Political Behavior

, Volume 31, Issue 1, pp 117–136

Gender Differences in Political Knowledge: Distinguishing Characteristics-Based and Returns-Based Differences

Authors

    • Department of Political ScienceUniversity of Missouri
Original Paper

DOI: 10.1007/s11109-008-9059-8

Cite this article as:
Dow, J.K. Polit Behav (2009) 31: 117. doi:10.1007/s11109-008-9059-8

Abstract

This study assesses whether gender-based differences in political knowledge primarily result from differences in observable attributes or from differences in returns for otherwise equivalent characteristics. It applies a statistical decomposition methodology to data obtained from the 1992–2004 American National Election Studies. There is a consistent 10-point gender gap in measured political knowledge, of which approximately one-third is due to gender-based differences in the characteristics that predict political knowledge, with the remaining two-thirds due to male–female differences in the returns to these characteristics. The methodology identifies the relative contribution of the predictors of political knowledge to each portion of the gap, and then uses this information to elucidate the underlying sources of the political knowledge gender gap and its prognosis. Education is the characteristic that most clearly enlarges the gap, with men receiving significantly larger returns to political knowledge from education than women. Group membership reduces the gap as women obtain gains in political knowledge from belonging to organizations that do not accrue to men. However, these gains are not sufficient to significantly reduce the gap.

Keywords

GenderPolitical knowledgeBlinder–Oaxaca decomposition

Introduction

One of the most robust findings in the study of political behavior is that men score higher than women on measures of political knowledge. First documented by Campbell et al. (1960) in the American Voter, gender-based differences in political knowledge persist through contemporary studies of the American electorate (Burns et al. 2001; Delli Carpini and Keeter 1996; Verba et al. 1997). These differences are not unique to the United States. Frazer and Macdonald (2003) report the same finding in the United Kingdom, as do several studies of the distribution of political interest and knowledge in Western Europe (Banducci and Semetko 2002; Inglehart 1981; Randall 1987). The political knowledge gender gap is real, and it exists across time and space.

This study addresses a basic question about gender and political knowledge that remains unanswered: How much of the political knowledge gender gap is explained by gender-based differences in observable characteristics such as education, employment status, and related attributes, and how much of the gap results from men and women receiving different returns to political knowledge for otherwise equivalent characteristics? To illustrate, suppose that political knowledge is a function of hours worked outside the home. Suppose that women and men receive equivalent returns to political knowledge for working hours. In this case, any male advantage in political knowledge accrues solely because men, on average, work more hours outside the home than women. In contrast, if men receive greater political knowledge returns for a given number of working hours, then even if women and men worked the same number of hours, the political knowledge gap would still exist.

There are several reasons why it is important to determine whether women and men receive different knowledge returns to otherwise equivalent characteristics. Foremost, the existing evidence suggests they do. Gender consistently predicts measured political knowledge controlling for other characteristics. Typically, the residual explanatory power of gender is rationalized in terms of differences in male-female socialization. But this explains little. If socialization accounts for the unexplained difference between female and male political knowledge, its effect must manifest through some characteristic or personal attribute that is associated with political knowledge. By estimating the proportion of the political knowledge gender gap that results from differences in returns to otherwise equivalent characteristics, I identify those characteristics that most clearly affect the size of this residual portion of the knowledge gap. This provides information that is useful for understanding the underlying sources of the gender gap in political knowledge and its long term prognosis.

To do so, I apply a statistical decomposition methodology to regression equations estimated using data obtained from the American National Elections Studies, 1992–2004. This approach, known as the Blinder-Oaxaca decomposition (Blinder 1973; Newmark 1988; Oaxaca 1973), has been extensively applied to the study of gender and racial wage discrimination. The method allows one to separate equivalent and non-equivalent returns to political knowledge as a function of predictor variables. The decomposition reveals how much of the knowledge gap is explained by gender-based differences in attributes that predict political knowledge, and how much of the gap owes to differences in returns to otherwise equivalent attributes. Following convention, I term the former figure the “explained” portion of the gap because this part of the gap is explained by differences in the levels of observable characteristics such as education possessed by men and women. The latter figure is termed the “unexplained” potion of the gap because it owes instead to gender-based differences in the returns to otherwise equivalent characteristics.

The study’s key finding may be summarized as follows: Most of the characteristics that predict political knowledge have equivalent effects for men and women. However, only about one-third of the political knowledge gap can be accounted for by these attributes. That is, about one-third of the knowledge gap is explained by differences in the levels of the predictors of political knowledge possessed, respectively, by men and women. The remaining two-thirds of the gap results from gender-based differences in political knowledge returns to two characteristics; education and membership in groups and organizations. Both Chow tests and the Blinder–Oaxaca decomposition show that men receive higher knowledge returns from education, while women receive higher knowledge returns from group membership. These two effects work in opposite directions—education enlarges the gap while group membership reduces it. However, the education effect is much larger. Simply put, the gender gap in political knowledge exists because for any given level of education, men learn and retain more factual knowledge about politics than women.

Theoretical Background

The Persistence of the Political Knowledge Gender-Gap

Studies of the political knowledge gender gap are largely devoted to making it go away. Since most people are unwilling to argue that there are innate differences between men and women such that one gender is less capable than the other of acquiring factual political knowledge and organizing political thoughts along abstract conceptual lines, the literature focuses on identifying variables that explain the gap and, by extension, identifying the ceteris paribus conditions under which it disappears.

The problem is that nobody has identified those ceteris paribus conditions. Mondak and Anderson (2004) explicitly argue this point by presenting a multivartiate analyses using a standard battery of explanatory variables to demonstrate the residual explanatory power of gender in predicting political knowledge. Comparable results are obtained by Delli Carpini and Keeter (1996, Table 4.1), Jamison (2000, Tables A91–A93), Jennings (1996, Table 3), Garrand et al. (2005), all of which are broadly consistent with studies of gender and political engagement, interest and efficacy. Although they are summarizing their own findings, Verba et al. (1997, p. 1060) could easily be characterizing the field when they conclude “The net result is a significant distinction between women and men in political…information even after controlling for gender differences in other factors associated with political engagement.” One can control for everything including the kitchen sink and the political knowledge gender gap remains.

Why Do Some People Know More About Politics than Others?

So why are some people more informed about politics than others, and why do women score lower on measured political knowledge than men? The literature speaks most directly to the first question, and Luskin’s (1990, p. 335) explanation will resonate especially well with those who appreciate Jack McCoy’s ability to convict criminals: To acquire political knowledge one must have the means, motive and opportunity to learn about politics. That is, one must have occasion or opportunity to be exposed to political information, have the ability to organize and process this information, and the motivation to engage political information.

The population distribution of political knowledge, thus, depends on the distribution of the means, motives and opportunity to obtain political information. The question is what personal attributes capture these characteristics, and how might these inform the political knowledge gender gap?

Raw intelligence certainly enhances ones ability to organize and retain political information. Likewise, education increases ones knowledge of politics; both by enhancing ones ability to acquire, organize and retain political information, and by increasing ones motivation to acquire such information in the first place. Education also directly contributes to ones store of political knowledge because the schools teach civics.1 Comparing intelligence and education, Luskin argues that cognitive ability is the more proximate predictor of political knowledge and the preferred variable. However, there are at least three reasons why education may be better suited for understanding the distribution of political knowledge, especially when ones focus is on gender. First, there are no gender differences in standard intelligence measures (see Halpern and LaMay 2000 for a review),2 and little reason to believe that intelligence translates in to political knowledge differently for women and men. Second, the only ANES measure of respondent intelligence is the interviewer’s assessment, and this assessment is based on a politically centered conversation. This potentially confounds political knowledge (the dependent variable) with native intelligence (the independent variable) (Delli Carpini and Keeter 1996, p. 195).3 Third, these characteristics may be simultaneously determined and difficult to disentangle empirically. Those of higher intelligence often obtain more formal education, but formal education may increase measured intelligence (Brody 1999). For these reasons, I follow much of the literature in using education as the primary predictor of ones political knowledge.

Political knowledge also derives from the civic skills obtained from community engagement (Burns et al. 2001). Putnam (1995, p. 343), for example, argues that “We learn about politics through casual conversation.” He is also clear about where these casual conversations take place: “Faith communities in which people worship together are arguably the single most important repository of social capital in America” (Putnam 1995, p. 66). Places of worship and membership organizations are not only important venues for developing political awareness and skills, it is this “social capital [that] allows political information to spread” (Putnam 1995, p. 343). Religiosity and group membership may also have gender specific implications. On average, women are more religious than men and are more likely to attend services (Kosman et al. 2001). Men and women also gravitate toward different types of groups. Men tend to join nationally and economically oriented associations, while women tend to join smaller, more community based groups (McPherson and Smith-Lovin 1982; Inglehart and Norris 2003; Gidengil et al. 2003). If group type has implications for political knowledge, then this may further increase the likelihood that membership translates in to political knowledge differently for men and women.

Work and occupation also affect the opportunity, motivation and ability to learn about politics. Lipset (1960, pp. 191–192) provides the classic statement: “Insight into complex social problems…seems to depend…more on the social experiences flowing from one’s work…The housewife is at a great disadvantage in this respect.” There are at least three ways in which work structures political knowledge. First, one learns about politics in the workplace. Casual conversation at the office water cooler is not unlike casual conversation at the Moose Lodge. Second, some occupations are particularly sensitive to politics, providing workers extra incentives to learn about government and policy within their professional domain (Popkin et al. 1976, Luskin 1990, Delli Carpini and Keeter 1996). Farmers are more aware of agriculture policy, and aerospace workers are more familiar with defense politics, respectively, than the public at large. Third, work and occupation has gendered implications for the distribution of political knowledge. On average, women work fewer hours outside the home than men, and are employed in less prestigious occupations. Employment, working hours, income, and occupational prestige all tap in to aspects of work that influence the acquisition of political knowledge.

Among personal characteristics, strength of partisanship is positively associated with political knowledge (Delli Carpini and Keeter 1996, pp. 172–173). McClurg (2006) suggests one reason for this is that partisanship enhances the quality of political conversations. He also finds that there are spillover effects in that these conversations enhance the political engagement and knowledge content of ones social network. Political interest also predicts political knowledge, but is scrutinized because political interest and knowledge are so closely related that they may part of the same class of behaviors (Zaller 1990). In addition, interest may not be entirely exogenous; interest drives one to acquire knowledge, and knowledge spurs increased interest. These are legitimate concerns, but political interest is too important to the distribution of political knowledge to ignore. To at least partially assuage these concerns, recall that political interest is shaped through childhood socialization, and may be sufficiently removed from current knowledge to be defensibly treated as exogenous.4 Also, the extent to which political interest is a product of socialization may help explain why men are more interested in politics than women (e.g. Verba et al. 1997, Table 1).

Finally, political knowledge increases in age (Glen and Grimes 1968). This likely owes to a combination of life learning, the possibility that one has more time to devote to politics as one ages, and because settling down and developing deeper community ties increases opportunities and motivation for political engagement. Marriage and children may also influence the acquisition of political knowledge. Verba et al. (1997) provide evidence that marriage increases political knowledge for men and women. While they don’t discuss the finding in detail, marriage (hopefully) provides increased opportunities for conversation, and some of these conversations will be politically oriented. In contrast, children, despite their many endearing qualities, impose time demands that detract from political engagement. Further, the joys of child rearing fall disproportionally on women, with potential consequences for the acquisition of political knowledge.

The Dependent Variable: Political Knowledge

Political knowledge is easy to recognize; an astute observer can often quickly assess the knowledge level of a person with whom one is having a politically centered conversation. The concept is somewhat harder to define and measure. Delli Carpini and Keeter (1996, pp. 10–11) define political knowledge as “the range of factual information about politics that is stored in long-term memory.” The key is ones possession of factual information that is of more than momentary interest, and of such a nature that when combined with beliefs, values and cognition, enable citizens to “think and act with greater autonomy and authority.” This definition has many virtues. It comports nicely with common understandings of political knowledge, and distinguishes core political knowledge from more trivial facts that primarily serve entertainment or related purposes. It also presents tractable measurement strategies. Finally, the definition emphasizes that political knowledge is both an objective in and of itself, and a means to a democratic end. While not wanting to engage the debate over how much knowledge is necessary for meaningful democratic participation, more information is certainly better than less, and gender differences in political knowledge raise legitimate normative concerns.

Measuring political knowledge from readily available data is a non-trivial problem, and the measurement strategy itself may have implications for assessing male-female differences in political knowledge. For example, Mondak and Anderson (2004) argue that if men and women use different response strategies to answer survey questions used to measure political knowledge, then the measurement strategy itself may generate gender-based differences in political knowledge when none exist.5 I approach this issue by constructing a measure of political knowledge that taps into the underlying elements widely viewed as comprising political knowledge, without over reliance on any single component, and while making use of information regularly contained in the National Election Studies series. As a result, my measure is not predicated on an overly narrow conception of political knowledge, and it will be comparable across time. Subject to the constraints of the specific instruments included in the 1992–2004 American National Election Studies, the measure builds on the knowledge index recommended by Delli Carpini and Keeter (1993). It is also comparable with suggestions offered by Luskin (1987, 1990; and Luskin and Bullock 2004) and other widely used measures of political knowledge and, more generally, sophistication.

The political knowledge measure is constructed from the summation of binary indicators (0, 1) that denote correct or otherwise appropriate responses to a series of National Election Studies questions that (1) solicit factual knowledge about major political figures,6 (2) request that the respondent self-place on a liberal-to-conservative ideological scale, and correctly identify the relative placements of the Democratic and Republican candidates and parties on this scale, and (3) request that the respondent identify the relative placements of the Democratic and Republican presidential candidates and parties on several issue scales. The specific questions used to construct the index are presented in the Appendix.7

To illustrate, the 1996 National Election Study (NES) survey requests respondents to identify, among others, Al Gore. If the respondent identifies Gore as the Vice President of the United States, the respondent is assigned a score of 1 on that question. Incorrect responses, including “don’t know,” are scored 0. The 1996 NES contains four factual knowledge questions of this type, and the respondent may obtain a summary score of 0–4 on the factual knowledge section of the measure. The NES also asks the respondent to place himself or herself on a liberal-to-conservative ideological scale, and to place the Democratic and Republican parties and the Democratic and Republican presidential candidates on the same scale. If the respondent self-places on the scale, we score him or her 1 on this measure, 0 otherwise. If he or she correctly places the Democratic Party to the left of the Republican Party on this scale, and the Democratic candidate to the left of the Republican candidate on this scale, we score each of these as 0 or 1 depending on the response. Thus, on the ideological scale portion of the measure, the respondent obtains a score of 0–3. The issue-knowledge scale is constructed similarly, except we score only party and candidate placement, not respondent self-placement. For example, if the respondent places the Democratic Party to the left of the Republican Party, and the Democratic candidate to the left of the Republican candidate on the NES seven point defense-spending scale, he or she will be scored 0, 1, or 2 depending on the number of correct responses. Suppose there are four issue scales in a given National Election Study. This means the scores for this component of the knowledge measure will range between 0 and 8. The sum of all three components is the measure of political knowledge, and in this example it will range between 0 and 15. To facilitate regression analysis, I transform this raw count to percentage form, so that the final measure of political knowledge ranges between 0 and 100 percent.

The Distribution of Political Knowledge, Explanatory Variables, and the Oaxaca Decomposition

Table 1 displays the mean values of political knowledge and variables that predict the distribution of political knowledge for the 1992–2004 election cycles. These are disaggregated by gender, and an asterisk denotes those variables in which the differences in male–female means are significant at the 0.05 level.
Table 1

Political knowledge and predictor variables: 1992–2004 means by gender

 

Male

Female

p < 0.05

Political knowledge

61.36

50.72

*

Age

46.32

47.65

*

Married

0.63

0.50

*

N children

0.59

0.58

 

Education

4.09

3.90

*

PID strength

1.80

1.84

 

Political interest

2.23

2.12

*

Working

0.75

0.68

*

Hours worked

34.83

26.56

*

Occupational prestige

46.26

45.04

*

Personal income

12.19

8.56

*

Group membership

0.44

0.43

 

Religious services

1.66

1.88

*

Note: * Indicates difference between male and female value is statistically significant at the 0.05 level (two-tailed). Difference of means test conducted on pooled, 1992–2004, sample. Number of observations: 1992 (1536), 1996 (1181), 2000 (1114), 2004 (740)

As expected, men and women display statistically significant differences in measured political knowledge, a gap averaging more than 10 points. Among the predictor variables, age, marital status, education, political interest, working status, hours worked, occupational prestige personal income and religious service attendance show statistically significant gender-based differences in their mean values. These differences, however, are often modest. On average, women in the sample are slightly older than the men, men enjoy slightly higher levels of education and are slightly less likely to attend religious services. Other gender based differences are more pronounced. This is especially true in the employment related variables. Men are more likely to work, work longer hours, and have higher incomes than women. Number of children, strength of partisan identification and group membership do not display statistically significant differences between men and women. In sum, Table 1 presents strong and consistent differences between men and women in measured political knowledge and in many, but not all, of the predictors of political knowledge.

The Oaxaca Decomposition

As the background section emphasized, the empirical literature has largely focused on identifying the ceteris paribus conditions that narrow or eliminate the knowledge gender gap. I believe this places emphasis in the wrong place for understanding differences in female–male political engagement. That gender retains residual explanatory power begs the question of which characteristics differentiate women and men in their respective returns to political knowledge, and whether these characteristics inform our understanding of the underlying source of the political knowledge gap. We know that at least some of the gap—indeed, most of the gap as it turns out—cannot be accounted for by gender differences in the possession of attributes that predict political knowledge because even if, hypothetically, differences between women and men in the possession of these attributes were to disappear, the gender knowledge gap would persist.

To say that gender retains residual explanatory power after controlling for all other factors is tantamount to saying that women and men must receive different knowledge returns for at least some characteristics. Otherwise, one could postulate that if women received as much income, worked as many hours, or some similar hypothetical, the gap would close. But it doesn’t, which means that there must be gender-based differences in the way at least some attributes translate into political knowledge.By isolating the attributes or characteristics that are indistinguishable in their political knowledge implications for men and women from those that have gender-specific implications, one obtains information useful for understanding the underlying sources of the political knowledge gender gap, the veracity of competing explanations for the gap, and its long-term prognosis. The following section outlines an empirical approach for estimating which predictors of political knowledge contribute, respectively, to the explained and unexplained portions of the political knowledge gap, and the sizes of the explained and unexplained proportions of the gap. The method is a modest extension of least squares regression and, thus, makes the study’s finding quite comparable with those in the extant literature.

Our objective is to estimate the proportion of the male–female political knowledge gap that exists because of gender-based differences in characteristics, and the proportion of the difference resulting from men receiving greater returns to otherwise equivalent characteristics, and which variables most influence the size of each component of the gap. The obvious approach is to estimate separate linear regressions of political knowledge on explanatory variables for men and women, and evaluate gender-based similarities and differences in the estimated coefficients. This method, used by Burns et al. (2001, Tables C11.11.1, C11.12, C11.13), is related to the underlying logic of the Oaxaca decomposition, which expands on this approach.

The Oaxaca decomposition uses information obtained from separate male–female regressions along with algebraic identities derived from the regression models to calculate how much of the gap in male and female political knowledge is due to differences in observable characteristics, how much results from men receiving higher returns to equivalent characteristics, and what variables account for each component of the difference. In the language of regression analysis, the decomposition approach shows in what combination the political knowledge difference between men and women results from differences in the amount of characteristics associated with political knowledge (differences in the X’s) and male–female differences returns to these characteristics (differences in the betas). For example, if level of education affects political knowledge, the decomposition shows how much of the political knowledge gap is due to differences in male–female education levels, and how much results from the way in which education translates into political knowledge for women and men.

The decomposition is fairly intuitive. It begins with standard regression notation. Let \( Y_{ig} \) be observed political knowledge for the i-th individual in group g, where g is male or female. One writes:
$$ Y_g = X_g \beta _g + \varepsilon _g \quad E(\varepsilon _g ) = 0,\quad g \in \{ {\text{male}},{\text{female}}\} $$
(1)
by definition, \(\bar Y_g = \bar X_g \hat \beta _g \) where the means are taken over all individuals in the group.
The mean difference in male and female political knowledge is written:
$$ D = \bar Y_M - \bar Y_F = \bar X_M ^\prime \hat \beta _M - \bar X_F ^\prime \hat \beta _F $$
(2)
Following Oaxaca (1973), Neuman and Oaxaca (2003), and Jann (2005), algebraic manipulation of these terms provides the following equivalent representation of this difference.
$$ D = \underbrace {(\bar X_m - \bar X_f )'\hat \beta ^* }_{{\rm Differences\;in\;characteristics}} + \underbrace {[\bar X_m ^\prime(\hat \beta _m - \hat \beta ^* ) + \bar X_f ^\prime (\hat \beta ^* -\hat \beta _f )]}_{{\rm Differences\;in\;returns}} $$
(3)
The arithmetic values of the first and second terms, respectively, measure the relative importance of differences in characteristics (the X’s) and differences in returns (the betas) in explaining the differences in mean levels of male–female political knowledge. For example, suppose the difference (gap), D, in political knowledge equals 8. If the first term in Equation 3 equals 6 and the second term equals 2, then 6/(6 + 2) = .75, or three-quarters of the difference in male–female political knowledge is accounted for by differences in male–female characteristics, and one-quarter of the difference results from men receiving higher returns to otherwise equivalent characteristics.

There is, however, ambiguity in Eq. 3: It holds for \(\beta ^* = \beta _m {\text{ and }}\beta ^* = \beta _f \), and the differences in political knowledge that result from differences in characteristics and returns depend on which coefficient vector, \(\beta _m {\text{ or }}\beta _f \), is used in the calculation. In some sense, the “correct” \(\beta ^* \) depends on the prevailing or “baseline” level of political knowledge in the population. It may be that women, on average, possess what might be considered the baseline or expected levels of political knowledge, and the higher level for males results from a preference for the entertainment value of political knowledge. Likewise, it may be that men, on average, possess this baseline level of political knowledge, and the lower level held by women reflects a distaste for politics. Regardless, determining the relative importance of characteristics and returns for male-female political knowledge requires a baseline from which to judge deviations. To establish this, researchers follow one of two strategies. The first, used here, calculates the decomposition separately for \(\beta ^* = \beta _f {\text{ and }}\beta ^* = \beta _m .\) Doing so gives the low and high range for the proportion of the gender knowledge gap that is due to differences in characteristics and the proportion that is due to differences in returns. These estimates can then be averaged to calculate each variable’s contribution to the explained and unexplained portions of the gap. The second strategy is to let \(\beta ^* = \beta _p \) where \(\beta _p \) is the vector of coefficient estimates from a pooled male-female regression. This also represents an averaging strategy. In practice, this ambiguity makes little difference in the approach used because all of these estimates are close and generate similar conclusions.

Gender and Political Knowledge 1992–2004

This section presents the regressions for estimating the effect of predictor variables on the distribution of political knowledge. It discusses the Oaxaca disaggregation applied to these estimates and its implications for understanding differences in male-female political knowledge.8

The regression equation is:
$$ \begin{gathered} {\text{PolKnowledge}}_i = \alpha + \beta _1 * {\text{Year}}92_i + \beta _2 * {\text{Year}}96_i + \beta _3 * {\text{Year}}00_i \; + \beta _4 * {\text{Age}}_i + \beta _5 * {\text{Age}}_i^2 \hfill \\ \quad \quad \quad \quad \quad \quad \; + \beta _6 * {\text{married}}_i + \beta _7 * {\text{nchild}}_i + \beta _8 * {\text{education}}_i + \beta _9 * {\text{PID strength}}_i \hfill \\ \quad \quad \quad \quad \quad \quad \; + \beta _{10} * {\text{Pol Interest}}_i + \beta _{11} * {\text{working}}_i + \beta _{12} * {\text{hourswork}}_i + \beta _{13} * {\text{occ prestige}}_i \hfill \\ \quad \quad \quad \quad \quad \;\;\;\; + \beta _{14} * {\text{personal income}}_i + \beta _{15} * {\text{groups}}_i + \beta _{16} * {\text{religious}}\_{\text{service}}_i + \;\varepsilon _i \hfill \\ \end{gathered} $$
Table 2 presents the male and female coefficient estimates for the combined 1992–2004 election cycles. For comparison purposes, it also presents a combined regression that pools the female and male samples. For each set of estimates, standard diagnostics reveal no estimation problems.
Table 2

OLS estimates: Political knowledge by gender, 1992–2004

 

Female

Male

Combined

Intercept

−10.43**

(4.00)

−11.90*

(4.21)

−8.02**

(2.90)

1992

−5.71**

(1.36)

−5.60**

(1.36)

−5.63**

(0.96)

1996

3.81**

(1.42)

2.08

(1.47)

3.15**

(1.02)

2000

−14.89**

(1.51)

−11.97**

(1.64)

−13.31**

(1.11)

Female

  

−7.07**

(0.68)

Age

0.35*

(0.15)

0.59**

(0.17)

0.48**

(0.11)

Age-squared

−0.01**

(0.00)

−0.01**

(0.00)

−0.01**

(0.00)

Married

1.56

(0.93)

−0.60

(1.06)

0.54

(0.68)

N children

−1.29**

(0.48)

−0.74

(0.48)

−1.02**

(0.34)

Education

5.09**

(0.36)

6.09**

(0.35)

5.68**

(0.25)

PID strength

3.94**

(0.47)

4.13**

(0.48)

4.07**

(0.33)

Political interest

10.74**

(0.66)

10.51**

(0.68)

10.64**

(0.48)

Working

3.85

(2.04)

3.53

(2.31)

3.64*

(1.51)

Hours worked

−0.14**

(0.05)

−0.08

(0.04)

−0.10**

(0.03)

Occupational prestige

0.13**

(0.04)

0.13**

(0.04)

0.13**

(0.03)

Personal income

0.46*

(0.10)

0.26*

(0.10)

0.32*

(0.07)

Group membership

4.46**

(1.00)

0.48

(0.99)

2.39**

(0.70)

Religious services

−0.34

(0.28)

−0.13

(0.30)

−0.23

(0.20)

N

2,280

2,291

4,571

Adjusted R2

0.44

0.42

0.45

Standard errors in parentheses

p < 0.05; ** p < 0.01

The consistent predictors of political knowledge, regardless of respondent gender, are age, education, strength of partisan identification, political interest, occupational prestige9 and personal income.10 Since the estimates are least squares, the interpretation is straightforward. For example, a one-point increase on the seven-point NES education scale produces about a six-point gain in political knowledge for men and about a five-point increase for women, all else equal. For women, a unit increase in partisanship strength, say a change from moderate to strong partisanship, increases the average political knowledge score by approximately four points, all else equal. Among the remaining variables, hours worked and number of children in the household appears to decrease political knowledge for women, but not men. Group membership appears to increase political knowledge for women but not for men, while religious service attendance has little effect on political knowledge for either gender.

The interpretation that marriage, number of children, working hours, group membership and education have gender specific implications for the acquisition of political knowledge follows from the regression coefficients in the two samples returning different levels of statistical significance. To formally assess whether the corresponding female and male regression coefficients are indeed statistically distinguishable, I calculate Chow statistics to test the null hypothesis that \(\beta _{fj} = \beta _{mj} \) against the alternative \(\beta _{fj} \ne \beta _{mj} \) (Gujarati 2003).11 These statistics are presented in Table 3, and show that only the regression coefficients for education and group membership are statistically distinguishable.
Table 3

Chow test of hypothesis that βm = βf

 

\(F_{(1,4537)} \)

1992

0.00

1996

0.71

2000

1.71

Age

1.11

Age-squared

0.47

Married

2.37

N children

0.65

Education

3.89*

PID strength

0.08

Political interest

0.06

Working

0.01

Hours worked

0.84

Occupational prestige

0.00

Personal income

1.83

Group membership

7.95**

Religious services

0.25

Asterisks indicate the level of statistical significance at which one can reject the null hypothesis that βm = βf

p < 0.05; ** p < 0.01

Knowing that the female and male coefficients for education and group membership differ is not sufficient, however, to determine how much these variables contribute to the gender knowledge gap because this does not directly incorporate information on the corresponding levels of attributes (the X’s) possessed by men and women. The Blinder–Oaxaca decomposition provides this information because it uses both the estimated betas and the levels of the X’s to separate returns from possession of attributes from returns to attributes. Recall that the knowledge gap averages just over 10 points and that, on average, men possess more of several attributes expected to predict political knowledge. Some of the gap must be explained by these differences in the possession of attributes. However, despite being statistically significant, these differences are often substantively modest. The male–female difference in education level, for example, is a fraction of a point on the NES scale, seemingly too small to drive anything but a minute portion of the double-digit gender difference in political knowledge. This indicates that there are gender-based differences in returns to some attributes, an expectation confirmed by the Chow tests, that when combined with female–male differences in the levels of these attributes, produces the knowledge gap.

It is this information that is presented in the fourth and final table. The four columns in Table 4 present the knowledge gap disaggregated in to explained and unexplained portions from the male and female base estimates, respectively. The row entries present each variable’s contribution to the explained and unexplained portion of the gap, with the Blinder–Oaxaca formula used to calculate these entries presented at the top of each column. The columns sum vertically to equal each component of the gap. The knowledge gap is 10.63; and using the female base estimates, 3.55, or about one-third, of the gap is explained by gender-based differences in characteristics, while 7.08, or two-thirds, results from differences in returns to characteristics. The male base estimates return a slightly lower explained portion of the gap and a slightly higher unexplained portion. In either case, the clear majority of the gender gap in political knowledge results from differences in returns to attributes, not the attributes themselves.
Table 4

Predictor variable contribution to explained and unexplained portions of mean gender-based differences in political knowledge

 

Female base

Male base

Explained

Unexplained

Explained

Unexplained

\(\beta _f \left( {\overline X _m - \overline X _f } \right)\)

\(\left( {\beta _m - \beta _f } \right)\overline X _m \)

\(\beta _m \left( {\overline X _m - \overline X _f } \right)\)

\(\left( {\beta _m - \beta _f } \right)\overline X _f \)

1992

−0.20*

0.04

−0.20*

0.04

1996

−0.06

−0.43

−0.03

−0.46

2000

0.37

0.68

0.30

0.75

Age

−0.46

11.22

−0.79*

11.55

Age-squared

0.79*

−3.84

1.05*

−4.10

Married

0.20

−1.36

−0.08

−1.08

N children

−0.02

0.33

−0.01

0.32

Education

0.91**

4.32*

1.09**

4.15*

PID strength

−0.16

0.34

−0.17

0.35

Political interest

1.09**

−0.52

1.06**

−0.50

Working

0.28

−0.24

0.26

−0.21

Hours worked

−1.13**

2.02

−0.65

1.54

Occ. prestige

0.16**

−0.17

0.16*

−0.17

Personal income

1.65**

−2.41

0.94*

−1.69

Group membership

0.06

−1.77**

0.01

−1.72**

Religious services

0.07

0.34

0.03

0.38

Intercept

 

−1.47

 

−1.47

Gap decomposition

3.55

7.08

2.96

7.67

Note: Column totals may not sum to decomposition totals because of rounding

p < 0.05; ** p < 0.01, where statistical significance is calculated in manner described by Jann (2005)

The characteristics that predict political knowledge may affect the size of the explained portion of the gap, the unexplained portion of the gap, or both. For example, political interest only contributes to the explained portion of the gap. The male advantage in political knowledge resulting from political interest accrues solely because men are more interested in politics, not because of differences in the way political interest translates into knowledge for each gender. Group membership, in contrast, only affects the unexplained portion of the gap. It decreases the gap, and does so only through differences between men and women in the ways that group membership translates into political knowledge. Finally, education disproportionally benefits male political knowledge because, on average, men are slightly better educated than women and, more importantly, because men receive higher knowledge returns to education than women.

What attributes are most responsible for the unexplained differences in male–female political knowledge? The Chow tests and decomposition rule out strength of partisanship and political interest, two major predictors of political knowledge. Demographic characteristics provide limited purchase on this question. Men obtain greater unexplained returns to political knowledge in early life, but the negative sign on age-squared coefficient means that as diminishing returns set in, and this may advantage women by reducing the size of unexplained portion of the gap. Finally, both the Chow test and the Oaxaca decomposition show that children are an equal opportunity drain on political knowledge.

The only two variables that stand out as predictors of the unexplained portion of the knowledge gap are education and group membership. Education enlarges the gap. There is something about the way that education translates into political knowledge that systematically advantages men relative to women. Loosely speaking, education accounts for about half of the unexplained portion of the knowledge gap. All else equal, women must possess associate level college degrees to obtain approximately the same level of political knowledge as men possessing high school diplomas. Group membership, on the other hand, benefits women in that it systematically reduces the knowledge gap. However, the impact of education on the gap is significantly greater than that for group membership.

Discussion and Conclusion

At least two-thirds of the gender difference in political knowledge results from female–male differences in returns to the predictors of political knowledge, with educational achievement accounting for most of the gap Group membership attenuates the gap a bit, but its effect is too small to significantly reduce the gender difference in political knowledge. These findings allow us to better understand what drives the knowledge gap and, hence, what might eventually reduce or eliminate it. As background, it is useful to recall Verba et al. (1997, p. 1070) conclusion to their study of gender and political engagement in which they write that “gender differences in political interest and information seem to reflect a genuine difference in the taste for politics.” The evidence suggests this is correct; women and men do appear to have different tastes for politics.

The question is, what is the underlying source of this taste difference, and is this explanation consistent with the finding that education is most closely associated with enlarging the political knowledge gender gap, while female membership in groups reduces it? The most obvious explanation for male–female differences in taste for politics is that it reflects the continuing importance of political socialization, however difficult the concept is to measure and empirically evaluate, in shaping male and female orientations toward the political process (Gilligan 1982; Trevor 1999). However, it’s not just any type of socialization that matters; it appears that socialization associated with education is most important. Is this plausible? Somewhat contrary to the received view from the first generation studies, current research shows that the schools do impact political knowledge; and perhaps more so for boys than for girls (Niemi and Junn 1998; see Galston 2001 for a review). While this is not the venue for a discussion of the learning styles of boys and girls, the empirics are consistent with the proposition that pedagogic approaches and curriculum impacts the ways boys and girls engage politics, with subsequent implications for male–female political learning.

Perhaps the more intriguing finding is that group membership affects the acquisition of political knowledge differently for men and women. Women clearly benefit from belonging to groups in ways that men do not. This suggests there are important, gender-based, differences in the effects of social setting and environment on civic engagement. Putnam (2000, p. 195) points to the particular importance of groups for women.

Women have traditionally invested more time than men in social connectedness. Although men belong to more organizations, women spend more time in them. Women also spend more time than men in informal conversations and other forms of schmoozing…

While, again, this is not the venue to discuss gender differences in group interactions, there appears to be something about the nature of face-to-face interactions in the groups that engage women that is not present in male dominated organizations. The irony is that women tend to belong to groups traditionally associated with so-called women’s issues and interests such as those affiliated with education, heath care, and the arts (Inglehart and Norris, 2003, Table 5.2). This gender segregation is criticized because it may attenuate female access to political and economic circles and knowledge which would otherwise benefit women. However, the empirics suggest the opposite is true, or at least that the story is more complex. To the extent that the beneficial effects of group membership on female political knowledge derive from the types of groups women tend to join, this segregation has positive effects at least so far as political engagement is concerned.

So what do these finding suggest for the prognosis for the gender gap in political knowledge? Unfortunately, the weight of evidence suggests it will not close anytime soon. Only one third of the gap explained by differences between men and women’s possession of the attributes that predict political knowledge, so even if this playing field were to suddenly level, most of the gap would remain. One could point to women’s organizational membership patterns as holding promise for closing the gap, but the trend has been for women to join fewer of these organizations (Putnam 2000, pp. 196–201), and it is hard to imagine this trend reversing in the near future. Education presents a more promising long-term picture if only because younger women outperform their male peers in most facets of educational achievement (National Center for Educational Statistics, 2005). These achievement differences are presently too small to significantly impact the knowledge gap, but do suggest that if female-male trends in educational achievement continue, perhaps accompanied by changes in curricular and pedagogic approaches, the political knowledge gap may narrow. Still, the most definitive thing one can say about the political knowledge gender gap is that it manifests primarily as a consequence of the way education translates in to political knowledge for men and women, and educational practices reflect deeply held norms and values. For this reason, the political knowledge gender gap will certainly be a fixture in the political landscape for the foreseeable future.

Footnotes
1

Although there is debate over whether the schools do so effectively (Niemi and Junn 1998; cf. Galston 2004).

 
2

This is by construction: “IQ” tests indicate academic ability, and contain few instruments that produce significant gender-based advantages or liabilities.

 
3

In addition, the ANES interviewer assessments also show a statistically significant difference in perceived respondent intelligence by gender. Men score higher than women on this assessment in each of the 1992–2004 studies.

 
4

Speaking of political knowledge, Jennings and Niemi (1974, p. 97), write that “We are arguing that homes where parents have higher levels of information are likely to be homes where the atmosphere is conducive for the child’s acquisition of political facts.” The same is most certainly true for political interest as well (see, for example, Jennings 1981, Table 4.1).

 
5

Specifically, if women are more likely to respond “don’t know” when presented a question to which they do not know the answer, while equally uninformed men are more likely to guess, a knowledge measure created from a summation of correct responses with no adjustment for guessing may produce higher scores for men when there is no underlying difference in knowledge. This conclusion, however, has not gone unchallenged. The question of how to treat “don’t know” responses in the measurement of political sophistication is the subject of an active debate (Barabas 2002; Luskin and Bullock 2005, 2006; Mondak 2001). While the details of these exchanges are not central to this study, I believe the weight of evidence supports coding these responses as incorrect answers.

 
6

Lupia (2006) cautions that responses to factual political knowledge questions such as those used in the ANES are sensitive to interviewing protocols, and may produce underestimates respondent’s actual knowledge. However, the dependent variable can withstand this scrutiny because it does not exclusively rely on factual knowledge questions of the type described by Lupia (e.g. “How long is the term of office for a US Senator?”). The dependent variable is also constructed in a manner similar to those used in the literature, so the findings may be easily compared with those from previous studies.

 
7

There is also concern that the ANES questions may focus on political figures and policies that are of more interest to men than women, with attendant consequences for measured political knowledge. There is evidence, for example, that women are more likely than men to correctly identify members of the local school board (Delli Carpini and Keeter 1996, pp. 207–208). However, more generally, there is little evidence that women are more knowledgeable about so-called “women’s issues” than men, and even if this were the case, it would do little to attenuate the importance of the knowledge-gap revealed by the ANES questions.

 
8

Race is not included as a variable because African American respondents are dropped from the sample. Race is a significant predictor of political sophistication, and it is customary to conduct analysis separately for racial groups (e.g. Burns et al. 2001, Table C11.10.1). Since my focus is on gender, the sample is limited to non-African American ANES respondents.

 
9

The occupational prestige scores are the raw scores that are redacted from current ANES releases, but available through institutional research board request. For details on their construction, see Nakao et al. (1990). “Computing 1989 Prestige Scores,” Chicago: NORC, 1990. GSS Methodological Report No. 70.

 
10

The mean personal income levels recovered from the 2000 ANES are less than those in the remaining years because this study uses a different income scale than in the previous or following years (compare ANES variables v960701 and v000997). This has few implications for the statistical analysis and interpretation because I am only interested in relative, not absolute, income.

 
11

I use the test statistic outlined in Gujarati (2003, pp. 306–310). The Chow test assumes the female and male regressions have equal error variances. Regression diagnostics reveal that these are approximately equal, and retesting using Wald statistics, which allows for unequal variances, revealed no significant differences in the statistical inferences.

 

Acknowledgments

I thank Kenneth Troske, Jeff Milyo and the anonymous reviewers for helpful comments on this study. I am responsible for all errors.

Copyright information

© Springer Science+Business Media, LLC 2008