INTRODUCTION
In much of the United States, elections are administered by partisan elected officials rather than nonpartisan bureaucrats. This sets the United States apart from other advanced democracies and leads many experts to worry that election officials give their party an unfair advantage. When asked whether election officials are impartial, election experts rank the United States 31 out of 34 OECD countries, ahead of only Hungary, Mexico, and Turkey (Norris and Grömping Reference Norris and Grömping2019). Many members of the public are also worried about the American way of conducting elections. According to an ABC News/Ipsos Poll conducted in 2021, 41% of Americans are not so confident or not at all confident in the integrity of the U.S. electoral system.Footnote 1 In the fall of 2020, Gallup reported that the share of people who were confident in the accuracy of U.S. elections matched its all-time low.Footnote 2 These widespread concerns about election integrity raise an important empirical question: do partisan local election officials give their party an advantage?
Political economy models of elections disagree about whether directly elected local election officials will advantage their party. Candidates improve their chances of winning by moderating their positions and may therefore run elections in a similar manner (Downs Reference Downs1957). On the other hand, relatively extreme candidates are more likely to run than moderates, and this may result in distinctive Republican and Democratic ways of administering elections that tend to benefit co-partisans (Besley and Coate Reference Besley and Coate1997; Osborne and Slivinski Reference Osborne and Slivinski1996). These standard models may do a poor job of describing the considerations of people running to be a local election official. For example, the set of qualified candidates may hold relatively similar views on election administration regardless of their party affiliation (Manion et al. Reference Manion, Anthony, Kimball, Udani and Gronke2021).
Sorting out how much of an advantage clerks give their party is difficult.Footnote 3 Democratic clerks are more likely to serve in places where more people vote for Democrats for president, congress, and statewide office. But this does not tell us that clerks advantage their party; voters may simply prefer candidates from the same party in many offices.
We overcome this problem using a close-election regression discontinuity design, comparing Democratic presidential vote share in counties that narrowly elected a Democratic clerk to those that narrowly elected a Republican clerk. To do so, we build an original dataset of 5,880 clerk elections in 1,313 counties from 1998 to 2018. This design ensures that the differences we observe arise from who administers elections rather than pre-existing differences in citizen preferences or local conditions. Using election results as our primary outcome also allows us to evaluate the downstream consequences of partisan clerk elections rather than infer them from changes in policy.
Despite widespread concern that partisan election officials advantage their party, we find that Democratic and Republican election officials oversee similar election outcomes when serving comparable counties. We estimate that partisan clerks give their party an advantage of less than 0.4 percentage points. Three of our four estimators can detect an effect of 1.7 percentage points or smaller with 80% power. While our year-by-year estimates are noisier, we find that the effect on Democratic vote share is similar in every presidential election from 2004 to 2020. We also present evidence that even clerks who win in a landslide do not noticeably advantage their party and that Democratic and Republican clerks from comparable counties oversee elections with similar turnout and policies.
Why do elected clerks not advantage their party? We provide evidence that clerks do not advantage their party even when they no longer face reelection, suggesting that the reelection incentive is not the primary moderating force on clerks. Clerks who are most able to independently affect statewide outcomes also do not advantage their party, suggesting that collective action problems may not be the main reason clerks fail to advantage their party. Instead, we explain our main findings by pointing to existing research that suggests clerks are more likely than the general public to agree on election administration issues across parties and that election administration may only modestly affect electoral outcomes.
While we find that Democratic and Republican election officials oversee elections with similar outcomes, we cannot rule out small differences between Democratic and Republican officials that could determine very close elections. We also cannot rule out rare but very large effects. If a few election officials dramatically change the outcomes of elections they oversee, the effect in those counties would make up a small share of the average effect and be drowned out by the many officials who do not advantage their party. Still, we find that the average effect of replacing a Republican local election official with a Democrat is small, suggesting that most local election officials are not meaningfully biasing elections in their party’s favor. Additionally, our results pertain only to county election officials in past elections. It is possible that partisan election officials at the state level or future county officials are able to bias elections in their party’s favor. Finally, our analysis does not imply that electing partisan officials is the best way to select local election officials. Nonpartisan appointed officials may perform better than partisan elected officials (Ferrer Reference Ferrer2022).
PARTISAN ADVANTAGE IN LOCAL ELECTION ADMINISTRATION
Should we expect clerks to advantage their party? Canonical theories of electoral competition reveal that candidates whose policies more closely resemble the median voter’s preferred policy are more likely to win reelection, which leads politicians from both parties to implement similar policies (Downs Reference Downs1957; Fearon Reference Fearon, Mann, Przeworski and Stokes1999). This reelection incentive is especially powerful for executives with meaningful discretion, like governor or mayor, who are especially likely to produce similar outcomes across parties because they make unilateral choices that directly affect their constituents’ lives (Mayhew Reference Mayhew1974). The role of clerk has many of these qualities: the elected official has considerable discretion over local election administration and citizens directly observe their performance when they vote or communicate with the office (Burden et al. Reference Burden, Canon, Lavertu, Mayer and Moynihan2013). However, elected partisan clerks must raise money for their campaign and win a partisan primary. These additional steps mean that candidates have to satisfy donors and primary voters who may prefer candidates that administer elections in their preferred way or even promise to tilt the scales in their party’s direction (Ansolabehere, Snyder, and Stewart Reference Ansolabehere, Snyder and Stewart2001; Brady, Han, and Pope Reference Brady, Han and Pope2007; Burden Reference Burden2004). This incentive to shift policy away from the median voter’s position may be especially strong in places where an overwhelming majority of citizens favor one party.
Citizen-candidate models point out that candidates with moderate policy preferences are unlikely to run if elections are costly because these potential candidates will often be nearly indifferent between the other candidates running (Besley and Coate Reference Besley and Coate1997; Osborne and Slivinski Reference Osborne and Slivinski1996). Candidates with more extreme policy positions will have relatively more reason to run. This is especially true when the office confers few benefits and running is costly.Footnote 4 Elected county clerks often receive modest pay (Adona et al. Reference Adona, Gronke, Manson and Cole2019), and running for office requires campaigning which many citizens might view as costly. Given these conditions, we would expect only committed partisans to run for clerk and then implement different policies across parties.
There are three potential countervailing forces within the citizen-candidate model leading clerks to not advantage their party. First, people with experience in election administration may have less polarized election policy views across parties than the public and elections may select for people with experience (Manion et al. Reference Manion, Anthony, Kimball, Udani and Gronke2021; Thompson Reference Thompson2020). Second, Democratic and Republican clerks may want to implement different policies, but if they were to do so they would not be able to noticeably influence turnout or partisan vote share (e.g., Gronke et al. Reference Gronke, Galanes-Rosenbaum, Miller and Toffey2008; Thompson et al. Reference Thompson, Wu, Yoder and Hall2020). Third, clerks may face costs for changing policies that are only worth bearing if they can influence who wins. This creates a collective action problem: each clerk wants to help their party win but shirks to avoid bearing the costs because they are not pivotal by themselves. This collective action problem is not present when the costs of influencing elections are low and the likelihood of influencing election outcomes is high.
Empirical research directly testing whether U.S. local election officials favor their party, which we review in Table A.1 in the Supplementary Material, is mixed. While some studies find that Democratic and Republican officials implement different policies and other studies find they do not, no study has a research design that can fully account for differences in the places that elect Democratic and Republican clerks that might lead to different policies regardless of which party controls the clerk’s office.
The risk of partisan election administration is not limited to the United States. While everyone agrees that election administrators ought to ensure “free and fair elections” (Hall Reference Hall, Herron, Pekkanen and Shugart2018), it is difficult to completely insulate election administration from partisan actors (James Reference James2012). Central election management bodies are most effective when they are independent of the executive (López-Pintor Reference López-Pintor2000), but in practice partisan actors are involved in virtually every system (Massicotte, Blais, and Yoshinaka Reference Massicotte, Blais and Yoshinaka2004). One notable example of partisan election administration comes from Ukraine, where party control of election management committees boosts that party’s vote totals by a few percentage points (Herron Reference Herron2020). While outright fraud is certainly a factor in many places (Alvarez, Hall, and Hyde Reference Alvarez, Hall and Hyde2008), practices that amount to a “soft perversion of the process” are even more common, such as appointing biased poll workers (Alvarez and Hall Reference Alvarez and Hall2006) and filtering out candidates from the opposing party (Szakonyi Reference Szakonyi2022). Independent election monitors may curtail election day fraud and violence (Asunka et al. Reference Asunka, Brierley, Golden, Kramon and Ofosu2019), but they may simply shift fraudulent practices to earlier in the process (Daxecker Reference Daxecker2014).
THE ROLE OF LOCAL ELECTION OFFICIALS
Across the United States, thousands of local election officials play a central role in the administration of elections. Clerk responsibilities include registering voters, maintaining an up-to-date list of registered voters, hiring and training poll workers, selecting poll locations, printing ballots, acquiring and maintaining election equipment, running early and absentee voting, educating and communicating with voters, overseeing election day, tabulating the votes cast, handling provisional ballots, and certifying election results (Kimball and Kropf Reference Kimball and Kropf2006). They also usually have the authority to hire staff and influence department funding levels.
Clerks administer elections within the bounds of complex and frequently changing federal, state, and local laws. They work in concert with a range of other officials to successfully conduct elections. Clerks typically serve at the county level, though in 10 mostly Northeastern states important responsibilities are carried out at the municipal level.
Building on the work of Kimball and Kropf (Reference Kimball and Kropf2006), we conduct a review of state and local election laws. Table A.2 in the Supplementary Material shows a simplified division of states into tiers based on how much authority is vested in a single partisan elected official. We identify 32 states that contain at least some jurisdictions with a partisan elected official tasked with election responsibilities. In many of these states, partisan elected officials share responsibilities with other local officials or with boards. In 21 of these 32 states, partisan elected clerks are the sole or primary election administrators. Our main analysis focuses on partisan elected officials in these 21 states.Footnote 5
Even among states that delegate considerable election administration authority to a partisan elected official, there are significant differences in clerks’ responsibilities and discretion. We describe this variation in Table A.3 in the Supplementary Material. For example, county clerks in Nevada have complete authority to register voters, maintain the registration list, site polling places, conduct early voting, and purchase voting equipment. They also have some discretion in recruiting poll workers and are not subject to any statewide training requirements. In contrast, probate judges in Alabama do not register voters or maintain registration lists. They are constrained by state law in recruiting poll workers, and both site polling places and select voting equipment in conjunction with the county commission.
Overall, most of the 21 states give registration and voting administration duties to the same partisan elected official. Most also entrust registration list maintenance and voting equipment decisions to this official. Partisan elected officials choose polling places in 14 states and administer early voting in 13 states, but are usually limited in their ability to hire poll workers, with most states requiring bipartisan appointments.
Clerks could plausibly affect election results with formal or informal practices. Using formal authority, clerks could attempt to increase participation and shift the composition of the electorate by siting many polling places in populated and accessible locations, providing extensive early voting options, ensuring that no eligible voters are removed from the voter roll, purchasing easy-to-use and reliable voting equipment, adequately resourcing polling locations with ballots and poll workers, and showing leniency in their acceptance of provisional and vote-by-mail ballots. Alternatively, officials might minimize participation and alter the composition of the electorate by siting polling places in inconvenient locations, providing limited early voting options, regularly purging voters from the rolls, maintaining old and difficult-to-use voting equipment, inadequately sourcing polling locations, and rejecting borderline provisional and vote-by-mail ballots.
Clerks might also undertake informal practices to reduce voter costs or do only what the law requires. Officials can conduct voter outreach campaigns, advertise how and where to register, maintain an active social media presence, and engage in extensive constituent communication. Alternatively, they could take none of these actions. Local election officials can engage in targeted practices by attempting to increase participation among co-partisans and reduce participation among citizens from the opposing party. Finally, officials could take illegal actions at the risk of litigation. These include siting fewer polling places than the statutory minimum,Footnote 6 following procedures that infringe upon the Voting Rights Act, and engaging in vote manipulation.
By estimating the effect of partisan election administrators on Democratic presidential vote share, we measure the sum total effect of all actions election officials take to influence elections.
STUDYING PARTISAN CONTROL OF LOCAL ELECTION OFFICES
In this section, we first describe our data including original data on the elections of local election officials, county-level election results and turnout for presidential and statewide offices from 2000 to 2020, and county-level administrative data on the number and location of polling places, the number of registered voters, the number of provisional ballots, and survey-reported wait times. Next, we discuss our close-election regression discontinuity design and how we improve the precision of our estimates by first predicting outcomes.
New Data on the Elections of Partisan Local Election Officials
We gather an original dataset of 5,880 elections of partisan local election officials in 1,313 counties and 21 states held between 1998 and 2018. We collect these results in three steps. First, we scrape state election websites for all county-level results. Next, we visit county election websites for results not available from states. Finally, we contact counties directly to request results not available on their websites.
Figure 1 shows the counties for which we have at least some data in light blue. Counties with partisan elected election officials where we are unable to find any election data are in dark blue. We use dark gray to denote counties where municipalities run elections, boards share responsibilities for elections, or election officials are appointed or nonpartisan. In Table A.4 in the Supplementary Material, we present descriptive statistics for the counties in and out of our sample, as well as out of scope counties. Missing counties tend to be less populous, located in the South, and have larger Black and Hispanic populations.Footnote 7
Figure 1. Map of Counties Included in Original Data on the Elections of Partisan Local Election Officials
Note: Out of 1,582 counties that elect a partisan election official, 1,313 appear in our dataset at least once. Alaska and Hawaii do not have elected partisan election officials. “Not in Scope” indicates jurisdictions that did not elect partisan local election officials between 1998 and 2018.
Notably, the correlation between Democratic presidential vote share and Democratic clerk vote share is very low. Among counties in our sample that election elect local election officials on a presidential year cycle, Democratic presidential vote share correlates with lagged Democratic presidential vote share with a coefficient of 0.89. By contrast, Democratic clerk vote share correlates with same-year Democratic presidential vote share with a correlation coefficient of 0.32. Figure A.1 in the Supplementary Material captures this pattern.
County-Level Election Results and Voter Participation
We obtain county-level presidential election results for 1996 to 2020 from Dave Leip’s Election Atlas.Footnote 8 We also compile data on every regularly scheduled governor election from 1994 to 2017 and every regularly scheduled U.S. senate election from 1994 to 2020 from Leip’s Atlas, as well as the number of votes counted in the race for the highest federal office on the ballot—either representative, senator, or president.Footnote 9
We measure turnout as the share of voting age residents who cast valid ballots for the highest office. Voting age population is measured using estimates from the National Cancer Institutes’s Surveillance, Epidemiology, and End Results Program.Footnote 10
County-Level Data on Election Administration
We assemble a set of indicators on how elections have been run over time and across counties using the Election Administration and Voting Survey (EAVS) from the U.S. Election Assistance Commission.Footnote 11 We use this survey to measure the following for each federal general election in every county: the number of polling places, provisional ballots cast, provisional ballots rejected, absentee ballots rejected, and the number of registrants removed from the voter roll. We use Dave Leip’s Election Atlas to measure the number of registered voters in each county and the share of registered voters listed as members of the Democratic party.
Additionally, we follow Pettigrew (Reference Pettigrew2017) in using the Cooperative Congressional Election Study to measure voter wait times.Footnote 12 We compute the share of voters who had to wait at the polls for more than 30 minutes for each federal general election between 2006 and 2018, except for 2010 when the CCES did not ask about wait times. We also use data from Chen et al. (Reference Chen, Haggag, Pope and Rohla2022) who measure wait times by tracking cell phone locations.
Empirical Strategy: Regression Discontinuity Design
We estimate the advantage election officials give their co-partisans using a regression discontinuity design, fitting regression equations of the form:
$$ \begin{array}{rll}{Y}_{ct+k}=\mu +\tau De{m}_{ct}+f({M}_{ct})+{\epsilon}_{ct+k,}& & \end{array} $$
where
$ {Y}_{ct+k} $
is Democratic presidential vote share in elections held k years after the election official was elected in county c, year t.
$ De{m}_{ct} $
is a dummy variable indicating a Democratic local election official winning the election.
$ f({M}_{ct}) $
is a flexible function of the margin
$ {M}_{ct} $
by which the Democratic local election official won (i.e., the share of the two-party vote they received minus 0.5).
$ {M}_{ct} $
ranges from −0.5 to 0.5 and is positive for a Democratic win, negative for a Republican win, and zero in an exact tie. We interpret
$ \tau $
as the average effect of electing a Democratic rather than Republican local election official in counties where the election was an exact tie. In other words, it is the effect of electing the next most likely or marginal Democrat to be a local election official rather than a Republican.
In our turnout and policy analyses, when each clerk election determines control of the office for multiple observations of the outcome, we cluster standard errors by clerk election (Abadie et al. Reference Abadie, Athey, Imbens and Wooldridge2017).
Our close-election regression discontinuity design ensures that, when we compare counties that elect a Republican to those that elect a Democrat, both sets of counties have a similar average partisan makeup, state political environment, preferences over election administration, and population, in addition to any other fixed and time-varying county factors. Our regressions identify the average effect of electing a Democratic rather than Republican election official in places with tied elections when the only thing that changes sharply at that point is which candidate was elected (Cattaneo, Idrobo, and Titiunik Reference Cattaneo, Idrobo and Titiunik2019; Imbens and Lemieux Reference Imbens and Lemieux2008; Lee and Lemieux Reference Lee and Lemieux2010).Footnote 13 We evaluate the plausibility of this assumption by comparing pre-election county-level characteristics in counties that narrowly elected Democratic officials to those that narrowly elected Republicans. We are most interested in the comparison of turnout and Democratic presidential vote share from before the local election official was elected because these are our primary outcomes of interest, and because they tend to correlate highly within a county over time. In Section A.6 in the Supplementary Material, we show that counties where a Democratic election official narrowly won are similar to counties where a Republican narrowly won on a large number of pre-treatment characteristics, including the lagged Democratic presidential vote share and lagged turnout. In Section A.6.2 in the Supplementary Material, we also show that Democrats and Republicans win close races at similar rates in counties controlled by Democrats at the time of the election and those controlled by Republicans.Footnote 14 These results serve as evidence to support our claim that the only difference between a district that narrowly elects a Democrat and a district that narrowly elects a Republican is the partisanship of the elected clerk.
Our intention is to estimate the effect of replacing a marginal Republican with a marginal Democrat, which is identified under the assumptions we mention above. Our design does not identify the effect of a candidate changing the party they associate with or the effect of replacing a typical Republican with a typical Democrat (Hall Reference Hall2019, Chapter 2; Marshall Reference Marshall2021).
We present results using a variety of regression specifications because of the bias-variance tradeoff that must be resolved in every regression discontinuity analysis. If the functional form of the running variable is not flexible enough, it can induce bias, mistaking a smooth curve in the outcome for a discontinuity. On the other hand, less flexible specifications that use more data and fewer degrees of freedom make the estimate more precise. Presenting multiple specifications ensures the robustness of our results across different functional forms of the relationship between Democratic election official vote share and our outcomes. Following Cattaneo, Idrobo, and Titiunik (Reference Cattaneo, Idrobo and Titiunik2019), our primary specification is a local linear regression using triangular kernel weights and the automated bandwidth selection procedure described in Calonico, Cattaneo, and Titiunik (Reference Calonico, Cattaneo and Titiunik2014).
Improving Precision by First Predicting Outcomes
One of the main challenges we face when estimating the advantage clerks give their party is statistical precision. Estimating discontinuities is difficult—across many applications, the common estimators produce large standard errors and do not have sufficient power to detect substantively interesting effects (Stommes, Aronow, and Sävje Reference Stommes, Aronow and Sävje2021).
We improve the precision of our estimates using a three-step procedure building on the recommendations of Lee and Lemieux (Reference Lee and Lemieux2010):Footnote 15
-
1. Using leave-one-out cross-validation, we select a regression specification that best predicts Democratic presidential vote share from lagged Democratic presidential vote share.Footnote 16 We use the full dataset for this exercise, not just the counties with competitive elections for their local election official. This procedure selects a prediction equation with state-year-specific coefficients on the lag and state-year-specific intercepts.
-
2. We compute the difference between predicted and observed Democratic vote share using the best-performing specification.
-
3. We use the residual from step 2 as the outcome in a standard regression discontinuity estimator.Footnote 17
We use this procedure to improve our power for our main findings and for studying voter turnout and election policies.
We conduct power analyses to evaluate whether this more precise estimator is powerful enough to detect substantively meaningful effects. We report the minimum effect detectable 80% of the time with a one-sided t-test at a 5% significance level (i.e.,
$ \alpha =0.05 $
and
$ \beta =0.20 $
). We discuss our approach to calculating power in Section A.5 in the Supplementary Material.
As we report in Table 1, our main estimators have a minimum detectable effect of Democratic election officials on Democratic presidential vote share of between 1.2 percentage points and 2.3 percentage points. That means our design has sufficient power to detect effects on partisan vote share that are about as large as running 50 television ads (Sides, Vavreck, and Warshaw Reference Sides, Vavreck and Warshaw2021; Spenkuch and Toniatti Reference Spenkuch and Toniatti2018) or 15% as large as the effect of nominating a moderate candidate (Hall Reference Hall2015). Our minimum detectable effect is also approximately half the size of the effect of Democratic local election officials on the Democratic share of turnout reported in previous research (Bassi, Morton, and Trounstine Reference Bassi, Morton and Trounstine2009). In Table 2, we report that our estimators have minimum detectable effects of Democratic election officials on turnout of between 1.0 percentage points and 1.1 percentage points. Our minimum detectable effect on turnout is less than half the size of a large TV advertising campaign in a presidential election (Green and Vavreck Reference Green and Vavreck2008).
Table 1. Effect of Democratic Election Officials on Democratic Presidential Vote Share
Note: Robust standard errors in parentheses. The outcome is first regressed on a state- and year-specific lag using all counties including those for which clerk election results are not available. The regression discontinuity is estimated using the residuals from that regression. The bandwidth row reports the maximum clerk win margin allowed for inclusion in each specification. CCT refers to Calonico, Cattaneo, and Titiunik (Reference Calonico, Cattaneo and Titiunik2014) bandwidth selection procedure. Min. detectable effect refers to the minimum effect that a one-sided test with a 0.05 alpha would have 80% power to detect.
Table 2. Effect of Democratic Election Officials on Turnout
Robust standard errors clustered by clerk election in parentheses. Rep. counties are those where the last Republican presidential candidate received more votes than the last Democratic presidential candidate. Dem. counties are all remaining counties. The outcome is first regressed on a state- and year-specific lag using all counties including those for which clerk election results are not available. The regression discontinuity is estimated using the residuals from that regression. The bandwidth row reports the maximum clerk win margin allowed for inclusion in each specification. CCT refers to Calonico, Cattaneo, and Titiunik (Reference Calonico, Cattaneo and Titiunik2014) bandwidth selection procedure. Min. detectable effect refers to the minimum effect that a one-sided test with a 0.05 alpha would have 80% power to detect. Unif. refers to a uniform kernel. Tri. refers to a triangular kernel.
CLERKS DO NOT MEANINGFULLY ADVANTAGE THEIR PARTY
Descriptive Graphical Evidence Suggests Clerks Do Not Advantage Their Party
First, we show descriptive graphical evidence that presidential candidates from the clerk’s party perform no better than expected based on historical election results. Figure 2 captures this result. In the top panel, we plot the regression of Democratic presidential vote share for each county-year on Democratic vote share in the previous presidential election. Counties with a Democratic clerk are colored blue and counties with a Republican clerk are colored red. We fit separate locally weighted regressions for counties with Democratic and Republican clerks.
Figure 2. Democratic and Republican Election Officials Conduct Elections with Similar Results
Note: The top panel presents the relationship between Democratic presidential vote share and lagged Democratic presidential vote share separately in counties with Democratic and Republican clerks. The relationship is nearly identical in both sets of counties. The bottom panel presents the distribution of the residuals from predictions of Democratic presidential vote share in counties with Democratic and Republican election officials. On average, Democratic clerks oversee elections that are slightly less favorable for Democratic presidents than expected.
Counties that vote overwhelmingly for Democratic presidents are also likely to elect Democrats to run their elections. We can see this by noticing that the upper-right quadrant of the plot is made up almost entirely of blue Ds and the bottom-left portion of the plot is primarily composed of red Rs.
Nevertheless, this plot suggests that local election officials are not giving their party a large electoral advantage. We can see this by noticing that the lines are nearly identical. Conditional on being elected in counties with similar historical Democratic vote shares, Democratic and Republican local election officials oversee similar elections. If clerks were advantaging their party and continuing to seek new advantages every cycle, we would expect the blue line to be higher than the red line. That is, Democratic presidential candidates would perform better in counties with Democratic clerks than with Republican clerks after accounting for the normal two-party presidential vote in that county. This figure provides us little reason to suspect that clerks are giving their party a substantial advantage in presidential elections.
The bottom panel of Figure 2 plots histograms of the residual of predicted Democratic presidential vote share for counties with Democratic and Republican clerks.Footnote 18 The histograms overlap substantially, although the histogram for Democrats is shifted slightly to the left and has a modestly wider dispersion.Footnote 19 If clerks were advantaging their party and continuing to seek new advantages each cycle, we would expect the central tendency of the distribution of blue residuals to be shifted to the right of the central tendency of the red residuals indicating that Democratic presidential candidates perform better in counties with Democratic clerks than with Republican clerks after accounting for the expected presidential vote in that county. This implies that Democratic clerks oversee elections that are getting worse, on average, for Democratic presidential candidates.
One important weakness of these plots is that the party of the clerk is often the same in the previous presidential election. If partisan control of the clerk’s office is constant over time and not increasing as the party holds the clerk’s office, this plot would tend to understate the effect partisan control of the clerk’s office on election results. We address this concern in the next section by using a regression discontinuity design which compares places with Democratic and Republican clerks that had an equal likelihood of having a Democratic clerk during the previous presidential election.
Regression Discontinuity Plot Suggests Clerks Do Not Advantage Their Party
Figure 3 captures our main result: local election officials do not improve their party’s vote share in presidential elections. On the horizontal axis, we plot the two-party Democratic vote share in the race for local election official. We subset to elections with a Democratic and Republican candidate both on the ballot and finishing in the top two places. This means that a Democratic official runs elections to the right of 0.5, and a Republican official runs elections to the left of 0.5. On the vertical axis is the residual of Democratic presidential vote share in each county in the first presidential election after the election official was elected. Each of the small gray points represents the election of a county election official and the subsequent presidential election result. The large black points are equal-sized binned averages made up of 25 elections each, computed separately for counties that elect a Democratic clerk and those that elect a Republican. The solid lines are simple linear regression lines fit separately for counties that elect Democratic election officials and those that elect Republicans. We plot data within the bandwidth selected by the automated procedure described in Calonico, Cattaneo, and Titiunik (Reference Calonico, Cattaneo and Titiunik2014).
Figure 3. Electing a Democratic Election Official Rather than a Republican Does Not Noticeably Increase Democratic Presidential Vote Share
Note: Two-party Democratic vote share for contested local election official elections is the running variable, making 0.5 the threshold above which a county elects a Democratic election official and below which they elect a Republican. Democratic presidential vote share in the following presidential election is plotted along the vertical axis. The large black points are equal-sized binned averages marking the average of 25 elections each. The binned averages are computed separately for each side of the 50–50 threshold. The black line is a linear regression fit separately on each side of the 50–50 threshold. The full tabular results are found in column 1 of Table 1.
We can learn about the effect of electing a Democrat rather than a Republican as local election official by focusing on the 50–50 point in the middle of the plot. To the left and right of 0.5, the average residual Democratic presidential vote share is nearly identical. If clerks were advantaging their party, we would expect the average vote share for Democratic presidential candidates to be higher in counties that narrowly elected a Democratic clerk compared to those that narrowly elected a Republican clerk. This would be visible as a vertical jump in the regression line on the plot with the line being noticeably higher on the right side of the 50–50 line than on the left side of the 50–50 line. This suggests that election officials do not noticeably advantage their party.
Regression Estimates also Suggest Clerks Do Not Advantage Their Party
In Table 1, we provide formal estimates of the effect of electing a Democrat rather than a Republican as election official on Democratic presidential vote share. Column 1 reports the estimate from a local linear regression with uniform kernel weights and the bandwidth selected by the procedure described in Calonico, Cattaneo, and Titiunik (Reference Calonico, Cattaneo and Titiunik2014). Column 2 reports estimates from the same procedure used in column 1 but with a bandwidth twice as wide. Column 3 reports estimates from the same procedure used in column 1 but with a bandwidth half as wide. Column 4, our primary specification, reports estimates from a local linear regression with triangular kernel weights and the bandwidth selected by the procedure described in Calonico, Cattaneo, and Titiunik (Reference Calonico, Cattaneo and Titiunik2014).
We find consistent evidence across all four specifications that local election officials do not meaningfully advantage their party’s candidate for president. The point estimates range from −1.1 to 0.3 percentage points, with three out of four point estimates falling below 0.1 percentage points. Across all four columns, our 95% confidence intervals include zero.
In the final row of Table 1, we present the minimum detectable effect. As we discuss in our Empirical Strategy section, three of our four estimators are able to detect partisan advantages as small as 1.7 percentage points with 80% power.
While Table 1 presents results across only four specifications, we estimate very similar effects across a much wider set of potential estimators. Section A.6.4 in the Supplementary Material shows that our estimates are similar for every choice of bandwidth from 0.02 to 0.25. In Section A.6.3 in the Supplementary Material, we demonstrate that, though our estimates are noisier when using outcomes that are not first residualized, they are substantively similar.
In Table A.14 in the Supplementary Material, we extend our data to include all governor, senate, and presidential election results. Despite adding more data, predicting governor and senate election results based on lagged results is more difficult than predicting presidential results, resulting in noisier estimates. Nevertheless, the point estimates are still substantively quite small, and a zero effect falls well within all of the 95% confidence intervals in the table.
Similar Findings across Time and States
This finding—that election officials do not noticeably advantage their party—is not limited to the early part of our study period, to states where officials have slightly less authority, or to regions with distinctive politics. In Figure 4, we present estimates of the effect of electing a Democratic local election official on Democratic presidential vote share in every presidential election since 2004. Despite the concern that election administration has become an increasingly salient and partisan issue, we do not find evidence that the marginal local election official advantaged their party in 2020 or in any previous election since 2004.
Figure 4. Clerks Provide Their Party Minimal Advantages Over Time
Note: Each dot represents a regression discontinuity-based estimate of the effect of electing a Democratic clerk on residual Democratic presidential vote share in a given presidential election. Vertical lines extending from each point represent 95% confidence intervals. Estimates come from regressions that mimic column 4 of Table 1 using local linear regression with traingular kernel weights. Full tabular results are found in Table A.8 in the Supplementary Material.
In the Supplementary Material, we also study three sets of states where we might expect clerks to give their party a larger advantage. Across all three sets, we find that clerks give their party little to no advantage. First, in Table A.9 in the Supplementary Material, we present estimates of the advantage clerks give their party in the 14 states where one partisan elected official handles all local election administration. Three of the four reported point estimates of partisan advantage are negative. Given the long tenure of clerks and the slow pace of the Southern realignment in local offices, we might expect that Democratic clerks in the South may favor the Republican party in statewide and national elections, especially in the first few elections in our data (Kimball et al. Reference Kimball, Kropf, Moynihan and Silva2013). In Table A.11 in the Supplementary Material, we report estimates of the partisan advantage clerks provide, removing counties in Southern states from the analysis. We find substantively similar point estimates, implying that our national estimates are not masking positive effects in places where clerks are most likely to favor national co-partisans. Finally, some counties in our data were subject to pre-clearance requirements under the Voting Rights Act prior to the 2013 Supreme Court ruling in Shelby County v. Holder. In Table A.12 in the Supplementary Material, we find that, even when omitting counties subject to the pre-clearance requirement, clerks do not appear to advantage their party. In Table A.13 in the Supplementary Material, we subset to counties previously covered under the pre-clearance provisions but in years after the Shelby County v. Holder decision, finding a similar pattern of results. In other words, there is no indication that local election officials have used their new discretion post-Shelby to advantage their party.Footnote 20 In addition to these more powerful tests, in Figure A.4 in the Supplementary Material, we also present evidence that clerks do not noticeably advantage their party in any of the eight states that we have sufficient data to study. This suggests that state-level laws are not the primary reason clerks do not advantage their party. Put together, these results suggest that clerks do not meaningfully advantage their party.
Generalizing beyond Close Clerk Elections
Using a regression discontinuity design, we find that clerks elected in close elections do not give their party a substantial advantage in presidential elections. Might clerks elected by wider margins give their party an advantage?
Our data suggest that, even when clerks win by a relatively large margin, they do not grant their party a sizable advantage. In Figure 2, we document the difference in Democratic presidential vote share between counties controlled by Democratic and Republican clerks. Though the majority of these clerks are elected by large margins or in uncontested races, the average Democratic clerk oversees an election with slightly lower Democratic presidential vote share than the average Republican clerk. This descriptive evidence suggests that our finding is not limited to counties with close clerk elections. In Section A.6.11 in the Supplementary Material, we present a more formal analysis of how local our estimates are drawing on the approach described in Angrist and Rokkanen (Reference Angrist and Rokkanen2015) and Hainmueller, Hall, and Snyder (Reference Hainmueller, Hall and Snyder2015). We find that, even including counties where the Democratic clerk candidate won as little as 25% or as much as 75% of the vote, partisan clerks do not appear to advantage their party on average.
Given this evidence, in our Mechanisms section, we consider explanations for clerks not advantaging their party that apply to all clerks rather than just those elected by very small margins.
Democratic and Republican Clerks Produce Similar Turnout and Policies
While conventional wisdom holds that high-turnout elections favor Democrats (Lijphart Reference Lijphart1997; Piven and Cloward Reference Piven and Cloward1988), some reforms that increase turnout do not noticeably increase Democratic vote share (see, e.g., Thompson et al. Reference Thompson, Wu, Yoder and Hall2020). Might local election officials successfully affect turnout but fail to offer their party an advantage?
Table 2 presents regression discontinuity estimates of the effect of electing a Democrat rather than Republican election official on turnout. The first two columns mirror columns 1 and 4 from Table 1. Across both specifications, we find that, after accounting for differences in where and when Democrats and Republicans run for office, members of both parties oversee similar levels of voter participation on average.
In the final row, we report the minimum detectable effect using each estimator. Both estimators can detect an effect as small as 1.1 percentage points with 80% power or greater. Even with these high-powered tests, we find no evidence that electing a Democratic rather than a Republican election official increases turnout on average.
While Democrats are often expected to pursue policies that increase turnout, vote-maximizing partisans will only work to increase participation when their party makes up a majority of the people affected by their policies (Burden et al. Reference Burden, Canon, Lavertu, Mayer and Moynihan2013; Kimball, Kropf, and Battles Reference Kimball, Kropf and Battles2006). Might Democratic clerks oversee lower turnout in Republican-majority counties and higher turnout in Democratic-majority counties?
Table 2 presents evidence that Democratic and Republican officials do not strategically increase turnout when their party makes up a majority and decrease turnout when their party is in the minority. Columns 3 and 4 report the effect of electing a Democratic clerk in Republican-majority counties. There, marginal voters are more likely to be Republicans, so we would expect vote-maximizing Democratic clerks to decrease turnout relative to Republican clerks. Instead, we find that Democratic and Republican clerks oversee similar turnout rates in these counties. Columns 5 and 6 report the effect of electing a Democratic clerk in Democratic-majority districts, where Democrats are most likely to make up a majority of marginal voters. Still, we find that Democratic and Republican clerks oversee similar levels of participation.
These results could arise if partisan clerks implement different policies that have very modest effects on turnout. Committed partisan clerks could pursue these policies anyway if they are unaware of their ineffectiveness or if they have ideological positions about how elections ought to be administered. In Section A.6.12 in the Supplementary Material, we present evidence that Democratic and Republican clerks representing comparable places make similar administrative decisions across many parts of the job, including the number of polling places sited per voting-age resident, the share of votes cast provisionally, the provisional ballot rejection rate, the registration rate, the registration removal rate, the partisan balance of registrants, and voter wait times.
Put together, the analyses presented in Table 2 and Section A.6.12 in the Supplementary Material cast doubt on the claim that partisan clerks are strategically changing turnout or policies while failing to convert those changes into noticeable advantages in election results. Instead, partisan clerks oversee similar turnout and policies even when it is in their party’s interest for them to increase or decrease turnout.
WHY DON’T CLERKS ADVANTAGE THEIR PARTY?
Why do elected clerks not advantage their party? Drawing on our discussion in the Theory section, we explore four explanations. The first explanation we explore is that clerks are elected officials and want to win reelection, so clerks from both parties work to satisfy the median voter in their county and produce similar policies and outcomes. The next three explanations are countervailing forces within the citizen-candidate framework that could lead clerks to not advantage their party: (1) qualified candidates hold similar views across parties, (2) administration has modest effects on turnout and outcomes, and (3) clerks face a collective action problem because elections are decided jointly by many counties. No single piece of evidence we present conclusively answers why clerks do not advantage their party, but we provide suggestive evidence against the reelection incentive and collective action problem as meaningful constraints and discuss existing research that favors preference convergence and the limited ability of clerks to influence electoral outcomes as explanations.
Reelection Incentives Do Not Noticeably Affect Partisan Advantage Clerks Provide
Might Democratic and Republican clerks oversee similar election outcomes because they are competing for the support of the median voter in their next election? This is the prediction of one class of standard political economy models of elections (Downs Reference Downs1957; Fearon Reference Fearon, Mann, Przeworski and Stokes1999). We study this question using election official term limits. Clerks in Indiana are allowed to serve for no more than two consecutive 4-year terms in a 12-year period.Footnote 21 If the threat of being thrown out of office is the main constraint on clerks advantaging their party, clerks should advantage their party more in their second term than their first term, since the reelection incentive is removed entirely. To test this prediction, we compare the change in Democratic presidential vote share from the first term to the second term of Democratic clerks to the same change for Republican clerks.
Table 3 presents our estimates. In the first column, we present the simple difference in means between Democratic and Republican clerks in how much more of their county’s presidential vote goes to the Democratic candidate in their second term than their first term. The second column presents regression estimates with year fixed effects to account for statewide changes in support for Democratic presidential candidates across years in our data. The third column presents regression estimates with lagged Democratic presidential vote share in addition to year fixed effects to account for any polarization across counties in voting trends over the years.
Table 3. Estimates of Increase in Partisan Advantage Provided by Term-Limited Clerks
Note: Robust standard errors clustered by county in parentheses. The data are limited to term-limited, incumbent clerks in Indiana. The outcome is the change in Democratic presidential vote share from the first term to the second term of the term-limited clerk.
Across all three regression specifications, we find that clerks do not give their party a bigger advantage when they are ineligible for reelection. While this simple analysis does not fully account for differences in trends in presidential vote across counties unrelated to the party of the clerk, which our regression discontinuity estimates do account for, we take this as suggesting that reelection incentives are not a key constraint limiting the advantage clerks give their party.
This result suggests that concerns about reelection are not the main reason clerks do not advantage their party, but it does not imply that elections fail to motivate clerks. Clerks seem to be held accountable for bad behavior in many cases. For example, in 2010, a lawsuit was filed against Boone County, West Virginia clerk Gary Williams alleging sexual harassment right after he was reelected without opposition.Footnote 22 He was challenged in the Democratic primary 6 years later and lost, receiving only 34% of the vote. Bosque County, Texas clerk Brigitte Bronstad was arrested for taking money from the county in 2002, right before the general election. Four write-in challengers quickly jumped into the race, successfully ensuring her defeat.Footnote 23 In other cases, election officials caught engaging in malfeasance retired rather than face the voters. This was the case for Montezuma County, Colorado clerk Carol Tullis in 2012, who faced a lawsuit alleging she demoted an employee for running against her,Footnote 24 and likely played a role in Whitman County, Washington auditor Eunice Coker’s retirement, who faced a lawsuit in 2018 alleging improper denial of employee medical leave, financial mismanagement, ballot irregularities, audit failures, discriminatory behavior, and politically partisan efforts to alter election outcomes.Footnote 25
Clerk Candidates May Have More Similar Preferences across Parties
Might Democratic and Republican clerks agree on how to run elections? Looking at the public, this seems unlikely. The average Democrat and Republican have meaningfully different views on issues like automatic voter registration, all-mail voting, and moving election day to a weekend (Stewart III Reference Stewart2021). On the other hand, candidates and winners often have experience in election administration and may have more similar policy views. Manion et al. (Reference Manion, Anthony, Kimball, Udani and Gronke2021) surveys members of the public and clerks, and compares their responses across parties. While Democratic and Republican clerks still have meaningfully different responses to some policy questions, their preferences are more similar than Democrats and Republicans in the public and fully converge on some policy issues. For example, Democratic and Republican clerks express equivalent levels of voter confidence in national elections, agree that voting is a duty, and believe that local, state, and federal elections should be consolidated. Like their co-partisans in the public, Democratic and Republican clerks are divided on the issue of voter ID but hold much more similar views across parties on expanded early voting than members of the public—a policy that many clerks have discretion over. This explanation only partially accounts for the similarity in policies, turnout, and vote shares in elections run by Democrats and Republicans serving similar counties, but it is consistent with our main findings and existing survey data of these individuals.
In Section A.6.12 in the Supplementary Material, we also document that clerks from both parties serving identical counties implement roughly the same policies. While we cannot rule out that they do this because they expect these policies would have minimal effects (as we discuss below), this is consistent with clerks agreeing more on election administration across parties than the public.
Clerks May Have Limited Ability to Affect Election Outcomes
Even if clerks are unconstrained by reelection incentives and want to offer their party an advantage, they may not be able to. As we discuss in the Clerk Responsibilities section, clerks are given wide latitude to make important decisions such as where to locate polling places and when to host in-person early voting. These decisions may make it easier or harder to vote and likely affect some groups more than others. However, these policies do not necessarily affect election outcomes. First, when the cost of voting goes up, citizens may simply find the next cheapest way to vote (Clinton et al. Reference Clinton, Eubank, Fresh and Shepherd2020). Second, even if more people vote when the cost goes down, the new voters may be similar in partisan composition to the people already voting (Burden et al. Reference Burden, Canon, Mayer and Moynihan2014).
This explanation is difficult to directly test. If clerks know that they cannot meaningfully affect outcomes, and they only care about changing policy if it affects outcomes, we may not observe partisan differences in policies or turnout because clerks never even try to advantage their party. Still, based on the existing work on the limited effect of election administration, it is reasonable to expect clerks are at least somewhat constrained by the modest effects these policies have on partisan outcomes.
Clerks Do Not Advantage Their Party More When It Is Less Costly or When the Stakes Are Higher
Suppose most election officials would like to see their party win and that they all have authority to advantage their party in their county. If they bare costs for tilting elections in their party’s favor, they would only want to advantage their party when it would plausibly change the statewide outcome. In this world, the fragmented nature of local election administration creates a collective action problem where partisan clerks would like to work together and swing the election in their party’s favor, but they know that every individual clerk would have a reason to shirk and avoid baring the costs. This collective action problem does not arise if an individual clerk could reasonably expect their decisions to be pivotal and worth the cost.
We offer suggestive evidence that even clerks who face the lowest costs to advantaging their party or have the greatest chance of swinging an election in their party’s favor do not advantage their party. We do this by identifying six related conditions that either make it less costly for an official to advantage their party or increase the value of the advantage they provide. The first two conditions—residential segregation and racial and ethnic diversity—make use of the fact that race and ethnicity are some of the most useful heuristics for guessing the party a citizen may vote for (Carlson and Hill Reference Carlson and Hill2022; Carmines and Stimson Reference Carmines and Stimson1989; Hersh Reference Hersh2015). Even if clerks are primarily motivated by providing their party an advantage, they may fail to do so if they cannot easily distinguish between members of their party and the opposing party. Accordingly, local election officials may have an easier time giving their party an advantage in counties that are more diverse and segregated. The third factor we consider is county-level partisan balance. As we discuss in Section A.7.3 in the Supplementary Material, we find using a stylized model that clerks serving counties evenly split between Democrats and Republicans will have a larger effect on election outcomes than clerks in places dominated by one party. The fourth factor we consider is the capacity of the office, which we proxy with population. We would expect clerks serving in larger counties to have greater capacity to affect election outcomes (Kimball and Baybeck Reference Kimball and Baybeck2013). The final two factors we consider—how close the last presidential election was in the state and whether the county is large enough to meaningfully alter the outcome—build on the prediction that election officials might be most motivated to advantage their party when it would be most likely to help their party win.
Figure 5 reports estimates of the effect of electing a Democratic local election official on Democratic presidential vote share in counties where we would expect clerks to be most likely to advantage their party if collective action problems were the primary barrier. Each point is an effect estimated using local linear regression with triangular kernel weights—the same specification we use in column 4 of Table 1. The lines extending out from the points are 95% confidence intervals. From top to bottom, the plot presents estimates using seven subsets of the data: (1) all counties, (2) segregated counties—that is, those with residential racial dissimilarity scores above the median, (3) counties where non-Hispanic white people make up less than 80% of the population, (4) counties in which the last Democratic presidential candidate won or lost the county by less than 15 percentage points, (5) counties with over one hundred thousand residents, (6) counties in states in which the last Democratic presidential candidate won or lost by less than 5 percentage points, and (7) counties with populations that are at least half as large as the margin by which the last Democratic presidential candidate won or lost in the state.
Figure 5. Clerks Do Not Advantage Their Party More When It Is Easier or Most Advantageous
Note: Each dot represents a regression discontinuity-based estimate of the effect of electing a Democratic clerk on residual Democratic presidential vote share for a subset of the data. The lines around each point represent 95% confidence intervals. Estimates come from regressions that mimic column 4 of Table 1 using local linear regression with triangular kernel weights. Segregated counties are those with residential racial dissimilarity scores above the median. Diverse counties are those less than 80% non-Hispanic white. Balanced counties are those in which the most recent Democratic presidential candidate won or lost by less than 15 percentage points. Large-population counties are those with over one hundred thousand residents. Competitive states are those in which the most recent Democratic presidential candidate won or lost by less than 5 percentage points. Determinative counties are those where the population of the county is at least half as large as the most recent Democratic presidential candidate’s margin of victory or loss at the state level. Full tabular results are found in Section A.7 of the Supplementary Material.
The estimates reported in Figure 5 are more consistent with clerks intending to administer elections in neutral ways than with a collective action problem preventing clerks from advantaging their party. If they want to advantage their party but fail due to a collective action problem, we might observe a partisan advantage in the cases where a county is closer to being pivotal or the cost of advantaging one party is lower. Instead, across the seven subgroups that we study, we cannot reject the null hypothesis that Democratic and Republican clerks fail to advantage their party. Our evidence suggests that clerks do not noticeably advantage their party even when they have the greatest ability to affect the statewide outcome and the lowest costs.
The regression specifications chosen and the rules used for including a county in each subgroup are somewhat arbitrary. In Section A.7 in the Supplementary Material, we present estimates using all four of our regression specifications for every outcome and estimates across many different rules for inclusion in each subgroup analysis. The results reported in Figure 5 are similar to those we estimate across our different specifications and subgroup inclusion rules.
CONCLUSION
The unusual American practice of electing partisan local officials to oversee elections concerns many experts and members of the public. When an official runs as a member of a party, it is natural to expect that they will use their authority to advance their party’s goals. Even some local election officials themselves report feeling uncomfortable running as partisans when they have a duty to be neutral.Footnote 26
Using a credible research design with new partisan clerk election data from 21 states, we find that partisan election officials do not typically offer a large advantage to their party. While we cannot be confident that partisan officials do not offer rare and large or very small but consequential advantages to their party, our findings make clear that clerks are not consistently providing their party a meaningful advantage to date.
While clerks do not advantage their party, this does not imply that we ought to use partisan elections to select election administrators. In many parts of the country and around the world, elections are run by appointed bureaucrats, and future work should consider how the benefits and costs of such a system weigh against the benefits and costs of the system we study in this article (Ferrer Reference Ferrer2022). Also, a recent survey of the public found that about 75% of both Democrats and Republicans support requiring that election officials be selected on a nonpartisan basis (Stewart III Reference Stewart2021). Future work should consider if even neutral partisan election administration leaves citizens suspicious that the election was unfair.
How concerned should we be that future changes in who runs and wins clerk races may lead to highly partisan election administration? Our explanation that election policies only have modest effects on electoral outcomes provides some reason for optimism. However, our explanation that clerks are neutral because they share more similar preferences across parties than the public does leave room for concern. If the next generation of election officials begins to exhibit higher levels of preference polarization, there is no guarantee that partisan election officials will continue to administer elections neutrally.
SUPPLEMENTARY MATERIAL
The supplementary material for this article can be found at https://doi.org/10.1017/S0003055423000631.
DATA AVAILABILITY STATEMENT
Research documentation and data that support the findings of this study are openly available at the American Political Science Review Dataverse: https://doi.org/10.7910/DVN/SHYPDU.
ACKNOWLEDGMENTS
For helpful discussion and comments, the authors thank Graeme Blair, Wiola Dziuda, Ted Enamorado, Bob Erikson, Anthony Fowler, Sandy Gordon, Paul Gronke, Andy Hall, Seth Hill, Will Howell, Greg Huber, Apoorva Lal, Hans Lueders, Will Marble, Toby Nowacki, Avshalom Schwartz, Clemence Tricaud, and Jesse Yoder; seminar participants at the Goldman School of Public Policy, the Harris School of Public Policy, and the UC Irvine Department of Political Science; and participants in the 2021 Election Sciences, Reform and Administration conference, the 2022 Southern Political Science Association conference, the 2022 Midwest Political Science Association conference, and the 2022 American Political Science Association conference. We would also like to thank David Kimball and Brady Baybeck for providing data from their 2008 survey of local election officials and the hundreds of local election officials who shared election data with us.
CONFLICT OF INTEREST
The authors declare no ethical issues or conflicts of interest in this research.
ETHICAL STANDARDS
The authors affirm this research did not involve human subjects.
Open access
