THE GENDER GAP IN U.S. PRESIDENTIAL ELECTIONS: WHEN? WHY?

THE GENDER GAP IN U.S. PRESIDENTIAL ELECTIONS: WHEN? WHY?

The Gender Gap in U.S. Presidential Elections: When? Why? Implications?1 Jeff Manza Pennsylvania State University Clem Brooks Indiana University Social scientists and political commentators have frequently pointed to differences between men and women in voting and policy attitudes as evidence of an emerging “gender gap” in U.S. politics. Using survey data for 11 elections since 1952, this study develops a systematic analysis of the gender gap in presidential elections. The authors find evidence that women’s changing rates of labor force participation explain the origins of the gender gap.

Additional analyses show that attitudes toward social service spending mediate the interrelationship of women’s labor force participation and vote choice. In the 1992 election, feminist consciousness also emerged as a signifi- cant factor shaping women’s voting behavior. INTRODUCTION When the right to vote was finally extended to women in the United States with the adoption of the 19th Amendment to the Constitution in 1920, there was considerable speculation about the possibility that a distinct women’s vote would emerge. Many feminists hoped—and some male politicians feared—that newly enfranchised women voters would back candidates supporting a wide range of “maternalist” social policies such as protective wage and hours laws, expansive health and housing policies, and other types of social provision for indigent women and families. But such a vote failed to materialize in the period after enfranchisement (Lemons 1973, pp. 157–58; Baxter and Lansing 1983, pp. 17–22; Klein 1984, pp. 13–21, 142–43; Alpern and Baum 1985; Skocpol 1992, pp. 505–6). 1 Authors’ names are listed in reverse alphabetical order; both contributed equally to the article. We thank Eric Plutzer for incisive comments on an earlier draft. Data from the American National Election Studies were provided by the Inter-University Consortium for Political and Social Research; the usual disclaimers apply. Direct correspondence to Jeff Manza, Department of Sociology, Pennsylvania State University, University Park, Pennsylvania 16802-6207. Email: manza@pop.psu.edu  1998 by The University of Chicago. All rights reserved. 0002-9602/98/10305-0003$02.50 AJS Volume 103 Number 5 (March 1998): 1235–66 1235American Journal of Sociology More recently, however, a gender cleavage has developed in U.S. electoral politics. When voting in presidential and in many congressional elections, women have disproportionately supported Democratic candidates in comparison with men. The size of this “gender gap” has often been larger than the margin of victory for Democratic candidates in congressional races as well as in the 1992 and 1996 presidential elections. Appeals designed to attract the women’s vote have become widespread, as have strategies aimed at maximizing or minimizing the size of the gender gap for Democratic and Republican candidates respectively.2 Other gender gaps in U.S. politics have also been observed (Conover 1988, pp. 985–86): for example, in turnout rates, partisan identification, policy attitudes, rates of office holding, and evaluations of political officeholders. In the case of voter turnout, researchers have found that men were significantly more likely to vote than women for decades after enfranchisement, although this gap had become reversed in the 1980s (Merriam and Gosnell 1924; Welch 1977; Baxter and Lansing 1983; Beckwith 1986; Firebaugh and Chen 1995). Women and men have been shown to hold different views on a range of domestic and foreign policy issues (e.g., Smith 1984; Shapiro and Mahajan 1986; Conover 1988), and those differences appear to influence their evaluations of elected officials (e.g., Gilens 1988). Women also continue to be significantly less likely to run for, or hold, elected political office, although the number of female elected offi- cials at all levels has risen sharply in recent years (Darcy, Welch, and Clark 1994; Witt, Paget, and Mathews 1994). Our focus in this article is on gender differences in the voting behavior of men and women in presidential elections. While discussions of the gen- 2 Some feminists have viewed the gender gap as a means of securing a new Democratic Party majority; see, e.g., Abzug (1984) and Smeal (1984). A couple of recent examples highlight the importance of gender-centered electoral strategies in the 1996 election, as reported by the New York Times. (1) “Republican governors . . . were given a sobering presentation of their party’s troubles with female voters, and asserted that a more compassionate tone and a focus on education were the cure. The presentation, at the closing session of a three-day meeting of the Republican Governors’ Association, came from Haley Barbour, the Republican national chairman, who said that according to the party’s own research, Bob Dole and Republican congressional candidates fared even worse among women on Election Day than had been suggested by exit polls taken by the news media” (“Gender Gap Wider Yet, GOP Session Is Told,” New York Times, November 27, 1996, p. A22 [National Edition]). (2) “Seeking to insure a strong turnout from a crucial voting bloc, President Clinton made a special appeal to women today by highlighting Government efforts to fight breast cancer and telling a rally audience in Virginia that the election is ‘not about party, it’s about you.’ . . . Clinton stressed the health care, education and child-based themes that have been central to his campaign and that have a particular resonance with women” (“Clinton Campaign Puts an Emphasis on Female Voters,” New York Times, October 28, 1996, p. A1 [National Edition]). 1236The Gender Gap der gap in electoral politics have become commonplace in election reporting and commentary (a veritable “national pastime,” according to Jennings [1988, p. 12]), there have been very few systematic analyses of its historical origins. We have identified at least 23 analyses that date the origins of the gap to the 1980 election.3 In that election, Republican candidate Ronald Reagan ran a campaign emphasizing his opposition to the Equal Rights Amendment (ERA) and abortion and his support for “traditional” family values and for an aggressive policy of military containment of Soviet-style communism. These themes, many of which have characterized the Republican Party’s policy agenda since Reagan, are presumed to have alienated a large number of women voters who have never returned to the Republican fold. Is the gender gap indeed as recent a development as assumed by most scholars? To anticipate one of the key findings of our analysis, we present in figure 1 two sets of estimates of the gender gap in presidential elections for the period between 1952 and 1992. The figure’s left-hand panel shows the changing size of the gender gap using raw data from the National Election Studies (NES) data (and the Voter News Service [VNS] data for the 1996 presidential election).4 Even without model fitting, the observed estimates suggest that the gender gap has grown over much of the NES series. Figure 1’s right-hand panel offers further corroboration, showing that, when competing models of change in the magnitude of the gender gap have been fitted to the data (details discussed below), the estimates from the preferred model reveal a growing trend since 1952, which peaked but did not begin in 1980. In this article, we seek to explain the causes of this evolving gender gap in vote choice. Our analysis is in four sections. In the first part of the 3 See Baxter and Lansing (1983, p. 179), Miller (1983, pp. 7–8), Abzug (1984, p. 1), Klein (1984, pp. 2, 163; 1985, p. 35), Bolce (1985, p. 373), Mansbridge (1985, pp. 165–66), Norris (1985, p. 192), Wirls (1986, p. 316; 1991, pp. 117–19), Heideprien and Lake (1987, p. 1), Klatch (1987, p. 3), Mueller (1988, pp. 16–17), Kenski (1988, pp. 38–39), Erie and Rein (1988, p. 174), Carroll (1988, p. 242), Miller, Hildreth, and Simmons (1988, p. 106), Hartmann (1989, p. 153), Faludi (1991, p. xx), Tolleson Rinehart (1992, p. 146), Delli Carpini and Fuchs (1993, p. 34), Bendyna and Lake (1994, pp. 238–39), Chaney, Alvarez, and Nagler (1996, pp. 1, 20). Writing in the mid-1980s, Bolce (1985, p. 376) noted that “all commentators seem agreed that [Reagan] and his policies are at the center of the gender gap controversy.” Viewing the gender gap as reflecting the unique historical circumstances of the Republican presidential victories beginning in 1980, a handful of analysts argued that the gender gap per se has no long-term significance for American politics; see, e.g., Poole and Ziegler (1985); Wirls (1986, 1991). 4 We discuss in detail the NES data and our quantitative measure of the gender gap (the standard deviation of the probability of Democratic vote choice among men and women) in the data and measures section below. 1237American Journal of Sociology Fig. 1.—The emergence of the gender gap in presidential elections, 1952–92. (Index scores are calculated by taking the SD of the observed or fitted probabilities of Democratic vote choice among women and men voting for a major party candidate.) article, we summarize the main causal arguments that have been developed to explain political differences between men and women. In the second part, we discuss the data and statistical models we use to adjudicate the competing claims about voting behavior implied by these arguments. In the third part, we present the results of our analyses, first establishing the trends and magnitude of the gender gap from 1952 to 1992 and then developing an analysis of the factors responsible for this trend. The article’s conclusion discusses the implications our research has for theories about the origins and development of gender divisions in American political life. THEORIZING THE GENDER GAP The classics of postwar political behavior research—such as Berelson, Lazarsfeld, and McPhee (1954), Stouffer (1955), Lane (1959), Lipset ([1960] 1981), Campbell et al. (1960), and Almond and Verba (1963)—did not 1238The Gender Gap view gender as a central factor (cf. Bourque and Grossholtz 1974; Carroll and Zerilli 1993, pp. 57–58). When gender differences were mentioned at all, women were often portrayed as lacking interest in politics and a sense of political efficacy (e.g., Berelson et al. 1954, p. 25; Campbell et al. 1960, pp. 489–92). If they did take an interest in politics, women were said to be more likely to “personalize” politics than men, who focused more on substantive issues (Greenstein 1965, p. 108) or to “focus . . . political attention upon persons and peripheral ‘reform’ issues” (Lane 1959, p. 216). These early studies asserted that women were more apathetic and less well informed than men, who had the benefit of greater exposure to public life (e.g., Almond and Verba 1963, p. 325; Lipset 1981, pp. 216–17), and were less politically tolerant than men (Stouffer 1955, pp. 131–49). When women did participate in elections, they were also viewed as likely to simply follow the lead of their husbands (Campbell et al. 1960, pp. 485– 86, 492–93). The resurgence of women’s liberation and second wave feminist movements beginning in the late 1960s and early 1970s, coupled with the accumulation of evidence that women were now turning out to vote at levels comparable to men, had by the early 1980s led to critical reassessments of the classical treatments of women and politics (Tolleson Rinehart 1992, pp. 13–14). Although there are multiple ways of carving up the recent literature on women and politics (cf. Clark and Clark 1986; Bennett and Bennett 1993), we propose that, for the purposes of understanding voting differences, four distinct theories can be identified. One theoretical perspective emphasizes differences in the political socialization of men and women, which are seen as shaping their public activities (including voting). A second set of theories links the growing autonomy of women voters to changing marital patterns and rising divorce rates. A third perspective gives primacy to rising levels of feminist consciousness among women since the 1960s. A final theory points to the importance of the increasing participation of women in the labor force as a factor influencing women’s political orientations. We discuss each of these approaches in turn.5 5 Not all of the scholarship we discuss below as exemplars of each theoretical approach seeks to explain gender differences in voting behavior per se. These arguments do, however, identify causal mechanisms that have substantive relevance to the phenomenon under investigation. We thus use the four distinct theories as a source of hypotheses about the causal mechanisms of differences in vote choice among women and men. It is also important to note that there is often some overlap in the causal factors emphasized in individual studies purusing multicausal explanations of the gender gap. The four causal arguments we identify should thus be viewed as ideal-types rather than literal descriptions of all the accounts offered by scholars who have written about the gender gap. 1239American Journal of Sociology Gender Socialization The first set of theories we consider emphasizes the importance of the different patterns of socialization experienced by women and men in shaping core values, political orientations, and behavior.6 Two distinct strands of gender-centered social psychological research can be identified. The first emphasizes the importance of childhood socialization (e.g., Chodorow 1978; Gilligan 1982; for applications to politics, see, e.g., Feltner and Goldie [1974], Hess and Torney [1978], and Kelly and Boutilier [1978]). These interpretations view gender differences in political behavior as stemming from the sex role conditioning experienced by girls and boys. These differences are assumed to endure over the life course, insofar as they are the product of a contrasting set of moral values imparted during childhood.7 Adult socialization approaches, by contrast, hypothesize that the effects of childhood socialization are mediated by adult roles, especially women’s experience of motherhood (Sapiro 1983; Ruddick 1989). Traditional familial arrangements are viewed as especially consequential, given that they result in a much higher likelihood that women will be connected to their roles as mothers, leading to the development of “separate spheres” of moral and political activity among men and women (Sapiro 1983, chap. 2). Since the New Deal, the Democratic Party has stood for a greater governmental role in insuring the welfare of individuals and families. Conversely, the Republican Party has—over much of this period—been viewed by many voters as the party more likely to use military force (and hence threaten the lives of young men and women in the armed forces) in the pursuit of geopolitical aims. The socialization approach predicts that women who have been socialized to value nurturing activities may react to these policy differences by viewing the Democratic Party more favorably. Despite focusing on different aspects of the life course, both types of 6 We caution that the socialization literatures reviewed in this section are heterogeneous in many of their assumptions; they nevertheless share common assumptions about the causal importance of socialization experiences in producing gender differences. 7 As Gilligan (1982, p. 100) puts it, “The moral imperative that emerges repeatedly in interviews with women is an injunction to care, a responsibility to discern and alleviate the ‘real and recognizable trouble’ of this world. For men, the moral imperative appears rather as an injunction to respect the rights of others and thus to protect from interference the right to life and self-fulfillment.” Some evidence suggests that when it comes to economic voting, women are more likely to be motivated by “sociotropic” concerns (i.e., concerns about the economic health of the nation as a whole) while men are more likely to be concerned with “egocentric” issues relating to their own (or their family’s) economic well-being (Welch and Hibbing 1992; Chaney et al. 1996). 1240The Gender Gap socialization approaches view political differences between men and women as stemming from general patterns of sex-role differentiation, not “interests” (Sears and Huddy 1990). As a consequence, no particular sociodemographic group of women is expected to hold political views different from those of men. As Elshtain (1984, p. 24) puts it, “The gender gap is real and it concerns what political scientists like to call ‘moral issues,’ which traditionally have been the purview of women and, for that reason, often in the past labeled ‘social questions,’ not properly political ones. . . . On questions of conscriptions, militarization, nuclear weapons, capital punishment, and environmental safeguards, statistically significant differences show up between the categories ‘all men’ and ‘all women.’ These differences cut across class and education and hold whether the woman in question is employed or not.” Similarly, Sears and Huddie (1990, pp. 274–75) argue that political divisions among women “go back in large part, if not completely, to political socialization in earlier life” and that their empirical work suggests that “interest-related differences were dwarfed by differences based on predispositions, as in much earlier work.” The socialization approach to the gender gap would therefore be challenged, albeit indirectly, by evidence that differences among women and between women and men are related to economic interests. Family Transformation and the Rising Autonomy of Women A second causal mechanism that may explain the growth of the gender gap is the hypothesized increase in women’s autonomy, resulting from changes in family structures that have significantly reduced the importance of marriage (and the percentage of adult women in intact marriages). Many of the classical works of political behavior assumed that married women tend to follow the political lead of their husbands. Some feminist scholars have reached similar conclusions but for different reasons, arguing that women’s interests are subordinated to those of men in marriage because of the routine functioning of the sexual division of labor (e.g., Hartmann 1981). The interdependence of men and women in traditional marriages may thus give them common material interests. As Carroll (1988, p. 241) puts it, “The wife and husband . . . have a common interest in the rate of inflation, the condition of the stock market, the husband’s veteran benefits, and the availability of public transportation.” Such common interests may thus override other sources of division between men and women to produce electoral convergence. Rising divorce rates since the 1960s, however, coupled with the increasing average age of first marriage, contribute to growing numbers of women who are independent of husbands and the institutional constraints 1241American Journal of Sociology of marriage (e.g., Spain and Bianchi 1996).8 Divorced, widowed, or single women may have very different material interests (and ultimately voting preferences) than those of married women.9 Some analysts have indeed found evidence to support this hypothesis. Frankovic (1982, p. 444) reports that “sex differences in support for Ronald Reagan in 1980 were smallest in two adult, one male, one female households. Consensus within a household in the course of living together does appear to minimize sex differences.” Carroll (1988) likewise suggests that the gender gap in the 1980 and 1982 national elections can be largely attributed to the growing number of autonomous women who are economically and/or psychologically independent from a husband (see also Rossi 1983, pp. 726–27). Feminist Consciousness A third causal thesis we consider in this article relates to the rise of feminist consciousness as a possible source of voting preferences (Gurin 1985; Conover 1988; Cook 1989, 1993; Cook and Wilcox 1991; Tolleson Rinehart 1992; Plutzer and Zipp 1996). While gender identification appears by itself to have a limited influence on policy attitudes (Miller et al. 1988, pp. 111–12), feminist identities may be of much greater consequence for policy attitudes and ultimately political behavior (Gurin 1985; Conover 1988; Cook 1993). Identification with feminism is typically measured in terms of the strength of women’s identification with other women, the women’s liberation movement or feminists, or (less commonly) by egalitarian attitudes toward gender roles. Conover (1988) argues that gender differences in policy attitudes between men and women can be traced to the greater prevalence of feminist identities among women. Cook’s (1993) recent work finds that net of partisanship, political ideology, and sociodemographic factors, feminists were significantly more supportive of Democratic presidential candidates in 1972, 1984, and 1988.10 Such findings suggest that political cleavages among women (and between women and men) 8 Approximately one-half of all marriages are predicted to end in divorce or separation (Cherlin 1992, pp. 23–25), and the percentage of adult women not currently married rose from 36.6% in 1960 to 47.2% in 1992 (Bianchi and Spain 1986; Spain and Bianchi 1996). 9 For example, consider the financial consequences of divorce: estimates of women’s economic decline after divorce range from 13% to 35% (Peterson 1996, p. 529); for overviews of the extensive research documenting the diverging material interests of men and women after divorce, see, e.g., Holden and Smock (1991); Sørensen (1992). 10 Plutzer and Zipp (1996) find evidence that in 1992 the appeals of women candidates running for statewide office among women voters were amplified by Democratic female candidates who were seen as “feminist.” 1242The Gender Gap may be a product of growing identification with the feminist goals of the women’s movement. Women’s Rising Labor Force Participation The final mechanism that we consider for explaining the gender gap comes from interpretations that stress the political significance of the increasing proportion of women in the paid labor force. Labor force participation rates for women have risen steadily throughout the 20th century (rising from 33.9% in 1950 to 58.8% in 1994; see Spain and Bianchi 1976, p. 81), but women’s wages have remained significantly below those of men overall, despite recent improvements for some women (Marini 1989; Bernhardt, Morris, and Handcock 1995; Petersen and Morgan 1995). There are a number of distinct processes through which labor force participation may shape women’s political behavior. First, some analysts have hypothesized that work exposes voters to discussions about candidates, policy debates, and other information about political campaigns (e.g., Lipset 1981, p. 217). The disappearance of the turnout gap between men and women is frequently viewed as a function of the increasing proportion of women in the workforce, along with increases in women’s level of education (Andersen 1975; Welch 1977; McDonough 1982; Baxter and Lansing 1983; Andersen and Cook 1985; Beckwith 1986; but see Wolfinger and Rosenstone 1980, p. 43; Clark and Clark 1986). Other analysts suggest that employment increases women’s support for feminist political goals and political activism by providing women with life experiences that call into question traditional gender roles (Rossi 1983; Klein 1984; Luker 1984, chap. 5; Gerson 1985, 1987). Further, paid employment directly exposes women to gender inequalities that they are less likely to experience as homemakers, while also providing them with a means of economic independence that may shape their political behavior. Finally, women are more dependent on the public sector (and the expansion of the welfare state) for employment than men (Erie and Rein 1988), and they also tend to depend more on social programs to support and subsidize their families (Piven 1985; Deitch 1988). The political significance of employment for women’s political attitudes and behavior remains controversial, however. Andersen and Cook (1985) report that employment influences attitudes toward work-related issues (such as whether women should be the first to be laid off) but not broader political values. Similarly, using General Social Survey data, Plutzer (1988) and Deitch (1988) both found that labor force participation has very modest effects on support for feminist political goals; each reports that other sociodemographic variables have a much larger effect. Our 1243American Journal of Sociology evaluation of the work-centered approach reconsiders these debates in the context of the emergence of the gender gap in voting behavior. Evaluating the Theories The gender socialization thesis cannot be tested directly using the crosssectional NES data that we analyze.11 If, however, this thesis is relevant to explaining the gender gap in presidential elections, we would expect that variables representing sociodemographic cleavages should not explain the gender gap, given that political differences between men and women are assumed not to be a function of interests. The women’s autonomy thesis predicts that the women who are economically and psychologically independent of men are the most likely to diverge from men in their political behavior. We test this claim by examining the effects of being married over the 40-year period covered by our analyses. The feminist consciousness thesis predicts that women who identify with the women’s movement are the most likely to vote Democratic. To the extent to which feminist consciousness is growing, we can expect it to help explain the emergence of the gender gap (as discussed earlier, this thesis makes no claims about whether feminist consciousness is concentrated among certain social groups, but we explore this possibility as well in the course of our analyses). The labor force participation thesis predicts that women in the workforce are the most likely to be Democratic and that the growth of the gender gap in electoral politics reflects the increasing proportion of working women. 11 Perhaps the most decisive test of the socialization hypothesis, as one of the AJS reviewers pointed out, would be to examine the differences between men and women on the use of force to attain foreign or domestic political goals. The disproportionate support for the use of force or military intervention in foreign policy among men is well established (e.g., Smith 1984; Ruddick 1989; Wilcox, Ferrara, and Alsop 1993), though generally fairly small. Shapiro and Mahajan (1986) reviewed public opinion polls from the 1960s to the 1980s and found an average of 6%–8% difference in opinion between men and women on the use of force. Such differences may lead to greater support for presidential candidates who advocate more moderate military postures— generally the Democrat—among women, and some analysts have suggested just such an interpretation for the Reagan period (e.g., Frankovic 1982; Poole and Ziegler 1985; Sears and Huddy 1990; see also Gilens [1988] on the gender gap in presidential evaluations). Other analysts have found little evidence that attitudes toward the use of force affect gender differences in voting (e.g., Chaney et al. 1996). Given the considerable changes over time in the foreign policies pursued by the parties (and their frequent convergence in the pursuit of a “bipartisan” foreign and defense policy), as well as the measurement problems associated with the variability of the specific “force” issues in a given election, we do not analyze them here. 1244The Gender Gap TABLE 1 Variables from the NES Variable Coding 1952–92 analyses: Partisan vote choice ……………………. GOP voter 5 0; Democratic voter 5 1 Gender ……………………………………….. Men 5 0; women 5 1 Linear year covariate* ………………… 1952 5 1, 1956 5 2, . . . , 1992 5 11 Class location ……………………………… Dummy variables for professionals, managers, routine white-collar workers, self-employed, skilled workers, and un- and semiskilled workers; reference 5 non-labor-force participant Class identification ……………………… Working class 5 1; all else 5 0 Labor force participation …………….. Nonparticipant 5 0; participant 5 1 Years of education* ……………………. 1 year 5 1, 2 years 5 2, . . . , 171 years 5 17 Marital status* ……………………………. Not married 5 0; married 5 1 Household income* …………………….. Scaled to constant 1992 dollars Race ……………………………………………. African-American 5 1; all else 5 0 Region ………………………………………… Dummy variables for residence in South, Midwest, or West; reference 5 Northeast Cohort ………………………………………… Dummy variables for 1950s, 1960s, and 1970s generations; reference 5 all else 1980–92 analyses: Gender role attitudes* ………………… Likert item: women should stay at home 5 1, …, should have an equal role 5 7 Views of social services* …………….. Likert item: cut spending/services 5 1, . . . , increase spending/services 5 7 Views of women’s movement* …….. Feeling thermometer score: 0°–100° * Continuous variable. DATA AND MEASURES We analyze data from the NES for presidential elections from 1952 through 1992 (Center for Political Studies 1995). The NES is the premier source of voting data for the United States and contains the items necessary to measure gender-based political cleavage in presidential elections over an extended time period. The dependent variable in our analyses of trends in the gender gap is major party vote choice (coded “1” for the Democratic and “0” for the Republican candidate). Table 1 summarizes the variables in our analyses. Gender is coded as a dichotomy (women 5 1). A number of analysts have found a relationship between the class cleavage and the gender gap (Goertzel 1983; Burris 1984; Wirls 1986, 1991), and in our multivariate analyses, we utilize three measures of class: objective class location, subjective class identification, 1245American Journal of Sociology and household income. Building upon recent work on class voting (Heath, Jowell, and Curtice 1985; Hout, Brooks, and Manza 1995; Manza, Hout, and Brooks 1995), our objective class location variables distinguish six class categories (professionals, managers, the self-employed, routine white-collar workers, foremen and skilled workers, and semi- and unskilled workers), as well non-full-time labor force participants who are employed less than 20 hours per week (this seventh category is treated as the reference in the regression models). We have recoded the NES occupational data for each election to conform to this class map.12 Our subjective class identification measure is a dichotomy (working-class identification 5 1). Household income is a continuous variable, scaled to 1992 dollars. In the models that do not include the six objective class dummy variables, we use a separate dichotomous variable to analyze the effects of labor force participation (employed more than 20 hours a week 5 1). The class and labor force participation variables enable us to test hypotheses about the role of work as a factor shaping women’s political attitudes and behavior. To assess the possibility that generational shifts in socialization patterns (Inglehart 1990; cf. Miller and Shanks 1996) may be related to the emergence of the gender gap, we include in our regression models three dummy variables for 1950s, 1960s, and 1970s cohorts. Cohort membership in the 1950s is defined as respondents born between 1933 and 1944; the 1960s cohort is defined as respondents born between 1945 and 1962; and the 1970s cohort is defined as respondents born since 1963. We evaluate the women’s autonomy thesis with a marital status variable, coded as a dichotomy (married 5 1). We also include a series of controls for years of education, race (African-Americans 5 1), and region (three dummy variables for South, Midwest, and West residence). Gender differences in policy attitudes may mediate the relationship between the various sociodemographic variables in our models and vote choice. A number of scholars have pointed to differences in attitudes toward domestic policy, especially the comparatively greater support of women for social provision than men (Piven 1985; Deitch 1988; Erie and Rein 1988) or a combination of issues (Chaney et al. 1996). Divergent views of social provision are a potentially important source of voting differences, given the long-standing differences in the major parties’ domestic policy platforms. Other analysts have located the gender gap in growing support for gender equality, including the ERA, among women (Klein 1984, pp. 157–64; Burris 1984, p. 338; Smeal 1984). Finally, some proponents of the feminist consciousness thesis assert that 12 Additional details about our occupational coding scheme can be found in Brooks and Manza (1997a). 1246The Gender Gap it is attitudes toward the women’s movement that lead to gender differences in electoral politics. We use three scales to measure policy attitudes and feminist consciousness. These variables are only available for our analyses of recent presidential elections (1980–92). However, as will become clearer in the course of our analyses, the results have implications for the earlier years in which we cannot measure these variables. The first item is a seven-point scale measuring support for social services in which respondents indicate their preference for increasing social service spending versus cutting spending by decreasing social services. The other two items measure attitudes that relate to the feminist consciousness thesis. Both are seven-point Likert items, which we treat as continuous variables. The first of these tests attitudes toward gender role equality by asking whether women and men should have equal roles in the family and the workplace. In this item, response category “1” indicates the greatest support for the view that “women’s place is in the home” and “7” indicates the greatest support for the view that “women should have an equal role with men in running business, industry, and government.” The second item is a 100-degree feeling thermometer that asks respondents to rate how warmly they feel toward the women’s movement, with higher thermometer scores indicating more favorable views of the women’s movement.13 Measuring the Gender Gap We conceptualize the gender gap in presidential elections as the average difference between men’s and women’s major party vote choice. When women’s and men’s votes are similar (i.e., both support the Democratic or Republican candidate at the same rate), the gap disappears, but when their votes diverge, the gap emerges. The gender gap varies from election to election, and the task at hand is to determine whether there is evidence that variation in its magnitude over time represents an emerging trend. As presented in equation (1), our measure of the gender gap (κ) is calculated as the standard deviation of the probability of Democratic vote choice ( j 5 1) among women and men (g 5 1 for women; 2 for men). This measure can be calculated using either the raw NES data (see fig. 1’s first panel), or using the fitted probabilities according to a particular 13 The wording of this item has changed during the four surveys we analyze. In the 1980 and 1984 surveys, the referent of this item is the “women’s liberation movement”; in 1988, the referent is “feminists”; and in 1992, the referent is the “women’s movement.” Given this change in question wording, we analyze each of these four surveys separately rather than analyze them jointly, using election year as a covariate in our models. 1247American Journal of Sociology statistical model (see fig. 1’s second panel). Either way, this index measures the average difference in voting behavior among men and women at a given year (t 5 1 for 1952, 2 for 1956, . . . , 11 for 1992), and by comparing index scores over time, we can thus infer the strength and direction of change in the gender gap over time:14 κt 5 √^ G g51 (Pˆtgj 2 Ptj) 2 G . (1) Models We derive the fitted probabilities of men’s and women’s vote choice from the coefficients of the logistic regression model that we select as the best description of the data. The models we consider predict the log odds of choosing the Democratic over the Republican presidential candidate, which we designate by yˆij for vote choice j for person i in sample size N (i 5 1, . . . , N). Because we are analyzing time trends, we have pooled the 11 NES studies for 1952–92 into a single data set, in which time is itself a covariate. The base model (model 1 in table 2) includes terms only for the main effect of election year (βtj).15 Insofar as this model does not include any gender-by-year interactions (or the main effect of gender), it serves as a useful comparison with models that parameterize trends in the gender gap: yˆij 5 βj 1 ^ T t51 βtjCit. (2) Our preferred model (model 6 in table 2) of change in the gender gap has a single additional parameter (βgj) for the (changing) effect of being 14 Our measure of the gender gap follows our other recent work on the measurement of social cleavages (see, e.g., Brooks and Manza 1997a, 1997b, 1997c; Manza and Brooks 1997). These measures have several desirable properties and avoid the serious biases that often affect cleavage variables that have more than two categories (such as religion and class). Note that because gender is measured as a dichotomy, the index we use in the current study is related to simpler, percentage difference measures (e.g., gender gap 5 percentage of female Democratic voters 2 percentage of male Democratic voters). For cleavage variables having only two categories, this percentage difference measure is simply twice that of our standard deviation measure. 15 In eq. (2), Cit represents dummy variables for election year. To identify the model, we set the lowest βtj (for the 1952 election) equal to zero. 1248The Gender Gap female (g 5 1) during the 1952 through 1992 elections.16 In equation (3), Di1 is a dummy variable for gender. The final term, Z2 i0, is a constant with fixed scores for year (0 5 1952, 1 5 1956, …,6 5 1976, and 7 5 years $ 1980). The superscript indicates that the constant is squared, and the constraint on time produced by this exponential function results in the distinctive trend line observed in the graphed estimates of figure 1’s right-hand panel. In the course of our analyses, we compare this model to competing specifications of change in the gender gap: yˆij 5 βj 1 ^ T t51 βtjCit 1 ^ 1 g51 βgjDigZ2 i0. (3) Once we have chosen a preferred model, we add a series of explanatory variables to the model to determine whether they can explain away the magnitude of, and change in, the gender gap. Our explanatory analyses of the entire 1952–92 series provide evidence that a single causal process (increases in women’s labor force participation) is responsible for the changing voting behavior of women (and thus the emergence of the gender gap). Taking advantage of the more detailed information available in recent NES surveys, we then conduct a finer-grained analysis, examining the attitudinal mechanisms that mediate the effect of labor force participation. Taken together, these analyses provide us with a portrait of the causes of gender differences and trends in the voting behavior of women and men. ANALYSES The Evolution of the Gender Gap, 1952–92 In tables 2 and 3, we present fit statistics for competing models of the gender gap.17 Our first task is to compare models that include competing specifications of trends in the gender gap. Once we have chosen a preferred model, we then attempt, using the models in table 3, to explain away the coefficient representing trends in the gender gap. If the coeffi- cient shrinks to statistical insignificance when explanatory variables are added to the model, it indicates that the trend has been explained away. By estimating the trend coefficient when different combinations of the explanatory variables have been added to the model, we test which of the latter are responsible for the emergence of the gender gap. 16 Given that the preferred model does not include a term for the main effect of gender, it represents a highly parsimonious model of (change in) the gender gap. 17 Models are discretely numbered within each table. 1249TABLE 2 Logistic Regression Models of Change in the Gender Gap, 1952–92 22 Log Likelihood βgender*(year′) 2 Model (df ) (SE) 1. Election year ………………………………………………………………. 17,683.97 ⋅⋅⋅ (13,070) 2. Model 1 1 gender 3 year1980/84/88 …………………………………. 17,665.95 ⋅⋅⋅ (13,069) 3. Model 1 1 gender 3 year1980/84/88/92 ………………………………. 17,660.73 ⋅⋅⋅ (13,069) 4. Model 1 1 gender ∗ year ……………………………………………. 17,658.73 ⋅⋅⋅ (13,069) 5. Model 1 1 gender ∗ (year)2 ………………………………………… 17,659.16 ⋅⋅⋅ (13,069) 6. Model 1 1 gender ∗ (year′) 2 ……………………………………….. 17,656.82 .006† (13,069) (.0001) 7. Model 6 1 gender 3 year ………………………………………….. 17,642.82 ⋅⋅⋅ (13,060) Note.—Linearly constrained interaction effects are designated by an asterisk, unconstrained interaction effects by a multiplication cross. Dependent variable is coded “0” for the choice of the Republican and “1” for the choice of the Democratic presidential candidate; N 5 13,081. † P 5 .05 (two-tailed test). TABLE 3 Logistic Regression Models Explaining Change in the Gender Gap, 1952–92 22 Log Likelihood βgender*(year′) 2 Model (df ) (SE) 1. Base model: election year, gender ∗ (year′) 2 ………………… 15,370.90 .005† (11,398) (.001) 2. Model 1 1 class location, class identification, household income, cohort, marital status, education, race, and region ……………………………………………………………………………. 14,286.17 .005† (11,381) (.001) 3. Model 2 1 significant two-way interactions: gender 3 marital status, women 3 1950s cohort ……………………….. 14,275.20 .005† (11,379) (.001) 4. Model 1 1 gender 3 labor force participation ……………. 15,356.23 ,.002 (11,397) (.001) 5. Model 4 1 labor force participation …………………………… 15,356.22 .003 (11,396) (.002) 6. Model 4 2 gender ∗ (year′) 2 ……………………………………….. 15,359.48 ⋅⋅⋅ (11,398) Note.—Linearly constrained interaction effects are designated by an asterisk, unconstrained interaction effects by a multiplication cross. Dependent variable is coded “0” for the choice of the Republican and “1” for the choice of the Democratic presidential candidate; N 5 11,410. † P 5 .05 (two-tailed test).The Gender Gap In table 2, model 2 parameterizes the gender gap as a product of only the 1980, 1984, and 1988 elections. This model improves the fit of model 1 in table 2 (which includes only the main effects of election year). Model 3 in table 2 measures the gender gap as also including the 1992 election, hypothesizing that the gender gap emerged as a single step by women toward Democratic candidates beginning with the 1980 election. Model 3 has the same degrees of freedom as model 2 but yields a considerably smaller 22 log-likelihood statistic, indicating its superiority to model 2. The trend parameter of model 4 (table 2) implies a very different hypothesis, that the gender gap emerged over the course of the entire 1952– 92 period. This model also has the same number of parameters as model 3, although its 22 log-likelihood statistic is smaller by 2.00. Using Raftery’s (1995) BIC index for comparing nonnested models, the BIC test provides positive evidence (BIC improvement 5 22) favoring model 4. Model 5 (table 2) tests the hypothesis that the growth of the gender gap during the 1952–92 period is best captured by an exponential function. Its fit is virtually identical to the fit of model 4, making comparisons using the BIC test inconclusive. The specification of the gender gap in model 6 (table 2) uses a modification of the year2 term in model 5. In this model, the constant for the year variable is coded the same except for the past four elections (which receive the same score of “7”). This constraint in the year scores has the effect of flattening out the voting trend among women in recent (i.e., 1980–92) elections, while still allowing for a curvilinear increase in their likelihood of Democratic vote choice from 1952 through 1976. Model 6 results in a smaller 22 log-likelihood statistic than either model 5 (by 2.34), model 4 (by 1.91), or model 3 (by 3.91) in table 2; the BIC test thus provides positive evidence favoring model 6 over models 1–5.18 The comparison between models 6 and 7 (table 2) examines the evidence for any residual change in the gender gap (as captured by the unconstrained interaction of gender by time). The 22 log-likelihood test (as well as BIC) easily favors model 6, making it our preferred model.19 18 We evaluated the fit of a model that adds to model 6 an additional term for the main effect of gender. The resultant reduction (3.36) in 22 log likelihood (for 1 df ) favors model 6, implying that all gender differences in vote choice are already accounted for in model 6. 19 We caution that the positive but not decisive evidence for our preferred model leaves the testing of hypotheses about the gender gap partially open in the face of future evidence (which would be available when our current analyses are supplemented with the NES data from the 1996 election). However, the observed estimates of the gender gap using the VNS data (see fig. 1’s left-hand panel) are consistent with our preferred model’s specification of the trends and perhaps even more so with the alternative linear specification used by model 4 (given that the VNS data for 1996 suggests the largest gender gap to date). In the absence of these data, the main point to be appreci- 1251American Journal of Sociology The .006 coefficient from table 2 for the gender gap indicates that, between 1952 and 1980, women’s log-odds of Democratic vote choice has increased by .29. This growth is not large in absolute terms, but its signifi- cance should be gauged relative to changes experienced by all voters (as captured by the election year coefficients in the model).20 As shown in figure 1’s second panel, this small but steady shift among women has thus resulted in the emergence of the gender gap in presidential elections. The constraint in our preferred model’s specification of change in the gender gap implies that the magnitude of this gap was stable between 1980 and 1992. Taken together, these results direct our attention to the causal factors that can explain these cumulative changes in the voting behavior of women relative to men. Explaining the Gender Gap We evaluate competing explanations of change in the gender gap in table 3. The models in table 3 add to the preferred model (model 6 in table 2) various subsets of our independent variables.21 Model 2 (table 3) adds the main effects of class-related factors, birth cohort, marital status, education, race, and region. While its improvement in fit is significant, the coef- ficient for the trend in the gender gap is not affected by the addition of these variables (i.e., it is .005 [SE 5 .001] in both models 1 and 2 of table 3). This means that the additional sociodemographic factors introduced in model 2 do not explain the emergence of the gender gap. Model 3 (table 3) adds two significant interaction effects, for gender by marital status and gender by 1950s cohort. While this model improves over the fit of both the base model and model 2, the coefficient for the gender gap is again unchanged. In model 4 of table 3, we test a different causal hypothesis, that it is women’s greater entrance into the paid labor force that explains the gender gap. Model 4 adds to the base model the interaction of gender and labor force status. Not only does model 4 improve the fit of the base model, it results in the coefficient for the gender gap shrinking to statistical insignificance (, .002; SE 5 .001), providing strong evidence that the changing ated in the current study is that the evidence supports the inference that the gender gap stretches back to the 1950s. More specifically, model 4 and especially model 6 are both preferred over model 3 using the BIC index for comparing nonnested models. 20 In 1952, the probabilities of Democratic vote choice among women and men are predicted as both being equal to .42. In 1972, these probabilities are .37 and .34, and in 1980 .47 and .4. 21 The sample size for the analyses of table 3’s models is somewhat smaller than in table 2, reflecting the presence of missing values for the independent variables in table 3’s models. 1252The Gender Gap rate of labor force participation among women explains the growth of the gender gap. In model 5 of table 3, we test an additional feature of labor force participation by adding to model 4 an additional parameter for the main effect of labor force participation (i.e., the effect of labor force participation among men). Model 5 does not improve over the fit of model 4. Taken in tandem with the comparison of the base model and model 4, this result implies that labor force participation has an effect on vote choice only among women. As a final test, we compare models 4 and 6 in table 3. We derive model 6 from model 4 by constraining the coefficient for the gender gap to be equal to zero. The comparison of their respective fits favors model 6 (at P 5 .05), showing that the inclusion of the interaction between women and labor force participation appears to be sufficient to explain away women’s changing voting behavior over the 1952–92 period. As a result, the inclusion of a trend parameter for the gender gap is now unnecessary in model 6. This result suggests the importance of work-related factors to the emergence of the gender gap. Mediating Role of Policy Attitudes To this point, our analyses provide evidence that changing rates of labor force participation among women explain the emergence of the gender gap. We now examine the impact of attitudes toward gender roles and social services to evaluate whether and how these attitudes may mediate the effects of labor force participation. For these analyses, we take advantage of data from recent NES surveys that have asked questions about gender roles, social service spending, and feminist consciousness. We consider the possible roles of attitudes toward gender roles and social service spending using the models and coefficients presented in table 4. While the data is for the 1984 through 1992 elections, our strategy is similar to the analyses of the longer 1952–92 series: we compare models with explanatory factors (relating here to attitudes toward gender roles and social service spending), presenting for each model the coefficient for the main effect of gender during the 1984–92 period.22 If the coefficient for gender becomes insignificant, it indicates that the independent variables for the model in question have explained away the gender gap in these three elections. (Note that labor force participation is not considered in table 4’s models but is analyzed separately in table 5.) 22 Given the stability of the gender gap since 1980 according to our preferred model (see fig. 1), it is sufficient to measure the gender gap during the 1984–92 period as the main effect of gender (rather than a gender-by-time interaction, as would be appropriate when elections prior to 1980 are analyzed). 1253TABLE 4 Fit Statistics and Select Logistic Regression Coefficients for the Effect of Policy Attitudes on the Gender Gap, 1984–92 22 Log Likelihood βgender Model (df ) (SE) 1. Election year …………………………………………………………………… 3,491.29 ⋅⋅⋅ (2,578) 2. Model 1 1 gender main effect ………………………………………… 3,482.15 .243† (2,577) (.080) 3. Model 2 1 gender role attitudes ……………………………………… 3,425.15 .233† (2,576) (.081) 4. Model 3 1 gender role attitudes ∗ gender ……………………….. 3,425.14 .205 (2,575) (.226) 5. Model 2 1 views of social service spending ……………………. 3,094.13 .041 (2,576) (.087) 6. Model 5 2 gender main effect ………………………………………… 3,094.35 ⋅⋅⋅ (2,577) Note.—Linearly constrained interaction effects are designated by an asterisk. Dependent variable is coded “0” for the choice of the Republican and “1” for the choice of the Democratic presidential candidate (N 5 2,581). † P 5 .05 (two-tailed test). TABLE 5 OLS Regression Coefficients for the Effects of Gender and Labor Force Participation on Social Spending Attitudes, 1984–92 Gender Labor Force Gender 3 Labor Only Participation Force Participation β0 ……………………………………….. 2.683* 2.761* 2.923* (.056) (.074) (.094) βgender …………………………………… .396* .378* .126 (.059) (.060) (.108) βlabor force participant ……………………… ⋅⋅⋅ 2.101 2.318* (.064) (.100) βgender 3 labor force participant ……………… ⋅⋅⋅ ⋅⋅⋅ .364* (.129) Note.—SEs are presented in parentheses. Dependent variable is the seven-category Likert spending item (higher scores indicate a preference for increased service spending). N 5 2,581. All models also control for the main effects of election year (coefficients not presented in table). R2 5 .02 for all three models. † P 5 .05 (two-tailed test).The Gender Gap In table 4, model 2 readily improves the fit of model 1, corroborating our earlier result that women’s and men’s vote choices have differed significantly during the 1984, 1988, and 1992 elections.23 Using model 3 (table 4), we test the hypothesis that the proximate source of the gender gap is women’s and men’s attitudes toward gender roles. While model 3’s fit is superior to model 2, the coefficient is largely unchanged across models. We also find no support for an interaction between gender role attitudes and gender (model 3 is preferred to model 4). Gender role attitudes, while an important source of voting preferences in their own right, do not appear to explain the gender gap in vote choice between men and women.24 The key lies with men’s and women’s contrasting preferences for social services spending. Model 5 (table 4) adds to model 2 the main effect of attitudes toward social service spending. Not only does it readily improve the fit of the model, the coefficient has now shrunk to a nonsignificant .041 (SE 5 .087). In turn, the improvement of model 6 (table 4) over model 5 shows that the gender main effect can be deleted from the model once malefemale differences in spending preferences have been taken into account. In table 5, we develop a more comprehensive portrait of the respective roles of our two explanatory factors (labor force participation and social services spending preferences). The most plausible causal interpretation is that labor force participation shapes working women’s (more favorable) views of social service spending, which then serves as the proximate cause of their voting differences with men. To consider the evidence for this interpretation, we adopt a simplified path-analytic strategy, treating the spending item as our new dependent variable (we choose an ordinary least squares [OLS] specification for our analyses of this Likert item). By comparing the coefficients for three models presented in table 5, we test whether the source of gender differences in views of social service spending is to be found in the distinctive preferences of working women. The results presented in table 5 support this interpretation. The first column’s coefficient shows that women as a whole favor increased social service spending (relative to men’s preferences). The second column shows the results of adding the main effect of labor force participation to the model, and this coefficient is not significant, corroborating our earlier results. The third column’s model uses the correct “gendered” specification 23 The .243 (SE 5 .080) coefficient for gender represents the phenomenon to be explained in subsequent models. 24 The explanation for this is that men’s and women’s gender role attitudes are, on average, quite similar; their respective means on the gender role item during the three elections in question are 4.44 and 4.49. Differences between men and women are thus considerably less pronounced than differences among men and women with respect to gender role attitudes. 1255American Journal of Sociology TABLE 6 The Effect of Feminist Consciousness on the Gender Gap, 1980–92 22 Log Likelihood βgender Model (df ) (SE) 1980: Gender main effect ………………………………………………………….. 1,128.47 .306* (829) (.14) Gender main effect 1 feelings toward the women’s liberation movement ……………………………………………………………… 1,060.03 .336* (828) (.15) 1984: Gender main effect ………………………………………………………….. 1,574.42 .340* (1,174) (.12) Gender main effect 1 feelings toward the women’s liberation movement ……………………………………………………………… 1,442.90 .322* (1,173) (.13) 1988: Gender main effect ………………………………………………………….. 1,340.79 .278* (974) (.13) Gender main effect 1 feelings toward feminists ………………. 1,238.27 .276* (973) (.14) 1992: Gender main effect

 

"Is this question part of your assignment? We Can Help!"