Officials versus ordinary citizens

Are local elected officials more confident, less confident, or no different from ordinary citizens with respect to the accurate counting of votes cast in the 2022 election? There are two reasons that officials might be more confident than the general public. First, all local officials, no matter their partisan identity, have greater exposure to members of the opposite party in their daily work than ordinary citizens. Interacting with opponents in person, rather than viewing them only as caricatures, can dampen extreme views and motivated reasoning, thereby reducing variance animated by partisanship. Likewise, local officials are also known to discount disagreeable opinions voiced by their constituents [12].

Additionally, if low public confidence is due to misinformation and the exaggeration of the frequency of electoral fraud [13], we might expect local officials to be more confident than the average citizen. This is because of their greater access to reliable information and the greater likelihood that they interact professionally or socially with local colleagues who are directly engaged in election administration. These observations lead us to hypothesize:

H1a: Elected local officials will be more confident in the 2022 election results than ordinary citizens.

Moreover, if the informational mechanism is in play, then this will have its greatest impact among officials in locales with primarily responsibility for election administration. This will mostly affect county officials, as counties have the primary responsibility for election administration in most US states. But there is considerable variation with many cities having primary responsibility, particularly in New England and in some larger cities such as Chicago, Milwaukee, and others.

H1b: Officials in locales that administer elections will have greater confidence in election outcomes than other officials.

On the other hand, most local officials are unpaid or part-time public servants, making them similar to engaged citizens they represent. From this perspective, public officials would have beliefs that are quite like those of ordinary citizens [14], and we might expect that we cannot reject the null hypothesis of no effect.

H1₀: Elected local officials will be no more confident in the 2022 election results than ordinary citizens.

Partisan effects on levels of confidence

Partisanship is a powerful predictor of virtually all political beliefs and attitudes in the contemporary United States. We theorize three mechanisms through which partisanship can impact officials’ confidence in elections.

Individual partisanship effects among officials

Recent surveys of the voting eligible public show that Republican citizens show lower confidence in elections than Democrats [15,16]. Republican officials may genuinely believe that fraud, technology, or incompetence results in substantial errors in the counting of votes, extrapolating from documented inefficiencies in the maintenance of voter rolls, and use of electronic machines that cannot produce a written record. [10]. Such legitimate worries were manipulated by President Trump in 2020, and amplified by other GOP officials, with numerous false claims about voter fraud introduced as well. If elected officials tend to have stronger partisan ties than the average citizen, then this should lead to substantial polarization as reflected by large associations between partisan identification and election confidence beliefs.

H2a: Local officials who identify as Republican have lower electoral confidence than local officials who identify as Democrats.

Alternatively, it is possible that due to differential sources of information, Democrats and Republicans sincerely perceive different realities with respect to the accuracy of tabulated votes in their communities and in the nation as a whole (e.g., [17]). This would also lead to hypothesis H2a.

Partisan contextual effects

In addition to individual partisan identification, partisanship may also operate through contextual mechanisms. Local elected officials are not only drawn from the local environment but also seek to send signals that they support their party’s prevailing view. The pressure to do so will be strongest in areas where one party is clearly dominant [18]. We therefore expect officials working in local contexts that voted heavily for Joe Biden in the 2020 election will express less election skepticism than officials in areas where Donald Trump did especially well. This contextual effect could operate above and beyond personal party identity.

H2b: As the share of the local vote won by Donald Trump increases, local officials’ electoral confidence will decline.

Contextual effects may be amplified for the subset of local officials who are open to or actively considering running for higher office. Political ambition increases officials’ attentiveness to local opinions generally (e.g., [19]), and they may be especially attentive to their party’s base. We expect politically ambitious Republican local officials to display their party’s election trust stance with greater intensity in areas of Republican voting strength.

H2c: As the share of the local vote won by Donald Trump increases, local officials’ electoral confidence will decline, with the effect being stronger among ambitious officials than non-ambitious officials.

Social trust

Confidence and trust are closely related concepts (e.g., [20–21]). High confidence in institutional performance reflects trust in the capability and impartiality of individuals running those institutions. Within most nations, individuals who are high in generalized social trust (e.g., think that most people can be trusted) have higher confidence in political institutions [21] and the people running those institutions [22–23]. In that light, confidence that votes were counted accurately and that election outcomes reflect the intentions of voters could be largely determined by, or simply a reflection of, social and institutional trust (see [24–26]). In this case,it is useful to see how strongly social trust predicts election confidence and it is important to account for social trust as a potential confounder. The survey we utilize has a good measure of social trust to explore this argument, and we therefore hypothesize:

H3: As generalized social trust increases, confidence in elections increases.

Study description

Our data come from a larger study, which fielded a national probability sample survey of elected and appointed municipal, county, and school district officials. This paper relies only on the subsample of elected officials – members of municipal and county councils, mayors, county executives (if elected), and a small number of other elected officials (mostly county sheriffs). The sample intentionally does not include local officials whose sole or primary responsibility is election administration (e.g., “Director of Elections,” “Elector Board,” “Registrar”). The sample is based on a two-stage process where we first selected a probability sample of 801 cities and 616 counties, where the largest places were selected with certainty, all but the smallest places were selected with probability proportional to size (PPS), and tiny places selected with a PPS oversample. The full details on the sampling design are provided in Appendix A in S1 Appendix. In a second stage, all officials in each selected locale with a relevant job title were sampled. This research was deemed exempt by the IRB of Penn State University (IRB study number 00020506).

Recruitment was based on the tailored design method [30]. For the 95% of officials for whom we had a valid email, we sent a pre-notification letter sent by post, an email invitation to participate, three email reminders, and a “study is closing soon” final email. The 5% lacking email received a pre-notification letter with a link to our web survey, a reminder postcard, and a “study is closing soon” postcard. Pre-notification letters were mailed July 3, 2023 and surveys were completed between July 5 and September 19, 2023 (97% of surveys were completed before August 16, but the survey remained open for very late returns). The first screen of the survey explained that the survey was voluntary, any question could be skipped and was anonymous. Respondents affirmed they were over 18 years old and currently serving as a local official by giving implicit consent (advancing to the next screen and first question). A more detailed description of the fieldwork and informed consent is contained in Appendix B in S1 Appendix. Officials completed a survey hosted by the corresponding author’s home institution Qualtrics platform. The median official completed the survey in 34 minutes (including any time away from the survey).

The overall response rate (AAPOR RR4) for city and county officials was 9.9%. As detailed in Appendix C in S1 Appendix, response rates were higher for officials with a valid email, and for officials in communities with higher levels of education. However, there were no systematic patterns of non-response based on community race, ethnicity, religion, or county level vote in the 2020 election (see S1 Appendix Tables C1 and C2).

The sample was designed to be self-representing with cities selected in proportion to the number of constituents (that is, the cities with a population of 100,000 had a probability of selection exactly double that of cities with a population of 50,000). Ignoring non-response, results generalize to constituents, not officials. Based on the propensity model reported in table C1 in S1 Appendix, we calculated a non-response adjustment weight, and all analyses in this paper are weighted.

Dependent variables

Our key dependent variables are answers to two questions that measure confidence in the election. These questions are the second and fourth items in the question set used in the MIT/Caltech Survey of the Performance of American Elections [31–32], a longstanding national survey of electoral performance from the citizen’s perspective. Our first item asks, “Thinking about last year’s mid-term elections in November, how confident are you that votes in your county or city were counted as voters intended?” Of the elected officials who completed our survey, 83% said they were “very confident” (see Table 2 for more details). Our second item similarly asks, “Now think about vote counting throughout the country. How confident are you that votes nationwide were counted as voters intended?” As shown in Table 2, only 54% said they were very confident and fully 19% said they were “not too confident” or “not confident at all.” The full question wording for this and all other survey questions used in this paper can be found in Appendix E in S1 Appendix.

In samples of the general public, public confidence as measured by these questions consistently shows greater confidence in local elections than national elections [32]. Moreover, answers to both items are sensitive to election outcomes, with citizens identifying with the losing party showing less confidence than when their candidate prevails [3,32]. By contrast, smaller shifts in these questions are associated with actual administrative performance (e.g., lower confidence expressed by residents of states that report the vote late) [33]. Our interest lies in understanding variation among officials regarding confidence in a specific election cycle, especially factors that operate after controlling for partisan identification.

Key independent variables

Partisan identification.

We utilize the seven-point party ID measure used in the American National Election Study and the General Social Survey, where 0 indicates strong Democrats and 6 indicates strong Republicans. Because partisanship is central to many of our core hypotheses, we show the sample distribution in Table 1. Notably, “strong” partisans represent a majority (56%), but with considerable variation in intensity.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Partisan identification among local elected officials.

https://doi.org/10.1371/journal.pone.0324794.t001

Social trust.

Social trust is measured with five questions we adapted from those used in the World Values Survey. Officials were asked to say how much they trust their family, neighbors, strangers, people from a different religion, and people from a different political party. For each, they could indicate they trusted “completely,” “somewhat,” “not very much,” or “not at all.” Exploratory factor analysis suggested that the items are well represented by a single latent factor and the resulting scale (based on factor scores using the Bartlett method) has an Omega reliability of 0.61 (Cronbach’s alpha = 0.60). The scale is standardized, with a mean of zero and a standard deviation of one (see Fig 1). Those with maximum scores display high trust in both close and distal contacts, those in the middle tended to show high trust in neighbors and family, but intermediate levels of trust in strangers and those from a different political party. The small number of very low scores are officials who distrust all groups, including their family and neighbors.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Distribution of social trust scale (N = 743).

https://doi.org/10.1371/journal.pone.0324794.g001

Control variables

Our multivariable analyses include a number of covariates that might confound the relationships among election confidence, partisanship, and trust. These include age, whether the official is married or in a civil union; whether they have children under 18 living in their household; whether they or someone in their household ever served in the U.S. military, reserves, or National Guard; highest level of school completed; whether the official identifies as Hispanic; a measure of community size grouped into quartiles; an ordinal measure of the official’s number of years in their current office; and whether the official is the highest elected official in their jurisdiction (e.g., Mayor, coded 1). (We use a measure of Hispanic for race/ethnicity because most non-white racial identities have low numbers of cases and, due to close overlap, African Americans are largely captured by the RELTRAD religious tradition classification of Black Protestant instead).

We also control for religiosity and religious tradition. Religiosity is a composite scale variable constructed from frequency of personal prayer other than at meals (daily, weekly, monthly, a few times a year, never), frequency of religious service attendance other than at weddings and funerals (every week, almost every week, once or twice a month, a few times a year, never), the extent that religion provides guidance in day-to-day life (none, some, quite a bit, a great deal), and extent they are a spiritual person (not at all, slightly, moderately, very). Each measure was converted to a z-score (M = 0, SD = 1) and then averaged. Religious affiliation is measured using a variation of the RELTRAD classification [35], with seven categories (Conservative Protestant, Mainline Protestant, Black Protestant, Catholic, Jewish, Other Religion, No affiliation). Note that the RELTRAD category of Black Protestant churches is nearly collinear with Black racial identity, and we cannot estimate models with both; as a result, we do not include a dummy variable for identifying as Black. Finally, gender and sexual identity are based on two questions. The first is “Are you a: Man, Woman, Something else,” and the second asks “Which of the following best represents how you think of yourself? Lesbian or gay, Heterosexual or straight, Bisexual, Prefer not to answer, [or] Don’t know.” Operationally we created a dummy variable for female identification, and a dummy variable coded 1 for those selecting Lesbian or gay or Bisexual on the sexual preference measure or selecting “something else” on the gender identity measure. (We are mindful that this operationalization lumps together heterogeneous sexual minority statuses. In a general population survey of similar size, best practices would make finer distinctions. However, this sample skews older than the general population and we have limited subsamples: n = 9 individuals who reported their sex as something other than male or female (2 of whom also identify as lesbian or gay and 1 who identifies as bisexual); 24 men and 10 women who identify as lesbian or gay; 7 men and 10 women who identify as bisexual. Because this paper employs these measures as control variables, we felt collapsing disparate identities into a single LGBQT indicator that encompasses answers of “something else” on the sex question was the best approach.)

Officials compared to registered voters

To address our first research question, we pooled data from our own survey of elected officials with comparable data from the November 2022 Survey of the Performance of American Elections and its sample of registered voters.

Comparing columns 1 and 2 of Table 2, we can see that 83% of elected officials were very confident that votes in their community were counted accurately. Only about 3.5% of elected officials expressed that they were “not at all confident” or “not too confident.” In contrast, more than three times as many – 11% – of registered voters expressed low confidence and only six in ten were very confident. Moving to columns 3 and 4, we see that both populations show lower confidence that votes in the nation as a whole were counted accurately than votes in their community. Only 37% of registered voters were very confident, substantially fewer than the 54% of local elected officials. We conducted a formal test by pooling the two data sets and estimating ordinal logistic regressions for each confidence question, with a source indicator taking a value of 0 for the elected official sample and 1 for the MIT/Caltech public sample. For both dependent variables the odds ratio was statistically significant and negative, indicating lower confidence among the public (for national confidence, OR =.597, t = − 6.10, p < .000; for local vote confidence OR =.314, t = − 10.34, p < .000). These results were similar when we also added state fixed effects. Thus, these results support hypothesis 1a.

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Confidence in local elections (columns 1 & 2) and confidence in national election (columns 3 & 4) for samples of local elected officials and sample of registered voters (column percentages).

https://doi.org/10.1371/journal.pone.0324794.t002

This exploration tells us two things. First, both groups are more confident that votes in their own communities were counted accurately than they are that votes elsewhere were. Second, elected officials are much more confident in elections than the voters they represent. (This pattern is not due to different geographical representation; in results reported in the appendix, we show that the same pattern results after we account for state fixed effects – indeed, state fixed effects have virtually no impact on the gap between officials and voters).

What drives confidence in local and national elections?

We now move on to test our hypotheses concerning mechanisms of partisanship as well as trust. Our empirical strategy relies on regressing each of our confidence measures on the array of key independent and control variables. Because each outcome is ordinal, we use ordinal logistic regression with survey corrected standard errors (using Stata’s Taylor linearization estimate of standard errors to account for weighting). When estimating the model for confidence in national elections, we include confidence in local elections as a control so that we can isolate effects on national confidence net of local confidence.

The results for the entire sample are summarized in Fig 2, where the dots represent the odds ratio associated with a one-unit change in each independent variable. Estimates for control variables are omitted from the graph and are reported fully in Table G.2 in S1 Appendix.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Ordinal logistic regression of confidence in elections.

https://doi.org/10.1371/journal.pone.0324794.g002

How does party elite denialism and political ambition impact Republican local office holders?.

In this section we focus on the subsample of Republican local elected officials. In contrast to the Democratic party, Donald Trump and other GOP elites have consistently sought to call attention to alleged electoral fraud and the “rigging” of election procedures – efforts that pre-date the events of January 6, 2001 by more than four years. Nearly a month before the 2016 election, Trump tweeted “The election is absolutely being rigged” [36]. Shortly after winning that election, he signed an executive order creating a Presidential Advisory Commission on Election Integrity. For this reason, two of our hypotheses concerning partisanship focused on GOP officials. We thus re-estimated our model for the subsample of Republicans (including leaners). The key estimates are summarized in Fig 3 (full results, including for controls, are reported in Table G3 in S1 Appendix).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Ordinal logistic regression of confidence in elections (Republicans only).

https://doi.org/10.1371/journal.pone.0324794.g003

We first note that the smaller sample (N = 256) results in greater uncertainty in the estimates. We also note that, compared to the full sample analyses, the effects of local responsibility and social trust are similar, the effect of individual partisanship is now diminished for confidence in local elections, but with a compensating increase in the effect of local Trump support.

We now turn to the hypotheses that are specific to Republicans. H4a follows from the substantial cross-state variation in election denial rhetoric and proposes that Republican office holders would take cues from party elites in their states. The plot shows that variation in party elite position had no impact on officials’ confidence in either local or national elections, so H4a is rejected.

Hypothesis H4b predicted an interaction between state elite denialism and ambition, reasoning that ambitious officials would have their eye on cues from elites who are serve as gatekeepers for higher office. But we found that this interaction was not statistically significant in predicting either local or national election confidence (p > 0.50 in both instances, as shown in Table G4 in S1 Appendix).

However, Fig 3 shows a main effect of ambition with ambitious GOP local office holders being less confident in national elections than their less ambitious counterparts. Indeed, the estimated probability that an ambitious GOP official expresses high confidence in the national election results is just 0.12.

Limitations

To our knowledge, our survey is the first national probability survey of local officials to explore levels of election confidence. That said, our cross-sectional research design has several implicit limitations.

First, our officials were interviewed eight months after the November 2022 election. The opinions of respondents to the MIT/Caltech survey might have shifted in those eight months so the comparison between the two surveys is not perfect. However, a major shift in the general public’s election confidence in eight months seems very unlikely. In the period from 2012 to 2022, the gap between the highest and lowest levels of this variable is only 15 percentage points (23% in 2014 versus the high point of 38% in 2020. Thus, we think that only a small portion of the 23 point gap between officials and the public could conceivably be explained by the eight-month span between the surveys.

Second, we cannot account for the possible role of election trust in citizens’ selection into the pool of public officials that we sampled. It is possible that those with high election confidence are more likely to seek public roles or be more likely to respond to our survey. On the latter, the analyses reflect weights that account for the propensity to respond to the survey and hopefully minimize this potential self-selection bias. It is possible that election winners naturally express confidence in election outcomes as a consequence of motivated reasoning.

Third, it is possible that confidence in elections is a stable trait learned earlier in life whose changes induce corresponding changes in social trust. If this were the case –that is, if election confidence were similar to confidence in parliament and other institutions [22–23] then our treatment of social trust as an independent variable would be unjustified.

Implications

We found that local office holders are substantially more confident in local and national election results than the general public. We suspect that the interactional nature of local government work, which occurs across partisan lines, along with greater access to information, both shape local officials’ relatively high amounts of election trust. Consistent with this, officials’ jurisdiction level predicted election trust in a straightforward way: officials in governments that administer elections were more confident than others. We attribute this to the familiarity that these officials are likely to have with election staff and operations in their city or county, even though our sample members are not directly involved. However, this knowledge does not appear to insulate officials from other sources of distrust and polarization that significantly impact confidence in election results outside of one’s own community.

A second important factor is social trust. Notably, our measure of social trust was completely uncorrelated with partisanship (r = 0.01, ns) and it showed that those who are wary of others – strangers, those of a different religion or political affiliation – are especially likely to doubt the results of the 2022 election. Social trust, at least for this outcome, extends its reach to confidence in institutions, even in a sample that is well educated and has access to information from their professional contacts and associations. Local elected officials could act as models of trust for local communities. Unfortunately, social trust has steadily declined since the early 1970s (e.g., [39]), and it is unlikely that local officials are immune to this long-term trend.

However, as with the general public, the most powerful predictor of election confidence was partisanship, with Republican local officials less likely to express confidence both local and national election results. Even in the context of a non-partisan, confidential survey, the large majority of officials followed the party line. However, our hypothesis that partisanship could influence officials through cueing by state party elites was not supported. Notably, differences in strength of partisan identity were also significant in the Republican sub-sample analyses, but only regarding national elections, suggesting the continued salience of commitment to the “big lie” within the party as a marker of Republican identity. This interpretation is supported by one other finding—that political ambition is negatively correlated with trust in national elections, but only among Republic officials.

Two conclusions may seem contradictory at first glance. We showed that Republican partisans have extremely low confidence that ballots were counted accurately outside of their own jurisdiction, but also that election confidence is higher among elected officials than ordinary citizens. These conclusions can both be true because, in July 2023, most Republican identifiers were weak identifiers or leaners who opted for a political label only after probing. By comparison, a similar survey of members of Congress would reveal only a single-digit presence of independents and weak identifiers (e.g., Senators Manchin and Sinema before declaring themselves Independents). The distribution of party identification simply appears different among local elected officials than among elected officials in higher levels of government, with a larger share of local officials displaying weaker partisan ties. Thus, local Republican officials may buffer the more extreme election denial rhetoric of their national leaders.

But there also is a less optimistic reading to this result. If moderates are increasingly forced out of their parties, discouraged from running for local office, or compelled to toe the party line to fulfill their ambitions, the moderate middle could become hollowed out. If that occurs, local officials will become increasingly polarized. Given the current positions of party leaders, this would result in election skeptics replacing confident officials over time.