INTRODUCTION
Six months after the 2004 election, the Washington Post reported that the gender gap “is alive and well.”1
Brian Faler, “Women Returning to Democratic Party, Polls Find,” Washington Post, 10 May 2005, sec. A.
The presence and prominence of a gender gap in any election is a complex phenomenon. We will define the gender gap as the percentage of women voting for the Democratic candidate minus the percentage of men voting for the Democratic candidate, a definition that mirrors previous usage (see Carroll 2006; Chaney, Alvarez, and Nagler 1998; C. Clark and J. Clark 1999; Kaufmann and Petrocik 1999; Wirls 1986). We argue that the electoral environment is paramount in understanding the variation in the gender gap. Gender gaps can arise from a combination of factors specific to individual campaigns, to the state in which the election takes place, and to the interaction of campaign and state factors that create this electoral environment.
Scholarly research on the gender gap has taught us much about why men and women behave differently politically and why these differences persist. Still, we agree wholeheartedly with Virginia Sapiro and Pamela Johnston Conover (1997) that “the theory of the gender difference in electoral behavior remains underdeveloped, especially in its ability to account for variation across elections, not to mention cross-nationally” (p. 497, emphasis ours). We therefore examine the gender gap at the subpresidential level, specifically Senate elections from 1988 to 2000.
PREVIOUS RESEARCH ON THE GENDER GAP
The gender gap in vote choice has been part of presidential elections since at least the 1950s (Manza and Brooks 1998). At its outset, it represented women favoring the Republican Party over the Democratic Party, relative to men. During the 1960s and 1970s, the small gender gap switched directions several times, with no gap at all in the 1968 Republican vote and the 1972 Democratic one (J. Clark and C. Clark 1996; C. Clark and J. Clark 1999). Thus, while it is incorrect to say (as many do) that the gender gap began in 1980, that year saw the dawning of a new gender gap.
Figure 1 plots Senate election data from 1988 to 2000 and shows that while the increase during the past decade has not been monotonic, there is no mistaking its overall growth. Increasingly, the different voting patterns of men and women have become decisive in many elections (CAWP 1997). Sex differences in voting have now reached proportions similar to other significant demographic influences on voting behavior—that is, income, education, and age (Clark and Clark 2000; Kaufmann and Petrocik 1999).
Average gender gap in Senate elections, 1988–2000.
The electoral gender gap is composed of many different components, each expressing its influence in different ways. Among these components are partisanship and ideology. Barbara Norrander (1999) contends that the gender gap in partisanship is a function of southern realignment and that men are more likely than women to identify as independents. Karen Kaufmann and John Petrocik (1999) demonstrate that partisanship explains more than half of the gender gap in vote choice in presidential elections. Men and women also differ on issue positions and intensity. Women traditionally have shown a stronger desire to avoid conflict with other nations (Conover and Sapiro 1993; Miller 1988). Domestically, women favor stronger gun control laws and also take different positions on issues that reflect compassion and cost bearing, such as education, health care and welfare (C. Clark and J. Clark 1999, 2000; Frankovic 1982; Shapiro and Mahajan 1986). Men and women also may hold similar opinions on issues yet place different weight on these issues when it comes time to vote (Kaufmann and Petrocik 1999; Sapiro and Conover 1997).
The increase in the electoral gender gap highlights the more active independent role women have come to play in politics. This pattern also highlights the coupling of important demographic and socioeconomic trends and the fact that women in the electorate are speaking in a different voice than are men (Box-Steffensmeier, De Boef, and Lin 2004; Schlozman et al. 1995; Trevor 1999). As the political influence of women continues to grow, the value in understanding gender-based differences grows as well.
CONTEXT MATTERS: MODELING THE GENDER GAP ACROSS ELECTIONS
The electoral context is important in determining the magnitude of the gender gap in vote choice.2
On the general issue of campaign context influencing the information voters gain, and by extension the choices they make, see Hutchings (2003) and Kahn and Kenny (1999).
Demographic factors specific to where the election is occurring also play a key role in determining the electoral environment. We look at four state factors relevant to Senate races: percentage of women below the poverty line, percentage of women in the workforce, percentage of African Americans in the state, and the ideological polarization of the state. Each factor creates the environment in which the election is conducted and hence can be influential in determining how large the gender gap will be.
Finally, campaign and state factors are not likely to act completely independently of one another. For example, the ideological polarization of a state may help determine the importance of certain issues. Thus, the interaction of these factors helps form the variation in the gender gap. In the following, we offer specific predictions about how and why these campaign factors, state factors, and their interactions affect the gender gap.
Research Design
We use the Senate election as our unit of analysis to isolate the electoral context's impact on the gender gap, offering unique leverage in understanding variation in the electoral gender gap across time and space. Studying Senate elections has many advantages over either House or presidential elections. First, we have more cases than we would have if we restricted our focus to presidential elections. As the Senate ranks above the House in prestige, these elections should be visible to voters, albeit less so than presidential elections. Senate elections also offer more variation in independent variables than presidential elections; for example, the sex of the presidential candidates has never varied. We increase our sample size and the variation in our variables by pooling all the Senate elections between 1988 and 2000. This does not cause us to lose contextual information about individual years or cases because these are included in the model as measures of campaign and state factors.
The gender gap is largely an aggregate phenomenon produced by factors that vary at the aggregate rather than the individual level. Campaign effects, for example, are constant across individual voters but vary across elections. Thus, individual analysis of campaign effects is often inconclusive because the model is dominated by noise created by unsystematic behavior; the noise washes out when we focus on the aggregate, allowing us to explain these macro-level systematic variations in electoral behavior (MacKuen, Erikson, and Stimson 1989).
THEORETICAL EXPECTATIONS
Our theoretical model distinguishes between variables affected by individual campaigns and those affected by the state-level context in which the election is held. Studies of the gender gap at the individual level provide insight into the performance of the aggregate; thus, we draw on individual-level research to construct expectations about aggregate behavior.3
We note that it is an ecological fallacy to draw conclusions about individuals from aggregate behavior; our research, which uses individual-level research to construct our expectations about aggregate behavior, is methodologically valid and theoretically sound.
Campaign-Level Hypotheses
Our first hypothesis examines candidate sex as a voting cue. Voting based upon sex-inferred traits about candidates, like partisanship, is a heuristic used in the absence of complete information about candidates and their positions. Especially in low-information elections, women are more likely to vote for a female candidate (Cook 1994, 1998; Dolan 1997, 1998, 2004; McDermott 1997; Paolino 1995), perhaps because voters infer ideology from a candidate's personal traits (Koch 2000).4
However, Cook (1998) suggests that these results are dependent on the type of election, with women more likely to vote for female candidates regardless of party in House and gubernatorial elections, but not in Senate elections.
Beginning from the “baseline” gender gap that exists in a male-male contest, we expect the gender gap to increase when the Democratic candidate is a woman. Then, women Republicans and Independents should be more likely to defect and vote for the Democrat. On the other hand, when a Republican woman runs, we expect defections among women Democrats and Independents to lead to a smaller-than-usual gender gap. Conversely, although group identity is rarely used to explain male political behavior, similar defections by male voters to support the opposite party's candidate of their sex should add to the hypothesized effect.
The presence of an incumbent, however, decreases the need to employ sex as a heuristic, and thus reduces the gender gap. When an incumbent is running, voters have more information about at least one of the candidates. Therefore, voters should be influenced more by evaluations of the incumbent and less by sex-based inferences or group identity.
Campaign issues generate a third set of hypotheses. One oft-cited reason for an increased electoral gender gap in 1992 was the Clarence Thomas–Anita Hill hearings, a hot button issue making gender issues salient (Sapiro and Conover 1997; Wilcox 1994). Moreover, Brian Schaffner (2005) finds larger gender gaps when campaigns focus on education, welfare, or child care. Thus, we expect a larger gender gap when issues that evoke the largest differences between men and women increase in saliency, issues such as crime, education, the environment, and health care (Clark and Clark 1999, 2000; Frankovic 1982; Shapiro and Mahajan 1986).5
The crime variable we use is a fairly broad measure encompassing the issues of crime (general), the death penalty, and gun control.
Although we do not expect the salience of abortion to have significant effects on the gender gap (abortion attitudes are influenced primarily by factors other than sex; see Cook, Jelen, and Wilcox 1992, 1993), we include it in the model to test this no-difference expectation. War-and-peace issues in individual-level gender gap studies (Conover and Sapiro 1993; Miller 1988) are important, but the paucity of subnational elections in which these issues are critical did not allow us to examine them.
The impact of campaign issues, however, should not be uniform across all circumstances. Thus, we consider the impact of these issues as they interact with other contextual forces, such as when a woman runs as a Democrat. We also expect a larger gender gap when a woman candidate articulates the issues most likely to differentiate men and women voters, when women in a state face more economic adversity, and in conservative states with a greater likelihood that policies are passed that run counter to “women's interests.”
Voter mobilization should also increase the gender gap for two reasons. First, in the event that it is a highly salient gender-based event (such as the Clarence Thomas hearings) that energizes voters and leads to the higher turnout, the resulting electorate becomes more gender conscious and produces a larger gender gap. Second, even when the increase in turnout is not gender related, higher-turnout elections commonly pull in voters who are less informed about candidates for lower offices (Campbell 1960; Teixeira 1992) and more likely to rely on heuristics. High-turnout elections will bring more less-informed women to the polls; they can be expected to use cognitive shortcuts and vote for the Democrat (whom they infer to be more liberal). High-turnout elections will also bring more less-informed men to the polls, who will likely use these heuristics and vote for the (perceived) more conservative Republican.
State-Level Hypotheses
The economic situation of women in a particular state should also affect the gender gap. Women are more likely to use governmental programs for jobs and economic support (Clark and Clark 1999; Manza and Brooks 1998). As poverty in a state is feminized, it may cause even those women who do not rely on government programs to vote on the basis of social welfare issues. Cal Clark and Janet Clark (1999) explain that women tend to demonstrate a greater compassion for the less fortunate in society, as well as a greater willingness to share in the cost. Therefore, we expect to find larger gender gaps in states with higher rates of women in poverty.
Women's increasing presence in the workforce also drives the growth of the gender gap (Manza and Brooks 1998). By participating in the workforce outside the home, women gain independence and exposure to new experiences that shape their political behavior and give them increased psychological and financial independence (Carroll 1988). Women are also exposed to inequality within the workforce (Erie and Rein 1988). The changing economic position of women then leads them to adopt more feminist views and thus widens the gender gap.
The degree to which a state is ideologically polarized should also affect the size of the gender gap. When Democrats in a state are very liberal and Republicans very conservative, their respective nominees should be polarized as well. Given ideological differences between men and women, this kind of election should pull their votes apart and lead to a larger gender gap.
Additionally, the size of the gender gap should increase as the state's African-American population increases. African-American voters are disproportionately female and liberal (Teixeira 1992). Issues that invoke compassion and cost bearing, differentiating male and female voters, may well be similar to those that concern the average African-American voter. Thus, in an election where women vote differently than men do, the effect should be most pronounced in states with large African-American populations, which will have larger proportions of liberal women voting.7
Sigelman and Welch (1984) show that African-American women are more likely to support both African-American and women candidates. African-American men are more likely to support African Americans, but show no sex-based pattern in support for candidates.
An Additional Hypothesis
Finally, we present one additional hypothesis, alluded to in Figure 1, showing an increase in the size of the gender gap over time (Clark and Clark 1999, 2000; Kaufmann and Petrocik 1999; Manza and Brooks 1998). We expect that this longitudinal increase will remain even when controlling for other factors in the model, because the factors separating men and women voters are large, complex, and beyond the reach of even a well-specified model.8
Two factors common to vote choice models do not appear in our model—the electoral competitiveness and the state's partisanship. We find no compelling theoretical reason to include them when looking at the gender gap in vote choice. A competitive election could produce a large gender gap (as in the 2000 presidential election) or a small gender gap (as in 2004). The factors making elections competitive or uncompetitive (i.e., issues, candidates, and context) influence the gender gap more than does the competitiveness of the election. Similarly, a measure of the state's partisanship would also produce inconclusive results, as the gender gap exists and varies across Democratic and Republican states. For example, a highly Democratic state might produce a small gender gap because both men and women are more likely to vote for the Democratic candidate.
DATA AND OPERATIONALIZATION
Our dependent variable is the gender gap in a Senate election. We have used all Senate elections between 1988 and 2000 for which we were able to obtain reliable exit polling data.9
We studied 203 elections in all. Availability of data was an issue in many 1988 elections; therefore, we are able to include only 20 elections from that year. Due to unavailability of data on ideological polarization, Alaska and Hawaii are not included in the analysis. Gender gap data come from CBS News Polls (1988), Voter Research and Surveys (1990–92) and Voter News Service (1994–2000). Data files for 1988–96 were made available through Inter-university Consortium for Political and Social Research; those for 1998 and 2000 are available at CNN.com.
Among the campaign variables, we test the hypothesis about the sex of candidates with two dummy variables, one for a female Democrat/male Republican race and one for a male Democrat/female Republican race. The omitted case is that of same-sex candidates for the two major parties.10
Out of 203 elections, 17 had a male Democrat opposing a female Republican, 27 had a female Democrat opposing a male Republican, and 159 had same-sex candidates. In all but one of these 159 cases, the two candidates were male.
To measure this variable, we read the election summaries in both sources and coded whether the particular issue was mentioned as a significant one in the campaign. We coded the issues as salient to the campaign if the summaries mentioned the issue; given the brief nature of these summaries, we expect that they would not spend time talking about issues not central to the campaigns. We believe that these sources adequately summarize major events and issues in campaigns and thus are appropriate to use for our purpose. Out of the 203 elections we studied, crime was salient in 42 of them, education in 30, the environment in 49, health care in 41 and abortion in 42.
To measure the increase or decrease in turnout in a state, we took each state's average voter turnout from 1984 to 2000 and subtracted it from the turnout in a given year. Positive figures indicate higher-than-average turnout, while negative figures indicate that turnout was lower than average. This measure allows us to measure the extent to which more peripheral voters participated in any given year's election.
Data for the percentage of women below the poverty level were calculated from the Current Population Survey (CPS). On the basis of reported household income, respondent's sex, number of people in the household, and the thresholds for poverty found on the Census Bureau's Website, we calculated the percentage of women living below the poverty level. For the 1988 elections, we used data from 1989 because household income was reported differently in 1988.
Data on the percentage of women in the workforce and the percentage of African Americans in each state were reported in The Statistical Abstract of the United States. We used Robert Erikson, Gerald Wright and John McIver's (1993) index of mass polarization to measure a state's ideological polarization. This measure subtracts the mean ideology of Republicans in a state from the mean ideology of Democrats. Higher scores indicate greater polarization.12
Although these data were constructed in 1988, and are somewhat dated, we do not expect that dramatic changes in state ideological polarization occurred during the decade.
Our results are calculated using multivariate regression, with the size of the gender gap as the dependent variable. We use a GLS (generalized least squares) model rather than ordinary least squares to correct for the heteroskedasticity that might occur due to states having different sample sizes in their exit polls and hence unequal error in the measurement of the dependent variable. We also use robust standard errors within clusters to account for potential autocorrelation among the error terms (such as error caused by idiosyncratic factors that occur within a particular state across each year of the observation). While a well-specified model does minimize many of the problems associated with violating the no-autocorrelation assumption, the robust standard errors within clusters provides some statistical damage control as well.
RESULTS OF THE EMPIRICAL MODEL
Campaign-Level Effects
Table 1 shows that gender gaps in Senate elections are a function of both campaign-level and state-level variables. The sex of the Republican candidate, but not the Democratic one, matters. With a female Democratic candidate, the gap increases by 1.5 points, slightly short of statistical significance. With a female Republican candidate, the size of the gender gap shrinks by almost 3 points, a statistically and substantively meaningful result. We believe that women voters in these elections are more likely to support women candidates, regardless of their party identification. These findings support previous research that finds circumstances in which women voters are more likely to support women candidates, regardless of party (Cook 1994, 1998; Dolan 2004; Seltzer, Newman, and Leighton 1997; Wilcox 1994).
Regression model for equation estimating size of gender gap, 1988–2000
Incumbency and the number of years served has a modest impact. Each full term an incumbent serves decreases the gender gap by almost half a point. Incumbency provides name recognition and establishes a record with the electorate; the more years a Senator spends in office, the less the electorate will have to rely on heuristics, such as sex, in the voting decisions. The impact is small, but it demonstrates that in these elections the gender gap is not an anti-incumbent bias expressed by women (but see Chaney, Alvarez, and Nagler 1998).
Although the salience of crime as a campaign issue shows a positive sign and is close to statistical significance, no other issue variables affect the size of the gender gap. This null result contradicts our hypotheses for education, environment, and health issues and confirms it for abortion. Abortion aside, the lack of significant results for the issue variables could indicate that salience alone may not be enough. We address this later.
The relationship testing our last campaign-level hypothesis, that an increase in turnout from the state average would increase the size of the gender gap, is statistically insignificant. Voters in a high-turnout election seem to vote in the very same patterns as those in a low-turnout election insofar as the gender gap is concerned.
State-Level Effects
Our hypotheses concerning the state-level variables achieve modest results. Increased percentages of women below the poverty line show no effect on the gender gap; this suggests two possible explanations. First, high levels of women below the poverty line as a contextual factor may not be enough to change voting behavior. Perhaps one needs to experience poverty directly before it can change behavior. Testing this is beyond the scope of our aggregate data and is better addressed at the individual level. Second, the context of feminized poverty does not in and of itself create an adequate environment to influence the gender gap in elections; it might create an environment where issues in a campaign play a larger role.
The percentage of women in a state's workforce also had no significant impact on the size of the gender gap. This could mean several things. Lee Ann Banaszak and Eric Plutzer (1993) report that while entering the workforce may make women more liberal and feminist, the net effect may be diminished by the tendency of women homemakers to become more conservative when women enter the workforce in greater numbers. This finding is consistent with Janet Box-Steffensmeier, Suzanna De Boef, and Tse-min Lin's (2004) argument that women's workforce involvement does not automatically translate into sex differences in political behavior.
The percentage of African Americans in a state shows a very significant, positive relationship with the gender gap in vote choice. Holding all else constant, states with larger numbers of African-American voters show larger gender gaps, an increase of 0.15 percentage points for each percentage increment in the number of African Americans. The mechanism by which this happens is hard to pinpoint exactly. It could result from African-American women voting more frequently than African-American men and, as a group, favoring the Democratic Party. Second, we are persuaded by Hutchings et al.'s (2004) argument that gender gaps increase in electoral environments where racial issues, and by extension compassion issues, are prominent. States with higher populations of African Americans likely present electoral environments where these issues play a larger role, increasing the size of the gender gap.13
We also examined whether this result would hold when we controlled for region, specifically for states in the old Confederacy (that today generally have large African-American populations). When we added a control variable for southern state (not shown), this new variable was insignificant, and the coefficient for percent African Americans in the state showed no change in magnitude or significance.
Ideological polarization shows an effect, albeit one that falls slightly short of conventional levels of statistical significance. More polarization in a state is associated with larger gender gaps. Studies show that partisan and ideological gender gaps grow where men support conservative/Republican positions and women the liberal/Democrat position (Trevor 1999; Wirls 1986). Thus, it is not surprising that in states with a larger liberal/conservative divide, one would also find a larger electoral gender gap.14
An alternative specification of this variable, in which we measured polarization using the absolute value of the deviation between the Democrat and Republican share of the two-party vote in the most recent presidential election, turned up insignificant results.
Finally, the time-trend variable is extremely significant even as we control for factors explaining the gender gap. Each election year, the average gender gap increased by 0.79 percentage points, making this the single most important variable in the model. Idiosyncratic explanations available in presidential elections (based on individual candidate attributes) do not explain our findings, because our trends are based on a wider range of lower-visibility elections, including off-year ones. This secular trend transcends the variables in our model and warrants further explanation—men and women are increasingly seeing politics differently.
The Interaction of State and Campaign Context
The performance of our issue variables warrants further examination. It may well be that issues are important only with priming from other campaign or state factors. While there are many different interactions we could test, we focus on three: presence of a female Democrat, female poverty rate, and state ideology. As Kathleen Dolan (2004) suggests, issues may become more important due to stereotypes held by voters; thus we explore the relationship among the sex of the candidate, issues, and the gender gap. We examine the impact on the female poverty rate because we expect that when a larger number of women are living below the poverty line, compassion and cost-bearing issues will become even more salient to voters. Finally, in ideologically polarized states, we expect that candidates will reflect the greater polarization, possibly increasing the impact these issues may have on the gender gap. The results of regression models for each of the factors are presented in Table 2.15
We note at the outset that some of these interactions will produce low numbers of cases; when we begin with few cases where women run as Democrats, and a minority of cases where any one issue is significant, the resulting interaction will yield a small number of cases. This low number of cases biases our results against statistical significance, thus providing a more stringent test for our hypotheses. They also suggest that caution be exercised when interpreting the results. The solution to this problem, of course, comes in adding cases (which is more likely to happen going forward in time than backward, when even fewer women ran).
Regression models for equations estimating size of gender gap with interaction terms for campaign issues × contextual factors
We begin with the sex and party of each of the candidates. (We confine our discussion here largely to the interaction terms in the respective models.) If the issues we have discussed are those on which men and women hold differing views, a woman candidate running for the Democrats ought to polarize such differences. The Democratic Party is most identified with the issues that women support in greater numbers than men. Political attitudes most differentiated by sex will be “activated” in the minds of voters when the Democrats run a woman.
The context in which the election occurs does matter: Table 2 demonstrates that under some conditions, certain issues do influence the size of the gender gap. Model 1 in Table 2 presents a test of these hypotheses. When interacted with the presence of a female Democratic candidate, the salience of crime as a campaign issue increases the size of the gender gap by 3.7 points. The impact of the education issue is an even larger 4.25 percentage points. These results could arise from two different behaviors. First, when a woman candidate is running and either crime or education is a salient issue, male voters (particularly conservative male voters) might fear the female Democratic candidate's liberal positions on these issues, or female voters might be more attracted to the Democratic candidate because of congruence with their own views. Lacking individual-level data, we are unable to make a definitive conclusion about why this is happening, but the pattern seems clear.
The salience of health care actually decreases the size of the gender gap by 3 percentage points when a Democratic woman runs. Women in the House and Senate, especially Democratic women, played a prominent role in the efforts of the Clinton administration to reform health care. Although we do not have the data to test this, we speculate that the liberal position on this issue might generate a stronger backlash among conservative voters (both male and female) than the other issues we examine, leading to a shrinking gender gap. Finally, the environment and abortion, once again, produce no meaningful effect.16
We would like to have run this interaction as well using female Republican and issues as a way to parse out the effects of candidate sex and party. However, our ability to do this is hampered by the extremely small number of cases where these issues are significant and a woman runs for the Republicans (between 2 and 5 for each issue). The results are thus too unstable to analyze, and are not reported here.
Next, we consider how the issue variables interact with state-level variables. We expect campaign issues to have their biggest impact in states where more women are poor, leading to increased displays of gender consciousness and compassion among women voters. When fewer women are poor, we have less reason to believe that these issues will activate gender-based patterns of voting. To test this, we partitioned states according to whether their female poverty rates during the year of the election were higher or lower than the national average. We then interacted this dummy variable with the issue salience variable.
Model 2 of Table 2 presents the model with these interaction terms. Only the salience of environmental issues reaches statistical significance; the gender gap increases by slightly more than 2 points when the environment is a salient issue in a state where female poverty rates are above average. We are somewhat puzzled by why this particular issue showed a significant result, as the relationship between the environment and the economic needs of women addressed by this interaction model seems somewhat tenuous. Still, this is some limited support for a context-based role of campaign issues. None of the other issue variables show any effect when interacted with the dummy variable for whether the state's female poverty rates were above average.17
We also ran a model that interacted the issue variables with the percentage of women in the workforce in a given state (not shown). None of the issue variables in the model were significant.
Finally, Model 3 shows the impact of issues interacting with state ideology.18
We measured the liberalness of a particular state by taking the state's two-party share of the vote for the Democratic presidential candidate in the two most recent presidential elections and subtracting it from the Democratic candidate's national two-party share of the vote. The difference in the most recent presidential election was multiplied by two and added to the result from the next-to-most recent election; we divided the sum by three. States with negative scores are more conservative than the national average.
In our model, only crime had a significant effect on the gender gap when interacted with the state ideology dummy variable. Crime's salience in the campaign increased the gender gap by over 2.5 points. The importance of crime suggests that the gender gap could be a function of different issue positions made central by the electoral context. Men and women tend to hold different views on the use of force domestically, which would include how crime is handled. These differences could make the Democratic candidate more attractive to women and the Republican candidate more attractive to men. The context of a conservative state may increase the attraction to the Democratic candidate for women, or the attraction to the Republican candidate for men. Once again, none of the other issue variables proved significant.
To summarize, the interaction of campaign-level and state-level variables with issue variables demonstrate that the context in which the election occurs matters when studying the gender gap. Four of the 12 such interactions were significant in the anticipated direction (one in the incorrect direction). This indicates that campaign issues can play a conditional role in the gender gap's magnitude. That is, elections occur within an environment that helps shape outcomes. The presence of a female candidate, percentage of women below the poverty level, and ideology of a state create a context in which certain issues are important. It is not just the issues that are raised but the context in which they are raised, that affects the size of the gender gap.
In evaluating these results, we must remember that in any election, the majority of the gender gap can be attributed to partisanship and is therefore relatively stable (Kaufmann and Petrocik 1999). Beyond that, our analysis indicates that context explains its variation across elections in both space and time. The campaign matters. The sex of the candidates, incumbency, and sometimes issues, are important, the latter largely dependent upon the state context in which they are raised. Additionally, aspects of the state population matter, again often in interaction with campaign issues. Thus, variation in the gender gap across different contests can be attributed to the context in which the election is occurring.
CONCLUSION AND IMPLICATIONS
We have sought to discover the determinants of the aggregate gender gap in Senate elections. Our discussion focused on both campaign characteristics and the state-level context. Both contribute to the gender gap individually and as interactions. The sex of the candidates and the presence and seniority of an incumbent are important campaign factors, while within the state context, the African-American population of a state increases the gender gap.
Moreover, we have determined that the salience of various campaign issues affects the size of the gender gap, particularly as these issues intersect with other campaign-level variables and with state context-level variables. In particular, we have demonstrated that the impact of campaign variables is slightly more pronounced when women run as Democrats, when poverty in the state is highly feminized, and when the state is conservative. The voice of women in the electorate is heard more loudly when a woman articulates the views on which women and men differ. And it is heard more loudly when women find themselves facing economic adversity and in a hostile political climate.
In an ideal world, we would be able to parse out the impact of each of these issues in numerous contexts—when women were running for the Democrats, when the state was more conservative, and when poverty in the state was highly feminized. Unfortunately, the data do not allow us the degrees of freedom to study all of these contingencies in the same model.19
Only 30 of our races had a woman running for the Democrats, only 10 of these women ran in conservative states, and in only two of these races was the female poverty rate above the national average.
Further research on the role context plays in shaping the gender gap in vote choice is called for. This project's use of the Senate election as the unit of analysis limits many of the conclusions we can draw about the mechanisms that are at work among individual voters. Just as we have built on work done at the individual level, new work needs to consider our aggregate findings in forming expectations about the gender gap in vote choice. In particular, future work at the individual level ought to consider further the context in which these individual-level choices are made.
Gender gaps determine elections as much as other factors over which far more ink has been spilled. Moreover, they continue to grow in importance with each passing year. Controlling for the increase in the number of women candidates, the increase in the percentage of women below the poverty level, and the increase in the number of women in the workforce, our model shows that the average gender gap has increased by nearly one full point in each electoral cycle. Despite the considerable amount of variance we have been able to explain, the puzzle of why men and women continue to vote so differently has grown ever more complex. As a field, we have developed strong models and a good understanding of why individual men and women vote differently. Now, we believe the time has come to extend these analyses from the individual level to the aggregate level. This allows us to determine more fully the impact of campaigns, contexts, and the combination of both on the size of the gender gap.
