Gender Bias and Selection Bias in House Elections

Public Choice105: 41–59, 2000.© 2000Kluwer Academic Publishers. Printed in the Netherlands.

41

Gender bias and selection bias in House elections

JEFFREY MILYO & SAMANTHA SCHOSBERG∗Department of Economics, Tufts University, Medford, MA 02155, U.S.A.

Accepted 10 March 1999

Abstract. We demonstrate that female incumbents are of higher average candidate qualitythan male incumbents. This quality difference is the result of barriers to entry faced by poten-tial female candidates, although the observed effects of this quality differential on vote shareare partially masked by the fact that female incumbents are also more likely to be opposed or tobe opposed by high quality challengers. Using data from House elections for 1984–1992, weestimate that the gender-based differential in candidate quality yields an extra six percentagepoints of vote share for female incumbents.

1. Introduction

Contrary to popular opinion, recent research has shown that female can-didates suffer no electoral or fund-raising disadvantage compared to malecandidates (Burrell, 1985, 1994; Darcy, Clark and Welch, 1994; Gaddie andBullock, 1995; Werner, 1997). A gender related fund-raising disadvantagemay have existed in Congressional elections prior to the mid-eighties, butthis seems to have disappeared or even reversed in recent years (Flammers,1997). Since 1984, female candidates for the House have been slightly moresuccessful than males at raising funds and at least as successful in raisingPAC contributions, large contributions and even early contributions (Burrell,1994). At the state level, female candidates also raise more money than theirmale counterparts (Werner, 1997). Consequently, the current dearth of femaleoffice holders is thought to be primarily the result of prior barriers to womenentering politics, the effects of which are still realized today because of thegeneric incumbency advantage (Uhlaner and Schlozman, 1986).1 Similarly,the recent successes of female candidates for Congress are seen as the nat-ural consequence of turnover: older and more male-dominated cohorts arebeing replaced by younger and less male-dominated cohorts. Consequently,the media-ballyhooed “Year of the Woman” in 1992 was not so much an in-dication of a sudden end to gender bias, but rather part of a persistent erosionof male domination in politics.

∗ Ms. Schosberg graduated from Tufts with a B.A. in 1998.

42

Of course, the absence of a fund-raising disadvantage does not implythat male and female candidates are on equal footing. Equivalent campaignspending may not produce equivalent results. It is possible that due to eithervoter bias or candidate inexperience, campaign spending by female candid-ates is somehow less productive in swaying the electorate. Several studieshave examined the impact of voter perceptions of candidates through surveysand experiments, but the evidence that gender bias translates into an electoraldisadvantage is at best mixed (e.g., Leeper, 1991; Kahn, 1996; Smith, 1997).Further, Herrick (1996) uses data from the 1988 through 1992 House elec-tions to test the proposition that the electoral effects of campaign spendingdiffer with the gender of the candidate. She finds that campaign spending byfemale challengers is less productive in terms of vote share, but that just theopposite is true for campaign spending by female candidates for open seats.Consequently, there is no consistent evidence of voter bias against femalecandidates.

It is tempting to conclude that recent cohorts of female candidates face no(material) bias. In contrast, we argue that there remain strong prejudices thatwork to the disadvantage of female candidates. Previous work has assumedthat the existence of a bias against women should be evidenced in poorerperformance, either in raising funds or in winning votes. This would be trueif candidates were chosen by a random process from among a pool of “qual-ified” individuals. We would then expect to see voter bias or donor bias tobe manifested in poorer electoral performances by female candidates, sincemale and female candidates would be otherwise comparable in their electoralprospects.

Under this “random nomination” hypothesis, it is not voter or donor biaswhich explains the dearth of female office-holders, but rather the less thanproportional representation of women in the pool of qualified candidates.However, as the educational and work opportunities of women improve overtime, the pool of qualified females will expand and more women will bothrun for office and win election, until the representation of women in officeapproaches their representation in the population. In this scenario, there isno important gender bias specific to the political process, as evidenced bythe comparable success of those few but increasing number of women who“make the grade.” Existing studies of female candidates seem to support theconclusion that there is no specific bias against qualified female candidates.However, the interpretation of existing evidence is called into doubt once weconsider the effect of barriers to entry in politics.

Suppose that there exists a barrier to entry for potential female candid-ates. Initially suppose that this barrier is the product of gender bias amongvoters, potential campaign donors or some combination. The exact source

43

of the bias is irrelevant except that it be specific to the political process andseparate from any additional socioeconomic bias that determines the size ofthe pool of qualified candidates. Further, suppose that the electoral successof a candidate depends on that candidate’s “quality,” as well as the directeffects of any voter or donor bias. If potential candidates choose to run whentheir expected performance exceeds some threshold (for now assume thatthe threshold is the same for men and women), then the average quality offemales candidates will behigher than that of male candidates. This qualitydifference arises because quality must compensate for the expected electoraleffects of gender bias among voters or donors in order for a female potentialcandidate to reach the threshold required to run for office. If researchers areunable to appropriately control for candidate quality then, this “cream of thecrop” effect – the presence of relatively few but relatively high quality femalecandidates – will serve to counter the observed effects of voter or donor biason candidate receipts and vote share.2

It is also possible that the candidate selection process is biased againstwomen; this would raise the minimum quality threshold required for womento become candidates, exacerbating the “cream of the crop” effect. If this werethe only gender bias present in the political process, then we would expectthat those few and high quality female candidates who run for office wouldraise and spendmoremoney and farebetter in their election bids than theirmale counterparts. However, the additional presence of voter and/or donorbias against female candidates works in the opposite direction. This makesthe effects of gender on the campaign finances and vote shares of candidatesambiguous. Consequently, the observation that female candidates suffer nonet fund-raising or electoral disadvantage is consistent with both the absenceand presence of gender bias.

Previous studies of the effects of gender in House elections have implicitlyassumed that the quality of candidates is either random (and therefore omittedfrom the set of control variables) or perfectly controlled by proxy measures.However, it is difficult to quantify candidate quality, so it is not possibleto control completely for this factor (Levitt, 1994). Further, even if we cancontrol for incumbent quality, the presence and quality of challengers is notrandom, but determined in part by the characteristics of incumbents (Milyo,1999). We address the first point by focusing on the experience of male andfemale incumbents, since when comparing across incumbency status the ef-fects of gender may be obscured (because incumbency is such a dominantdeterminant of fund-raising and electoral success). Further, more is knownabout incumbent quality (tenure, committee and leadership positions, etc.),which mitigates the problem of unobserved heterogeneity among incumbents.Initially, we address the second point by controlling for the presence and

44

observed quality of challengers. However, we also model the selection pro-cess for challengers and estimate the effects of this selection process on thefund-raising and electoral success of female incumbents.

We show that if non-gender determinants of fund-raising and electoralsuccess are ignored, then female incumbents appear to run under-fundedcampaigns against well-heeled challengers and consequently fare worse interms of vote share than their male counterparts. As we control for otherfactors (including incumbent quality) these gender differences become smal-ler. However, once we control for the presence and quality of challengers,we find absolutely no gender effect in the campaign finances of female in-cumbents or their challengers, but female incumbents do enjoy a statisticallysignificant electoral advantage (about two percentage points of the vote). Fur-ther, female incumbents are significantly more likely to face a challenger, amajor party challenger or a high quality challenger. Once this bias in thepresence and quality of challengers is considered, the electoral advantageenjoyed by female incumbents rises to five or six percentage points, or abouttwice the electoral advantage attributable to the benefits of holding office inthe House (Levitt and Wolfram, 1997). This suggests that there does indeedexist important unobserved individual heterogeneity among incumbents andthat female incumbents are of higher quality than male incumbents. The factthat female incumbents are nevertheless more likely to be opposed suggeststhat potential challengers systematically underestimate the quality of femaleincumbents. This contention is supported by the additional finding that oncewe control for the selection of challengers, female incumbents still do notraise or spend more money, but their challengers spend farlessmoney than dochallengers of male incumbents. Apparently, potential campaign donors arenot as optimistic about defeating female incumbents as are their challengers.While we can not rule out the existence of gender bias among either or bothvoters and campaign donors, our results do suggest that the most import-ant effects of gender in House elections are realized through the barriers toentry into politics for women and the irrational tendency for challengers tooverestimate their likely success against female incumbents.

2. Data and methods

We examine the electoral success of House incumbents for the period 1984to 1992. We pool the data over these elections, since over this time periodthere are only 108 instances of female incumbents running for reelection.3

We focus on incumbents elected prior to 1992 (a.k.a., “Year of the Woman”)for two reasons. First, the period chosen makes our results more comparableto those found in previous studies (e.g., Burrell, 1994). Second, we focus

45

on incumbents since it is easier to control for variations in candidate qual-ity among incumbents.4 Future work should explore whether the surprisingfindings reported here carry over to more recent elections. For example, asthe socioeconomic barriers to entry in politics fall, do the specific politicalbarriers for women also fall? If so, we would expect the average quality offemale candidates to fall as more women enter politics.

Descriptive statistics for all of the variables employed in this study arereported in Tables A1 in the Appendix. All campaign finance data are fromthe Federal Election Commission and are adjusted for inflation (1990 = 100);information on challenger quality is taken from the Congressional QuarterlyWeekly Report. All other data is from the Almanac of American Politics.

We assume the following structural model of incumbent vote share,incumbent campaign finance and challenger campaign finance:

(Incumbent Vote Share)i = vs1∗(Incumbent Expenditures)i (1)

+ vs2∗(Challenger Expenditures)i + X1i VS + u1i

(Incumbent Receipts)i = r1∗(Incumbent Vote Share)i + X2i R + u2i (2)

(Incumbent Carry Over)i = s1∗(Incumbent Vote Share)i + X3i CO + u3i (3)

(Incumbent Expenditures)i = (Incumbent Receipts)i (4)

− (Incumbent Carry Over)i

+ (Incumbent Carry Over)−1,i

(Challenger Expenditures)i = (Challenger Receipts)i (5)

− (Challenger Carry Over)i

+ (Challenger Carry Over)−1,i

(Challenger Receipts)i = r1∗(Incumbent Vote Share)i + X6i CR + u6i (6)

(Challenger Carry Over)i = cs1∗(Incumbent Vote Share)i (7)

+ X7i CCO + u7i

46

In the model above, the u’s are disturbance terms, the other terms in italicsare structural parameters and the subscript ranges from one to the numberof incumbents running for reelection. The X’s are matrices of exogenousdeterminants of each structural equation. “Carry Over” is the savings or debtcarried over to the next election, while “Carry Over−1” is the same variablelagged one election (it is the savings or debt carried over to the current elec-tion). Finally, we assume for the time being that the presence and quality ofchallengers is determined exogenously. Later, we will treat the presence andquality of challengers as endogenous variables.

Fortunately, we do not need to estimate the structural parameters of thismodel, since identification of the structural parameters of this kind of systemof equations has been the bugaboo of the literature on the electoral effects ofcampaign spending (Milyo, 1999). We are only interested in the net effectsof gender on the dependent variables of interest (Incumbent Vote Share, In-cumbent Receipts, Incumbent Expenditures, and Challenger Expenditures).Consequently, we estimate reduced form regressions of the following type:

(Dependent Variable)i = Xi DV + ui (8)

In this reduced form equation, X is composed of all of the exogenous vari-ables that appear in the other independent variables in (1–7) above and theparameter estimates (DV) are the net effects of each of these exogenousfactors on the dependent variable. Consequently, regressions on each of thedependent variables employ the same set of independent variables.

We take an agnostic approach to the exact specification of (8) and reportregression results for transformed (log or square root) and untransformedvalues of the dependent variables. Further, we employ four nested sets of in-dependent variables. The first includes only a constant and a dummy indicatorof incumbent gender (Female). The second adds party and year (interacted)effects as well as a measure of district level partisan affiliation. Our districtmeasure is the district vote share for the presidential candidate of the incum-bent’s party in the most recent election (vote share is calculated using only themajor party candidates).5 In the third specification, we add controls for tenure(tenure, tenure squared and a dummy for first term incumbents) and positionsof influence in the House. The position controls are a dummy variable forleaders (the Speaker and other party leaders, members of the Committeeon Rules, and committee or subcommittee chairs) and a dummy variableindicating membership on three of the more powerful House committees (Ap-propriations, Ways and Means, and Energy and Commerce).6 The positioncontrols are also interacted by party. The fourth specification includes severaldummy indicators for the presence of a primary challenger, the presence ofany opponent in the general election, the presence of a major party opponent

47

and the presence of a “high quality challenger.” We define “high quality” asprevious experience in elective office, service as a major party official, othergovernment service (Congressional staffer, judge, sheriff, etc.) or some localcelebrity status (professional athletes, television news anchors, etc.). Thismeasure of challenger quality is admittedly simplistic, but it is common inthe literature (e.g., Jacobson and Dimock, 1994; Milyo, 1997a,b).7 Further,this simple dichotomous measure becomes an advantage when we treat thepresence and quality of challengers as dependent variables of interest, sinceit permits us to use a simple probit analysis.

Initially, we do not treat “Carry Over−1” as an independent variable in thereduced form estimation; as a lagged endogenous variable, “Carry Over−1”is correlated with other time invariant independent variables, like gender.Because we are specifically interested in the net effects of gender in elec-tions, we do not include this variable as a control.8 However, once we treatthe presence and quality of challengers as dependent variables, then in somespecifications we include “Incumbent Carry Over−1” (as well as lagged voteshare) as a determinant of opposition to the incumbent.9

3. Incumbent gender and campaign finance

In this section, we estimate various specifications of reduced-form regres-sions of incumbent campaign receipts and incumbent spending in Houseelections. This exercise demonstrates that the apparent campaign financeadvantage enjoyed by female incumbents disappears once non-gender de-terminants of campaign finances are considered.

First, we examine the campaign receipts of House incumbents. The firstcolumn of Table 1 shows that absent other controls, female incumbents raiseabout $50,000 more than male incumbents (or 10% of the average receipts foran incumbent during this time period). As additional controls are added, thisestimate becomes smaller, both in absolute terms and relative to the standarderror. Once the controls for the presence and quality of challengers are ad-ded in specification (4), the female effect on receipts completely disappears.This same pattern is evidenced even more dramatically in specifications (5–8)where the dependent variable is the natural logarithm of incumbent receipts.The estimated gender effect in specification (5) for an incumbent with averagereceipts is about $85,000 (close to 20% of the average receipts for an incum-bent during this time period), but this completely disappears once controlsare added for the presence and quality of challengers in specification (8).

The results in Table 2 confirm the existence of similar magnitudes andpatterns for incumbent expenditures (1–4) and the natural logarithm of in-cumbent expenditures (5–8). Consequently, despite significant differences in

48

Table 1. Effects of incumbent gender on campaign fund-raising (All incumbents running forreelection, 1984–1992; n=1924)

Campaign receipts

Explanatory variables (1) (2) (3) (4)

Female (n=108) 50,865∗ 42,902∗ 38,688 13,633

(27,138) (25,627) (24,826) (24,518)

R2 = .00 R2 = .08 R2 = .13 R2 = .19

Log of campaign receipts

(5) (6) (7) (8)

Female (n=108) .163∗∗∗ .149∗∗∗ .120∗∗∗ .068

(.050) (.047) (.046) (.044)

R2 = .00 R2 = .08 R2 = .16 R2 = .22

Party, year and district No Yes Yes Yes

Tenure and position No No Yes Yes

Challenger controls No No No Yes

Notes: ∗(p<.10), ∗∗(p<.05), and∗∗∗(p<.01). Heteroscedastic consistent standard errors inparentheses (White’s method).

mean campaign finances,ceteris paribusthere is no significant difference inthe campaign finances of female and male incumbents.

A similar pattern emerges for challenger expenditures (Table 3). How-ever, this dependent variable does present two additional complications: about10% of the incumbents in this time period are unopposed (n=197), and onlythose candidates spending more than $5,000 are required to file campaignfinance disclosure reports with the FEC (a total of 289 challengers do notmeet the reporting threshold). We have tried to account for these cases inseveral ways. The results we report treat the absence of a challenger as aninstance of zero challenger spending, while non-reporting challengers areassumed to have spent $2,500 (following Erickson and Palfrey, 1993).10 InTable 3, the previous pattern remains the same: absent other controls, femaleincumbents face higher spending challengers, but this difference disappearsonce the presence and quality of challengers are included as controls. Theonly statistically significant effects are seen in the logarithmic specifications,where female incumbents face challengers who spend almost $250,000 morethan the average challenger’s spending during this time period (an increaseof almost 200%). However, as above, this estimated gender effect completelydisappears once the presence and quality of challengers is added to the list ofcontrol variables.

49

Table 2. Effects of incumbent gender on campaign spending (All incumbents running forreelection, 1984–1992; n=1924)

Campaign expenditures


Female (n=108) 60,120∗∗ 44,865∗ 41,250 8,955

(28,447) (26,339) (25,840) (25,204)

R2 = .00 R2 = .12 R2 = .17 R2 = .25

Log of campaign expenditures

(5) (6) (7) (8)

Female (n=108) .216∗∗∗ .180∗∗∗ .139∗∗ .059

(.064) (.060) (.060) (.058)

R2 = .00 R2 = .12 R2 = .19 R2 = .28





Table 3. Effects of incumbent gender on challenger campaign spending (All Incumbentsrunning for reelection, 1984–1992; n=1924)

Challenger campaign expenditures


Female (n=108) 29,139 27,828 11,616 –10,025

(23,182) (22,223) (20,751) (21,265)

R2 = .00 R2 = .10 R2 = .13 R2 = .23

Log of challenger campaign expenditures

(5) (6) (7) (8)

Female (n=108) 1.089∗∗∗ .951∗∗∗ .753∗∗∗ –.051

(.260) (.248) (.248) (.168)

R2 = .00 R2 = .08 R2 = .09 R2 = .82





50

Table 4. Effects of incumbent gender on vote share (All incumbents running for reelection,1984–1992; n=1924)

Incumbent vote share


Female (n=108) –.022∗∗ –.020∗ –.014 .020∗∗(.011) (.009) (.009) (.008)

R2 = .00 R2 = .17 R2 = .18 R2 = .72

Square root of incumbent vote share

(5) (6) (7) (8)

Female (n=108) –.012∗ –.010∗ –.006 .013∗∗∗(.007) (.005) (.006) (.005)

R2 = .00 R2 = .17 R2 = .18 R2 = .68





4. Incumbent gender and incumbent vote share

The absence of a gender difference in campaign spending has inspired someattempts to test whether campaign expenditures by female candidates areless productive than expenditures by male candidates (e.g., Herrick, 1996).However, we find that once other relevant factors are taken into considera-tion, female incumbents have a significant electoral advantage over their malecounterparts.

As in most studies of Congressional elections, we measure electoralsuccess by the incumbent’s share of the vote received by the top two vote-getters.11 The first column of Table 4 shows that absent other controls, femaleincumbents receive significantlylessvote share (2.2 percentage points) thanmale incumbents. As more controls are added in specifications (2) and (3),this estimate becomes smaller both in absolute terms and relative to thestandard error. Once the presence and quality of challengers are included asindependent variables, the female effect on vote share becomes positive andsignificant (2 percentage points). This is about half the size of recent Levittand Wolfram’s (1997) estimate of the office-specific portion of the incum-bency advantage itself. Specifications (5–8) show this same pattern when thedependent variable is the square root of vote share.

51

The evidence presented up to this point demonstrates that the estim-ated effects of gender in House elections can change dramatically depend-ing on whether the presence and quality of challengers are held constant.Consequently, these variables merit further attention.

5. Incumbent gender and the presence and quality of challengers

In Table 5, we present the results of probit analysis on each of three nes-ted dichotomous variables: whether the incumbent is opposed, whether theincumbent is opposed by a challenger who is endorsed by a major party,and whether the incumbent is opposed by a high quality challenger (definedabove). We use the same nested set of independent variables, with one ex-ception: our fourth specification now differs from the third in that we addthe incumbent’s lagged carry over (or the incumbent’s “war chest”) and theincumbent’s previous vote share as determinants of the presence and qualityof challengers. For each of these specifications, gender has a statistically sig-nificant influence on the presence and quality of challengers. For example,using the estimated coefficients in specification (4) and considering a maleincumbent with the average probability of being opposed in the general elec-tion [(popposed|male) = .90], a change in gender raises this probability by sevenpercentage points [(popposed|female) – (popposed|male) = .07], which is an 8%increase in the baseline probability. Similar calculations for the probabilityof being opposed by a major party challenger or the probability of beingopposed by a high quality challenger produce changes of twelve and sevenpercentage points, respectively [(pmajor-party|female) –(pmajor-party|male) = .12and (phigh-quality|female) – (phigh-quality|male) = .07]; these represent increasesof 14% and 39% over the respective baseline probabilities.12

This gender related increase in the presence and quality of challengerssuggests that the electoral advantage of female incumbents is understatedin the regression results reported in Table 3, above (since the presence andquality of challengers were treated as exogenous). Consequently, we nowtreat the presence and quality of challengers as endogenous variables.

6. Gender effects in a sample selection model of House elections

Assume the following two-stage process: first, a high quality challengerchooses whether to run against an incumbent, then given that a high qualitychallenger runs we estimate the gender effect on the other dependent variablesof interest. The first stage estimate that we use is identical to specification (12)in Table 5 (identification comes from the lagged Carry Over and lagged Vote

52

Table 5. Effects of incumbent gender on the presence and quality of challengers (Allincumbents running for reelection, 1984–1992; n=1924)

Opposed in general election


Female (n=108) .671∗∗∗ .623∗∗ .601∗∗ .601∗∗(.251) (.263) (.275) (.275)

Major party opponent

(5) (6) (7) (8)

Female (n=108) .903∗∗∗ .901∗∗∗ .871∗∗∗ .878∗∗∗(.250) (.263) (.265) (.274)

High quality opponent

(9) (10) (11) (12)

Female (n=108) .332∗∗ .348∗∗ .274∗∗ .244∗(.134) (.140) (.141) (.143)



(Carry over)−1 and (vote share)−1 No No No Yes

Notes: ∗(p<.10),∗∗(p<.05), and∗∗∗(p<.01). Standard errors in parentheses.

Share variables). The second stage regression is run on the selected sampleof only those incumbents with high quality challengers (n=340); we correctfor the sample selection bias by including the inverse mills ratio from thefirst stage as an independent variable (Heckman, 1979). In Table 6 we reportthe second stage estimated gender effect for the campaign finances and voteshare variables (transformed and untransformed). Gender still has no net ef-fect on the finances of the incumbent in specifications (1)–(4), but the effecton the other variables is more pronounced than in previous estimates. Theselection corrected estimate of the electoral advantage of female incumbentsis six percentage points in both the linear (5) and square root specifications(6); p < .01 and p< .05, respectively.13 This electoral advantage is aboutthree times as large as our previous estimate. The selection corrected effectof gender on challenger expenditures is also significant and about ten timeslarger than our previous estimate. For both the linear (7) and log (8) cases, theestimated incumbent gender effect is equal to about one-third of the standarddeviation in challenger expenditures; p< .05 and p< .10, respectively.14

Some caution is in order; we have already stated that challenger qualityis particularly difficult to measure (indeed, that is part of our motivation for

53

Table 6. Selection corrected estimates of effects of gender (All incumbents opposed by a highquality challenger, 1984–1992; n=340)

Dependent variables

Explanatory variables Receipts Log(Rec.) Expenditures Log(Exp.)

Female (n=29) –31,658 -.048 –22,900 –.037

(46,082) (.098) (66,487) (.105)

R2 = .17 R2 = .19 R2 = .18 R2 = .18

Mean and standard deviation of the dependent variable

641,039 13.23 644,656 13.22

(362,277) (.54) (391,112) (.59)

Dependent variables

Square root of challenger

Vote share Vote share Expenditures Log(Ch. Exp)

Female (n=29) .061∗∗∗ .038∗∗ –111,713∗∗ –.52∗(.017) (.010) (53,198) (.297)

R2 = .24 R2 = .30 R2 = .20 R2 = .21


.61 .78 292,548 11.85

(.09) (.06) (306,744) (1.56)

Party, year, district,

tenure and position Yes Yes Yes Yes

Notes: ∗(p<.10), ∗∗(p<.05), and∗∗∗(p<.01). heteroscedastic consistent standard errors inparentheses (White’s method).

examining the gender of incumbents). Further, only 29 female incumbentsface high quality challengers during the period we examine. So, as a checkon these findings, we estimate a more conservative version of this same two-stage model. Rather than attempt to identify “high quality” challengers, wefeign ignorance on this point and simply distinguish between challengerswho are endorsed by a major party and those who are not (a more gross, butincontrovertible measure of challenger quality). Now the first stage estimateis identical to specification (8) in Table 5. In the second stage, we includeonly those incumbents with a major party challenger (n=1638) and as before,we correct for sample selection bias.

The results of this exercise are found in Table 7; our findings are verysimilar to those reported in Table 6. There is no gender effect on incumbentfinances, but there is a significant gender effect on the incumbent’s vote shareand the challenger’s expenditures.15 While the estimated effects are a bit more

54

Table 7. Alternative selection corrected estimates of effects of gender (All incumbentsopposed by a major party challenger, 1984–1992; n=1638)

Dependent variable

Explanatory variables Receipts Log(Rec.) Expenditures Log(Exp.)

Female (n=105) –25,502 –039 –25,668 –.059

(28,323) (.054) (24,744) (.067)

R2 = .14 R2 = .18 R2 = .17 R2 = .21


501,364 12.95 473,040 12.84

(303,876) (.62) (327,905) (.72)

Dependent variable

Square Root of Challenger

Vote share Vote share Expenditures Log(Ch.Exp.)

Female (n=105) .054∗∗ .033∗∗∗ –44,299∗ –.514∗∗∗(.009) (.005) (23,939) (.194)

R2 = .38 R2 = .36 R2 = .15 R2 = .20

Mean and standard error of the dependent variable

.663 .812 145,819 10.56

(.098) (.098) (231,022) (1.91)

Party, year, district,

tenure and position Yes Yes Yes Yes

Notes: ∗(p<.10), ∗∗(p<.05), and∗∗∗(p < .01). Heteroscedastic consistent standard errors inparentheses (White’s method).

modest, this is to be expected since we have chosen to allow more unobservedheterogeneity among challengers by ignoring our measure of high qualitychallengers.

7. Conclusion

Potential challengers systematically underestimate the electoral strength offemale incumbents; this gender bias is irrational in that those challengers fareworse in terms of vote share than comparable challengers. This appears to beat least partly recognized by potential campaign donors as the challengers offemale incumbents raise and spend less money than comparable challengers.Finally, female incumbents do not raise or spend more money than male in-cumbents, despite their higher average quality (and vote share). However, this

55

does not necessarily imply that female incumbents suffer some disadvantagein fund-raising, since they presumably have less demand for campaign funds(owing to the fact that they typically win their races with higher vote sharesthan male incumbents and they face relatively under-funded challengers).

The gender bias in the challenger selection process has gone unrecognizedin the literature on women in politics. This selection bias is significant in itsown right (a 10 to 30% increase in the probability of being opposed, opposedin a primary, opposed by a major party challenger or opposed by a highquality challenger), but it is all the more important in that it confounds the es-timation of the gender effect on candidate vote shares and campaign finances.By considering the selection of challengers in House races, we have shownthat the “cream of the crop” effect causes female candidates to be of higherquality than males (as evidenced by the five to six percentage point genderadvantage in vote share). Further, once this quality difference is recognized,the lack of a gender difference in fund-raising is consistent with the existenceof gender bias on the part of donors (contrary to what previous studies haveconcluded).

It should also be noted that to the extent that there exists additional genderbias among voters, our findings understate the true electoral advantage attrib-utable to the higher quality among female incumbents. Consequently, futurework on gender and elections should take into consideration the large qualitydifference between female and male candidates demonstrated by this study.

Notes

1. Levitt and Wolfram (1997) estimate that the office-specific benefits of incumbency trans-late into about 3.7 percentage points of the two-party vote for House elections in the1980’s.

2. Similar “cream of the crop” effects have been observed in the wages of women and minor-ities, as well as in SAT scores. For example, Blau and Beller (1988) find that increasedlabor force participation by women has caused a decrease in the female to male wageratio (all else constant); this is the “cream of the crop” phenomenon in reverse: as genderbarriers in employment fall, the quality (hence wage) of the average female worker alsofalls.

3. Because of the small number of female incumbents, we treat each instance of a femaleincumbent running for election as an independent observation. This is not an uncommonassumption in the literature on Congressional elections (e.g., Ansolabehere and Snyder,1996b; however, also see Milyo, 1998). As more women serve in the House, it will be bothpossible and desirable to estimate individual fixed effects in addition to gender effects.

4. For a discussion of the difficulties and importance of measuring candidate quality, seeLevitt (1994) and Milyo (1998, 1999).

5. Lagged vote share is sometimes used as a proxy for district partisanship, but see thediscussion below on lagged dependent variables.

56

6. Our results are not sensitive to the exact definitions of the “positions of influence” vari-ables. The list of powerful committees mimics that in Ansolabehere and Snyder (1996b);also see Groseclose and Stewart (1998).

7. It is well documented that experienced challengers fare better than political novices(e.g. Jacobson, 1990). For more detailed treatments of challenger quality, see especiallyGreen and Krasno (1988), Krasno and Green (1988), Levitt (1994), Squire (1995) andAnsolabehere and Snyder (1996b).

8. Lagged endogenous variables can be useful as proxies for omitted and time-invariantfactors, so we have explored the impact of including Incumbent Carry Over−1; in short,our findings are slightly more modest, but still consistent with the effects of genderreported below.

9. Several studies have examined the relationship between incumbent war chests and thepresence of challengers; for evidence of the deterrent effects of war chests, see Sorauf(1988 and 1992), Squire (1989), Hersch and McDougall (1994), Goodliffe (1995) andBox-Steffensmeier (1996). However, see Ansolabehere and Snyder (1996a,b) and Milyo(1998) for evidence to the contrary.

10. We have experimented with setting a different amount for non-reporting challengers butthis has little substantive affect on the results reported in Table 3. Tobit estimates do pro-duce larger estimates of incumbent gender on challenger expenditures for specifications(1)–(3) and (5)–(7), but do not affect the pattern across specifications and do not producesubstantively different estimates for gender in the specifications of most interest, (4) and(8).

11. During this time period, there are simply too few incumbents defeated to use that asa measure of electoral success. However, probit analysis of incumbent wins does notcontradict the findings presented here (the estimated gender effects have similar signs,but are not significant). As a dependent variable, vote share does have its drawbacks: ithas an upper bound of one (197 incumbents are unopposed in the sample). However, tobitestimates for vote share (and square root of vote share) produce very similar estimates ofthe gender effect.

12. We conduct a similar analysis of primary opposition; for specification (4), the gendereffect is twelve percentage points [(pprimaryk-opponent|female) – (pprimary-opponent|male)= .12] and is statistically significant (p< .05); this represents a 39% increase in theprobability of being opposed in the primary [(pprimary-opponent|male)=.31].

13. As before, in the non-linear specifications, the estimated (untransformed) gender effect iscalculated for a male incumbent with an average vote share or with average challengerexpenditures.

14. There are 17 high quality challengers that do not report any expenditures, as before, weset challenger expenditures at $2,500 for these cases.

15. There are 231 major party challengers that do not report any expenditures, once again, weset expenditures at $2,500 for these cases.

References

Ansolabehere, S. and Snyder, J. (1996a). The inter-election dynamics of campaign finance:U.S. House elections, 1980 to 1994. Unpublished manuscript, presented at the 1996meetings of the American Political Science Association in San Francisco, CA.

57

Ansolabehere, S. and Snyder, J. (1996b). Money, elections and candidate quality. Unpublishedmanuscript; presented at the 1996 meetings of the Midwest Political Science Associationin Chicago, IL.

Blau, F. and Beller, A. (1988). Trends in earnings differentials by gender, 1971–1981.Industrial and Labor Relations Review41: 513–529.

Box-Steffenmeier, J. (1996). A dynamic analysis of the role of campaign war chests incampaign strategy.American Journal of Political Science40: 352–371.

Burrell, B. (1985). Women’s and men’s campaigns for the U.S. House of Representatives.American Politics Quarterly13: 251–272.

Burrell, B. (1994).A woman’s place is in the House. Ann Arbor: University of Michigan Press.Darcy, R., Welch, S. and Clark, J. (1994).Women, elections and representation. Lincoln:

University of Nebraska Press.Erickson, R. and Palfrey, T. (1998). Campaign spending and incumbency: An alternative

simultaneous equations approach.Journal of Politics60: 355–373.Flammers, J. (1997).Women’s political voice. Philadelphia: Temple University Press.Fox, R. and Smith, E. (1998). The role of candidate sex in voter decision-making.Political

Psychology19: 405–419.Gaddie, R. and Bullock, C. (1995). Congressional elections and the year of the wo-

man: Structural and elite influences on female candidates.Social Science Quarterly76:749–762.

Goodliffe, J. (1995). War chests and challenger quality. Unpublished manuscript, presented atthe 1995 meetings of the Midwest Political Science Association in Chicago, IL.

Green, D. and Krasno, J. (1988). Salvation for the spendthrift incumbent: Reestimating theeffects of campaign spending in House elections.American Journal of Political Science32: 884–907.

Groseclose, T. and Stewart, C. (1998). The value of committee seats in the House, 1947–1991.American Journal of Political Science42: 453–474.

Heckman, J. (1979). Sample selection bias as a specification error.Econometrica47: 153–161.Herrick, R. (1996). Is there a gender gap in the value of campaign resources?American Politics

Quarterly24: 68–80.Hersch, P. and McDougall, G. (1994). Campaign war chests as a barrier to entry in

congressional races.Economic Inquiry32: 630–641.Jacobson, G. (1990).The electoral origins of divided government. San Francisco: Westview

Press.Jacobson, G. and Dimock, M. (1994). Checking out: The effect of bank overdrafts on the 1992

House elections.American Journal of Political Science38: 601–624.Kahn, K. (1996).The political consequences of being a woman. New York: Columbia

University Press.Krasno, J. and Green, D. (1988). Preempting quality challengers in House elections.Journal

of Politics50: 920–936.Leeper. M. (1991). The impact of prejudice on female candidates.American Politics Quarterly

19: 248–261.Levitt, S. (1994). Using repeat challengers to estimate the effects of campaign spending on

election outcomes in the U.S. House.Journal of Political Economy102: 777–798.Levitt, S. and Wolfram, C. (1996). Decomposing the sources of incumbency advantage in the

U.S. House.Legislative Studies Quarterly22: 45–60.Milyo, J. (1997a). The electoral and financial effects of changes in committee power: GRH,

TRA86 and the money committees in the House.Journal of Law and Economics60: 93–112.

58

Milyo, J. (1997b). The economics of campaign finance: FECA and the puzzle of the not verygreedy grandfathers.Public Choice93: 245–270.

Milyo, J. (1998).The electoral effects of campaign spending in House elections. Los Angeles:Citizens’ Research Foundation.

Milyo, J. (1999). What do candidates maximize (and why should anyone care)? Unpublishedmanuscript, presented at the 1999 meetings of the Public Choice Society in New Orleans,LA.

Squire, P. (1989). Competition and uncontested seats in U.S. House elections.LegislativeStudies Quarterly14: 281–295.

Squire, P. (1995). Candidates, money and votes.Political Research Quarterly48: 891–917.Uhlaner, C. and Schlozman, K. (1986). Candidate gender and campaign receipts.Journal of

Politics48: 30–50.Werner, B. (1997). Financing the campaigns of women candidates and their opponents:

Evidence from three states, 1982–1990.Women & Politics18: 81–97.

59

Appendix

Table A1.Descriptive statistics

All incumbents running for reelection, 1984–1992; n=1924

Continuous variables

Variable name Mean Standard deviation

Incumbent receipts (1990 dollars) 474,768 295,817

Incumbent expenditures (1990 dollars) 439,430 319,957

Challenger expenditures (1990 dollars) 124,350 219,266

Incumbent vote share (top two vote getters) .709 .144

Tenure (years) 10.45 7.82

Presidential vote share∗democrat∗1984 .07 .19





Presidential vote share∗republican∗1984 .05 .18





Dichotomous variables

Variable name Mean Variable name Mean Variable name Mean

Female .056 Primary .295 1986∗republican .082

Republican .394 Year 1986 .201 1988∗republican .083

Good Committee .307 Year 1988 .207 1990∗republican .080

Leaders .127 Year 1990 .207 1992∗republican .070

Freshman .138 Year 1992 .176

Documents

Gender Bias and Selection Bias in House Elections