How We Conducted This Research

Between July 16 and Aug. 5, 2024, we conducted seven focus groups in partnership with YouGov. Participants were recruited from a YouGov panel and were selected because they met all these conditions:

• They indicated they planned to vote in the upcoming election;
• When given a five-point scale to measure their interest in politics, they responded with a 2 or higher, indicating they were at least “slightly interested” in politics; and
• When given a five-point scale to measure how often they think about their personal safety and security throughout the process of voting, they responded with a 3 or higher, corresponding to “sometimes” or more frequently.
  Participants were sorted into seven groups: white Democrats, nonwhite Democrats, white independents, nonwhite independents, white Republicans, nonwhite Republicans, and gender-nonconforming individuals of all political parties.

The focus group of nonwhite Democrats was run twice, with different sets of participants, because of a technology failure during the first focus group. Insights from both groups are incorporated into analyses.

Each group had six or seven participants. Groups were conducted on Zoom and lasted about an hour. To encourage comfort and openness among the participants, a white, female moderator directed the groups of white participants and the group of gender-nonconforming participants; a woman of color directed the groups of nonwhite participants.

Our pre-election survey was conducted Sept. 23–30, 2024. We surveyed 4,016 American adults. We assessed their opinions about voting safety, their experiences with voting safety in the past, and their expectations for the 2024 election. The margin of error for this survey is ±1.7%.

Our post-election survey was conducted Nov. 7–19, 2024. We surveyed a separate group of 4,017 American adults who reported that they were registered to vote and whose data could be matched to the national voter file. Again, we assessed their opinions about voting safety and their experiences with the 2024 election. We also conducted a short experiment with survey participants in which we tested the effect of various messages about elections on election safety-related outcomes. The margin of error for this survey is ±1.6%.

Respondents were recruited and surveys distributed in partnership with YouGov. Analyses were weighted using standard YouGov weights to approximate the national demographic characteristics of the appropriate target populations. In the pre-election survey, we used a set of weights to match all survey respondents to the demographics of all Americans in analyses that examined only pre-election data. We also used a set of weights to adjust only registered voters who were matched to the national voter file to that population. This set of weights was used in analyses that compared the pre-election survey data with the post-election survey data. In the post-election survey, we used a single set of weights to match respondents to the population of registered voters who were matched to the national voter file.

To identify gender-nonconforming participants, we used a series of questions. We first asked whether respondents identified as man, woman, non-binary, or other. We also asked whether they identified as transgender. Respondents who did not identify as a man or woman, or who did identify as transgender, were classified as gender-nonconforming. Then, in the pre-election survey, we determined who identified as LGBTQ+ by asking about sexual orientation and identifying those who were “lesbian/gay woman,” “gay man,” or “bisexual.” In the post-election survey, we directly asked whether respondents identified as “lesbian, gay, bisexual, transgender, queer, non-binary, etc.” To then identify a fuller sample of gender-nonconforming respondents, we followed up in December 2024 with participants from both surveys who did not identify as men or women and who did identify as LGBTQ+, and we asked them whether they identified as gender-nonconforming.

This method of classifying gender-nonconforming respondents resulted in a sample of 81 gender-nonconforming respondents in our pre-election survey and 103 in our post-election survey. We identified gender-nonconforming respondents in our focus groups using the same method as in the post-election survey.

Identifying respondents who exaggerate

In the post-election survey, we included a question to identify respondents who may have exaggerated to make a point rather than answering the survey questions honestly:

“We have asked a lot of tough questions about threats and violence during the voting process. Do you tend to answer these questions honestly with what you firmly believe, or give exaggerated answers because it feels good?”

They answered on a five-point scale from “Completely exaggerate because it feels good” to “Completely what I firmly believe,” with the midpoint labeled “Both equally.” We ran all analyses with and without respondents who said they exaggerated, and examined significant differences in response patterns between respondents who said they exaggerated and those who didn’t.

Of our 4,017 respondents, only 75 (1.9%) said they completely exaggerated and only 87 (2.2%) said they mostly exaggerated. General response patterns remained the same with and without these respondents.

There were significant differences in responses between those who exaggerated and those who didn’t, in the expected direction. Respondents who said they exaggerated also indicated that they felt less safe voting, took more safety precautions, experienced more incidents of violence, thought others experienced more incidents, and thought violence was committed based on more political reasons.

Our state legislator survey was conducted Aug. 6–Sept. 1, 2024. In partnership with CivicPulse, we surveyed 679 state legislators and candidates running for state legislative office in 2024. Respondents were recruited from comprehensive lists of all state legislators and candidates in 2024. The data relating to this project was part of a larger survey project. For analyses in this report, we filtered the sample to 208 state legislators, excluding challengers, and excluding respondents from U.S. territories.

For analyses, we weighted responses based on the population of the county in which each legislator served, as well as the number of legislators in each respondent’s census division. Data on the number of state legislators by census division was drawn from the National Council on State Legislatures. Data on county population for weights for all surveys was drawn from the U.S. Census Bureau’s 2023 data.

Our pre-election survey was conducted Oct. 2–Nov. 4, 2024. We surveyed 441 law enforcement officers, including 223 executives of law enforcement agencies and 218 lower and mid-level officers. Our post-election survey was conducted Nov. 19–Dec. 16, 2024. We surveyed 402 law enforcement officers, including 199 executives and 203 lower and mid-level officers. Both surveys were conducted in partnership with CivicPulse.

For both surveys, law enforcement executives were randomly sampled and recruited from a comprehensive list of leadership of municipal and county law enforcement in communities with populations of 1,000 or more across the country. Low- and mid-level officers were randomly sampled and recruited from a national list of officers from a third-party vendor, excluding officers from agency or jurisdiction types that do not play a role in elections.

For analysis purposes, low and mid-level officers were classified together as (non-executive) staff. Again, we weighted responses based on the population of the county in which each officer served, as well as the number of officers employed in each respondent’s census division. The pre-election staff survey had no respondents from the Middle Atlantic division, and the post-election staff survey had no respondents from the East South Central division.

Data on the number of officers employed by census division was taken from the Bureau of Labor Statistics and based on employment numbers of Police and Sheriff’s Patrol Officers in 2023. The sample of staff and the sample of executives were weighted separately given different recruitment methods.

Our pre-election surveys of election officials and election workers were conducted Oct. 2–Nov. 5, 2024. We surveyed a total of 1,948 election administrative staff, including 308 election officials and 1640 election workers. Our post-election surveys were conducted Nov. 14–Dec. 6, 2024. We surveyed 646 election administrative staff, including 223 election officials and 423 election workers. Both surveys were conducted in partnership with CivicPulse.

For both the pre-election and post-election surveys, election officials were randomly selected and recruited from a comprehensive contact list of election officials across the United States, using data obtained from web scraping of contact information on official government websites. Election workers were recruited through election officials who were randomly selected and then agreed to distribute the survey links to election workers in their areas. Election officials recruited to take surveys themselves, and election officials randomly selected to be asked to distribute the election worker survey, were randomized separately. Some respondents for the election worker survey were also recruited through county party chairs who were randomly contacted and agreed to distribute the link, or through email requests to third-party vendors with contact information for election workers.

For analysis purposes, we weighted responses based on the county population in which each respondent served, as well as the number of election officials or election workers in each respondent’s census region. The post-election survey of election workers had no respondents from New England or the West North Central or West South Central divisions.

Data on the number of election officials by census region was provided by CivicPulse, based on its comprehensive contact list of election officials. Data on the number of election workers by census region was taken from the 2022 Election Administration and Voting Survey dataset. The sample of election workers and the sample of election officials were weighted separately given different recruitment methods.

All error bars are 95% confidence intervals.

In keeping with best research practices, we classified independent voters who reported “leaning” toward either the Democratic or Republican party as partisans. Therefore, we defined “independents” as those respondents who professed no partisan attachments whatsoever.