Voter Roll-Off When Women Run for Congress

Michael G. Miller Assistant Professor Barnard College, Columbia University Department of Political Science

WORKING PAPER: Please do not cite without author permission.

Acknowledgments: This paper has previously been presented at the 2010 and 2014 annual meetings of the Midwest Political Science Association. I thank Gary C. Jacobson for sharing useful data. All mistakes and oversights rest with me alone. Abstract I advance a theoretical context in which abstention can affect the ordering of voter preferences when a woman is on the ballot. I analyze data from more than 300,000 election precincts over five presidential election cycles in 40 states, tracking trends in voter roll-off during elections when women run for the U.S. House. I find mixed evidence for a relationship between candidate gender and voter roll-off. Compared to male-only elections, roll-off is significantly lower when two women run against each other. When Democratic women run against Republican men, I find some evidence that roll-off trends lower; however, this trend is particularly strong for Democratic women in Republicanleaning areas. Similarly, when Republican women run against Democratic men, roll-off is lower than exclusively male elections, but only in Republican-friendly precincts. If anything, considering all precincts in the sample, I find that roll-off overall may trend slightly higher when Republican women run.

1

1

Introduction

There is no shortage of scholarship on reasons for the relative dearth of women in American legislatures. Much of that work has considered whether women’s underrepresentation has electoral roots, and has yielded important findings. For instance, women do seem to face stereotypes among voters (e.g., Dolan, 2014) and a different treatment from media (e.g., Falk 2010; Fowler and Lawless 2009). Yet, evidence of women facing a systematic disadvantage in terms of votes received has been scarce. Experimental assessments offer some reason to believe that voters are less likely to choose women over men in some conditions (Fox and Smith, 1998; Paul and Smith, 2008). However, observational studies of both survey responses (Dolan, 2014) and election returns (Darcy and Schramm, 1977; Smith and Fox, 2001) offer little reason to believe that women experience a pervasive gender-related impediment with voters. That is good news for women candidates, and it is of course worthwhile to consider how the presence of a woman on the ballot affects how voters choose between candidates. Yet, most examinations of gender-related voting behavior do not explicitly evaluate the potential that candidate gender can also affect a given voter’s decision to vote at all in a given contest. Particularly in studies examining two-party vote percentage or binary vote choice (i.e., choosing a Democrat over a Republican or a man over a woman), this issue could obscure gender-related voting behavior in an area that has not yet been fully examined: the decision of whether to abstain from voting in a given contest. Put another way, if we consider only how people voted, we miss an opportunity to evaluate whether women candidates receive either an advantage or disadvantage when voters who have already turned out decide to vote–or not–in their election. In this paper, I analyze data from more than 300,000 election precincts over five presidential election cycles in 40 states, tracking trends in voter roll-off during elections when women run for the U.S. House. I find mixed evidence for a relationship between candidate gender and voter roll-off. This said, in line with previous work (e.g., Dolan 2014), my find-

2

ings suggest that party plays a strong role in shaping the relationship between candidate gender and voting behavior. When Democratic women run against Republican men, I find some evidence that roll-off trends lower; however, this trend is particularly strong for Democratic women in Republican-leaning areas. Similarly, when Republican women run against Democratic men, roll-off is lower than exclusively male elections in Republican-friendly precincts. However, considering all precincts in the sample, I find that roll-off overall may trend slightly higher when Republican women run. Finally, when two women run against each other, I find evidence that roll-off is significantly lower. In extending analysis of gender and voter support to the realm of abstention, this paper therefore contributes to a growing literature on how voters view women.

2

Gender and Voter Abstention

2.1

The Decision to Abstain

Aggregate voter abstention in a given election contest is best operationalized as ballot rolloff, or the percentage of voters who turn out and then fail to cast a vote. Roll-off is an omnipresent feature of American elections: It is not uncommon for citizens to roll-off at rates of 20% or higher in low-information elections such as those for the judiciary (Hall, 2007), municipal office Bullock and Dunn, 1996, or initiative questions (e.g. Feig 2007; Vanderleeuw and Engstrom 1987). Yet, roll-off has been an understudied phenomenon compared to other areas of political behavior such as turnout or preference formation. In short, many more pages have been devoted to questions about why people vote than why they do not. Roll-off has received more attention in the last decade, however. The well-documented issues in the 2000 presidential election have fueled a growing body of work engaging the question of why a substantial percentage of voters expend the effort to travel to their polling place and then fail to vote in all contests on the ballot. A number of studies have shown that attentive administration (Stein et al., 2008; Ansolabehere and Stewart, 2005; Buchler,

3

Jarvis and McNulty, 2004; Tomz and Houweling, 2003; Sinclair and Alvarez, N.d.), effective ballot design (Kimball and Kropf, 2005; Reilly and Richey, 2011), or system warnings (Nichols and Strizek, 1995) can reduce residual votes. Thus, some part of what appears to be “abstention” is likely due to error on the part of voters or election administrators. That said, there is also considerable evidence suggesting that much roll-off is attributable to intentional voter abstentions. Survey data in Knack and Kropf (2003) suggest not only that a large proportion of undervoting in presidential elections is a purposeful act, but also that older, poorer voters are particularly likely to intentionally abstain. In a field experiment, Herrnson et al. (2006) found that while it is relatively common for voters to make errors on their ballot, under-voting does not usually result from voter mistakes. Furthermore, two studies of a unique election policy in Illinois–audible alerts informing voters that they may have forgotten to register a preference in a given election–found that it had practically no impact on aggregate roll-off in that state, which is consistent with voters choosing to skip those races (Miller, 2013; Miller, Tuma and Woods, 2015). These findings should not be particularly surprising given classic rational choice models of voting behavior that suggest voters should be more likely to abstain in elections that they view as unimportant (Downs, 1957; Riker and Ordeshook, 1968). In low-information settings when voters know little more than the names of the candidates, the odds that they will vote “incorrectly” increase substantially, which raises the costs of casting a vote (See: Lau and Redlawsk 2001). Wattenberg, McAllister and Salvanto (2000) suggest that voters may therefore hedge against voting incorrectly by abstaining when they lack sufficient knowledge to make an informed decision in a given contest. Thus, rather than seeking to reduce roll-off via changes to election policy, the best mechanism might be finding ways to raise the available information about and/or perceived salience of a given election (Bullock and Dunn, 1996). For instance, several previous studies have examined whether the racial context of an election can contribute to reduced roll-off. This work has yielded two important findings for the present study. First, roll-off among increasingly black voting populations decreases

4

when African-Americans hold prominent positions in local government, or when elections are expected to be competitive (see: Vanderleeuw and Liu 2002; Vanderleeuw and Sowers 2007). Moreover, referenda on Southern racial issues appear to influence roll-off of both blacks and whites: Feig (2007) found evidence that while the roll-off gap between blacks and whites did not shrink in the instance of a ballot initiative on interracial marriage in the Deep South, roll-off overall diminished. Second, studies or race and roll-off suggest that candidate characteristics also have the potential to influence voter abstention rates. For instance, black voters are less likely to roll-off when African-American candidates appear on the ballot (Harris and Zipp, 1999), especially when they run against white candidates (Vanderleeuw and Utter, 1993; Engstrom and Caridas, 1991). This effect appears to be particularly strong when black candidates make focused mobilization efforts (Vanderleeuw and Liu, 2002). Coupled with the evidence above that much roll-off is intentional, these results indicate that voters might be less likely to skip a contest on their ballot if they perceive it to be especially important, or if they share an identity trait with a candidate.

2.2

Competing Hypotheses

If co-racial identity can spur diminished roll-off, it is conceivable that candidate gender may also affect voters’ propensity to abstain from voting. Even basic personal information such as incumbency status, occupational background, or endorsements can be useful heuristics on which to base a decision (e.g., McDermott 2005; Lau and Redlawsk 2001; Popkin 1994). Gender is often inferred easily from the candidate’s name on the ballot, so even low-information voters should be able to glean the candidate’s gender in many cases. Indeed, even with no knowledge of the election at all, a voter will typically know two pieces of information about a given candidate: her party, and her gender. Voters very well may react to a candidate’s gender by voting for her (or her opponent). If this is the case, gender bias would presumably be detectable in the candidate’s share of the two-party vote: All else equal, if women attract support, they should win a larger

5

share of the vote, and vice versa. Indeed, previous work has engaged a number of related questions, including whether women garner more or fewer total votes (e.g., Brians 2005; Darcy and Schramm 1977) or support among poll respondents (e.g., Dolan 2014; Smith and Fox 2001). This work has yielded little evidence for a systemic gender disadvantage at the ballot box. For instance, once incumbency and other appropriate factors are held constant, women perform at-par with men in congressional general elections (Dolan, 2004), as well as those for state and municipal offices (Lawless and Fox, 2010). Women also do about as well as men in terms of vote percentage in congressional primary elections (Lawless and Pearson, 2008).1 However, much previous work has underscored that a voter’s party identification adds a wrinkle to the relationship between candidate gender and voting behavior (for a full exposition of this idea, see Dolan 2014). Party identification is a powerful determinant of voting behavior (e.g., Lewis-Beck et al. 2009), and even if a voter had strong feelings about supporting (or opposing) a woman, party loyalty may swamp this desire. Or, the opposite could be true: feelings about a candidate’s gender may be more important for a given voter than party. Much existing research (Brians, 2005; Darcy and Schramm, 1977; Smith and Fox, 2001; Dolan, 2014) has considered gender-driven voting behavior as a binary choice between either a man and a woman or a Republican and a Democrat. This work marks a valuable contribution toward understanding how voters react to women candidates; however, I argue that to fully understand the potential for candidate gender to affect voting behavior, the possibility of voter abstention should be included in the calculus. This shift has both theoretical and methodological implications. With regard to the first of these, the addition of an abstention choice accounts for the reality that voters face at the ballot box, at which they are not compelled to cast votes in any given contest. Thus, considering abstention as a possibility is a fuller representation of the actual voting decision. Methodologically, examining abstention is important because many of the most obvious means of evaluating support for women may not detect voters who react to the 1

Lawless and Pearson do find, however, that women face more primary competition, which means that they have to be better candidates to achieve the same result. I return to this point below.

6

presence of a woman candidate by opting out of (or into) voting in a given race on the ballot. For instance, an analysis of the two-party vote in an election relies only on input from people who voted, and does not account for the fact that not voting may itself be driven by the combination of the candidate’s gender and party. Likewise, asking survey respondents to choose between a Democrat and Republican–while either omitting a “no choice” category or omitting respondents who chose it–does not fully capture the range of voting decisions in an actual election. My theoretical framework therefore includes the following courses of action for a voter in a given single-member partisan election between two candidates: Casting a ballot for a candidate of the voter’s preferred party (Vp ), casting a ballot for a candidate of the opposite party, (Vo ), or abstaining in that particular contest (a). Since abstaining deprives both the voter’s preferred party and the opposite party of a vote, in a sterile ranking of preferences (or in one where gender preferences are relatively weak), all else equal partisan voters should always prefer both the candidate of their preferred party and abstention to casting a vote for the opposite-party candidate:2 Vp > a > V o 2.2.1

Women Candidates and Higher Abstention Propensity

Women candidates face stereotypes and double standards that could affect the preference ordering of voters’ choices–including the choice to abstain. For instance, the media have often held women to different standards, and their campaigns have been covered in a more negative tone (e.g., Fowler and Lawless 2009; Fox 1997; Kahn 1996). When women run for president, the media are less likely to describe them as strong candidates, and coverage of their campaigns contains less substance (Falk, 2010). These conditions have obvious implications on voting decisions; if media effects result in beliefs that women are less substantive candidates, voters may take their campaigns less seriously, and might therefore be more likely to abstain from voting. 2

The underlying assumptions for the logic in this section are exhibited in the Appendix.

7

Worse, media coverage may not even be necessary to trigger a gender-related abstention decision, as there is ample evidence that voters view candidates of both genders through a stereotyped lens. Indeed, voters are more likely to view women as skilled in “women’s issues” such as education, child care, or reproductive rights (Burrell, 2008; Herrnson, Lay and Stokes, 2003; Koch, 1999; Kahn, 1996; Plutzer and Zipp, 1996). These gender stereotypes might influence voting behavior as voters weigh their image of a candidate against their own preferences (Dolan, 2010, 2004; Lawless, 2004; Sanbonmatsu, 2002). For instance, a consequential percentage of voters appears to prefer “masculine qualities” in candidates (Dolan, 2004; Lawless, 2004; Rosenwasser and Dean, 1989). Furthermore, voters tend to view women candidates as more liberal than men, which may make them a less desirable choice for some voters Koch (2002). These conditions may fuel a reluctance to support women that in some cases could overcome a given voter’s partisan inclinations. During both the 2008 and 2016 Democratic presidential primaries–in which Hillary Clinton was a serious candidates–a number of polls consistently found that a significant minority of Americans believed felt that the country was not yet “ready” for a woman president.3 Moreover, as recently as 2015, 8% of Americans told Gallup pollsters that they would not vote for a “generally well-qualified” woman running for president, which was approximately equal to the percentage refusing to vote for AfricanAmerican, Hispanic, or Jewish candidates.4 Thus, candidate gender has clear potential to impact voting behavior–to the detriment of women candidates.

2.2.2

Women Candidates and Lower Abstention Propensity

There are also good reasons to expect that fewer voters will abstain when there is a woman on the ballot. For one, there is growing evidence that women might be better candidates than men on average. Previous work has described a widespread belief among the voting public that women who run for office face disadvantages relative to men (Dowling and 3

For instance, in the 2008 primary see: http://www.cbsnews.com/stories/2006/02/03/opinion/polls/main1281319.shtml In 2016: http://www.cnn.com/2016/03/04/politics/female-president-poll-cnn-orc/index.html 4 http://www.gallup.com/poll/4729/Presidency.aspx?g_source=woman%20qualified%20president&g_medium=search&g_camp

8

Miller, 2015). Unsurprisingly, would-be candidates appear to share this belief; prospective women candidates downgrade their own chances of winning, which contributes to lower political ambition among women–and lower rates of women entering politics to begin with (e.g. Fulton et al. 2006; Lawless and Fox 2010). Possibly because they believe they must be especially prepared if they are to run well, women appear to be particularly strategic candidates, entering politics when conditions are especially conducive to their success (Fulton et al., 2006). One drawback to strategic entry is that women draw more competition in the primary election than men (Lawless and Pearson, 2008). Nonetheless, women likely wait longer to run due to a recognition of the headwinds described above; in order to perform “as well” as men, women may recognize that they must possess more experience, connections, and skill as candidates. Indeed, there is evidence that women are higher-quality candidates on average (Fulton, 2012, 2014) and devote more personal effort to their campaigns when they run (Miller, 2015). A woman on the ballot, in other words, may simply be a proxy for a high-quality, hard-working candidate who is likely to mobilize more voters. Direct linkages between candidate gender and voting behavior have proven elusive in both observational (McDermott, 1997) and experimental (King and Matland, 2003; Thompson and Steckenrider, 1997) settings, though some experimental work suggests that voters may be less willing to support women (Fox and Smith, 1998; Paul and Smith, 2008). That said, there is more recent evidence that voters do incorporate candidate gender into their vote decision, with the important caveat that party appears to moderate candidate gender effects. For instance, Seltzer, Newman and Leighton (1997) find evidence of women voters being more likely to support women candidates, with Democratic women candidates receiving significantly more support. This is consistent with Brians (2005), which found that Republican-leaning women voters will “cross over” to support Democratic women challenging Republican men. Dolan (2008) also reports party-moderated effects; women feel significantly more positively toward Democratic women candidates than Democratic men, but there is no such

9

difference for Republicans. One explanation for these results is that both women in office (Kahn and Gordon, 1997) and women voters (Howell and Day, 2000) are more liberal than their male counterparts. Thus, if Republican women voters are closer to the political center than Democratic women voters, they should be more likely to choose gender over party. Or, as Dolan (2008) notes, “when faced with a female Republican, many women may experience cross-pressures,with gender considerations taking them in the opposite directions from the candidate’s political party.” Party appears to remain a more powerful determinant of voting behavior Dolan (2014). However, if gender truly can moderate the effect of party on voting decisions, women who run for office might be rewarded in certain conditions.

2.2.3

Expectations

It is analytically tempting to view a vote decision for the eight percent of voters who refuse to support a woman (“sexist voters”) as a binary choice: voting for a man is always preferable to voting for a woman (conditional on turnout). In this framework, women nominated by their least-preferred party should not impact sexist voters’ behavior at all, but they will defect to the opposite party when their own party nominates a woman. I posit that such an assumption represents a theoretical blind spot in the literature. In deciding whether to support a candidate of a given party who happens to be a man (m) or woman (w), a partisan voter with sexist motivations will discount a woman candidate by a factor of s, which takes a value between 0 and 1. If the discount is sufficiently large, the voter might consider abstention (a) as a better choice than supporting a woman candidate of his own party, because the voter’s preferences regarding support for the voter’s preferred (Vp ) and non-preferred (Vo ) parties are ordered as such:5 Vp |m > a > Vp s|w > Vo |m > Vo s|w Similarly, there may exist “affinity voters” with a relatively strong preference for supporting women candidates.6 That is, s>1. As s grows sufficiently large, the utility of casting 5

See Appendix for greater exhibition of this idea. To clarify, the “affinity” terminology above implies that most such voters are likely to be women, but it is certainly possible for a man (perhaps conscious of underrepresentation of women in legislatures) to 6

10

a vote for any woman–regardless of party–also becomes sufficiently large that preferences can be ordered as:7 Vp s|w > Vp |m > Vo s|w > a > Vo |m In short, there are clear reasons to expect candidate gender to affect voters’ propensity to abstain in a given election–and that these effects may differ by the candidate’s party. Yet, the effect of candidate gender on voter roll-off has not been fully evaluated in a partisan context. Streb, Frederick and LaFrance (2009) find that in partisan judicial elections, voters are less likely to roll-off when women are running; however, that study does not supply information about which party, if any, is systematically “advantaged.” In the context of a relatively high profile race, such as one for U.S. House, it seems less likely (at least compared to local or judicial elections) that voters will lack sufficient information to make an informed decision. Moreover, voters are likely to perceive an election to a federal office as more salient than other races that appear farther down the ballot. Thus, in an observational setting, congressional elections provide a promising theater in which to examine the effect of candidate gender on aggregate voter participation rates. In the following section, I describe the data and method I employ to evaluate these competing hypotheses.

3

Data and Method

Voter roll-off is a relatively straightforward concept: It is simply the percentage of voters who turned out but failed to vote in a given contest on the ballot. However, accurate calculation of voter abstention in legislative elections presents one important challenge: While it is usually straightforward to obtain vote counts for a given contest, it can be difficult to tabulate the number of ballots that were actually cast within a legislative district. One solution is to locate the number of votes cast in a presidential election tabulated by district, but even this approach leads to difficulty because most states maintain publicly accessible vote totals only at the county level, and it is not uncommon for legislative districts to cross disproportionately support women candidates. See: Dolan and Sanbonmatsu (2008). 7 See Appendix.

11

Figure 1: Count of Precincts With Usable Data in Present Election Only, and Both Present and Previous Election, by Year

county borders. While a number of almanacs maintain some measure of presidential vote percentage within a given congressional district, their figures often ignore vote receipts by third parties, or rely on geo-coding or some other estimation strategy.8 Considering the importance of accurate vote counts to the calculation of voter roll-off, I avoid secondary district-level sources, and instead examine voter abstention rates at the level of the election precinct.

I employ data from the Precinct-Level Election Data Project (“PLEDP,” Ansolabehere and Rodden 2012), which includes precinct returns in most states from as early as 1996 to 2012. I obtained files from 40 states that include both presidential and congressional vote totals from at least one election cycle during this time period, and retained data from 8 For example, since the mid-1970s, the National Journal’s Almanac of American Politics has provided useful data on each Congressional district and representative, including presidential vote totals. However, it is general practice for the almanac to include only major-party presidential vote totals. Calculations based on the almanac’s data therefore understate the number of presidential votes in a given state, which leads to roll-off appearing to be lower than it actually was.

12

precincts in which it is possible to vote in only one congressional election. The PLEDP tags each precinct for congressional district, which allowed for data about each congressional election, including expenditures, previous vote percentages, contestedness, and incumbency to be merged into the file.9 The finished dataset includes usable vote counts from more than 340,000 precinct tabulations over five elections; of these, more than 120,000 appear in multiple election cycles.10 A count of precincts with usable data in a given election, as well as in both a given election and the previous one, is displayed in Figure 1. I examine roll-off patterns using mixed-effects regression models of election precincts (level 1) within congressional districts (level 2), employing random intercepts for the latter.11 The aggregate congressional voter abstention rate, or “roll-off,” is the dependent variable, calculated as: H R = 100 − 100 ∗ C 



Where R is ballot roll-off, H is the number of votes cast for U.S. House in a given district, and C is the number of votes for president.12 Larger values of roll-off therefore reflect a higher percentage of voters who failed to cast ballots for U.S. House. As noted above, previous scholarship has found that candidates’ partisan affiliation can mediate voters’ propensity to support women candidates. As such, the predictors of interest are indicator variables coded 1 for the presence of Democratic, Republican, and/or two women candidates 9

I obtained most data on congressional elections from Gary C. Jacobson’s congressional election file. I included measures for Kernell’s (2009) measure of district balance from Bonica’s (2013) Database on Ideology, Money in Politics, and Elections. Finally, I coded the gender of each Democratic and Republican general-election congressional candidate included in the data. 10 Each state and year in the PLEDP is contained in a separate file. There is a wide range of practices for data standardization among the states. For instance, some employed standard precinct codes that made the process of identification straightforward, while others captured only the county and precinct name. Furthermore, in many cases the precinct name changed slightly from one cycle to the next, usually due to altered abbreviations (“Fire Dept” becomes “FD,” for instance). In these cases I cleaned each individual file, applying an algorithm to find and correct these changes. Given the dynamics of the data, more than half of the precincts come from the 2004 and 2008 elections. 11 District IDs account for redistricting. 12 While replacing P with the number of ballots cast in a given jurisdiction is a technically more accurate method of calculating roll-off, this information is not available in most files in the PLEDP, and I assume that most voters who do cast ballots do so with the intention of voting for president. Defining roll-off in this way therefore facilitates a relatively uniform calculus in a much larger number of states.

13

in a given U.S. House race. Thus, if voters are more (less) likely to abstain when these conditions occur, the models should return positive (negative) effects. I specify two models to test whether there is a significant association between the presence of a woman and voter roll-off in congressional elections. The first model utilizes all usable cross-sectional data that resulted from cleaning the PLEDP files.13 Since a number of other factors beyond candidate gender are also likely to affect roll-off, I include a number of both district- and precinct-level fixed effects. At the district level, the models hold constant Kernell’s (2009) measure of partisan preference, the presence of Democratic or Republican incumbents, (logged) total campaign expenditures, and an indicator for whether the election was contested by both major parties. The precinct-level fixed effects include the Democratic candidate’s vote share and (logged) total votes for president.14 I also include the abstention rate in the previous presidential election. However, because this information is available for fewer than half of the precincts, I include two model specifications–one including lagged roll-off, and one not. Finally, to test for the possibility that party preferences will moderate any effect on abstention of the presence of a woman candidate of a particular party, I include specifications including interaction terms between the candidate gender indicators and precinct-level Democratic presidential vote share. The second model leverages variation in precinct-level abstention through time to calculate the same-precinct difference-in-differences in roll-off as the presence of women on the ballot changes. In other words, the model seeks to determine whether the same voters abstained at different rates when faced with a woman candidate vs. a man. The dependent variable in precinct p in this model is a first-differenced measure accounting for trends 13

In an effort to reduce the potential for recording error to affect results, I exclude from analysis all precincts reporting negative roll-off, as well as those for which roll-off was more than 58.24%, or more than three standard deviations above the mean. 14 These variables capture partisan preferences and rough precinct size. I assume that most people vote for president before voting for House. The inclusion of Democratic presidential vote share may be problematic in many models of voting behavior, such as turnout, because the decision to turn out would have been made prior to the vote for president being cast. However, since here roll-off is the dependent variable, and the vast majority of voters vote for president before considering the race for Congress, in these models I proceed with measures of same-year presidential support as a proxy for the partisan balance of the district. It is important to control for precinct size because a single abstaining voter in a ten-voter precinct causes a swing of ten percentage points in roll-off, while the same abstainer in a 100-voter precinct would cause only a one-point shift.

14

between time t − 1and time t, calculated as: ∆R = Rpt − Rpt−1 . The drawback of this approach is that repeated measures are available only for a subset of the data. However, differencing eliminates time-invariant precinct characteristics as potential confounders; leaving the necessity of controlling only for factors of the election, district, or precinct that were different at times t − 1 and t. The most obvious of these is the presence of women. I therefore differenced the number of Democratic and Republican women in the same fashion. The resulting quantities take the values of -1 (a woman was present at time t − 1 but not t, 0 (a woman was either present or absent in both elections), or 1 (a woman was present at time t but not t − 1). Given the structure of the differenced dependent and independent variables, positive model coefficients would indicate that roll-off is higher when a woman is on the ballot. The same is true of the other differenced variables I include as controls in the second model, including district level measures of changes in the number of Democratic and Republican incumbents, two-party contestedness, and open seat status. At the precinct level, I include the difference in Democratic vote share. I also include the percentage change in the number of voters, the logged number of votes cast for president, and their interaction.

4 4.1

Findings Summary Statistics

I begin by examining mean levels of roll-off in congressional elections across four conditions: No women running, a Democratic woman running against a Republican man, a Republican woman running against a Democratic man, and two women running against each other. Figure 2 depicts point estimates and 95% confidence intervals for these four scenarios, pooled across the five elections in the sample, and weighted by the number of voters who cast ballots in a given precinct. Figure 2 shows mean precinct-level roll-off in congressional elections ranging between 4.75 and 6.25 percent. Moreover, there is evidence that candidate gender

15

might affect abstention rates. Compared to the baseline condition of two men running against each other, Figure 2 yields evidence of significantly lower roll-off rates when women are on the ballot. Abstention rates in races where a Democratic woman ran against a Republican man are comparable to those in which two women ran against each other; in both cases, roll-off is about one point lower than all-male elections. Roll-off trended lower still in elections where a Republican woman ran against a Democratic man; abstention rates in those elections was nearly two points lower than the man vs. man baseline, and about half a point lower than that for races in which Democratic women were present. Thus, Figure 2 offers preliminary support for the notion that the presence of a woman on the ballot may encourage a small percentage of voters to register a preference in a congressional election when they might not have otherwise done so.

Figure 2: 95% Confidence Intervals, Precinct-Level Abstention When Women Run for Congress

16

4.2

Model Results

I now turn to the results of the mixed-effects (precincts within districts) models, which examine congressional roll-off while employing random district intercepts and both precinct and district-level fixed-effects. Full results from these models are contained in Table 1. I first consider Models 1 and 3 in Table 1, which include indicators for elections in which a either a Democratic or Republican woman ran, as well as an indicator for elections in which two major-party women candidates ran against each other (two-man elections are therefore the reference category). Thus, the coefficients for these variables can be interpreted as the effect on congressional roll-off of women running in these conditions, compared to elections in which only men ran. Model 1 utilizes data from all precincts, while Model 3 employs only the subset of the precincts in which roll-off from the previous presidential election was retrievable. Thus, Model 1 has the advantage of using more data, but Model 3 controls for each precinct’s previous propensity for aggregate roll-off. Coefficients from the gender indicators derived from these models are depicted in Figure 3. Three trends are notable. First, contrary to the pooled estimates discussed above, both models return statistically significant positive effects for Republican women. Relative to elections in which two men ran, those in which a Republican woman ran appear to have abstention rates about one point higher. Second, the point estimate for Democrats is negative in both models, but the coefficient in Model 1 is effectively zero. However, in the model controlling for lagged abstention rate, the coefficient for Democratic women is statistically significant and suggests that the presence of Democratic woman is associated with a reduction in roll-off of about 1.25 percentage points. Third, both models return an negative effect on abstention of about 1.5 points when two women run, relative to when two men run. In total, these findings suggest that the partisan context of an election is likely an important factor when it comes to the relationship between candidate gender and voter abstention.

17

Table 1: Mixed-Effects Regression Coefficients and Standard Errors: Determinants of Voter Abstention in Congressional Elections (1)

(2)

(3)

(4)

Fixed Effects, District-Level: Democratic Woman

Democratic Woman X Dem. Vote Perc.

-0.04

-3.35*

-1.22*

-2.67*

(0.05)

(0.13)

(0.10)

(0.32)

-

0.07*

-

0.03*

(0.00) Republican Woman

Republican Woman X Dem. Vote Perc.

(0.01)

1.13*

-0.12

0.68*

(0.09)

(0.26)

(0.14)

(0.44)

-

0.03*

-

-0.07*

(0.00) Two Women

4.77*

(0.01)

-1.37*

-5.65*

-1.49*

-2.63

(0.14)

(0.46)

(0.28)

(1.39)

-

0.05*

-

0.03*

-0.01*

-0.03*

(0.00)

(0.00)

(0.00)

(0.00)

-0.22

-0.20

-0.92*

-0.66*

(0.23)

(0.23)

(0.27)

(0.27)

-1.82*

-1.83*

-3.09*

-3.10*

(0.07)

(0.07)

(0.13)

(0.13)

-0.71*

-0.80*

-2.12*

-2.30*

(0.06)

(0.06)

(0.09)

(0.10)

(Logged) Total Campaign Expenditures

-0.89*

-0.82*

-2.29*

-2.31*

(0.03)

(0.03)

(0.05)

(0.05)

Uncontested Election

12.52*

12.48*

15.22*

15.36*

(0.05)

(0.05)

(0.12)

(0.12)

-0.14*

-0.14*

-0.17*

-0.17*

(0.01)

(0.01)

(0.02)

(0.02)

-

-

0.13*

0.13*

(0.00)

(0.00)

Two Women X Dem. Vote Perc.

(0.01) Dem. Vote Perc.

District Partisanship

Democratic Incumbent

Republican Incumbent

(0.02) 0.02*

0.02*

Fixed Effects, Precinct-Level: (Logged) Total Votes for President

Lagged Abstention Rate

Constant

16.12*

16.17*

37.31*

36.82*

(0.52)

(0.52)

(0.80)

(0.81)

30.66*

31.33*

25.12*

25.70*

(1.94)

(1.99)

(1.87)

(1.91)

Number of Precincts

342,419

342,419

129,323

129,323

Number of Districts

604

604

394

394

64,108

65,636

26,933

27,070

Random Effects: Constant (District)

Wald Chi Square

* p<0.05. Standard errors in parentheses. 18Models also include year fixed effects (not shown).

Models 2 and 4 explore this possibility in greater depth. Both models feature interactions between the gender indicators and the Democratic candidate’s share of the two-party vote in the election in which roll-off was observed. As above, Model 2 employs all available data, while Model 4 includes a control for lagged precinct roll-off, which reduces the data from which the model is constructed. All three interaction effects are statistically significant in both models, suggesting that the partisan preference of a difference does moderate the relationship between candidate gender and roll-off. However, the substantive import of significant interactions is often difficult to grasp from a table alone. As such, Figure 4 depicts the predicted abstention rate for elections featuring Democratic, Republican, and two women running across a range of Democratic vote percentage, derived from Table 1, Model 2.15

15

Predictions are calculated from the fixed part of the model.

19

Figure 3: Predicted Effects on Voter Abstention of Women Running for Congress

20

Figure 4: Predicted Abstention Rate, by District Partisanship

21

Indeed, Figure 4 tells a nuanced story of the relationship between candidate gender and voter abstention. When two major-party women run against each other (bottom pane), predicted roll-off in highly Republican precincts (i.e., where Democrats won 20% of the two-party vote) is approximately the same as in a race where two men run against each other. However, as the precinct’s Democratic vote share increases, predicted roll-off in the women-only race decreases sharply–roughly three percentage points as the Democratic vote share rises to 80%–while the predicted roll-off in male-only races diminishes by less than one point. This trend suggests that the presence of two women can decrease roll-off, especially in Democratic-leaning areas. When Democratic women run against Republican men (top pane) the model predicts that roll-off will be about the same as in male-only races when the Democratic vote share is about 60%. The presence of a Democratic woman appears to decrease roll-off in precincts where the Republican party is stronger, however. Indeed, the roll-off gap is over two points until about the point at which Democrats win 40% of the vote. While these data cannot support conclusions about individual decisions, this pattern is consistent with previous work that has reported a higher propensity of Republican women to “cross-over” to support Democratic women (Brians, 2005). Finally, the middle pane of Figure 4 depicts predicted roll-off when Republican women run against Democratic men. Once again, the model predicts that roll-off is lower when Republican women run in Republican-leaning areas. However, the gap shrinks to zero in precincts where the Democratic candidate won 50% of the vote. Notably, this trend would be observed if Republican women candidates benefited from an increasing voting propensity by Republican, but not Democratic, voters. Once again then, Figure 4 offers evidence of party-moderated gender effects on roll-off, with voters in Republican-leaning precincts apparently most likely to be activated by the presence of either a Republican or Democratic woman running against a man of the opposite party.

22

Figure 5: Predicted Changes in Same-Precinct Voter Abstention

4.3

Difference in Differences

I next consider how same-precinct trends in roll-off are affected when women become–or cease to be–available for selection in congressional elections. Models 5 and 6 (Appendix, Table 2) regress the difference in roll-off (the lagged roll-off rate subtracted from the rolloff rate) on the difference in the presence of Democratic and Republican women on the ballot in a given precinct. If a woman running causes a decrease in roll-off, the coefficients for the gender differences should be negative. The coefficients for the gender difference variables from Model 5 are depicted in Figure 5. As was the case in the previous set of models (of non-differenced roll-off) described above, the coefficient for Democratic women is negative, while the one for Republican women is positive. However, in neither case do the coefficients achieve statistical significance, and in both cases their substantive size is

23

small–less than one percentage point. The interaction effects in Model 6 (Appendix Table 2) likewise fail to achieve statistical significance, suggesting that the partisan preference of a given precinct does not moderate the relationship between candidate gender and aggregate voter abstention.

Conclusion Large-sample quantitative studies employing observational election data are not capable of determining the preference ordering of any single voter, and that is certainly the case here. However, an examination of aggregated voter decisions–observed via the roll-off percentage– does allow us to look for evidence that the presence of a woman in a given contest affects the number of people who decide to vote in it. To the best of my knowledge, this paper marks the first analysis of the relationship between candidate gender and aggregate propensity for voter abstention. The models I present above offer mixed findings with respect to the relationship between candidate gender and voter roll-off in congressional elections. That said, there are at least two results worth reiterating. First, I find consistent evidence that roll-off is significantly lower when two women run against each other. Moreover, the effect on roll-off of two women running is substantively large, at about 1.5 percentage points. Mean precinct roll-off in the sample was 6.53%. Thus, two women running appears to reduce roll-off by about 15% Second, my results underscore findings from previous work which has found party affiliation to be an important factor shaping the relationship between candidate gender and voting behavior. Here, my paper offers a general pattern suggesting that Democratic women–but not Republican women–may benefit in the form of higher rates of voter participation. Point estimates for Democratic women are consistently negative in my models, even if they do not always achieve statistical significance. The opposite is true for Republicans. Third, the presence of a woman of either party on the congressional ballot appears to have the greatest implications for reduced roll-off in precincts where Republicans win

24

progressively more of the vote. Indeed, in precincts where Republicans won 40% of the vote, the effect on roll-off was about one point when Republican women ran, and nearly two points when Democratic women ran. It is worth noting that these findings are consistent with previous work that has found Democratic women to be the most likely beneficiaries of cross-over voting (Brians, 2005; Dolan, 2008; Seltzer, Newman and Leighton, 1997). There are limitations of this study beyond the above-mentioned ecological inference problem. Since running for office is a choice influenced by personal as well as environmental factors, women candidates almost certainly do not appear randomly. It is therefore prudent to consider the degree to which other observed covariates might be related to whether a district saw a woman candidate. For instance, it seems plausible that women candidates are more likely to emerge in districts where they are more likely to perform better, such as those with higher household incomes and education levels, or with less socially conservative leanings. Indeed, recent research confirms that districts matter to the success of women running for Congress (Palmer and Simon 2008). In other words, it is likely that the covariate distributions between the two district groups will display suboptimal balance. Future versions of this paper will employ techniques to deal with this problem. In advancing the study of candidate gender and voting behavior to the realm of voter abstention, this project contributes to a developing literature. I believe that there are two clear directions for future research into gender-driven voter abstention. First, a controlled experiment gauging individual voting decisions offers great promise not only to overcome concerns about ecological inference that could stem from an aggregated observational approach, but also to gain better traction on the internal validity of the relationship I report above. Second, a study of primary elections holds great promise to better understand why there is a gender-related pattern in voter roll-off. Specifically, while in this paper I assume that roll-off is lower in Republican-leaning areas because of party-moderated affinity effects reported elsewhere (e.g., Dolan 2008), that assumption is worth re-considering. It could also be true that such candidates raise the salience of the election for voters who desire to cast votes against Democratic women as well. While, as noted, existing scholarship would

25

seem to discount this possibility, it exists nonetheless. Accordingly, future studies should go further to untangle the intermingled effects of gender and party on roll-off and election outcomes.

26

References Ansolabehere, Stephen and Charles Stewart Stewart. 2005. “Residual votes attributable to technology.” Journal of Politics 67(2):365–389. Brians, Craig Leonard. 2005. “Women for women? Gender and party bias in voting for female candidates.” American Politics Research 33(3):357–375. Buchler, Justin, Matthew Jarvis and John E McNulty. 2004. “Punch card technology and the racial gap in residual votes.” Perspectives on Politics 2(03):517–524. Bullock, Charles S and Richard E Dunn. 1996. “Election roll-off: A test of three explanations.” Urban Affairs Review 32(1):71–86. Burrell, Barbara. 2008. “Political parties, fund-raising, and sex.” Legislative women: Getting elected, getting ahead. Boulder, CO: Lynne Rienner . Darcy, Robert and Sarah Slavin Schramm. 1977. “When women run against men.” Public Opinion Quarterly 41(1):1–12. Dolan, Kathleen. 2008. “Is there a gender affinity effect in American politics? Information, affect, and candidate sex in US House elections.” Political Research Quarterly 61(1):79–89. Dolan, Kathleen. 2010. “The impact of gender stereotyped evaluations on support for women candidates.” Political Behavior 32(1):69–88. Dolan, Kathleen. 2014. When Does Gender Matter?: Women Candidates and Gender Stereotypes in American Elections. Oxford University Press, USA. Dolan, Kathleen A. 2004. Voting for women: How the public evaluates women candidates. Westview Pr. Dolan, Kathleen and Kira Sanbonmatsu. 2008. “Gender stereotypes and attitudes toward gender balance in government.” American Politics Research . Dowling, Conor M and Michael G Miller. 2015. “Can Information Alter Perceptions about Women’s Chances of Winning Office? Evidence from a Panel Study.” Politics & Gender 11(01):55–88. Downs, Anthony. 1957. “An economic theory of political action in a democracy.” Journal of Political Economy 65(2):135–150. Engstrom, Richard L and Victoria M Caridas. 1991. “Voting for judges: Race and roll-off in judicial elections.” Political participation and American democracy 92:171–191. 27

Falk, Erika. 2010. Women for president: Media bias in nine campaigns. University of Illinois Press. Feig, Douglas G. 2007. “Race, Roll-Off, and the Straight-Ticket Option.” Politics & Policy 35(3):548–568. Fowler, Linda L and Jennifer L Lawless. 2009. “Looking for sex in all the wrong places: Press coverage and the electoral fortunes of gubernatorial candidates.” Perspectives on Politics 7(03):519–536. Fox, Richard L and Eric Smith. 1998. “The Role of Candidate Sex in Voter Decision-Making.” Political Psychology 19(2):405–419. Fox, Richard Logan. 1997. Gender dynamics in congressional elections. Vol. 2 Sage. Fulton, Sarah A. 2012. “Running backwards and in high heels: The gendered quality gap and incumbent electoral success.” Political Research Quarterly 65(2):303–314. Fulton, Sarah A. 2014. “When gender matters: macro-dynamics and micro-mechanisms.” Political Behavior 36(3):605–630. Fulton, Sarah A, Cherie D Maestas, L Sandy Maisel and Walter J Stone. 2006. “The sense of a woman: Gender, ambition, and the decision to run for congress.” Political Research Quarterly 59(2):235–248. Hall, Melinda Gann. 2007. “Voting in state supreme court elections: Competition and context as democratic incentives.” Journal of Politics 69(4):1147–1159. Harris, Jamie M and John F Zipp. 1999. “Black candidates, roll-off, and the black vote.” Urban Affairs Review 34(3):489–498. Herrnson, Paul S, J Celeste Lay and Atiya Kai Stokes. 2003. “Women running as women: Candidate gender, campaign issues, and voter-targeting strategies.” The Journal of Politics 65(1):244–255. Herrnson, Paul S, R Niemi, M Hanmer, B Bederson, Fred Conrad and Michael Traugott. 2006. The not so simple act of voting: an examination of voter errors with electronic voting. In Annual meeting of the Southern Political Science Association, Washington, DC. Howell, Susan E and Christine L Day. 2000. “Complexities of the gender gap.” Journal of Politics 62(3):858–874. Kahn, Kim Fridkin. 1996. The political consequences of being a woman: How stereotypes influence the conduct and consequences of political campaigns. Columbia University Press.

28

Kahn, Kim Fridkin and Ann Gordon. 1997. “How women campaign for the US Senate: Substance and strategy.” Women, media, and politics pp. 59–76. Kimball, David C and Martha Kropf. 2005. “Ballot design and unrecorded votes on paper-based ballots.” Public opinion quarterly 69(4):508–529. King, David C and Richard E Matland. 2003. “Sex and the grand old party: An experimental investigation of the effect of candidate sex on support for a Republican candidate.” American Politics Research 31(6):595–612. Knack, Stephen and Martha Kropf. 2003. “Roll-Off at the Top of the Ballot: International Undervoting in American Presidential Elections.” Politics & Policy 31(4):575–594. Koch, Jeffrey W. 1999. “Candidate gender and assessments of senate candidates.” Social Science Quarterly pp. 84–96. Koch, Jeffrey W. 2002. “Gender stereotypes and citizens’ impressions of House candidates’ ideological orientations.” American Journal of Political Science pp. 453–462. Lau, Richard R and David P Redlawsk. 2001. “Advantages and disadvantages of cognitive heuristics in political decision making.” American Journal of Political Science pp. 951–971. Lawless, Jennifer L. 2004. “Women, war, and winning elections: Gender stereotyping in the post-September 11th era.” Political Research Quarterly 57(3):479–490. Lawless, Jennifer L and Kathryn Pearson. 2008. “The primary reason for women’s underrepresentation? Reevaluating the conventional wisdom.” The Journal of Politics 70(1):67–82. Lawless, Jennifer L and Richard L Fox. 2010. It still takes a candidate: Why women don’t run for office. Cambridge University Press. Lewis-Beck, Michael S, William G Jacoby, Helmut Norpoth and Herbert F Weisberg. 2009. The American voter revisited. University of Michigan Press. McDermott, Monika L. 1997. “Voting cues in low-information elections: Candidate gender as a social information variable in contemporary United States elections.” American Journal of Political Science pp. 270–283. McDermott, Monika L. 2005. “Candidate occupations and voter information shortcuts.” Journal of Politics 67(1):201–219. Miller, Michael G. 2013. “Do Audible Alerts Reduce Undervotes? Evidence from Illinois.” Election Law Journal 12(2):162–178. Miller, Michael G. 2015. “Going all-in: gender and campaign commitment.” Research & Politics 2(3):2053168015605105. 29

Miller, Michael G, Michelle D Tuma and Logan Woods. 2015. “Revisiting Roll-Off in Alerted Optical Scan Precincts: Evidence from Illinois General Elections.” Election Law Journal 14(4):382–391. Nichols, Stephen M and Gregory A Strizek. 1995. “Electronic voting machines and ballot roll-off.” American Politics Quarterly 23(3):300–318. Paul, David and Jessi L Smith. 2008. “Subtle sexism? Examining vote preferences when women run against men for the presidency.” Journal of Women, Politics & Policy 29(4):451–476. Plutzer, Eric and John F Zipp. 1996. “Identity politics, partisanship, and voting for women candidates.” Public Opinion Quarterly 60(1):30–57. Popkin, Samuel L. 1994. The reasoning voter: Communication and persuasion in presidential campaigns. University of Chicago Press. Reilly, Shauna and Sean Richey. 2011. “Ballot question readability and roll-off: The impact of language complexity.” Political Research Quarterly 64(1):59–67. Riker, William H and Peter C Ordeshook. 1968. “A Theory of the Calculus of Voting.” American political science review 62(01):25–42. Rosenwasser, Shirley Miller and Norma G Dean. 1989. “Gender role and political office: Effects of perceived masculinity/femininity of candidate and political office.” Psychology of Women Quarterly 13(1):77–85. Sanbonmatsu, Kira. 2002. “Gender stereotypes and vote choice.” American Journal of Political Science pp. 20–34. Seltzer, Richard, Jody Newman and Melissa Voorhees Leighton. 1997. Sex as a political variable: Women as candidates and voters in US elections. Lynne Rienner Publishers. Sinclair, Betsy and R. Michael Alvarez. N.d. “Who Overvotes, Who Undervotes, Using Punchcards? Evidence from Los Angeles County.” Political Research Quarterly. Forthcoming. Smith, Eric RAN and Richard L Fox. 2001. “The electoral fortunes of women candidates for Congress.” Political Research Quarterly 54(1):205–221. Stein, Robert M, Greg Vonnahme, Michael Byrne and Daniel Wallach. 2008. “Voting technology, election administration, and voter performance.” Election Law Journal 7(2):123–135. Streb, Matthew J, Brian Frederick and Casey LaFrance. 2009. “Voter rolloff in a low-information context: Evidence from intermediate appellate court elections.” American Politics Research 37(4):644–669.

30

Thompson, Seth and Janie Steckenrider. 1997. “The relative irrelevance of candidate sex.” Women & Politics 17(4):71–92. Tomz, Michael and Robert P Van Houweling. 2003. “How does voting equipment affect the racial gap in voided ballots?” American Journal of Political Science 47(1):46–60. Vanderleeuw, James M and Baodong Liu. 2002. “Political empowerment, mobilization, and black voter roll-off.” Urban Affairs Review 37(3):380–396. Vanderleeuw, James M and Glenn H Utter. 1993. “Voter roll-off and the electoral context: A test of two theses.” Social Science Quarterly pp. 664–673. Vanderleeuw, James M and Richard L Engstrom. 1987. “Race, referendums, and roll-off.” The Journal of Politics 49(4):1081–1092. Vanderleeuw, James M and Thomas E Sowers. 2007. “Race, Roll-Off, and Racial Transition: The Influence of Political Change on Racial Group Voter Roll-Off in Urban Elections.” Social Science Quarterly 88(4):937–952. Wattenberg, Martin P, Ian McAllister and Anthony Salvanto. 2000. “How Voting is Like Taking an Sat Test: An Analysis of American Voter Rolloff.” American Politics Quarterly 28(2):234–250.

31

Appendix Exposition of Theory The aim here is not to prove the conditions under which abstention becomes more or less likely relative to partisan voting, but rather, that it is reasonably possible. Assume that a voter derives utility from voting (V ). Assume the utility derived from casting a vote for a candidate of her preferred party (Vp ) is 2.7 times greater than the utility from casting a vote for a candidate of her non-preferred party (Vo ).16 Let Vp = 10. Then, Vo = 3.7. Assume that the utility derived from abstention (a) is equidistant toVp and Vo : a = 6.85. Assume now that Vp and Vo are multiplied by a factor of s when the candidate is a woman (w) compared to when a candidate is a man (m). A voter has no gender preference when s = 1. A voter prefers voting for men over women when 0 ≤ s < 1. A voter prefers voting for women over men when s > 1. Thus, a voter prefers abstention to voting for a woman candidate of her own party when (Vp )w < a, or w < .685, and a voter prefers voting for a women candidate of her nonpreferred party when (Vo )w > a, or w > 1.84. Considering the utility of party voting relative to abstention, When w < .685 : Vp |m > a > Vp s|w > Vo |m > Vo s|w When w < .685 < 1.84 : Vp > a > V o When w > 1.84 : Vp s|w > Vp |m > Vo s|w > a > Vo |m

16

This assumption is derived from the 2012 National Election Study. Democrats viewed their own party 2.8 times more favorably than the Republican Party. Republicans viewed their party 2.6 times more favorably than the Democratic Party. Including party leaners does not change the figure for Democrats, but Republicans (including leaners) viewed the Republican Party 2.3 times more favorably than the Democratic Party, on average.

32

Table 2: Mixed-Effects Regression Coefficients and Standard Errors: Determinants of Changes in Voter Abstention in Congressional Elections (5)

(6)

Fixed Effects, District-Level: ∆ Democratic Women ∆ Dem. Women X ∆ Dem. Pres. Vote Share

-0.24

0.10

(0.69)

(0.74)

-

-0.05 (0.04)

∆ Republican Women ∆ Rep. Women X ∆ Dem. Pres. Vote Share

0.64

0.34

(1.18)

(1.22)

-

0.04 (0.06)

∆ Dem. Incumbent ∆ Rep. Incumbent ∆ Open Seat Status ∆ Contestedness Perc. Change in Total Campaign Expenditures

-0.65

-0.94

(3.03)

(3.04)

0.71

0.44

(2.93)

(2.95)

2.49

2.05

(3.01)

(3.03)

13.33*

13.47*

(0.93)

(0.93)

-0.26

-0.25

(0.14)

(0.14)

Fixed Effects, Precinct-Level: ∆ Dem. Pres. Vote Share Perc. Change in Number of Voters (Logged) Number of Votes Cast for President Perc. Change in Number of Voters X Log Pres. Votes Constant

-0.03

-0.03

(0.02)

(0.02)

0.09*

0.09*

(0.02)

(0.02)

0.17*

0.17*

(0.03)

(0.03)

-0.01*

-0.01*

(0.00)

(0.00)

-1.78

-1.82

(1.34)

(1.36)

32.37*

32.17*

(3.02)

(3.00)

Number of Precincts

120,293

120,293

Number of Districts

247

247

384.6

388.0

Random Effects: Constant (District)

Wald Chi Square

* p<0.05. Standard errors in parentheses. Models also include year fixed effects.

33