Background & Summary

Elections are the cornerstone of democratic governance, conferring legitimacy to authorities and fostering trust in political institutions, leaders, and systems. The perceived fairness and transparency of electoral processes are vital to ensuring public acceptance of outcomes and promoting social and political stability1,2. When elections are seen as illegitimate, however, democratic norms and systems may face significant challenges, including increased partisan polarization and conflict, ideological extremism and intolerance, public unrest, and even political violence. Understanding when and why people accept electoral outcomes, and its consequences for civic engagement, social protest, democratic norms, electoral outcomes, and more, is vital to the study of political psychology, mass politics, and democratic governance. In this data brief, we present an original dataset which includes a wide variety of outcome and explanatory measures that uniquely enables researchers and practitioners to investigate perceived electoral (il)legitimacy in the 2020 U.S. Presidential Election and 2022 U.S. Midterm Elections. The dataset builds upon the findings of other studies in the literature on election legitimacy, while including many novel political and psychological constructs as well. We anticipate that these data will complement existing work being done by researchers and practitioners.

Recent polls indicate that the majority of Republicans believe that the results of 2020 election were illegitimate3. Why did some Americans reject the outcome of this election and what consequence might it have for political attitudes and behavior? The 2020 U.S. Presidential Election was uniquely contentious and complicated, occurring amid a pandemic, widespread social and political upheaval, pervasive misinformation, and exaggerated allegations of voter fraud4,5,6,7. Conspiracy theories about voter fraud and foreign interference, for example, may have undermined confidence in the process; that these false claims were often amplified by political leaders raises vital questions about the antecedents and consequences of perceived illegitimacy in this electoral context.

Despite its clear theoretical and practical importance, however, much of the existing research on electoral legitimacy has predominantly been cross-sectional and focused on a limited set of predictors and outcomes. One particular advantage of using panel data is to study pre- and post-election outcomes, such as the effect of being a “winner” or a “loser” in the election on attitudes toward vote count accuracy and trust in government8,9. While important advances in the literature on election (il)legitimacy used panel data from representative samples to measure pre- and post-election attitudes8,9,10, researchers are largely limited by what questions are included in these large representative studies. These limitations in existing panel datasets may mean that many important theoretical relationships are understudied. For example, prior studies have highlighted the influence of partisan identity, conspiratorial thinking, and group prejudice on perceptions of electoral fairness and legitimacy11,12,13,14,15,16. The current dataset allows for a rigorous replication of existing insights and novel investigations of the interplay of these factors with a broad range of theoretically important individual differences, psychological constructs, and political beliefs, values, and behaviors, over time.

Critically, few existing studies of electoral legitimacy have employed longitudinal designs to examine how perceptions evolve before and after election results are announced, and its consequences for subsequent elections (i.e., 2022 U.S. Midterm Election). The current data brief helps to address these limitations by making available data from two panel surveys, one with a two wave design and another with a four wave design. That is, many of our measures and constructs are repeated across waves, allowing researchers to investigate stability and change in political beliefs and behaviors as a function of individual differences, psychological constructs, and the outcome of the 2020 and 2022 elections. By observing the same units over time, panel data enable statistical techniques like difference-in-differences and lagged variable models, which can help establish temporal ordering and mitigate endogeneity concerns. These techniques allow researchers to control for unobserved heterogeneity through fixed or random effects, thereby reducing omitted variable bias. However, because causal claims still require strong assumptions or natural experiments to rule out confounding factors, we nonetheless caution against overinterpreting causal relationships in these data.

Thus far, researchers utilizing these data have examined the role of need for chaos, big five personality, and conspiratorial thinking in pandemic psychology17,18,19, how collective narcissism predicted perceived illegitimacy of the 2020 election20, the independent influence of political and cognitive sophistication on political belief21, and how support for democratic norms shape affect towards Black or Blue Lives Matter22. Nonetheless, there remain many other potential relationships and research questions to explore with these data.

In summary, these two independent panel surveys provide a valuable resource for examining the factors shaping perceptions of electoral legitimacy and their implications for democratic societies. In the context of growing polarization, misinformation, and conflict, these data offer an opportunity to better understand the dynamics of electoral legitimacy and its role in shaping political behaviors and democratic societies.

Methods

We collected two samples of voting-aged Americans: (1) a sample from Amazon Mechanical Turk (MTurk) and (2) a sample from Forthright. Both samples were conducted online through Qualtrics, and participants were only recruited using online methods. For each sample, we collected data at multiple time periods before and after the 2020 and 2022 national U.S. elections (two waves for the MTurk sample and four waves for the Forthright sample). IRB approval from Stony Brook University (Approval Number: IRB2020-00608) and Lehigh University (Approval Number: 1957520-2) was obtained prior to data collection. We obtained informed consent from our subjects and they specifically consented to their answers being shared.

Regarding the first sample, although MTurk is a convenience sample and is not representative of any populations, it is more diverse than student samples23,24 and still allowed us to study causation25. Participants on MTurk were recruited using CloudResearch Turk Prime participant targeting. They saw an advertisement for the study and then could choose to enroll. We ran two waves in MTurk. The first wave was conducted between late October and early November, prior to the 2020 presidential election (N = 1090). The second wave was conducted after the election, from the middle of November until the end of the month (N = 960).

As for the second sample, we used Forthright, an online research panel affiliated with Bovitz, Inc., which is a market research firm that specializes in building and maintaining panels of participants that match national-level demographic information. While not a nationally representative sample, statistical tests show that Forthright holds up well against probability-based samples in approximating national representativeness26. Participants were recruited through the survey firm and randomly selected to participate. We ran four waves. The first wave was fielded between late October and early November, prior to the 2020 presidential election (N = 1127), and the second wave ran in the middle of November after the 2020 election (N = 769). The third wave was collected during the 2022 midterm elections (N = 506), and the fourth wave ran following the election until the end of November (N = 453). We include a timeline of when we ran out studies in Fig. 1.

Fig. 1
figure 1

Timeline of when we ran our studies. Wave 1 from both Sample 1 (MTurk) and Sample 2 (Forthright) were fielded prior to the 2020 presidential election. Wave 2 from both samples occurs after the 2020 presidential election. Wave 3 from Sample 2 occurs before the 2022 midterm election, and Wave 4 occurs after.

Full size image

Participants

Only English-speakers and people living in the United States over 18 years of age were recruited to participate in our survey. After participants were recruited, they received a link to complete the survey via Qualtrics. Participants completed the approximately 15-minute survey and received compensation. In the MTurk sample, they earned $2.00 for the first wave and $1.00 for completing the second wave. Participants in the Forthright sample received $5.17 for their initial participation and then $3.14, $5.17, and $3.14 for each subsequent wave respectively.

Tables 1, 2 summarize key demographic information across the various waves of our two samples. Table 2 shows statistics on the age, gender, race and ethnicity, educational attainment, income, and ideology of respondents in the MTurk sample. Likewise, Table 1 shows the same descriptive statistics for respondents from the Forthright sample.

Table 1 Descriptive demographic information from Sample 2 (Forthright).
Full size table
Table 2 Descriptive demographic information from Sample 1 (MTurk).
Full size table

Materials

Our surveys used multiple validated scales and unique measures to assess relevant political and psychological constructs. We used the following scales in Sample 1 (Mturk): System identity threat27; support for legal, illegal, or violent activism22,28; religiosity29; conspiratorial predispositions4; democratic norms12,22,30; Covid-19 and election Beliefs31; beliefs in voter fraud and election conspiracies15; fraud and interference beliefs4; perceived electoral fairness31; partisan strength and identity32; learned helplessness22,33; right-wing authoritarianism34; ambivalent sexism35; egalitarianism36,37; immigrant resentment15; bias awareness38; racial resentment39; conformity to masculine norms40; moral foundations41; need for chaos17,28; need for structure42; big five inventory19,43; anti-intellectualism18,44; authoritarianism45; cognitive reflection test46; national identity47; patriotism15; collective narcissism20,48; and intellectual humility49. We also developed and administered novel questions to ask about political interest and engagement, political knowledge, media use, Covid-19 conspiracies18,19,21, ideology, voting intentions and preferences, trust22, vaccine uptake, behaviors related to Covid-1919, outcome fairness and acceptance, satisfaction22, internal and external efficacy22, perceived vulnerability to disease, illusion of explanatory depth, white identity, and loser status. We also included feeling thermometers on Liberals, Conservatives, Republicans, Democrats, Donald Trump, Joe Biden, Mitch McConnell, Nancy Pelosi, QAnon, Antifa, Black Lives Matter, Blue Lives Matter, Black Americans, White Americans, Latino Americans, Asian Americans, Muslims, Christians, Jews, and Atheists.

In the Forthright sample (Sample 2), we included the following measures: System identity threat27; support for political violence50; religiosity29; conspiratorial predispositions4; democratic norms12,22,30; Covid-19 and election Beliefs31; beliefs in voter fraud and election conspiracies15; fraud and interference beliefs4; perceived electoral fairness31; partisan strength and identity32; learned helplessness22,33; right-wing authoritarianism34; egalitarianism36,37; immigrant resentment15; bias awareness38; racial resentment39; moral foundations41; big five inventory19,43; anti-intellectualism18,44; authoritarianism45; cognitive reflection test46; national identity47; patriotism15; collective narcissism20,48; support for legal, illegal, or violent activism22,28; and intellectual humility49. We also developed and administered questions about political interest and engagement, political knowledge, media use, Covid-19 conspiracies18,19,21, ideology, voting intentions and preferences, trust22, vaccine uptake, behaviors related to Covid-1919, outcome fairness and acceptance, satisfaction22, internal and external efficacy22, perceived vulnerability to disease, illusion of explanatory depth, white identity, loser status, attitudes toward January 6th, issue positions, and attitudes toward misinformation. We also included feeling thermometers on Liberals, Conservatives, Republicans, Democrats, Donald Trump, Joe Biden, Mitch McConnell, Nancy Pelosi, QAnon, Antifa, Black Lives Matter, Blue Lives Matter, Black Americans, White Americans, Latino Americans, Asian Americans, Muslims, Christians, Jews, and Atheists.

Finally, a wide range of demographic information was collected across both samples, including age, gender, race and ethnicity, income, educational attainment, etc. We also collected zip-codes as geographic indicators in both samples. For Sample 2 (Forthright), we also collect regional variables, like rural, urban, etc. Full question wording (including response scales and coding) can be found in the Codebook via OSF.

Question blocks

Our goal was to include many types of scales and questions (as seen above) at different waves. For Sample 1, most questions were asked in both waves. Some questions were only included in the first wave, and all participants responded to them (Covid-19 conspiracies, political interest and engagement, political knowledge, media use, and religiosity). We also wanted to include many additional scales and questions but we were limited with our sample size and survey length. To increase the number of questions captured, we decided to break our samples into two blocks. Subjects were randomly selected into one of the two blocks. Each block contained a different set of questions. Note that we did not have any experimental hypotheses, we were trying to maximize the responses we received on various types of questions. In Block 1 of Sample 1, participants responded to the following questions: right-wing authoritarianism, egalitarianism, perceived vulnerability to disease, immigrant resentment, bias awareness, racial resentment, conformity to masculine norms, and ambivalent sexism. In Block 2, participants received a different set of questions: big five inventory, moral foundations questions, anti-intellectualism, illusion of explanatory depth, authoritarianism, cognitive reflection test, national identity, patriotism, white identity, collective narcissism, intellectual humility, need for chaos, and need for structure. Note that both randomized blocks were only run in the first wave of the survey. We used the same randomized block method for Sample 2. Here, Block 1 contained questions on right-wing authoritarianism, egalitarianism, perceived vulnerability to disease, immigrant resentment, bias awareness, and racial resentment. Block 2 contained questions on the big five, moral foundations, anti-intellectualism, illusion of explanatory depth, authoritarianism, cognitive reflection test, national identity, patriotism, white identity, and collective narcissism. Both Wave 3 and 4 in Sample 2 included new questions but all these questions were asked to the entire sample, not to randomized blocks.

Data cleaning

Once we received our data via Qualtrics, we proceeded to anonymize the datasets by removing identifying information like IP Address, location, etc. Each wave from each sample came in a different datafile. Each variable in the datafiles was renamed to match the intent of the variable. Scales were labelled so that readers can differentiate between the first, second, third, etc. items. Most of our variables were recoded from 0 to 1—coding information on each variable is included in the Codebook. Additionally, missing responses were recoded to “NA”. Once each datafile was cleaned, we then merged all waves of each sample together to create one large dataset per sample. Variables are marked with “_w1” if they were asked in Wave 1, “_w2” if found in Wave 2, and so forth. When there were multiple items for a scale, we created those scales in each wave that they appeared. For your reference, we also include information on the Cronbach’s alpha, mean, and standard deviation of each scale in the codebook. Within samples, we also run t-tests to check for differences across waves in our scales. We report the t-stat, degrees of freedom, and p-value in the Codebook as well.

Data Records

The dataset is available at Open Science Framework (OSF)51, with this section being the primary source of information on the availability and content of the data being described.

All materials corresponding to this project can be found on Open Science Framework (OSF), linked here: https://osf.io/r9w85/. The folder titled Raw Data contains all six data files associated with this project (two MTurk waves and four Forthright waves). All raw data files have been anonymized and are saved as CSV files.

We used R Studio to clean the datasets and run statistical analyses. We used the following R packages: tidyverse52, car53, dplyr54, and psych55. The folder R Scripts contains important R Scripts: First, we uploaded a general clean up file for each sample (Mturk Cleaning and Forthright Cleaning)—this code renames variables in an intuitive way, drops missing observations, recodes variables, and creates scales. This code also merges waves together into a combined CSV file for each sample. The next R Script contains code to test the scales created in each sample (Scale Testing)—each scale has code for Cronbach’s alpha, mean, and standard deviation. We also run t-tests to check for differences across waves. The Descriptive Information R Script includes code for descriptive statistics on the demographic variables captured (this code was used to generate Tables 1 and 2). Finally, the last R Script includes code measuring attrition across our waves using chi-squared and t-tests (Attrition Analysis).

We have a Cleaned Data folder that contains cleaned data from our two samples. The last folder is titled Codebook and Other Files. First, we have the codebook, which contains question wording as well as the results of the statistical tests for scales generated. Next, we have the Attrition Analysis which shows differences in the demographic information across each wave of both samples.

All the files a part of this packet have been cleaned and prepared for immediate data analysis. Data files are saved as CSV files so researchers can use a wide variety of programs for data analysis, including Excel, R Studio, Stata, etc. Clean up files and statistical testing files were generated using R Studio, so we recommend R Studio for data analysis. We also recommend referencing our Codebook for question wording as well as more detailed information on scale reliability and other statistical tests.

For the sake of subject privacy, we have removed key identifying information from the datasets.

Technical Validation

To validate the data we collected, we ran several tests using both samples. First, we checked the reliability of all scales generated in each wave. We report Cronbach’s alpha for each scale in our codebook, as well as provide code for this type of testing in our materials. We also run t-tests comparing our scales across waves. In Sample 1, we compared scales in Wave 1 to Wave 2. For Sample 2, we ran t-tests between each wave that a scale was measured. For example, System Identity Threat was captured in all four waves, so we looked for differences from Wave 1 to Wave 2, Wave 1 to Wave 3, Wave 1 to Wave 4, Wave 2 to Wave 3, Wave 2 to Wave 4, and Wave 3 to Wave 4. These results are reported in the Codebook.

Since we use multiple waves to create panel data, we also assessed the attrition across our samples. In Sample 1, 12% of the sample from Wave 1 drops out by Wave 2. We experienced more significant attrition across the waves for Sample 2: 29% drops from Wave 1 to Wave 2, 34% drops out from Wave 2 to Wave 3, and 10% drops from Wave 3 to Wave 4. From Wave 1 to Wave 4, Sample 2 decreases by 58%. We ran t-tests and chi-squared tests to look at differences in the demographic characteristics at different waves. These results are reported in the Attrition Analysis document.