18
The Lingering Effect of Scandals in Congressional Elections: Incumbents, Challengers, and Voters Rodrigo Praino, University of Connecticut Daniel Stockemer, University of Ottawa Vincent G. Moscardelli, University of Connecticut Objective. We have two goals. First, we investigate both the short- and long-term electoral impact of involvement in scandals on reelection margins of incumbents in U.S. congressional elections. Second, we evaluate the impact of scandals on district- level turnout. Methods. We model the impact of involvement in a political scandal on incumbents’ electoral margins in the election cycle in which the scandal comes to light, as well as in future election cycles. We also model the impact of scandal on district-level turnout. Results. Involvement in a scandal exerts not only an im- mediate, negative effect on incumbents’ margins, but one that also lingers beyond the initial reelection cycle. Elections involving incumbents embroiled in scandals experience a small boost in turnout. Conclusion. In tandem, these results implicate the mobilization of previous nonvoters intent on “throwing the bum out” as one mechanism through which incumbent vote share is depressed in scandal elections. Scandals involving sitting members of Congress of both parties have sparked renewed interest in the impact of such behavior on incumbent reelection rates, margins of victory, and decisions to retire. The dramatic swings of recent elections notwithstanding, the number of House seats actually in play in any given election remains small (Engstrom, 2012). Given this context, the role of scandals in congressional elections is an important topic of study, in part because scandals have emerged as one of the few forces capable of dislodging incumbents who represent otherwise safe seats (Herrnson, 2012:252–55). It is not surprising, then, that scandals have received continued attention from students of congressional elections over the years. Since the first systematic efforts to understand this relationship appeared, analysts have consistently identified involvement in a scandal as a factor that depresses the vote share of Direct correspondence to Vincent G. Moscardelli, Department of Political Science, Uni- versity of Connecticut, Storrs, CT 06269–1024 [email protected]. The authors will share all requests for data and coding for replication purposes. The authors thank Robin Kolodny for helpful comments on an earlier version of this article at the Annual Meeting of the Northeastern Political Science Association, November 17–19, 2011. Moscardelli thanks Briana Bardos and Jeremy Galtieri for research assistance and gratefully acknowledges the support of T. A. Borradaile and the UConn’s Alan R. Bennett Fund. SOCIAL SCIENCE QUARTERLY, Volume 94, Number 4, December 2013 C 2013 by the Southwestern Social Science Association DOI: 10.1111/ssqu.12046

The Lingering Effect of Scandals

Embed Size (px)

Citation preview

  • The Lingering Effect of Scandals inCongressional Elections: Incumbents,Challengers, and Voters

    Rodrigo Praino, University of Connecticut

    Daniel Stockemer, University of Ottawa

    Vincent G. Moscardelli, University of Connecticut

    Objective. We have two goals. First, we investigate both the short- and long-termelectoral impact of involvement in scandals on reelection margins of incumbents inU.S. congressional elections. Second, we evaluate the impact of scandals on district-level turnout. Methods. We model the impact of involvement in a political scandalon incumbents electoral margins in the election cycle in which the scandal comesto light, as well as in future election cycles. We also model the impact of scandalon district-level turnout. Results. Involvement in a scandal exerts not only an im-mediate, negative effect on incumbents margins, but one that also lingers beyondthe initial reelection cycle. Elections involving incumbents embroiled in scandalsexperience a small boost in turnout. Conclusion. In tandem, these results implicatethe mobilization of previous nonvoters intent on throwing the bum out as onemechanism through which incumbent vote share is depressed in scandal elections.

    Scandals involving sittingmembers of Congress of both parties have sparkedrenewed interest in the impact of such behavior on incumbent reelection rates,margins of victory, and decisions to retire. The dramatic swings of recentelections notwithstanding, the number of House seats actually in play in anygiven election remains small (Engstrom, 2012). Given this context, the roleof scandals in congressional elections is an important topic of study, in partbecause scandals have emerged as one of the few forces capable of dislodgingincumbents who represent otherwise safe seats (Herrnson, 2012:25255). Itis not surprising, then, that scandals have received continued attention fromstudents of congressional elections over the years. Since the first systematicefforts to understand this relationship appeared, analysts have consistentlyidentified involvement in a scandal as a factor that depresses the vote share of

    Direct correspondence to Vincent G. Moscardelli, Department of Political Science, Uni-versity of Connecticut, Storrs, CT 062691024 [email protected]. The authorswill share all requests for data and coding for replication purposes. The authors thank RobinKolodny for helpful comments on an earlier version of this article at the Annual Meeting of theNortheastern Political Science Association, November 1719, 2011.Moscardelli thanks BrianaBardos and Jeremy Galtieri for research assistance and gratefully acknowledges the support ofT. A. Borradaile and the UConns Alan R. Bennett Fund.

    SOCIAL SCIENCE QUARTERLY, Volume 94, Number 4, December 2013C 2013 by the Southwestern Social Science AssociationDOI: 10.1111/ssqu.12046

  • 1046 Social Science Quarterly

    incumbents who choose to run for reelection (Rundquist, Strom, and Peters,1977; Peters andWelch, 1980; Abramowitz, 1991;Welch andHibbing, 1997;Brown, 2006; Long, 2011). But despite consensus on the impact of scandal onincumbents general election margins, many questions remain unanswered. Inthis article, we investigate two of them.First, existing studies (e.g., Welch and Hibbing, 1997; Brown, 2006) have

    limited themselves to modeling the impact of scandal on incumbents mar-gins in the election cycle in which the scandal comes to light. However, itseems plausible that involvement in a political scandal might have a lingering,negative effect on an incumbents electoral performance in future elections aswell. Just as cultivating a personal brand (Mayhew, 1974:49; Lott, 1991)may help insulate incumbents from the ebbs and flows of the partisan na-tional tide, damage to that brand might cut into electoral margins beyondthe current electoral cycle. We propose a method for estimating the impact ofinvolvement in a scandal in elections beyond the proximate one and apply itin an analysis of all contested U.S. House races involving incumbents between1972 and 2006. We find that, controlling for other factors, the impact ofinvolvement in a scandal does extend beyond the election cycle in which thescandal comes to light; incumbents recover much of their lost margins in theirfirst postscandal election, but it is not until four to six years after the scandalthat their predicted vote share approaches the level one would expect had theynever been involved in a scandal in the first place.Second, we investigate the impact of scandal on district-level turnout. We

    find that districts in which scandal-plagued incumbents run for reelectionexperience a small, marginally statistically significant, increase in turnout.However, this boost in turnout accrues entirely to challengers. Followinga brief discussion of the possible processes through which scandal mightsimultaneously depress incumbents margins and increase turnout amongsupporters of the challenger, we conclude that any effort to isolate the causalmechanism(s) through which scandals reduce incumbents electoral marginsmust account for the mobilization of previous nonvoters intent on throwingthe bum out.

    Scandals and Elections

    The existing literature on scandals involving members of Congress (MCs)can be divided into three separate lines of inquiry. The first deals with scandalsas an object of study per se. It seeks to classify and understand different typesof scandals, their causes, and their historical development (see, e.g., Thomp-son, 1995). The second, concerned with understanding career decisions ofrepresentatives, treats involvement in a scandal as a possible cause of strate-gic retirement. These analyses conclude that involvement in a scandal notonly affects members decisions to retire instead of seeking reelection, but alsovirtually precludes their seeking higher office afterwards (Kiewiet and Zeng,1993; but see Groseclose and Krehbiel, 1994).

  • The Lingering Effect of Scandals in Congressional Elections 1047

    The third strand of research deals with the electoral consequences of scan-dals. Beginning with Rundquist, Strom, and Peterss (1977) experimentalanalysis of electoral support for corrupt politicians, analysts have consistentlyidentified scandal as a factor that depresses the vote share of incumbents whochoose to run for reelection (Peters and Welch, 1980; Abramowitz, 1991;Welch and Hibbing, 1997; Brown, 2006).1 The type of scandal in whichone is involved (e.g., moral, financial, or abuse of power) can also affect onesmargin of victory, although this impact may be conditioned by ones party(Brown, 2006; Doherty, Dowling, and Miller, 2011).These findings have proven robust to model specification as well as op-

    erationalization and measurement of both the dependent and independentvariables. However, extant studies have not examined whether the impact ofscandals lingers beyond one election cycle or not. There are reasons to be-lieve it might. While Mayhew (1974:49) uses the concept of brand nameas a metaphor, the marketing literature on product-harm crises and brandcrises documents the damage adverse events inflict on firms and the stepsmanagement must take to recover (Siomkos and Kurzbard, 1994; Siomkos,1999). Such crises have been shown to cut into sales (VanHeerde, Helsen, andDekimpe, 2007), consumer confidence (Dawar and Pillutla, 2000), and stockvaluations (Pruitt and Peterson, 1986; Chen, Ganesan, and Liu, 2009). But forfirms that survive, the damage done by a crisis can be undone with time (Vas-silikopoulou et al., 2009:178). In other words, time heals all wounds (Vas-silikopoulou et al., 2009:174) because, eventually, consumers tend to forgetabout the crisis and its effects (Vassilikopoulou et al., 2009:178). Conceptual-izing scandals as a sort of brand crisis suggests that the initial damage inflictedby a scandal should dissipate over time rather than disappear immediately.Beyond the unaddressed questions surrounding the lingering effects of

    scandals on incumbents involved in them, we also revisit the relationshipbetween scandal and turnout in House elections. Specifically, we examine ifdistricts with scandal-ridden incumbents experience systematically higher orlower levels of voter turnout. Peters and Welch (1980) find that more seriouscharges of corruption will stimulate a net increase in turnout (1980:706)in congressional districts. But since theirs, to our knowledge no study hasreevaluated the relationship between an incumbents involvement in a scandaland voter turnout at the district level.2 Therefore, we investigate this link inthe second part of the article. However, we first turn to our examination ofthe relationship between scandals and electoral margins of incumbent MCs.

    1Some research suggests that involvement in a scandal may exert its largest effect on incum-bents electoral fortunes in primary elections. See, e.g., Jacobson and Dimock (1994), PetersandWelch (1980), and Brown (2006) (but seeWelch andHibbing, 1997:232). Boatright iden-tifies scandal as the second most frequently cited motivation for primary challenges between1970 and 2008 (2013:67).

    2Bowler and Karp (2004) demonstrate that scandals involving legislators can have a neg-ative influence on their constituents attitudes toward institutions and the political process(2004:271; see also Maier, 2011), and such attitudes have been shown to depress politicalparticipation at the individual and macro levels (Anderson and Tverdova, 2003; Kostadinova,2009).

  • 1048 Social Science Quarterly

    TABLE 1Scandals, Incumbents, and Outcomes, 19722006

    Outcome Scandal Races Nonscandal Races

    Won general election 43 (48.9%) 6,657 (86.5%)Defeated in general election 12 (13.6%) 304 (4.0%)Resigned/retired 23 (26.1%) 653 (8.5%)Defeated in primary election 10 (11.4%) 79 (1.0%)Total 88 (100.0%) 7,693 (100%)

    SOURCE: Information on House Ethics Committee actions taken from Historical Sum-mary of Conduct Cases in the House of Representatives, 17982004 (available athttp://ethics.house.gov/publication/historical-summary) and Summary of Activities fromvarious years (available at http://ethics.house.gov/reports/summary-activities). Data onelectoral outcomes of scandal cases were collected by the authors. Data on nonscandalraces were computed by the authors from Ornstein, Mann, and Malbin (2008:57, Tables27).

    Scandals and Electoral Margin: Concepts, Data, and Methodology

    For purposes of this analysis, an MC is involved in a scandal in the electioncycle in which any of the following occur:

    1. The House Ethics Committee votes to open a formal investigation of,or inquiry into, member activity.

    2. The House Ethics Committee is directed to open a formal investigationof, or inquiry into, member activity by a resolution of the full House.

    3. The House Ethics Committee appoints an outside or special counsel toinvestigate member activity.

    4. The House Ethics Committee indicates that it received a complaintabout a member but chose to delay opening an investigation due topending state or federal criminal proceedings.

    5. The House Ethics Committees investigation of an incident (as opposedto an individual) results in a member being singled out in the final reporton the results of that investigation.

    This Ethics-Committee-based operationalization ensures both the repli-cability of our study and the inclusion of only high-profile and reasonablyfounded allegations of professional misconduct. This approach yielded 88 in-stances of scandal, 65 of which involved incumbents who went on to seek re-election (U.S. Congress, House Committee on Ethics, 2004, n.d.). As Table 1reveals, between 1972 and 2006, incumbents involved in scandals were aboutthree timesmore likely to resign or retire, three timesmore likely to be defeatedin the general election, and 11 times more likely to be defeated in a primarythan other incumbents. Figure 1 graphs the relationship between scandal andthe electoral margins of incumbents who sought reelection. In the bivari-ate, the immediate impact of the opening of an investigation by the House

  • The Lingering Effect of Scandals in Congressional Elections 1049

    FIGURE 1Mean Electoral Margin of Incumbent Candidates Involved in Scandals, 19722006

    NOTE: The line depicts the mean margin of victory for the subset of incumbents involvedin scandals over time, with a value of 2 on the x-axis representing the margin in theelection immediately preceding the scandal, 0 representing the margin in the election cycleduring which the House Ethics Committee opens its investigation, and subsequent valuesrepresenting margins in two subsequent elections.

    Ethics Committee is substantial. On average, incumbents election marginsdrop almost 12 percentage points below their preinvestigation levels once aninvestigation is formally opened (from an average of about 33 points to anaverage of about 21 points).3 Furthermore, the scandal-plagued incumbentswho survive do not recover immediately, regaining only about 5 percentagepoints of their lost margins in the subsequent election. It is not until thesecond postscandal reelection bid (year 4) that the average scandal-plaguedincumbent returns to his or her prescandal margins.In the multivariate analysis that follows, our dependent variable, Electoral

    Margin, is the difference between the share of the vote obtained by incumbentsand the share of the vote obtained by opponents in each contested U.S. Houseelection between 1972 and 2006. To capture the immediate and lingering

    3Applying a more expansive operationalization of scandal to the 2006 and 2008 electioncycles as a validity check, we find that incumbents reelection margins drop by comparableamounts when scandal hitsfrom an average of about 33 points to an average of about 18suggesting that while identifying fewer cases, our operationalization effectively captures theunderlying concept of interest. In those cycles, we reviewed the biographies of all incumbentsin CQs Politics in America to identify those involved in scandals (see Boatright, 2013 for asimilar approach). This method yielded 27 cases of scandal in those two cycles, as comparedto 16 using our Ethics-Committee-based method.

  • 1050 Social Science Quarterly

    effects of scandals on vote margins depicted in Figure 1, we operationalizescandals in two ways. In the first, a dummy variable, Scandal Immediate Ef-fect, is coded 1 for every member of Congress who meets one of the criteriaoutlined above and 0 otherwise. For computational reasons that become clearbelow, this variable is also coded 1 for surviving scandal-plaguedMCs for everyelectoral contest subsequent to the opening of the investigation. The variablecaptures the immediate impact of involvement in a scandal on incumbentsreelection margins and should thus carry a negative sign. We also include asecond variable, Scandal Lingering Effect, that allows the impact of scandal todiminish (i.e., predicted margins to increase) over time relative to the baselineestablished by Scandal Immediate Effect. Scandal Lingering Effect is a countervariable that takes the value of 0 for the scandal cycle and all cycles beforethat, 2 for the election two years after the scandal, 4 for the election fouryears after the scandal, and so on. It is also coded 0 for incumbents who arenever the subject of an Ethics Committee investigation. Since the recoveryfunction necessarily tapers off over time, we take the natural log of ScandalLingering Effect. Because (ln)Scandal Lingering Effect captures scandals effecton election margins relative to the baseline established by Scandal Immedi-ate Effect, its coefficient should be positive, and its impact should diminishover time.Following Abramowitz (1991), we include several controls. District Par-

    tisanship, which captures the districts underlying partisanship, is calculatedas the district-level margin of victory or defeat of the incumbents partysmost recent presidential candidate, adjusted by subtracting the margin heor she obtained nationwide. Personal Advantage, our measure of the personalelectoral advantage enjoyed by incumbents, is the margin obtained by the can-didate in the previous electoral contest. Incumbent Expenditures andOpponentExpenditures are the amount of money the incumbent and his or her oppo-nent, respectively, spent in each election in 2006 U.S. dollars. Both are alsosquared to capture the nonlinear effect of campaign spending on incumbentmargins. Adjusted ADA Distance captures the ideological distance between anincumbent and the most ideologically extreme member of his or her partycaucus using Groseclose, Levitt, and Snyders (1999) adjusted ADA ratings asour metric. For Republicans, this variable takes the value of the incumbentsadjusted ADA score (AADA) minus the minimum AADA for congressionalRepublicans in that congress. For Democrats, it is calculated as the maximumAADA value for congressional Democrats minus the incumbents score.4

    We also control for seniority (Years in the House), freshman status(Freshman), membership on a reelection committee (Reelection Committee

    4Abramowitz (1991) uses conservative coalition support scores to capture the frequency withwhich MCs defect from their partys position on controversial roll call votes. Our measurealmost certainly taps into a similar dynamic since moderates of both partiesthose most likelyto defect in the way Abramowitz has in mindwill have the highest scores on our measure.

  • The Lingering Effect of Scandals in Congressional Elections 1051

    Member)5, and the quality of the incumbents challenger (Opponent Score).6

    In conclusion, to capture national party tides at specific points in time, wecalculate a dedicated Partyyear variable for every election year included in theanalysis. It is coded +1 for Democrats running in that year, 1 for Republi-cans running in that year, and 0 for both Democrats and Republicans runningin any other year (Abramowitz, 1991).

    Scandals and Electoral Margin: Model Estimation and Results

    We estimated our electoral margin model as a cross-sectional time-seriesgeneralized least squares (GLS) regression with random effects and robuststandard errors clustered by individual members of Congress. The results arepresented in Table 2.The model performs well (2 = 4,939, df = 30) and fits the data nicely

    (R2 = .58). Turning first to the immediate effects of scandal, the resultscorroborate the relationship we observed in the bivariate. On average, in-volvement in a scandal reduces an incumbents margin of victory by about14.5 points.7 More importantly, once we control for other factors, our resultsprovide strong support for our expectation that the electoral impact of involve-ment in a scandal lingers beyond an incumbents initial, postscandal reelectionbid. Incumbents involved in scandals recover much of their lost vote share,but not all of it, in the election immediately following the scandal election. Tofacilitate the interpretation of the coefficient associated with (ln)Scandal Lin-gering Effect, the solid line in Figure 2 plots the predicted margin of victory forincumbents embroiled in scandals at five points in time: their last scandal-freeelection bid, the scandal cycle, and the three subsequent postscandal electioncycles. The dotted line traces the same values for scandal-free incumbentsover a comparable period of their careers as a comparison.8 After the initial,

    5Reelection Committee Member is a dummy variable coded 1 for those incumbents who weremembers of one or more reelection committees in the Congress leading up to the election(Smith and Deering, 1997:64). We thank Charles Stewart for making the data from 1993forward available on his website (Stewart and Woon, 2011).

    6Opponent Score is an ordinal variable that measures the quality and experience of theincumbents opponent in each election. It is coded 1 if the opponent never held elective officein the past, 2 if the opponent held elective office, 3 if the opponent was a state legislator, and4 if the opponent was a former U.S. Representative. We thank Gary Jacobson for making hisdata on challenger quality available to us.

    7We should remind the reader that our dependent variable is reelection margin of victory incontested races. The mean of this variable from 1972 to 2006 is about 31, which correspondsto 65.5 percent of the vote share in a two-candidate race. So, for the average scandal-plaguedincumbent, the coefficient of 14.5 would correspond to a 16.5-point margin, or 58.25percent of the vote in a two-candidate race.

    8In our data, scandals emerge, on average, during an incumbents seventh term. Previousresearch suggests that while incumbents electoral margins tend to increase during their firstfew electoral successes, they also tend to taper off and decrease after a few terms in office (Prainoand Stockemer, 2012a, 2012b) Thus to facilitate comparison, the dotted line traces predictedmargins of victory for scandal-free incumbents in their sixth through tenth terms.

  • 1052 Social Science Quarterly

    TABLE 2Immediate and Lingering Effects of Scandal on Incumbent Reelection Margins,

    19722006

    Independent Variables Coefficient Standard Error

    Scandal immediate effect (dummy) 14.468 (2.132)(ln)Scandal lingering effect 2.990 (0.630)District partisanship 0.339 (0.015)Personal advantage 0.102 (0.009)Years in the House 0.142 (0.035)Freshman (dummy) 0.668 (0.475)Reelection committee member 0.610 (0.462)Incumbent expenditures 0.003 (0.001)Incumbent expenditures2 0.000001 (0.0000004)Opponent expenditures 0.023 (0.002)Opponent expenditures2 0.000004 (0.0000009)Opponent score 2.825 (0.244)Adjusted ADA distance 0.054 (0.012)Party 1972 0.312 (0.750)Party 1974 14.208 (0.829)Party 1976 3.972 (0.700)Party 1978 1.284 (0.784)Party 1980 2.627 (0.731)Party 1982 7.550 (0.733)Party 1984 2.978 (0.559)Party 1986 4.802 (0.701)Party 1988 0.999 (0.650)Party 1990 1.562 (0.829)Party 1992 0.921 (0.711)Party 1994 7.137 (0.552)Party 1996 2.894 (0.594)Party 1998 2.909 (0.569)Party 2000 2.209 (0.567)Party 2002 0.786 (0.744)Party 2004 2.618 (0.514)Party 2006 7.630 (0.552)Intercept 37.113 (1.685)R2 0.584Chi-square 4,939.43N 5,453

    NOTE: Dependent variable is incumbent margin of victory (or defeat) in U.S. House elections,19722006. Numbers in second column are random effects GLS regression coefficients.Robust standard errors, clustered by individual members of Congress, are reported in thethird column.p 0.1; p 0.01; p 0.001 (two-tailed tests).

    steep decline of over 14 points, incumbents involved in scandals recover asubstantial amount (almost nine points) of their lost margins in the subse-quent electionenough to move the average incumbent out of any dangerzone into which he or she might have fallen in the wake of the scandal. Butafter that, the rate of recovery slows, and it is not until the scandal-plagued

  • The Lingering Effect of Scandals in Congressional Elections 1053

    FIGURE 2Mean Predicted Electoral Margin of Incumbent Candidates Involved in Scandals,

    19722006

    NOTE: The solid line plots the mean predicted margin of victory (derived from the modelspecification presented in Table 2 while holding other variables constant) for incumbentsinvolved in scandals over time, with a value of 2 on the x-axis representing the margin inthe election immediately preceding the scandal, 0 representing the margin in the electioncycle during which the House Ethics Committee opens its investigation, and subsequentvalues representing margins in three subsequent elections. The dotted line depicts the samevalues for incumbents not involved in scandals over comparable periods in their careers.See text for further details.

    incumbents third postscandal reelection bid (year 6) that the lines approachconvergence, indicating that he or she has returned to the margin of victoryexpected for scandal-free incumbents with (on average) comparable seniority.Our control variables behave as expected and the coefficients associated

    with these variables corroborate previous efforts to model individual Houseelection outcomes (Jacobson, 1981; Abramowitz, 1991). Reelection marginsare higher for incumbents who represent districts in which their partys presi-dential nominee performed well, and a representatives margin in the previouselection is positively associated with his or her margin in the current one.Margins taper off over the course of a representatives career, and the presenceof a quality challenger significantly reduces an incumbents margin of victory(or defeat). The spending variables also behave as expected; all four are sta-tistically significant and in the expected direction, and challenger spendingexerts a larger impact on incumbent margins than does spending by the in-cumbent. Finally, the election-specific variables included to capture nationalforces (coded so that a positive coefficient would capture a national swingtoward the Democrats) corroborate conventional wisdom: for example, the

  • 1054 Social Science Quarterly

    largest positive coefficient is associated with the Watergate Babies electionof 1974, and the largest negative coefficient is associated with the RepublicanRevolution election of 1994.Having demonstrated that involvement in a scandal carries a lingering

    electoral cost for incumbentsa cost that carries forward two to three electioncycles beyond the scandal itselfwe now turn to the question of what, ifany, costs incumbent misbehavior might have for the system. Specifically, weinvestigate the relationship between incumbent scandals and voter turnout atthe district level.

    Scandals and Turnout: Concepts, Data, and Methodology

    In order to understand the impact of scandals on district-level turnout,we specified and estimated a model of turnout, controlling for a varietyof theoretically informed socioeconomic, political, and geographical district-level covariates: age (Flinders et al., 2009; Franklin, 2004; Jankowski andStrate, 1995; Clark et al., 2006), education (Comber, 2003; Flanagan, 2003),income (Barkan, 2004; Burns, Schlozman, and Verba, 2001), percent of thepopulation that is urban (Monroe, 1976), percent of the population thatworks in blue-collar jobs, percent of the population that is an ethnic/racialminority (Paulsen, 1991), total campaign spending in the district (Cox andMunger, 1989), a concurrently held Senate or gubernatorial race (Cox andMunger, 1989), the quality of the incumbents challenger, and region. We alsoincluded in the model election-specific dummy variables for each election,leaving 2006 as our baseline.9

    Our dependent variable, Turnout, is the total number of votes cast in eachcongressional district as a percentage of that districts voting-age population.The independent variable of interest, Scandal Allegation, is a dichotomous vari-able coded 1 if an MC is involved in a scandal (using the Ethics-Committee-based approach described above) in that cycle and 0 otherwise. The descriptivestatistics presented inTable 3 reveal that the only indicator of turnout forwhichwe observe a statistically significant difference between scandal-free districtsand districts in which scandal is present is the mean number of votes cast forthe challenger, which increases by 11,503.

    9In tables reporting results of the turnout models, we have tried to use variable labelsthat make the operationalizations and measurements of these variables transparent. Variousdistrict characteristics like Age 1824 and College Educated capture the percent of the districtpopulation that falls into each category according to the Census for that decade.Opponent Scoreand Personal Advantage take the same form they took in the model of incumbent reelectionmargins. Our regional dummies utilize the ICPSR categorization scheme, and we estimate themodel with the East region as the omitted category. The election-specific dummy variablesare included in the model but omitted from the table reporting the results. Since the CensusBureau only collects data every 10 years, for elections held between censuses, we assumed alinear increase (or decrease) of each socioeconomic variable and included the adjusted valuefor each year. We also estimated a model with reported, rather than adjusted, values on thesevariables, and our results were unchanged.

  • The Lingering Effect of Scandals in Congressional Elections 1055

    TABLE 3Scandal and District-Level Turnout, 19722006

    Indicator Scandal- Incumbent Involvedof Turnout Free Districts in Scandal Difference

    Mean percentage(voting-agepopulation)

    46.0% 46.27% +0.27

    Mean N of votes cast(all candidates)

    180,932 184,020 +3,088

    Mean N of votes cast(incumbents only)

    118,178 109,763 8,415

    Mean N of votes cast(challengers only)

    62,754 74,257 +11,503

    Bivariate difference of means (independent samples t-test) statistically significant atp 0.01 (two-tailed test).

    Scandals and Turnout: Model Estimation and Results

    To evaluate the impact of scandals on district-level turnout, we estimatea multilevel model (MLM) of turnout.10 Diagnostics reveal autocorrelationwithin the error terms due to the fact that districts with high turnout in oneelection tend to have high turnout in the next. We eliminate this first-orderautocorrelation through the addition of random district intercepts (OConnelland McCoach, 2008).11 Our results are presented in Table 4 in the columnlabeled Turnout (Model 1).Holding everything else constant, the model predicts that turnout increases

    moderatelyjust under 1.6 percentage pointsin elections in which theincumbent is the subject of an investigation by the House Ethics Committee.While the estimate falls short of statistical significance at traditional levels(p = 0.081; two-tailed test), this finding is consistent with Welch and Peters(1980), who report that severe instances of scandal (which we believe arecaptured in our operationalization of the concept) are associated with modestincreases in turnout in the subsequent election. We also note that all of ourcontrol variables (except one region indicator) are statistically significant andhave the expected sign.Of course, the real-world impact of a 1.58 percentage point increase might

    seem rather modest, especially if it were spread approximately equally acrossboth candidates in the race. According to U.S. census figures, the averagecongressional district in the 1990s had a voting-age population of 424,164.

    10The same specification, estimated using GLS regression with robust standard errors andCochrane-Orcutt regression, yields similar results.

    11So-called null models, which partition variance into within- and between-district variance,confirm that there is significant between-district variance, which we capture in the modelthrough the random intercepts.

  • 1056 Social Science Quarterly

    TABLE 4Impact of Scandal on Turnout in Congressional Elections, 19722006

    Turnout Incumbents Challengers(Model 1) (Model 2) (Model 3)

    Fixed EffectsScandal allegation 1.580 6435.983 11,915.420

    (0.904) (2931.878) (2277.387)Urban population 0.093 55.761 78.748

    (0.010) (32.022) (22.807)Blue-collar workers 0.097 218.912 46.197

    (0.027) (86.623) (65.167)Nonwhite population 0.208 659.272 508.308

    (0.010) (32.777) (23.772)Age 1824 0.727 2,052.251 922.764

    (0.081) (260.079) (190.273)College educated 0.218 1,088.327 241.548

    (0.032) (102.520) (75.755)Median family income (ln) 2.771 13,807.070 4,096.976

    (1.103) (3,555.491) (2,590.255)Total campaign spending ($millions)

    1.584 11,535.21 19,424.48(0.251) (813.082) (629.046)

    Total campaign spending2($ millions)

    0.150 1,062.386 1,768.081(0.036) (115.645) (89.696)

    Concurrent Senateelections

    0.699 2,923.255 462.649(0.170) (550.573) (429.604)

    Concurrent gubernatorialelections

    2.588 6,355.458 4,829.735(0.353) (1,140.243) (859.213)

    Years in the House 0.027 153.925 24.623(0.012) (39.819) (30.605)

    Opponent score 0.213 2,707.957 3,354.555(0.123) (398.017) (309.003)

    Personal advantage 0.013 124.460 141.725(0.004) (12.571) (9.716)

    South 0.396 578.672 5,261.038(0.623) (1971.758) (1244.027)

    Midwest 4.621 12,943.490 8,153.422(0.624) (1971.546) (1224.646)

    West 5.162 8,411.411 9,235.041(0.654) (2065.833) (1283.149)

    Intercept 20.059 284,929.900 13,452.180(11.282) (36,338.810) (26,410.830)

    Random EffectsIntercept 18.634 1.84e+08 6.32e+07

    (1.451) (1.46 10+07) (5,618,433)Residual 32.215 3.39 10+08 2.07 10+08

    (0.648) (6,818,308) (4,159,845)Log-likelihood 17,651.978 61,376.275 59,855.271N 5,445 5,445 5,438

    NOTE: Congress-specific dummy variables are omitted from the table; complete resultsavailable from the authors upon request. Dependent variable in Turnout Model is thepercent of a districts voting-age population that cast a ballot in the congressional race.Dependent variables in the Incumbent and Challenger models are the number of votescast for each. Numbers in cells are maximum likelihood coefficients of multilevel modelswith random district intercepts, estimated using Statas xtmixed command (standard errorsin parentheses).p 0.1; p 0.01; p 0.001 (two-tailed tests).

  • The Lingering Effect of Scandals in Congressional Elections 1057

    Turnout in congressional elections in presidential election years in that decade(1992, 1996, and 2000) averaged 50.46 percent (214,033 votes cast) in thenations 435 congressional districts. So, a bump of 1.58 percentage points to52.04 percent in a presidential election year would represent an increase ofonly about 6,702 in the total number of votes cast (to 220,734) in the averagecongressional race in those years.However, the descriptives presented in Table 3 suggest that modeling

    turnout at the aggregate level may mask interesting patterns present in thedisaggregated data. So, we also estimate separate multilevel models of the rawnumber of votes cast for incumbents and for challengers. The results of thesemodels (presented in the final two columns in Table 4) reveal that the boostin turnout accrues entirely to challengers. In fact, involvement in a scandalreduces an incumbents vote total by an estimated 6,436 (p 0.01) votesonce we control for other factors (Model 2). In contrast, challengers runningagainst incumbents involved in scandals enjoy an increase of 11,915 (p 0.001) votes over those running against scandal-free incumbents (Model 3).So, while races involving incumbents embroiled in scandals appear to experi-ence a small boost in turnout, disaggregating the analysis reveals that it is thesubstantial increase in the number of voters casting their ballots for challengersthat is entirely responsible for this result.

    Discussion

    Taken together, the results of the two parts of our analysis beg a simple, yetsurprisingly unaddressed, question: What is the causal process through whichscandal-plagued incumbents electoral margins decline by such considerableamount?Does the decline in support stem primarily from reduced turnout dueto the demobilization of previous supporters of the incumbent? Do challengersmobilize new voters intent on throwing the bumout?Doprevious supportersof the incumbent convert enmasse and cast their lot with the challenger?Or arethe aggregate numbers we observe driven by some complex and unrecoverablecombination of all of these mechanisms? While Peters and Welch (1980:699700) provide a thorough discussion of the various mechanisms through whichscandals might affect incumbents vote shares, they ultimately conclude thatthis question is unanswerable absent unavailable individual-level behavioraldata (including longitudinal data on turnout, awareness of the scandal, andvote choice). However, some of the possible causal mechanisms they identifyare difficult to reconcile with patterns we find in the district-level data onmargins and turnout. For example, to the extent that elections involvingincumbents embroiled in scandals experience a modest bump in turnout, ourresults reveal that this boost in turnout accrues exclusively to challengers.Furthermore, the number of votes gained by challengers far outstrips thenumber lost by scandal-plagued incumbents. Thus, even the implausible totalconversion scenario, in which all of the voters who abandon the incumbent

  • 1058 Social Science Quarterly

    switch their allegiances to the challenger, would still require themobilization ofthousands of previous nonvoters to generate the estimated bump in challengervotes.12 And of course the more realistic scenario in which some previoussupporters of the scandal-plagued incumbent convert to support the challengerbut many simply chose too stay home implicates even more strongly themobilization of previous nonvoters intent on throwing the bum out as one ofthemechanisms throughwhich scandal reduces incumbents electoralmargins.

    Conclusion

    Consistent with previous studies, we find that incumbents involved inscandals suffer at the polls when they choose to run for reelection. Usingthe literature on brand crises to frame our expectations, we offer the firstsystematic test of the hypothesis that the impact of scandal on an incumbentselectoral margin might linger beyond the election cycle in which the scandaloccurred. We find evidence of this lingering effect and demonstrate thatincumbents involved in scandals do not return to their scandal-free predictedmargins of victory until four to six years after the scandal cycle. However,incumbents who survive their initial reelection bid do recover a substantialamount (just under two-thirds) of the immediate loss after just one cycle,quickly moving the average incumbent back out of any danger zone intowhich he or shemight have tumbled in the immediate aftermath of the scandal.We also find that turnout rates increase slightly in the wake of a scandal, butthis boost is not reflected in an increase in the number of votes cast for theincumbent. In fact, the number of votes cast for scandal-plagued incumbentsdeclines significantly. Instead, this turnout bonus appears to accrue exclusivelyto challengers, leading us to conclude that any satisfactory description ofthe causal mechanism through which scandals depress incumbents electoralmargins must incorporate the mobilization of previous nonvoters.Future research on this topic should attempt to address at least two limi-

    tations of existing studies. The first limitation involves the specification andestimation of empirical models. Analysts should explore the possibility thatthe impact of scandal on incumbents electoral margins is felt both directly andindirectly (through quality challenger emergence, and possibly other mech-anisms).13 Also, given issues of survivor bias that plague any study of the

    12Here, we follow Peters and Welch, who reason that it would be highly unlikely thatcharges of corruption against a candidate would cause voters to defect to the allegedly corruptcandidate from a position of abstention or support for the noncorrupt candidate (1980:699).

    13Brown (2006:157) provides a depiction of this sort of relationship, but does not providea corresponding statistical model, nor does she conduct an investigation of the extent towhich the univariate and bivariate patterns in the data are consistent with it. Nonetheless, herpresentation could prove to be a useful starting point for such an investigation. Swearingenand Jatkowski (2011) find that the presence of a scandal does increase the amount of moneyspent by the party challenging an open seat (2011:319) but does not improve the quality of thecandidate nominated by the challenging party (2011:318). Our data reveal the same patternholds for races involving incumbents embroiled in scandals: both challenger and incumbentspending increase by over 80 percent when the incumbent is embroiled in a scandal but thereis no difference in challenger quality in the presence or absence of a scandal-ridden incumbent.

  • The Lingering Effect of Scandals in Congressional Elections 1059

    electoral impact of scandals, we believe explicit modeling of retirements, de-feats, and margins would prove fruitful.Second, while experimental studies (e.g., Doherty, Dowling, and Miller,

    2011) hold much promise to discern with confidence the relative importanceof mobilization of previous nonvoters in favor of the challenger, demobi-lization of previous supporters of the incumbent, and conversion by formersupporters of the incumbent to the challenger, analysts will need access toindividual-level, multi-election panel data on vote choice, turnout, and aware-ness of scandals that do not exist at this time. Until those data exist, however,our best bet is to continue exploring the impact of scandal through the sortof aggregate analysis we present here. The swings of the last several electioncycles notwithstanding, the number of House seats actually in play in anygiven election remains relatively small. As a result, scandals have emerged asone of the few causal forces capable of dislodging incumbents who representotherwise safe seats. They therefore merit our continued attention despite theevidentiary and inferential difficulties they present.

    REFERENCES

    Abramowitz, Alan I. 1991. Incumbency,Campaign Spending, and theDecline ofCompetitionin the U.S. House Elections. Journal of Politics 53(1):3456.

    Anderson,Christopher J., andYuliyaV.Tverdova. 2003. Corruption, Political Allegiances, andAttitudes Toward Government in Contemporary Democracies. American Journal of PoliticalScience 47(1):91109.

    Barkan, Steven E. 2004. Explaining Public Support for the Environmental Movement: ACivic Voluntarism Model. Social Science Quarterly 85(4):91337.

    Boatright, Robert G. 2013, Getting Primaried: The Changing Politics of Congressional PrimaryChallenges. Ann Arbor, MI: University of Michigan Press.

    Bowler, Shawn, and Jeffrey A. Karp. 2004. Politicians, Scandals, and Trust in Government.Political Behavior 26(3):27187.

    Brown, LaraMichelle. 2006. Revisiting the Character of Congress: Scandals in the U.S. Houseof Representatives, 19662002. Journal of Political Marketing 5(1/2):14972.

    Burns, Nancy, Kay Lehman Schlozman, and Sidney Verba. 2001. The Private Roots of PublicAction: Gender, Equality, and Political Participation. Cambridge, MA: Harvard University Press.

    Chen, Yubo, Shankar Ganesan, and Yong Liu. 2009. Does a Firms Product-Recall StrategyAffect Its Financial Value? An Examination of Strategic Alternatives During Product-HarmCrises. Journal of Marketing 73(6):21426.

    Clark, Nick, Clive Barnett, Paul Cloke, and Alice Malpass. 2006, Globalising the Consumer:Doing Politics in an Ethical Register. Political Geography 26(3):23149.

    Comber, Melissa K. 2003. Civics Curriculum and Civic Skills: Recent Evidence. Circle:The Center for Information and Research on Civic Learning and Engagement. Available athttp://www.civicyouth.org/PopUps/FactSheets/FS_Civics_Curriculum_Skills.pdf.Cox, Gary W., and Michael C. Munger. 1989. Closeness, Expenditures, and Turnout in the1982 U.S. House Elections. American Political Science Review 83(1):21731.

  • 1060 Social Science Quarterly

    Dawar, Jiraj, and Madan M. Pillutla. 2000. The Impact of Product Harm Crises on BrandEquity: The Moderating Role of Consumer Expectations. Journal of Marketing Research37(2):21526.

    Doherty, David, Conor M. Dowling, and Michael G. Miller. 2011, Are Financial or MoralScandals Worse? It Depends. PS: Political Science and Politics 44(4):74957.

    Engstrom, Erik J. 2012. Congressional Elections: Electoral Structure and Political Represen-tation. Pp. 4561 in Jamie L. Carson, ed., New Directions in Congressional Politics. New York:Routledge.

    Flanagan, Constance A. 2003. Developmental Roots of Political Engagement. PS: PoliticalScience & Politics 36(2):25761.

    Flinders, Matthew, Andrew Gamble, Colin Hay, and Michael Kenny. 2009. The OxfordHandbook of British Politics. Oxford, MI: Oxford University Press.

    Franklin, Mark. 2004. Voter Turnout and the Dynamics of Electoral Competition in EstablishedDemocracies. Cambridge, MA: Cambridge University Press.

    Groseclose, Timothy, and Keith Krehbiel. 1994. Golden Parachutes, Rubber Checks, andStrategic Retirements from the 102d House. American Journal of Political Science 38(1):7599.

    Groseclose, Timothy, Steven Levitt, and James Snyder. 1999. Comparing Interest GroupsScores Across Time and Chambers: Adjusted ADA Scores for the U.S. Congress. AmericanPolitical Science Review 93(1):3350.

    Herrnson, Paul S. 2012. Congressional Elections: Campaigning at Home and in Washington, 6thed. Washington, DC: CQ Press.

    Jacobson, Gary C. 1981. Incumbents Advantages in the 1978 U.S. Congressional Elections.Legislative Studies Quarterly 6(2):183200.

    Jacobson, Gary, and Michael Dimock. 1994. Checking Out: The Effects of Bank Overdraftson the 1992 House Elections. American Journal of Political Science 38(3):60124.

    Jankowski, Thomas B., and John M. Strate. 1995. Modes of Participation Over the AdultLife Span. Political Behavior 17(1):89106.

    Kiewiet, D. Roderick, and Langche Zeng. 1993. An Analysis of Congressional Career Deci-sions, 19471986. American Political Science Review 87(4):92841.

    Kostadinova, Tatiana. 2009. Abstain or Rebel? Corruption Perceptions and Voting in EastEuropean Elections. Politics and Policy (37)(4):693714.

    Long, Nicholas Chad. 2011. The Impact of Incumbent Scandals on Senate Elections, 19742008. Presented at the Annual Meeting of the Western Political Science Association (March16). Available at SSRN http://ssrn.com/abstract=1767272.Lott, John R. 1991. Does Additional Campaign Spending Really Hurt Incumbents?: TheTheoretical Importance of Past Investments in Political Brand Name. Public Choice 72(1):8792.

    Maier, Jurgen. 2011. The Impact of Political Scandals on Political Support: An ExperimentalTest of Two Theories. International Political Science Review 32(3):283302.

    Mayhew, David R. 1974. Congress: The Electoral Connection. New Haven, CT: Yale UniversityPress.

    Monroe, AlanD. 1976. Urbanism andVoter Turnout: ANote on SomeUnexpected Findings.American Journal of Political Science 21(1):7178.

  • The Lingering Effect of Scandals in Congressional Elections 1061

    OConnell, Ann A., and D. Betsy McCoach, eds. 2008. Multilevel Modeling of EducationalData. Charlotte, NC: Information Age Publishing.

    Ornstein, Norman J., Thomas E. Mann, and Michael J. Malbin 2008. Vital Statistics onCongress 2008. Washington, DC: Brookings.

    Paulsen, Ronnelle. 1991. Education, Social Class and Participation in Collective Action.Sociology of Education 54(1):96110.

    Peters, John G., and Susan Welch. 1980. The Effects of Charges of Corruption on VotingBehavior in Congressional Elections. American Political Science Review 74(3):697708.

    Praino, Rodrigo, and Daniel Stockemer. 2012a. Tempus Fugit, Incumbency Stays: Measuringthe Incumbency Advantage in the U.S. Senate. Congress & the Presidency 39(2):16076.

    . 2012b. Tempus Edax Rerum: Measuring the Incumbency Advantage in the U.S.House of Representatives. Social Science Journal 49(3):27074.

    Pruitt, Stephen W., and David R. Peterson 1986. Security Price Reactions Around ProductRecall Announcements. Journal of Financial Research 9(2):11322.

    Rundquist, Barry S., Gerald S. Strom, and John G. Peters. 1977, Corrupt Politicians andTheir Electoral Support: Some Experimental Observations. American Political Science Review71(3):95463.

    Siomkos, George J. 1999. On Achieving Exoneration After a Product Safety Industrial Crisis.Journal of Business & Industrial Marketing 14(1):1729.

    Siomkos, George J., and Gary Kurzbard. 1994. The Hidden Crisis in Product-Harm CrisisManagement. European Journal of Marketing 28(2):3041.

    Smith, Steven S., and Christopher J. Deering. 1997. Committees in Congress. Washington, DC:CQ Press.

    Stewart, Charles III, and Jonathan Woon. 2011. Congressional Committee Assignments,103rd to 112th Congresses, 19932011: [House of Representatives], [16 June]. Available athttp://web.mit.edu/cstewart/www/data/data_page.html.Swearingen, C. Douglas, and Walt Jatkowski. 2011. Is Timing Everything? Retirementand Seat Maintenance in the U.S. House of Representatives. Legislative Studies Quarterly36(2):30930.

    Thompson, Dennis F. 1995. Ethics in Congress: From Individual to Institutional Corruption.Washington, DC: Brookings Institution.

    U.S. Congress, House Committee on Ethics. 2004. Historical Summary of Conduct Casesin the House of Representatives, 17982004. Available at http://ethics.house.gov/publication/historical-summary.. n.d. Summary of Activities (various years). Available at http://ethics.house.gov/reports/summary-activities.Van Heerde, Harald, Kristiaan Helsen, and Marnik G. Dekimpe. 2007. The Impact ofProduct-Harm Crisis on Marketing Effectiveness.Marketing Science 26(2):23045.

    Vassilikopoulou, Aikaterini, George Siomkos, Kalliopi Chatzipanagiotou, and Angelos Pan-touvakis. 2009. Product-Harm Crisis Management: Time Heals All Wounds? Journal ofRetailing and Consumer Services 16:17480.

    Welch, Susan, and John R. Hibbing. 1997. The Effects of Corruption on Voting Behavior inCongressional Elections, 19821990. Journal of Politics 59(1):22639.

  • Copyright of Social Science Quarterly (Wiley-Blackwell) is the property of Wiley-Blackwelland its content may not be copied or emailed to multiple sites or posted to a listserv withoutthe copyright holder's express written permission. However, users may print, download, oremail articles for individual use.