survey _______________________________________________________________________________ T-statistic for Women Men Mean Difference _______________________________________________________________________________ 2000 ANES Scale General Knowledge: 0 1 2 3 4 5 6 Mean 2002 Louisiana Survey Items General Knowledge: 0 1 2 3 4 5 6 7 8 9 Mean National Political Knowledge: 0 1 2 3 4 Mean State Political Knowledge: 0 1 2 3 4 5 Mean 6.3% 11.5% 10.3% 13.0% 17.8% 16.1% 10.8% 10.2% 2.6% 1.4% 3.923 28.3% 35.2% 19.9% 12.8% 3.9% 1.288 7.5% 16.1% 17.1% 30.9% 21.7% 6.8% 2.634 2.6% 6.5% 9.9% 9.4% 15.1% 16.1% 12.2% 14.0% 7.5% 6.8% 4.875 17.9% 26.5% 21.6% 17.7% 16.4% 1.881 3.9% 11.4% 15.6% 30.9% 26.8% 11.4% 2.995 -4.27*** -7.80*** -6.69*** 22.5% 16.3% 23.4% 18.6% 14.1% 4.1% 1.0% 2.017 18.6% 8.5% 18.7% 20.1% 18.9% 8.7% 6.5% 2.642 -8.00***

32 Table 4. OLS estimates for models of political knowledge, 2000 American National Election Study.

____________________________________________________________________________ (1) (2) --------------------------------------------b t b t ____________________________________________________________________________
Intercept Gender Socioeconomic / Demographic Variables Age Education Family income Black Hispanic Home ownership Political Attitudes and Engagement Partisan identification Folded partisanship Ideology Folded ideology Interest in politics Follows politics Opinionated Political efficacy Political trust Antigovernment attitudes Polarized party placement Media Exposure National television news exposure Local television news exposure Newspaper news exposure 0.451 -0.426 0.003 0.235 0.009 -0.682 -0.283 0.126 -0.015 0.059 -0.060 0.044 0.289 0.287 0.047 0.119 -0.013 0.181 0.097 0.037 -0.013 0.051 2.00** -5.60*** 0.98 8.64*** 0.79 -5.12*** -1.58* 1.48* -0.68 1.44* -2.33*** 0.93 4.31*** 5.64*** 1.01 2.19** -0.24 3.35*** 3.59*** 2.19** -1.39 3.65*** 0.482 -0.325 0.002 0.234 0.003 -0.629 -0.255 0.103 -0.014 0.054 -0.059 0.039 0.283 0.289 0.046 0.113 -0.008 0.177 0.099 0.036 -0.014 0.048 1.83** -2.46*** 0.71 8.57*** 0.25 -4.68*** -1.42* 1.17 -0.61 1.32* -2.24** 0.82 4.21*** 5.67*** 0.99 2.09** -0.15 3.27*** 3.67*** 2.14** -1.46 3.43***

Personal Life Circumstances Married --0.052 0.60 Child under 18 ---0.073 -0.57 Single mother ---0.430 -2.11** Employed --0.064 0.46 Gender * employed ---0.040 -0.24 Child under 18 * employed --0.039 0.55 Gender * child under 18 * employed ---0.075 -0.92 _____________________________________________________________________________________ N R2 F Prob (F) ***prob < .01 ** prob < .05 * prob < .10 983 0.426 33.98 0.000 982 0.432 25.88 0.000

33 Table 5. Difference in means for men and women on various predictors of political knowledge, 2000 American National Election Study. _______________________________________________________________________________________ Mean Values ----- ------------------Variable Women Men Difference t-ratio _______________________________________________________________________________________ Political knowledge Socioeconomic / Demographic Variables Age Education Family income Black Hispanic Home ownership Political Attitudes and Engagement Partisan identification Folded partisanship Ideology Folded ideology Interest in politics Follows politics Opinionated Political efficacy Political trust Antigovernment attitudes Polarized party placement Media Exposure National television news exposure Local television news exposure Newspaper news exposure Personal Life Circumstances Married Child under 18 Employed 2.017 47.788 4.186 6.292 0.123 0.054 0.659 2.564 1.815 3.340 1.419 1.039 1.468 1.638 -0.031 -0.011 -0.157 2.924 3.302 6.083 3.147 0.486 0.363 0.587 2.642 46.461 4.421 7.325 0.105 0.048 0.677 2.935 1.814 3.464 1.522 1.113 1.936 1.711 0.040 0.014 0.200 2.943 3.283 5.518 3.809 0.558 0.363 0.714 -0.625 1.327 -0.235 -1.033 0.018 0.006 -0.018 -0.372 0.001 -0.224 -0.103 -0.073 -0.468 -0.073 -0.072 -0.025 -0.357 -0.019 0.019 0.565 -0.661 -0.072 -0.000 -0.127 -8.00*** 1.65* -3.06*** -5.39*** 1.18 0.57 -0.82 -3.78*** 0.03 -2.74*** -2.54*** -2.18*** -9.69*** -1.78** -1.77** -0.62 -8.55*** -0.25 0.14 2.56*** -4.80*** -3.06*** -0.02 -5.63***

_______________________________________________________________________________________ N 1017 790

Note: The N reported here is the maximum for men and women. The sample size varies from one item to the next, depending on the numbers of missing cases. ***prob < .01 ** prob < .05 * prob < .10

34 Table 6. OLS estimates for models of political knowledge, 2002 Louisiana Survey.

____________________________________________________________________________ (1) (2) --------------------------------------------b t b t ____________________________________________________________________________
Intercept Gender Socioeconomic / Demographic Variables Age Education Family income Black Home ownership Political Attitudes and Engagement Partisan identification Folded partisanship Ideology Folded ideology Interest in politics Political efficacy Political trust Antigovernment attitudes Media Exposure National television news exposure Local television news exposure Newspaper news exposure 0.496 -0.729 0.024 0.277 0.085 -0.393 0.341 0.106 0.305 -0.059 -0.124 0.431 -0.093 0.094 0.317 0.449 0.127 0.342 1.14 -4.51*** 4.50*** 4.83*** 2.00** -1.83** 1.65** 2.73*** 3.76*** -1.34 -1.66 3.52*** -0.79 0.79 3.00*** 2.38*** 0.64 2.15** 0.767 -0.890 0.028 0.268 0.088 -0.415 0.287 0.111 0.332 -0.045 -0.159 0.451 -0.088 0.080 0.329 0.386 0.177 0.375 1.62 -3.08*** 4.71*** 4.55*** 2.06** -1.95** 1.39* 2.86*** 4.08*** -1.01 -2.13** 3.66*** -0.74 0.67 3.13*** 2.03** 0.89 2.36***

Personal Life Circumstances Employed full time ---0.321 -1.10 Gender * employed --0.589 1.62 Children over 18 ---0.532 -2.48*** Child under 18 ---0.218 -0.77 Gender * child under18 ---0.157 -0.41 Gender * child under 18 * employed ---0.492 -1.24 _______________________________________________________________________________________ N R2 F Prob (F) ***prob < .01 ** prob < .05 * prob < .10 576 0.384 20.43 0.000 567 0.405 16.09 0.000

35 Table 7. Difference in means for men and women on various predictors of political knowledge, 2002 Louisiana Survey. _______________________________________________________________________________________ Mean Values ----- ------------------Variable Women Men Difference t-ratio _______________________________________________________________________________________ Political knowledge Socioeconomic / Demographic Variables Age Education Family income Black Home ownership Political Attitudes and Engagement Partisan identification Folded partisanship Ideology Folded ideology Interest in politics Political efficacy Political trust Antigovernment attitudes Media Exposure National television news exposure Local television news exposure Newspaper news exposure Personal Life Circumstances Child(ren) under 18 Child(ren) 18 and over Employed 3.923 45.095 4.220 4.297 0.271 0.709 2.383 2.155 3.490 1.592 1.184 0.696 -0.299 -0.101 0.716 0.788 0.571 0.312 0.413 0.459 4.875 45.204 4.345 5.091 0.168 0.781 3.012 1.994 3.691 1.698 1.317 0.724 0.051 0.182 0.743 0.730 0.589 0.340 0.366 0.623 -0.953 -0.109 -0.125 -0.794 0.103 -0.072 -0.629 0.160 -0.191 -0.106 -0.133 -0.028 -0.081 -0.283 -0.027 0.058 -0.018 -0.029 0.047 -0.164 -6.69*** -0.10 -1.38* -4.84*** 3.86*** -2.56*** -4.15*** 2.39** -1.55* -1.51* -3.07*** -0.64 -1.86** -5.52*** -0.96 2.17** -0.56 -0.97 1.53* -5.20***

_______________________________________________________________________________________ N 725 385

Note: The N reported here is the maximum for men and women. The sample size varies from one item to the next, depending on the numbers of missing cases. ***prob < .01 ** prob < .05 * prob < .10

36 Table 8. OLS estimates for separate models of national and state political knowledge, 2002 Louisiana Survey. ____________________________________________________________________________________ National Knowledge State Knowledge --------------------------------------------b t b t ____________________________________________________________________________________ Intercept Gender Socioeconomic / Demographic Variables Age Education Family income Black Home ownership Political Attitudes and Engagement Partisan identification Folded partisanship Ideology Folded ideology Interest in politics Attention to national news about politics Attention to local news about politics Political efficacy Political trust Antigovernment attitudes Media Exposure National television news exposure Local television news exposure Newspaper news exposure -0.381 -0.451 0.010 0.176 0.038 -0.224 0.082 0.066 0.140 -0.019 -0.059 0.162 0.162 -0.040 -0.046 0.067 0.195 0.136 0.067 0.209 -1.42 -4.76*** 2.88*** 5.21*** 1.55* -1.83** 0.69 2.95*** 3.00*** -0.74 -1.38 1.91** 2.53*** -0.70 -0.68 0.99 3.26*** 1.24 0.59 2.29** 0.784 -0.227 0.018 0.093 0.047 -0.191 0.247 0.044 0.176 -0.024 -0.095 0.134 -0.048 0.132 -0.065 0.027 0.122 0.203 0.049 0.147 2.72*** -2.23** 4.79*** 2.54** 1.78** -1.45* 1.93** 1.84* 3.51*** -0.86 -2.06** 1.46* -0.70 2.11** -0.88 0.37 1.89** 1.71** 0.40 1.50*

Personal Life Circumstances Employed full time -0.040 -0.41 0.020 0.19 Children over 18 -0.261 -2.13** -0.308 -2.32** Child under 18 -0.361 -3.24*** -0.143 -1.19 _______________________________________________________________________________________ N R2 F Prob (F) ***prob < .01 ** prob < .05 * prob < .10 567 0.384 13.38 0.000 567 0.287 9.93 0.000

37 Table 9. Binary logit estimates for effects of gender on correct responses to specific political knowledge questions, 2000 American National Election Study and 2002 Louisiana Survey.

_______________________________________________________________________ Correct / All other responses -----------------------b t _______________________________________________________________________ 2000 ANES Items
Trent Lott William Rehnquist Tony Blair Janet Reno Bush home state Gore home state 2002 Louisiana Survey Items Dick Cheney Tom Daschle William Rehnquist Dennis Hastert Mary Landrieu Charles Dewitt John Breaux Fox McKeithan Mike Foster -0.250 -0.944 -0.456 -1.596 -0.078 -0.389 -0.489 -0.265 -0.982 -0.97 -4.25*** -2.10** -5.12*** -0.32 -1.33* -1.99** -1.26 -2.19** -0.906 -1.133 -0.188 -0.609 -0.760 -0.480 -3.36*** 4.54*** -1.11 -3.82*** -2.37*** -2.68***

_______________________________________________________________________
Note: For the sake of brevity, we report only the coefficients for the gender variable, obtained from a full binary logit model. The coefficients for control variables are not shown. ***prob < .01 ** prob < .05 * prob < .10

38 Table 10. Multinomial logit estimates for effects of gender on correct, incorrect, and don’t know responses to specific political knowledge questions, 2000 American National Election Study and 2002 Louisiana Survey.

______________________________________________________________________________ Incorrect / Correct Don’t know / Correct --------------------------------------------b t b t ______________________________________________________________________________ 2000 ANES Items
Trent Lott William Rehnquist Tony Blair Janet Reno Bush home state Gore home state 2002 Louisiana Survey Items Dick Cheney Tom Daschle William Rehnquist Dennis Hastert Mary Landrieu Charles Dewitt John Breaux Fox McKeithan Mike Foster 0.099 0.754 0.194 1.535 -0.041 0.011 1.311 -0.338 1.831 0.22 2.36*** 0.58 3.70*** -0.11 0.03 2.44*** -1.15 1.98** 0.308 1.015 0.538 1.599 0.146 0.479 0.294 0.547 0.805 1.07 4.25*** 2.32*** 5.08*** 0.49 1.60* 1.11 2.34*** 1.52* 0.455 0.990 0.241 0.410 0.651 0.343 1.62* 3.82*** 0.92 2.09** 1.51* 1.55 1.426 1.386 0.193 0.836 0.929 0.639 4.99*** 5.13*** 1.09 4.13*** 2.01** 2.74***

_______________________________________________________________________
Note: For the sake of brevity, we report only the coefficients for the gender variable, obtained from a full multinomial logit model. The coefficients for control variables are not shown. ***prob < .01 ** prob < .05 * prob < .10

39 Table 11. OLS estimates for stereotype threat models of political knowledge, 2000 ANES.

____________________________________________________________________________ (1) (2) --------------------------------------------b t b t ____________________________________________________________________________
Intercept Gender Gender * interviewer gender Interviewer gender Socioeconomic / Demographic Variables Age Education Family income Black Home ownership Political Attitudes and Engagement Partisan identification Folded partisanship Ideology Folded ideology Interest in politics Follows politics Opinionated Political efficacy Political trust Antigovernment attitudes Polarized party placements Media Exposure National television news exposure Local television news exposure Newspaper news exposure 0.473 -0.564 0.279 -0.002 0.232 0.002 -0.625 0.102 -0.011 0.055 -0.060 0.035 0.284 0.288 0.051 0.112 -0.007 0.177 0.099 0.036 -0.013 0.048 1.82 -3.07*** 1.86** -0.77 8.52*** 0.22 -4.65** 1.16 -0.50 1.35* -2.29** 0.74 4.24*** 5.67*** 1.11 2.08** -0.13 3.26*** 3.70*** 2.11** -1.36 3.40*** 0.344 -0.418 0.105 0.174 0.002 0.231 0.002 -0.620 0.103 -0.012 0.054 -0.058 0.036 0.294 0.289 0.048 0.112 -0.006 0.175 0.097 0.035 -0.012 0.048 1.21 -1.85*** 0.48 1.11 0.71 8.46*** 0.20 -4.61*** 1.18 -0.54 1.31* -2.22** 0.76 4.25*** 5.69*** 1.04 2.07** -0.11 3.23*** 3.61*** 2.07** -1.25 3.42***

Personal Life Circumstances Married 0.052 0.62 0.053 0.63 Children under 18 -0.030 -0.82 -0.031 -0.85 Single mother -0.480 -2.50*** -0.482 -2.51*** Employed 0.097 0.74 0.091 0.70 Gender * employed -0.102 -0.65 -0.094 -0.60 _______________________________________________________________________________________ N R2 F Prob (F) ***prob < .01 ** prob < .05 * prob < .10 982 0.434 27.06 0.000 982 0.434 26.14 0.000

40 Table 12. OLS estimates for stereotype threat models of political knowledge, by gender, 2000 American National Election Study.

____________________________________________________________________________ Women Men --------------------------------------------b t b t ____________________________________________________________________________
Intercept Interviewer gender Socioeconomic / Demographic Variables Age Education Family income Black Hispanic Home ownership Political Attitudes and Engagement Partisan identification Folded partisanship Ideology Folded ideology Interest in politics Follows politics Opinionated Political efficacy Political trust Antigovernment attitudes Polarized party placement Media Exposure National television news exposure Local television news exposure Newspaper news exposure 0.184 0.331 0.005 0.207 0.005 -0.814 -0.255 0.083 0.020 0.046 -0.062 0.000 0.273 0.252 0.010 0.096 0.005 0.157 0.086 0.035 -0.015 0.017 0. 52 2.25** 1.14 5.60*** 0.32 -4.42*** -1.08 0.70 0.65 0.83 -1.79* 0.00 2.98*** 3.72*** 0.16 1.27 0.07 2.21** 2.40*** 1.53* -1.17 0.86 0.030 0.173 -0.002 0.264 -0.004 -0.407 -0.207 0.121 -0.050 0.070 -0.052 0.076 0.257 0.355 0.096 0.102 0.002 0.233 0.105 0.033 -0.009 0.072 0.08 1.04 -0.33 6.34*** -0.21 -2.02** -0.75 0.91 -1.46 1.14 -1.30 1.06 2.54*** 4.54*** 1.36* 1.28* 0.02 2.71*** 2.55*** 1.32* -0.59 3.47***

Personal Life Circumstances Married 0.059 0.50 0.086 0.62 Child under 18 -0.108 -1.97** -0.012 -0.22 Single parent -0.408 -2.00** 0.289 1.23 Employed 0.026 -0.23 0.026 0.17 ________________________________________________________________________________________ N R2 F Prob (F) ***prob < .01 ** prob < .05 * prob < .10 509 0.384 12.05 0.000 473 0.431 13.55 0.000

41 Appendix 1: Variable definitions ________________________________________________________________________________________ Variable Description ________________________________________________________________________________________ Political knowledge 2000 ANES: Six-point (additive) scale of political knowledge, based on respondents’ ability to identify correctly the following six political knowledge items: (1) Trent Lott, Senate Majority Leader; (2) William Rehnquist, Chief Justice of the U.S. Supreme Court; (3) Tony Blair, Prime Minister of England; (4) Janet Reno, Attorney-General of the United States; (5) Texas as the home state for George W. Bush; and (6) Tennessee as the home state for Al Gore. For each item, 1 = respondent exhibits knowledge of subject; 0 = all other respondents. 2002 Louisiana Survey: Nine-point (additive) scale of political knowledge, based on respondents’ ability to identify correctly the following nine political knowledge items: (1) Dick Cheney, Vice-President of the United States; (2) William Rehnquist, Chief Justice of the U.S. Supreme Court; (3) Tom Daschle, U.S. Senate minority leader; and (4) Dennis Hastert, Speaker of the U.S. House of Representatives; (5) Mary Landrieu, junior member of the U.S. Senate from Louisiana; (6) John Breaux, senior U.S. senator from Louisiana; (7) Mike Foster, Governor of Louisiana; (8) Fox McKeithan, Louisiana Secretary of State; and (9) Charles DeWitt, Speaker of the Louisiana House of Representatives. For each item, 1 = respondent exhibits knowledge of subject; 0 = all other respondents. Gender Age Education 1 = female respondent; 0 = male respondent. Respondent's age (in years). Seven-point scale representing number of years of formal education completed, ranging from 0 (less than 9th grade completed) to 6 (advanced degree). 2000 ANES: 24-point scale for yearly family income, ranging from 1 ($2,999 per year or less) to 24 ($105,000 per year or greater). 2002 Louisiana Survey: Eight-point scale for yearly family income, ranging from 0 (under $10,000 per year) to 7 (over $70,000 per year). Race: black Race: Hispanic Partisan identification Folded partisanship 1 = black respondent; 0 = all other respondents. 2000 ANES: 1 = Hispanic respondent; 0 = all other respondents. Seven-point scale of partisan identification, ranging from 0 (strong Democrat) to 6 (strong Republican). Four-point scale of strength of partisanship, ranging from 0 (pure independent) to 3 (strong partisan).

Family income

42 Appendix 1 (continued) ________________________________________________________________________________________ Variable Description ________________________________________________________________________________________ Ideological orientation Folded ideological orientation Seven-point scale of ideological orientation, ranging from 0 (strong liberal) to 6 (strong conservative). Four-point scale of strength of partisanship, ranging from 0 (pure independent) to 3 (strong partisan). 2000 ANES: Three-point scale of interest in “political campaigns,” ranging from 0 (not very much interested) to 2 (very much interested). 2002 Louisiana Survey: Three-point scale of interest in “government and politics,” ranging from 0 (not much interested) to 2 (very much interested). Attention to national news about politics 2002 Louisiana Survey: Five-point scale of how much attention respondents pay to national news about “government affairs and politics,” ranging from 0 (none) to 4 (a great deal). 2002 Louisiana Survey: Five-point scale of how much attention respondents pay to local news about “government affairs and politics in the state of Louisiana,” ranging from 0 (none) to 4 (a great deal). 2000 ANES: Four-point scale of the degree to which respondent follows “government and public affairs,” ranging from 0 (hardly at all) to 3 (most of the time). 2000 ANES: Five-point scale of the degree to which the respondent considers him- or herself more opinionated than others, ranging from 0 (a lot fewer than average) to 4 (a lot more than average). 2000 ANES: Respondents' level of external political efficacy, based on factor scores obtained through principle components analysis of the following two items: (1) degree of agreement with statement that "I don't think public officials care much what people like me think;" and (2) degree of agreement with statement that “People like me don't have any say about what the government does. Eigenvalue = 1.41, variance explained = 70.5%. 2002 Louisiana Survey: Respondents’ level of external politial efficacy, based on “how much attention” respondent feels that “Louisiana government pays to what people think when it decides what to do.” Variable is measured as a three-point scale, ranging from 0 (not much) to 2 (a good deal).

Interest in politics

Attention to local news about state politics

Follows politics

Opinionated

Political efficacy

43 Appendix 1 (continued) ________________________________________________________________________________________ Variable Description ________________________________________________________________________________________ Political trust 2000 ANES. Respondents' level of political trust, based on factor scores obtained through principle components analysis of the following four items: (1) the degree to which respondents think that government wastes money paid in taxes; (2) whether respondents would say that government is run by a few big interests; (3) how much of the time respondents would say that they can trust the government in Washington to do what is right; and (4) the extent to which respondents think that the people running the government are crooked. Eigenvalue = 1.93, variance explained = 48.4%. 2002 Louisiana Survey: Respondents’ level of political trust, based on factor scores obtained through principal components analysis of the following three items: (1) the degree to which respondents think that government wastes money paid in taxes; (2) how much of the time respondents would say that they can trust the government in Washington to do what is right; and (3) the extent to which respondents think that the people running the government are crooked. Eigenvalue = 1.80, variance explained = 60%. Anti-government scale 2000 ANES. Scale based on a principle components analysis of the following forced-choice items: (1) One, the less government, the better; or two, there are more things that government should be doing; (2) One, we need a strong government to handle today's complex economic problems; or two, the free market can handle these problems without government being involved; and (3) One, the main reason government has become bigger over the years is because it has gotten involved in things that people should do for themselves; or two, government has become bigger because the problems we face have become bigger. Variable is rescaled so that a high value represents the anti-government position. (Eigenvalue = 1.98; variance explained = 66%). 2002 Louisiana Survey. Scale based on same items as above. (Eigenvalue = 1.85; variance explained = 62%). Polarized party placement 2000 ANES. Absolute value of the difference in respondent's placement of the Republican and Democratic parties on the seven-point liberal-conservative scale.

44 Appendix 1 (continued) ________________________________________________________________________________________ Variable Description ________________________________________________________________________________________ National television news exposure 2000 ANES. Number of days in which respondent watches national television news, ranging from 0 to 7. 2002 Louisiana Survey. 1 = respondent has watched national television news in past 24 hours; 0 = otherwise. Local television news exposure 2000 ANES. Sum of number of days in which respondent watches early evening local television news and number of days in which respondent watches late night local television news, ranging from 0 to 14. 2002 Louisiana Survey. 1 = respondent has watched local television news in past 24 hours; 0 = otherwise. Newspaper news exposure 2000 ANES. Number of days in which respondent reads daily newspaper, ranging from 0 to 7. 2002 Louisiana Survey. 1 = respondent has read daily newspaper in past 24 hours; 0 = otherwise. Married Child under 18 Child over 18 Single mother Single father Employed Interviewer gender 2000 ANES. 1 = respondent is married; 0 = otherwise. 1 = respondent has children under 18; 0 = otherwise. 1 = respondent has children 18 and over; 0 = otherwise. 1 = respondent is female unmarried parent; 0 = otherwise. 1 = respondent is male unmarried parent; 0 = otherwise. 1 = respondent is employed full-time; 0 = otherwise. 1 = female interviewer; 0 = male interviewer.

45

REFERENCES Alvarez, R. Michael and Edward J. McCaffery. “Is There a Gender Gap in Fiscal Political Preferences?” paper presented at 2000 annual meeting of the American Political Science Association. Banducci, Susan A. “Gender Differences in Measures of Psychological Political Engagement: The Role of Interviewer Effects,” unpublished paper, 2003. Bartels, Larry. “Uninformed Votes: Information Effects in Presidential Elections,” American Journal of Political Science (February 1996: 194-230). Bennett, Linda and Stephen Earl Bennett. “Enduring Gender Differences in Political Interest: The Impact of Socialization and Political Dispositions,” American Politics Quarterly (January 1989: 105-22). Bennett, Stephen Earl. “Know-Nothings Revisited: The Meaning of Political Ignorance Today,” Social Science Quarterly (1988: 476-90). Bennett, Stephen Earl, “Trends in Americans’ Political Information, 1967-1987,” American Politics Quarterly (1989: 422-35). Bowen, William and Derek Bok, The Shape of the River: Long-Term Consequences of Considering Race in College and University Admissions. Princeton, New Jersey: Princeton University Press, 1998. Claibourn, Michele and Virginia Sapiro. “Gender Differences in Citizen-Level Democratic Citizenship: Evidence from the Comparative Study of Electoral Systems,” paper presented at the 2001 annual meeting of the Midwest Political Science Association. Claibourn, Michele and Virginia Sapiro, “Gender Differences in Citizen-Level Democratic Citizenship: Evidence from the Comparative Study of Electoral Systems,” paper presented at the meeting of the Comparative Study of Electoral Systems, Berlin, Germany, February 2002. Darcy, Robert, Susan Welch, and Janet Clark, Women, Elections, and Representation. New York: Longman, 1987. Delli Carpini, Michael X. and Scott Keeter, “Stability and Change in the U.S. Public’s Knowledge of Politics,” Public Opinion Quarterly (Winter 1991: 583-612). Delli Carpini, Michael X. and Scott Keeter, "Measuring Political Knowledge: Putting First Things First," American Journal of Political Science (November 1993: 1179-1206). Delli Carpini, Michael X. and Scott Keeter, What Americans Know about Politics and Why it Matters, New Haven, Connecticut: Yale University Press, 1996. Garand, James C. and Andre Blais, “Understanding Joint Support for Government Spending and Taxes: Linking Benefits and Costs in the Mass Public,” paper presented at the 2003 annual meeting of the American Political Science Association.

46

Garand, James and Marci Glascock Lichtl, “Explaining Divided Government in the United States: Testing an Intentional Model of Split-Ticket Voting,” British Journal of Political Science (January 2000: 173-91). Gidengil, Elisabeth, Elizabeth Goodyear-Grant, Neil Nevitte, Andre Blais, and Richard Nadeau. “Gender, Knowledge and Social Capital,” paper presented at Gender and Social Capital Conference, 2003. Hess, Robert and Judith V. Torney, The Development of Political Attitudes in Children. Chicago: Aldine, 1967. Holbrook, Thomas and James C. Garand, "Homo Economus? Economic Information and Economic Voting," Political Research Quarterly (June 1996: 351-76). Jennings, M. Kent, “Political Knowledge Over Time and Across Generations,” Public Opinion Quarterly (Summer 1996: 228-52). Jennings, M. Kent and Richard Niemi, The Political Character of Adolescence. Princeton, New Jersey: Princeton University Press, 1981. Kenski, Kate, “Women and Political Knowledge During the 2000 Primaries,” The Annals of the American Academy of Political and Social Science (2000: 26-28). Kenski, Kate and Kathleen Hall Jamieson. “The 2000 Presidential Campaign and Differential Growths in Knowledge: Does the “Knowledge Gap” Hypothesis Apply to Gender as Well as Education?” paper presented at 2001 annual meeting of the American Political Science Association. Kretschmer, M. “Gender and Political Knowledge,” paper presented at the 2001 annual meeting of the American Association for Public Opinion Research, Montreal, Quebec, 2001. Lambert, Ronald, James Curtis, Barry Kay, and Steven Brown, “The Social Sources of Political Knowledge,” Canadian Journal of Political Science (1988: 359-74). Lupia, Arthur and Matthew McCubbins, The Democratic Dilemma: Can Citizens Learn What They Need to Know? Cambridge: Cambridge University Press, 1998. Luskin, Robert, “Measuring Political Sophistication,” American Journal of Political Science (November 1987: 856-99). Mondak, Jeffery, “Newspapers and Political Awareness,” American Journal of Political Science (May 1995: 513-27). Mondak, Jeffery, and Belinda Creel Davis, “Asked and Answered: Knowledge Levels When We Won’t Take ‘Don’t Know’ for an Answer,” Political Behavior (September 2001: 199-224). Mondak, Jeffery J. “Reconsidering the Measurement of Political Knowledge,” Political Analysis (1999: 57-82). Mondak, Jeffery, “Developing Valid Knowledge Scales,” American Journal of Political Science (January 2001: 224-38).

47

Nadeau, Richard and Richard G. Niemi, “Educated Guesses: The Process of Answering Factual Knowledge Questions in Surveys,” Public Opinion Quarterly (Autumn 1995: 323-46). Niemi, Richard and Jane Junn, Civic Education: What Makes Students Learn. New Haven, Connecticut: Yale University Press, 1999. Popkin, Samuel, The Reasoning Voter: Communication and Persuation in Presidential Campaigns. Chicago: University of Chicago Press, 1991. Putnam, Robert, Bowling Alone: The Collapse and Revival of American Community. New York: Simon and Schuster, 2000. Steele, Claude M. “A Threat in the Air: How Stereotypes Shape Intellectual Identity and Performance,” American Psychologist (1997: 613-29). Steele, Claude M. and Joshua Aronson, “Stereotype Threat and the Intellectual Test Performance of African Americans,” Journal of Personality and Social Psychology (1995: 97-111. Steele, Claude M. and Joshua Aronson, “Stereotype Threat and the Test Performance of Academically Successful African Americans,” in Christopher Jencks and Meredith Phillips (eds.), The Black-White Test Score Gap. Washington, D.C.: Brookings Institution Press, 1998. Vars, Frederick and William Bowen, “Scholastic Aptitude Test Scores, Race, and Academic Performance in Selective Colleges and Universities,” in Christopher Jencks and Meredith Phillips (eds.), The Black-White Test Score Gap. Washington, D.C.: Brookings Institution Press, 1998. Verba, Sidney, Nancy Burns, and Kay Lehman Schlozman, “Knowing and Caring About Politics: Gender and Political Engagement,” Journal of Politics (November 1997: 1051-72).


				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:58
posted:5/30/2009
language:English
pages:49