The Racial Gap in Voter Confidence: How Trust in Ballot Counting Varies Across Racial and Ethnic Groups

Trust in elections is critical for a functional democracy. Electoral administration is complex, however, which means there are many places where trust can waiver. Recent years have seen online circulation of election rumors intended to stoke voter mistrust. Some political elites have even shared these messages, sowing doubt about how ballots are processed and counted.

All of this is to say that, when assessing a democracy's overall health, one useful indicator is the degree of trust voters have in elections.

In our paper, "Who Trusted the 2024 Election? Assessing Voter Confidence in Ballot Counting Within and Across Racial and Ethnic Groups," we focus on one facet of voter trust: voter confidence. Voter confidence is a measure of the degree of trust voters have in how ballots are counted at the individual, county, state, and national levels.

Our paper's main contribution is to highlight a systematic gap in voter confidence across and within racial and ethnic groups among American voters. But are that many voters unconfident in how ballots are counted in elections? And why focus on race? To answer these questions, read along for a summary of the key findings of our paper.

How common is low voter confidence?

In general, most voters are pretty confident in how ballots are counted. Even if some individuals are skeptical of election processes, most are at least somewhat confident that ballots in 2024 were counted as voters intended. The plot below shows the distribution of how confident voters are in their own vote being counted as intended.

Image Lightbox

Download Full Image

Figure 1: Weighted distribution of voter confidence across 2024 CMPS sample

So, if most voters are reasonably confident in how ballots are counted, why should we care about the marginal bunch of voters who aren't so confident? For one, elections are only as useful as the degree to which people trust them, so we should probably be concerned if more than a handful of voters are not confident in how elections are administered. Close to 10% of voters is more than a handful. Furthermore, there is a contagion effect to mistrust. False election rumors prey on the fact that a small but meaningful proportion of the electorate is distrustful enough to share information that depresses turnout. We should therefore worry about those who don't trust how votes are counted, not just because of their lack of voter confidence, but because they could also stoke electoral skepticism among other voters.

But the first step to mending a lack of voter confidence is to understand which voters are confident and which are not—which is exactly what we quantify within this paper.

The racial gap in voter confidence

Early studies of voter confidence focused primarily on testing how individual-level attributes correlate with voter confidence. For example, one prominent influence on voter confidence is what political scientists call the winner's effect: voters whose preferred candidate or party wins an election tend to be more confident in how ballots were counted than those voters whose candidate lost.

One persistent but less prominent finding is a gap in the confidence of white and nonwhite voters. Many previous studies have "controlled" for race, meaning researchers adjusted their estimates of voter confidence for voters' race and ethnicity, even if it wasn't their main focus. This statistical adjustment allows us to see the independent relationship between race and voter confidence in their models. What is clear across a number of studies is that voter confidence is generally higher among white voters than nonwhite voters.

However, our understanding of this phenomenon largely stops there. Most studies focus on other predictors of voter confidence, so not many attempt to explain why there's a racial gap. Nor do they tell us much about the nuances of voter confidence across, let alone within, racial groups. Most studies group respondents into white/nonwhite buckets, or if they disaggregate the groups, they focus on a small subset, such as comparing white and Black respondents.

Why the lack of attention to such a persistent finding? We argue it is likely an artifact of the realities of studying U.S. voters with nationally representative survey data.

For example, a typical representative survey of U.S. voters might include 1,000 to 2,000 respondents. According to the U.S. Census Bureau, the racial composition of the United States is approximately 60% white, 20% Latino, 14% Black, and 7% Asian American. The composition of the samples will reflect this distribution, so a perfectly representative panel of 2,000 voters would include 1,200 white voters but only 400 Latino, 280 Black, and 140 Asian American voters. Researchers could reliably estimate the confidence of white voters in the sample, but probably not that of other groups, given their smaller sample size. Even if they were interested in the voter confidence racial gap, the other researchers' methods were not optimal for further exploring it.

We overcome this common limitation through the use of Bayesian multilevel models. Bayesian multilevel models are particularly good at detecting patterns within smaller subsamples of surveys like the ones typically used to study voter confidence. Bayesian models are more efficient in these scenarios due to their use of partial pooling. Partial pooling works by calculating an average for the full sample (because everyone should have some things in common), but also allowing the estimates for individual groups to vary from the broader average based on what we see in their data. In doing so, we "borrow" information about the voter confidence of a smaller group in the sample from information about the voter confidence of a larger group. We get more statistical bang for our buck by using this approach, and our estimates are not only more precise for smaller groups, but we can also quantify how certain we are about those estimates.

Our analyses also get a boost from using a survey dataset with uniquely large numbers of nonwhite voters. The Collaborative Multiracial Post-Election Survey (or CMPS for short) is a collaborative survey conducted in presidential election years with the goal of broadening access to high-quality national survey data with large samples of racial and ethnic groups that are often underrepresented in U.S. survey research. The 2024 CMPS included large oversamples of Asian American, Black, and Latino voters, so we can produce even more precise estimates of these groups' voter confidence.

Our findings: Racial gaps in voter confidence are real and consistent

Finding 1: Voter confidence varies by race and ethnicity

Image Lightbox

Download Full Image

Figure 2: Average marginal predicted probability of low confidence in own vote being counted as intended, by racial group (50% and 95% CI)

As we expected, there are differences in the predicted probability of low voter confidence across racial and ethnic groups.

White voters are the most confident group, with an approximately 9% probability of low voter confidence. Black and Latino voters are the least confident groups, at just over 15% and 13% probabilities of low voter confidence, respectively. Asian American voters fall in the middle, with about a 10% probability of low voter confidence.

While there are some potential overlaps in the predicted voter confidence of Asian Americans and whites in our sample, there are clear differences between whites and Black and Latino voters. We are therefore confident in saying there is a racial gap in voter confidence that persists through at least 2024, and that much of this gap is driven by differences between white and Black/Latino voters.

Finding 2: Even when all else is equal, there are racial gaps in voter confidence

We established there's a racial gap in voter confidence, but a question remains: How certain are we that these results are due to race? While our point estimates of voter confidence are useful, one thing they can't tell us is whether that gap is driven by race itself or by other attributes that correlate with race, such as age or partisanship.

The voter confidence racial gap could stem from race-based differences like exposure to systemic racial marginalization. Historical and contemporary attempts to reduce turnout of minoritized groups would be a compelling explanation for why Black voters, for example, are less confident in ballot counting than white voters. However, these differences could also be driven by some other predictor that happens to also vary across racial and ethnic groups. Take age as an example. If younger people are less confident, and we know that Latino voters have a lower average age than white voters, then what appears to be a gap between Latino and white voters due to race could actually be due to age.

To evaluate this possibility, we computed average marginal contrasts by race for low voter confidence. What this means in practice is that we estimated the probability that a voter is low in voter confidence when all other predictors in the model are held at their observed value. We took each voter's actual characteristics, like age, and then calculated what their predicted probability of low voter confidence would be if they were (1) white or (2) Asian American, Black, or Latino. We then took the average difference between these two quantities across all voters to produce an estimate of the effect of race on voter confidence when all other characteristics are equal.

In doing so, we found that differences by race indeed persist even when all else is equal. One way to interpret our results is that approximately 2.4 of the 7-percentage point difference between Black and white voters can be attributed to race on its own. Similarly, 2 percentage points of the 5-percentage point difference between Latino and white voters can be attributed to race, in comparison to only 0.2 of the 1.5-percentage point difference between Asian American and white voters.

A main takeaway is that there are real racial gaps in voter confidence, and this gap cannot be fully explained by other demographic trends like differences in average age or voter turnout.

Finding 3: Voter confidence varies across Asian American and Latino national origin subgroups

Finally, we estimated how voter confidence varies across the national origin subgroups that make up the broader panethnic categories of Asian Americans and Latinos. While we didn't have strong expectations about which groups would be more confident than others, lots of research points to nuanced differences in their political attitudes and behavior, so we did expect some level of differentiation.

Among Asian Americans, the most confident subgroups are Chinese, Indian, and Japanese American voters. All three groups have an approximately 9% probability of low confidence that their ballot was counted as intended. The Asian American subgroup with the highest predicted probability of low voter confidence is Vietnamese Americans at about 13%, and all other groups fall somewhere between.

The most confident Latino voter subgroup is Cuban Americans, with only an approximately 8% probability of low voter confidence. The least confident subgroup is Puerto Ricans, with an over 15% probability of low voter confidence. The largest subgroup in the sample, Mexican Americans, has an approximately 13% probability of low voter confidence.

Therefore, if all these groups are relatively close in voter confidence, why does having subgroup estimates matter? In short, these national origin estimates illuminate important within-group differences otherwise obscured by broader group-level estimates. For example, Latino voters collectively have a 13% predicted probability of low voter confidence. However, Cuban Americans are much higher in voter confidence than we might expect based on the panethnic estimate, while Puerto Ricans are much higher. We wouldn't know this without disaggregating these groups in the first place, which shows the value of combining methods like Bayesian multilevel modeling with rich survey datasets like the CMPS.

Why this matters for American democracy

Healthy elections require voters to trust in their administration, and voter confidence is one metric of this trust. Low voter confidence is associated with a whole host of negative factors, including lower levels of civic engagement and lower voter turnout. The logic here is relatively clear—if you don't trust that ballots are being counted in the way voters intend, why participate in a political system at all?

Our research demonstrates voter confidence varies across racial groups in the United States, and that this gap cannot be attributed solely to differences in age, voter history, or other demographic factors. We suspect lower nonwhite voter confidence reflects the accumulated effects of racial marginalization in the American political system.

As a result, what should scholars and practitioners do with these results? Explaining what causes the gaps requires knowing where the gaps exist, and our research provides the most precise estimates to date of the state of voter confidence across racial and ethnic groups.

Finally, those interested in increasing voter confidence should pay special consideration to these gaps, as not all groups of voters start from the same place in their confidence. For example, imagine an intervention that increased voter confidence by 5% across all groups of voters. Voter confidence would look higher across the broader population, and this would be a great success overall. But if the work stopped there, Asian American, Black, and Latino voters would still be lower in confidence than white voters. If nonwhite voters continued to be less confident in ballot counting than white voters, this could additionally contribute to racial turnout gaps.

In short, voter confidence varies in important and consistent ways across racial and ethnic groups of voters. White voters are more confident than nonwhite voters, and there are nuanced differences in voter confidence across the Asian American and Latino panethnic umbrellas. Armed with this newly robust evidence of a racial gap in voter confidence, we hope additional research will dissect what drives these patterns.

Interested in subscribing to receive LCSSP's latest updates?

If you are external to Caltech, fill out this form.

If you are on the Caltech network, use these links to subscribe: LCSSP Mailing List, LCSSP BioPolicy Initiative, and LCSSP Democracy Mailing List.

For help or questions, please reach out to [email protected]!