Big Five personality traits and voting: A systematic review,meta-analysis,and mega-analysis

Personality and voting

Despite theoretical reasons to expect associations between Big Five traits and voting behaviors, empirical evidence has been inconsistent, with most studies indicating small, non-significant, or mixed effects. Here, we review existing research on the Big Five and voting, while highlighting methodological limitations that could explain mixed results.

Potential reasons for mixed evidence on personality traits and voting

Overall, the existing evidence paints a very unclear picture of the role the Big Five personality traits may play in voting intentions and behavior. It is possible that there are robust effects that are masked by methodological problems, such as low statistical power or measurement-related issues. Mixed findings may also indicate the presence of moderators such as the distinction between voting intentions and actual voting behaviors, and the sociodemographic characteristics of the study participants, including variables like age, gender, and cultural background. We review each of these potential factors presently.

Synthesizing existing results

To estimate the small effects of Big Five personality traits on voting intentions and behaviors and to assess methodological influences, robust and integrative data analysis methods are essential. Our approach encompasses two key methodologies: meta-analysis and mega-analysis.

Meta-analysis, a well-established technique, aggregates effect sizes from various studies at the sample level. Its strengths lie in the ability to combine findings from a broad spectrum of studies without requiring raw data, as relevant information is typically accessible through published articles and supplementary materials. However, meta-analysis is susceptible to biases like selective reporting and issues with handling outliers or rounding errors, which can be particularly challenging when dealing with small effect sizes (Borenstein et al., 2021; Cafri et al., 2010; Hohn et al., 2019). Thus, while meta-analysis provides a valuable perspective on the personality-voting connection, it may yield somewhat limited results because the quality of the studies may vary and because of potential publication biases.

Mega-analysis, on the other hand, involves pooling individual-level data, standardizing measures across studies, and applying uniform analytical techniques to the raw data from existing studies. This approach offers increased statistical power, facilitating a more accurate effect size estimation and enhanced capability for moderation testing (see, e.g., Beck & Jackson, 2022). Critically, the rigorous use of mega-analytic techniques requires sufficient raw data and the standardization of different measures across samples, which can introduce variability and complicate the interpretation of mega-analytic results (Burke et al., 2017; Curran & Hussong, 2009). Consequently, while mega-analysis addresses some limitations of meta-analysis by offering greater power, it also presents its own challenges, such as the exclusion of studies where data access is unfeasible, or when measures are incompatible with the research questions.

Given that both meta- and mega-analytic approaches have strengths and limitations, a combined approach should prove particularly effective in addressing their individual limitations, especially when dealing with small effect sizes and when complete access to all individual studies is not feasible. Here, we adopted a dual-study approach, applying both meta-analysis and mega-analysis to a dataset gathered through systematic review. The integration of these two approaches aims to provide a richer understanding of how the Big Five correlate with voting intentions and behavior, including the exploration and testing of hypotheses about potential moderators. The intention of this integrated analysis is to generate robust estimates, enhancing their reliability and the potential for replication in subsequent individual studies.

The present study

In this study, we used data from a systematic review to investigate the connections between the Big Five personality traits and voting. Our research methodology and analyses involved three distinct phases: (1) a systematic review process, (2) meta-analyses of the links between the Big Five and voting, and (3) mega-analyses of the links between the Big Five and voting. We consider the combination of meta- and mega-analyses as analogous to a sensitivity analysis (Nosek et al., 2022; Simonsohn et al., 2020). Our primary goal was to identify robust patterns of results that hold true across both analytical approaches, thus providing the most comprehensive and replicable estimates of the Big Five-voting links.

Our literature review suggested that results were consistently positive for openness and extraversion when significant. Conscientiousness and agreeableness showed similar trends, albeit with sporadic negative correlations. The evidence was more mixed for neuroticism, with a 2:1 ratio of negative to positive correlations. Using a benchmark of r ∼.05 and .10 for very small to small effects, respectively (see Bosco et al., 2015; Gignac & Szodorai, 2016), we thus preregistered the following hypotheses (see https://osf.io/4z6s9):

• H1a–H1e: The correlations between Big Five personality traits and voting will be small (r ≥ .05). The correlations with (a) openness, (b) conscientiousness, (c) extraversion, and (d) agreeableness will be positive and the correlation with (e) neuroticism will be negative.

Beyond estimating zero-order correlations, we also examined the moderating effects of various factors. These included the type of assessment tool (brief versus longer), the conceptualization and operationalization of voting (intention versus behavior, dichotomous versus continuous scaling), and sociodemographic variables (nationality, gender, and age). We additionally preregistered the following hypotheses:

• H2a–H2e: Personality-voting correlations will be strongest for voting attitudes and continuous scales, less strong for voting intent, and weakest for reported voting and dichotomous scales.

In further considering the assessment tool used (brief/longer) as a potential moderator, we did not anticipate finding larger effects with longer measures. However, we anticipated that these more comprehensive tools might reduce standard errors, thereby increasing statistical power. Regarding the potential moderating roles of nationality and gender, we did not formulate specific predictions. Instead, we planned to explore their potential influence on the personality-voting link.

After estimating all meta-analytic and mega-analytic models, we consolidated the evidence to test hypotheses. If both types of analysis concurred on a significant effect, such as a small positive correlation (r ≥ .05) between openness and voting, we deemed our hypothesis confirmed. If only one type of analysis suggested such an effect, we viewed our hypothesis as partially supported. If both analyses indicated no effect or contradictory effect sizes, we regarded our hypothesis as unsupported.

Transparency and openness

We report how we determined our sample size, all data exclusions (if any), all manipulations, and all measures in the study.

We adhered to the PRISMA 2020 guidelines for systematic reviews (Page et al., 2021). All data, analysis code, and research materials (including our coding scheme) are available on OSF (https://osf.io/mq2n5). The review project was also preregistered on OSF, with an initial version (https://osf.io/3abwz) and a subsequent revised version (https://osf.io/u9d68).

We adhered to the MARS guidelines for meta-analytic reporting (Appelbaum et al., 2018). All meta-analytic data, analysis code, and research materials (including our coding scheme) are available on OSF (https://osf.io/mq2n5). The mega-analytic harmonization and analysis code are available in the same OSF project. The mega-analytic dataset cannot be shared, as it was compiled from multiple databases and repositories across different countries, each subject to its own legal regulations. We obtained permission to use these data exclusively for the purposes of this study and formalized this agreement through the required documentation. Nonetheless, access to the original datasets should be attainable for other researchers interested in conducting independent analyses. Data were analyzed using R, version 4.3.0 (R Core Team, 2023) and the packages metafor, version 4.4-0 (Viechtbauer, 2023), clubSandwich, version 0.5.10 (Pustejovsky, 2023), and lme4, version 1.1-35.1 (Bates et al., 2023). The meta-analysis and mega-analysis were also preregistered on OSF (https://osf.io/4z6s9). The preregistration was submitted after conducting the systematic review, effect size coding, and data retrieval but before undertaking the meta-analysis and mega-analysis.

Identification of studies

The systematic review targeted empirical, quantitative studies published in either English or German—the working languages of the authors. The studies were sourced through a search of article titles, abstracts, and indexed keywords in Scopus (primary database), along with the titles and keywords indexed in BASE (grey literature).

The following search terms were used: (personality OR “Big Five” OR “five factor model” OR HEXACO OR extraver* OR introver* OR surgency OR agreeable* OR conscientious* OR openness OR “open to experience” OR “open to experiences” OR intellect OR neurotic* OR “emotional stability” OR “emotionally stable” OR emotionality OR honesty OR humility) AND (volunteeri* OR charit* OR philanthrop* OR donat* OR “don* blood” OR “blood don*” OR “don* organ*” OR “organ don*” OR “vot” OR “politic* particip*” OR “politic* behav*” OR “politic* involve*” OR “politic* engage*” OR vaccin*). ⁴ Two search queries were executed, the first on February 25, 2022, and the second on March 1, 2023. The second query was necessary after recognizing the omission of certain relevant cited articles due to the initial search terms’ lack of comprehensiveness. Specifically, the following terms pertaining to voting were deemed missing from the first query: “politic* particip*,” “politic* behav*,” “politic* involve*,” and “politic* engage*.”

These specific search terms were chosen because they were commonly used in key prior studies on personality and voting (e.g., Mondak et al., 2010; Mondak & Halperin, 2008). Some terms (e.g., “politic* engage*”) were not central to our core constructs but were included to avoid missing studies in which voting intentions or behaviors were only secondary outcomes or control variables. We tested the final set of keywords repeatedly to ensure that it captured all studies already known to the author team.

The search results were acquired automatically from Scopus through Christopher Belter’s Scopus Search API (https://github.com/christopherBelter/scopusAPI), while BASE search results were gathered manually. The R script that was used for the study retrieval is available on OSF (see https://osf.io/x9pzh). A total of 4,677 studies were identified on Scopus and 1,343 on BASE. Of these, 1,403 were automatically excluded as they were flagged as conference reviews, editorials, letters, notes, reviews, musical notations, maps, audios, images/videos, software, datasets, or because they were duplicates of already included studies.

Screening, eligibility, and inclusion of studies

Each of the remaining 4,617 studies was independently screened by the first author or one of four graduate students. Their focus at this stage was to ascertain whether the studies included measures of Big Five or HEXACO personality traits and civic engagement, with a preference for inclusion when in doubt. Both self-reported and peer-reported data were eligible for inclusion, and civic engagement indicators could be measured using dichotomous or continuous scales, as long as they corresponded to behavioral attitudes, intentions, or actual behaviors. The study design could also vary, provided it was either correlational (cross-sectional, longitudinal, and panel studies) or experimental (lab and field studies). Meta-analyses, clinical trials, and single-case studies were excluded, as well as studies reporting on economic games, such as the Dictator Game or Ultimatum Game, to prevent redundancy with another recent meta-analysis (Thielmann et al., 2020).

The screening was primarily based on the titles, keywords, and abstracts, turning to the full text only when uncertainties arose. During this process, the raters remained blind to publication metrics provided by Scopus and BASE, which included journal titles, study authors, and citation counts. To ensure high interrater consistency, a calibration process was undertaken prior to the screening phase. This process involved the repeated assessment of 100 study batches until the raters achieved an agreement rate exceeding κ = 95%. This calibration process was carried out three times, resulting in the collective appraisal of 300 studies. In the end, an additional 4,390 studies were excluded due to the absence of measures pertaining to both Big Five or HEXACO personality traits and civic engagement, or because they failed the other eligibility criteria.

Of the remaining 227 studies, three were excluded because they were unattainable despite best efforts, 11 were excluded because they were not available in English or German, and 87 were excluded because they were duplicates of previously included studies. The first author and two of the graduate students who had been involved in the previous screening stage independently assessed the full texts of the remaining 126 studies, ensuring that each study was reviewed by all three raters. During this process, an additional 21 studies were excluded as they were based on populations that were expected to have high scores in one or all civic engagement indicators (for example, samples only comprising individuals who had voted in a previous election) or because they reused a sample already included in a different study. The final selection comprised 105 studies, with 27 of these including measures of voting. Table 1 gives an overview of these studies, including demographic information, the conceptualization/operationalization of voting, and the reported direction of effects, if any were detected. The complete data sheet detailing study codes, variables, and effect sizes is also available on OSF (see https://osf.io/zk6m8).

OPEN IN VIEWER

Table 1.

Summary of studies identified in the systematic review, categorized by inclusion in meta-analysis, mega-analysis, or both.

No.	Code	Study	Country	N	Women	Age M	Concept.	Operat.	N	E	O	A	C	Use	Data	Sampling
1	854	Schoen and Steinbrecher (2013)	DEU	2,115	52.8%	50.2	Behavior	Dichotomous	-	0	0	+	0	Both	GLES 2009	Random
2	877	Gerber et al. (2013)	USA	1,444	55.2%	31.0	Intention	Continuous	0	0	0	0	0	Both	Study sample S1	Convenience
2	877	Gerber et al. (2013)	USA	3,382	58.9%	60.5	Behavior	Dichotomous	-	0	0	-	-	Both	Study sample S2	Convenience
3	928	Wang (2014)	USA	2,606	57.8%	53.1	Intention	Continuous	-	0	0	0	0	Both	ANES 2010	Representative
4	1660	Furnham and Cheng (2019)	GBR	9,790	50.7%	50.0	Behavior	Dichotomous	+	+	+	+	+	Both	NCDS	Population
5	2217	Sindermann and Montag (2023)	DEU	3,325	60.3%	37.7	Intention	Dichotomous						Both	Study sample S3	Convenience
6	2453	Sindermann et al. (2021)	DEU	4,286	53.8%	38.0	Intention	Dichotomous	0	0	+	+	0	Both	Study sample S4	Convenience
7	2524	Brandt et al. (2022)	DEU	2,687	61.6%	39.8	Intention	Dichotomous	-	+	+	+	0	Both	BIJU: Wave 8	Random
8	2750	Gerber et al. (2011)	USA	3,367	51.3%	50.2	Intention	Dichotomous	0	0	0	0	0	Both	CCAP 2008	Weighted
8	2750	Gerber et al. (2011)	USA	3,367	51.3%	50.2	Intention	Dichotomous	0	0	0	0	0	Both	CCAP 2008	Weighted
8	2750	Gerber et al. (2011)	USA	3,367	51.3%	50.2	Behavior	Continuous	-	0	0	-	-	Both	CCAP 2008	Weighted
8	2750	Gerber et al. (2011)	USA	3,367	51.3%	50.2	Behavior	Continuous	0	0	0	0	0	Both	CCAP 2008	Weighted
9	3038	Smith et al. (2019)	USA	800	51.2%	47.6	Behavior	Continuous						Both	Study sample S5	Representative
10	758	Gallego and Oberski (2012)	ESP	3,459	52.2%	48.6	Behavior	Dichotomous						Mega	PEALE 2009	Weighted
11	2785	Verhulst (2012)	USA	1,349	61.2%	57.2	Intention	Continuous						Mega	MTPS	Convenience
12	2843	Weinschenk (2013)	USA	1,500	52.4%	48.4	Intention	Dichotomous						Mega	AB 2010	Weighted
13	3024	Weinschenk et al. (2019)	DEU	4,096	58.4%	23.0	Behavior	Dichotomous						Mega	TwinLife Wave 1	Random
14	579	Anderson (2009)	USA	818			Behavior	Dichotomous	-	0	0	0	0	Meta
15	678	Mattila et al. (2011)	FIN	212		36.0	Behavior	Continuous	0	+	0	+	0	Meta
15	678	Mattila et al. (2011)	FIN	212		36.0	Behavior	Continuous	0	+	0	+	+	Meta
16	1421	Wang et al. (2017)	KOR	1,245	54.0%	50.5	Behavior	Continuous	0	0	0	0	0	Meta
17	1912	Wang et al. (2019)	TWN	839	50.0%	44.2	Behavior	Dichotomous	0	0	0	0	0	Meta
18	2707	Mondak and Halperin (2008)	USA	404			Behavior	Dichotomous	0	0	0	0	0	Meta
18	2707	Mondak and Halperin (2008)	USA	1,338			Behavior	Dichotomous	0	0	0	0	0	Meta
19	2732	Mondak et al. (2010)	USA	1,062			Behavior	Dichotomous	+	0	+	0	0	Meta
20	2788	Cooper et al. (2013)	USA	760	68.0%	47.6	Behavior	Dichotomous	0	0	0	+	+	Meta
21	3115	Huber et al. (2021)	BRA	1,086			Behavior	Continuous						Meta
21	3115	Huber et al. (2021)	KOR	944			Behavior	Continuous						Meta
21	3115	Huber et al. (2021)	RUS	1,145			Behavior	Continuous						Meta
21	3115	Huber et al. (2021)	GBR	1,064			Behavior	Continuous						Meta
21	3115	Huber et al. (2021)	USA	1,161			Behavior	Continuous						Meta
22	903	Turska-Kawa (2013)	POL	726	52.6%		Behavior	Continuous						None
23	1385	Weinschenk and Panagopoulos (2014)	USA	724	46.0%	34.0	Intention	Continuous	0	0	0	0	0	None
24	1467	Weinschenk and Dawes (2018)	USA	724	46.0%	34.0	Attitude	Continuous	-	+	+		+	None
25	1690	Pruysers et al. (2019)	CAN	371	58.0%	49.2	Attitude	Continuous						None
26	1880	Ivanchenko et al. (2019)	UKR	1,247	59.0%	25.8	Behavior	Continuous	+	0	0	0	+	None
27	2504	Lynch (2008)	USA	63	70.0%	22.0	Behavior	Dichotomous						None
27	2504	Lynch (2008)	USA	63	70.0%	22.0	Intention	Continuous						None

Coding of studies

From the 27 studies identified in the systematic review, the first author and the two graduate students used a standardized coding manual (see https://osf.io/4hrz7) to code the (1) sample size, (2) effect size, (3) the type of effect size statistic, (4) the number of items used for the Big Five measurement, (5) the conceptualization of voting (attitude/intention/behavior), (6) the operationalization of voting (dichotomous/continuous), (7) the geographic region where the sample was collected, (8) the percentage of female participants, and (9) the average age of participants. In the single longitudinal study identified, we only included the initial measurements of personality traits measured at age 33 and voter turnout at age 36, while excluding subsequent personality assessments at ages 42 and 50 (Mattila et al., 2011). All coded variables and statistics were then cross-checked amongst the coders for consistency. Discrepancies in coding were resolved through discussion until a consensus was reached.

Raw effect sizes suitable for meta-analysis, that is, Pearson’s correlations, standardized mean differences, or odds ratios, were not reported in 24 studies (88.89%). In such instances, attempts were made to contact the respective authors or calculate the missing effect sizes from the data and supplementary materials available. The R script that includes these calculations is available on OSF (see https://osf.io/x9pzh). Effect sizes were obtainable from the authors in nine instances and could be computed from the available data in six cases. However, effect sizes could not be procured for the remaining nine studies, leaving 18 studies as the provisional basis for the meta-analysis. Upon further analysis of the moderator variables, we noticed that most studies employed measures of voting intention or behavior, whereas only one study used an attitude measure of voting (Weinschenk & Dawes, 2018). To maintain consistency in voting operationalization and avoid confusion, we decided to exclude this study. Consequently, the meta-analysis was based on a final sample of 17 studies that included measures of voting intention or behavior.

Summary of the literature

Table 2 summarizes key statistics for the 17 studies included in the meta-analysis. These studies encompass 27 datasets, owing to some studies reporting different Big Five measures or multiple voting measures (e.g., Gerber et al., 2011; Gerber et al., 2013; Mattila et al., 2011).

OPEN IN VIEWER

Table 2.

Descriptive statistics of data included in the meta-analysis, mega-analysis, or both.

	Meta-analysis	Mega-analysis	Included in both types of analyses
			Systematic review data	Mega-analytic data
Total sample size (N)	65,036	44,206	42,598	33,802
Total study size (k)	17	13	9	9
Individual dataset size (n)
Mean (M)	1,871	3,158	2,811	3,380
Standard deviation (SD)	1,716	2,192	1,915	2,472
Minimum (Min)	212	800	800	800
Maximum (Max)	7,135	9,790	7,135	9,790
Percentage of women (%)
Mean (M)	56.0	54.8	54.6	54.4
Standard deviation (SD)	6.0	4.0	4.7	3.9
Minimum (Min)	50.0	50.7	50.0	50.7
Maximum (Max)	68.0	61.6	62.0	61.6
Average age (years)
Mean (M)	42.8	46.2	42.5	46.8
Standard deviation (SD)	9.3	9.5	10.5	7.9
Minimum (Min)	22.8	23.0	22.8	31.0
Maximum (Max)	52.4	60.5	52.4	60.5
Number of items
Mean (M)	6.4	4.4	4.9	4.6
Standard deviation (SD)	6.8	3.3	3.6	3.4
Minimum (Min)	1.0	1.0	1.0	1.0
Maximum (Max)	36.0	10.0	10.0	10.0
Voting conceptualization as intention (%)	22.2	46.7	38.5	54.5
Voting operationalization as dichotomous scale (%)	48.1	66.7	46.2	63.6
Geographic locations in the U.S. (%)	51.9	50.0	61.5	50.0

The average sample size across individual studies was 1,871 participants (SD = 1,716, Range = 212–7,135). Gender distribution was approximately balanced in most studies, with 56.0% women on average and a maximum of 68.0% in Cooper et al. (2013). Participants were mostly middle-aged adults, with an average age of 42.8 years, and a minimum age of 22.8 years in one study (Brandt et al., 2022). Most studies used scales with one or two items per trait (i.e., the FIPI or TIPI; Gosling et al., 2003) or the BFI (eight to ten items per trait; John et al., 1991) to measure the Big Five personality traits. An exception was the study by Mattila et al. (2011), which employed a 36-item-per-trait version of the NEO inventory. There was a slight preference for measuring voting behavior over voting intentions in the included studies. However, the distributions of voting operationalization (dichotomous/continuous) and geographic location (U.S./other) were relatively balanced. Outside the United States, datasets were sourced from various countries, including Germany (Brandt et al., 2022; Schoen & Steinbrecher, 2013; Sindermann et al., 2021; Sindermann & Montag, 2023), Great Britain (Furnham & Cheng, 2019; Huber et al., 2021), Finland (Mattila et al., 2011), South Korea (Huber et al., 2021; Wang et al., 2017), the Island of Taiwan (Wang et al., 2019), Brazil (Huber et al., 2021), and Russia (Huber et al., 2021).

Analysis plan

Consistent with broader research trends, the systematic review indicated that the literature about the links between the Big Five and voting can be best understood through a three-level model (see Cheung, 2014; Pastor & Lazowski, 2018). In this model, individual data are nested in measures of different personality traits or voting conceptualizations/operationalizations, which are then nested within individual studies. ⁵ We aimed to mirror this nested structure to separate and quantify variations both within individual studies and across the larger research corpus. In the meta-analytic framework, we hence implemented and tested three-level models (Assink & Wibbelink, 2016; Cheung, 2014).

First, we transformed all raw effect sizes into Fisher’s z-scores and estimated their standard errors (Borenstein et al., 2021; Fisher, 1921). Then, as a first sensitivity analysis, we fitted five intercept-only three-level Correlated and Hierarchical Effects (CHE) models with robust variance estimation to the data, setting different constants for within-study correlations (ρ = .00, .20, .40, .60, .80). This approach was chosen due to the presence of multiple effect sizes per study and the cultural diversity of the studies, making CHE models more suitable than random-effects models (Assink & Wibbelink, 2016; Cheung, 2014). The analysis showed that varying the constant had minimal impact on the model estimates, with differences emerging only at the third decimal place. However, given the likelihood of some correlation between effect sizes within the same study, and considering that measurement error can produce small correlations even when the true correlation is zero, we opted for the ρ = .20 model, which assumes within-study correlations of the Big Five traits to settle around .20, due to common-method effects (Borenstein et al., 2021; Schmidt & Hunter, 2014).

Subsequently, we fitted a series of three-level CHE models with robust variance estimation to the data, using restricted maximum-likelihood estimation (REML) alongside Hartung-Knapp-Sidik-Jonkman adjustment to account for potential overestimation of effect size precision due to variability among studies (Borenstein et al., 2021; Knapp & Hartung, 2003; Sidik & Jonkman, 2002). Additionally, we corrected the confidence and prediction intervals by employing a sandwich estimator for standard errors and the Satterthwaite approximation for critical values (Satterthwaite, 1946; Tipton & Pustejovsky, 2015). The initial model was structured to treat the Big Five personality traits as categorical moderators influencing voting outcomes, with each trait represented by a separate dummy variable (0 or 1) to specify the corresponding effect size in a long-format dataset. We then progressively included additional moderators in the analysis: (1) the number of items used for personality measurement (brief/longer), (2) the conceptualization and operationalization of voting (intention/behavior, and dichotomous versus continuous scaling), and (3) sociodemographic variables (nationality, gender, age). As a final step, we converted the model statistics back into correlations and reported point estimates, along with corrected 95% confidence intervals and further model diagnostics, such as I² (Borenstein et al., 2021; Fisher, 1921).

Robustness checks and deviations from preregistration

All of our analyses were preregistered and grounded in a limited set of a priori hypotheses, and thus we did not apply formal multiple-comparison corrections (e.g., Bonferroni adjustments). In our baseline models, we tested five primary trait–outcome associations—one for each Big Five personality trait—each specified in advance. Although our moderator models involve additional parameters (including interaction effects), these analyses were also determined a priori, thereby minimizing the risk of spurious findings that commonly occur in exploratory examinations (Perneger, 1998; Rothman, 1990). In this context, overly conservative adjustments would unduly elevate the risk of Type II errors (i.e., false negatives), potentially obscuring genuine theoretical effects (Gelman et al., 2012; McShane et al., 2019). Consequently, we emphasize effect size estimation by reporting point estimates along with robust 95% confidence intervals, and we employ robust variance estimation to account for dependencies among effect sizes (Tipton & Pustejovsky, 2015). This approach is consistent with meta-analytic best practices, which favor estimation over reliance on dichotomous significance thresholds (Borenstein et al., 2021; Schmidt & Hunter, 2014), thereby ensuring that our inferences reflect both the magnitude and precision of the observed effects.

To assess the possible influence of outliers, we visually inspected forest plots and calculated Cook’s distance and DFFITS for each Big Five trait. Due to the lack of established cutoffs for influential outliers, we performed sensitivity analyses with and without studies showing extreme deviations or high Cook’s distance and DFFITS values (Viechtbauer & Cheung, 2010). We retained these studies in our models when sensitivity analyses yielded similar results (these ancillary analyses are available on OSF; https://osf.io/xjuqm). When significant differences, like effect direction changes or new statistical significance, were observed, we included both sets of analyses in our results. For small-study effects, we examined contour-enhanced funnel plots and Egger’s regression test for each trait (Egger et al., 1997; Peters et al., 2008). In cases of asymmetry or indicated small-study effects, we applied the PET-PEESE method for bias-corrected correlations (PET-PEESE; Borenstein et al., 2021; Stanley & Doucouliagos, 2014).

Initially, our plan included examining the moderating effects of reliability, gender, and age (see https://osf.io/4z6s9). However, challenges arose due to data limitations. Many studies used only one or two items per Big Five trait and did not report reliability indices. This lack of data reduced our sample to just three studies for the reliability analysis, leading to inflated standard errors and the subsequent decision to exclude this moderator from the meta-analysis. We encountered similar problems with gender and age, where the available data narrowed down to only 10 and 8 studies, respectively. This significant reduction in sample size also resulted in larger standard errors, rendering these analyses unfeasible in the meta-analytic framework. Nonetheless, the results from these preliminary estimations are available on OSF (https://osf.io/xjuqm).

Results

Figure 2 shows the forest plot of effect sizes from the studies considered in the meta-analysis. Table 3 shows the main effects of the meta-analytic correlations between the Big Five personality traits and voting, without considering the influence of moderators. Figure 3 illustrates the meta-analytic correlations between the Big Five personality traits and voting across various moderator models, including corrected 95% confidence intervals. Individual statistics, including 95% prediction intervals for the correlations, are accessible at OSF (https://osf.io/57kgd).OPEN IN VIEWER

OPEN IN VIEWER

Table 3.

Main effects of the meta-analytic and mega-analytic personality-voting correlations.

	Meta-analysis		Mega-analysis
	r	95% CI	r	95% CI
Neuroticism	−.086	[−.112, −.060]	−.035	[−.044, −.025]
Extraversion	.052	[.024, .079]	.099	[.079, .119]
Openness to experience	.053	[.025, .080]	.105	[.084, .125]
Agreeableness	.065	[.038, .092]	.100	[.080, .121]
Conscientiousness	.048	[.020, .075]	.093	[.072, .113]

Note. Number of effect sizes in meta-analysis = 132, k_{Meta-analysis} = 17, I²_Within = 71.4%, I²_Between = 9.9%; N_{Mega-analysis} = 41,077, k_{Mega-analysis} = 13.

OPEN IN VIEWER

Figure 3.

Meta-analytic personality-voting correlations, with corrected 95% confidence intervals. Note. ES = Number of effect sizes; k = Number of studies; I²_w = Within-study heterogeneity; I²_b = Between-study heterogeneity. In order of complexity, the reference conditions of the moderator models are (1) the Ten Item Personality Inventory (Gosling et al., 2003), contrasted with the Big Five Inventory (John et al., 1991); (2) voting intent and use of a dichotomous scale, contrasted with reported voting and use of a ratio scale; and (3) samples from outside the U.S., contrasted with samples from the U.S.. In each line plot, the limits denote the corrected 95% confidence intervals. Black points indicate point estimates where the confidence interval excludes zero. Confidence intervals are corrected using a sandwich estimator for the standard errors and Satterthwaite approximation for the critical values (Satterthwaite, 1946; Tipton & Pustejovsky, 2015).

Individual data retrieval and harmonization

Of the studies identified in the systematic review, several were derived from publicly accessible data, and were thus retrievable for a mega-analysis. Table 1 includes a column labeled “Data” that lists these 14 datasets, spread across 13 studies. Of these 14 datasets, nine originated from established panel studies using comprehensive sampling techniques, whereas the remaining five were convenience samples tailored to the specific goals of each study. We initially planned to incorporate data from three additional panels identified in the systematic review: The 1992 Jyväskylä Longitudinal Study of Personality and Social Development (JYLS; Mattila et al., 2011), the 2012 Korean General Social Survey (KGSS; Wang et al., 2017), and the 2006 Cooperative Congressional Election Study (CCES; Mondak et al., 2010). Unfortunately, despite extensive efforts, access to these datasets was not feasible, resulting in 14 datasets used for the mega-analysis.

Before pooling the data, the various instruments used to measure personality and voting were harmonized by the first author. The detailed R script for this process is accessible on OSF (https://osf.io/v7j69). For the Big Five personality measures, we computed the trait scores independently rather than using the provided scores. This approach was feasible in every case except for the German Longitudinal Election Study (GLES; Schoen & Steinbrecher, 2013), which utilized the one-item-per-trait TIPI measure (Gosling et al., 2003). Missing values in any item were excluded in a pairwise manner, and scores indicating emotional stability were inverted to reflect neuroticism. Subsequently, all Big Five scores were converted to “Percentages Of the Maximum Possible” scores (POMPs; Cohen et al., 1999). POMPs are advantageous over other standardization methods, as they retain crucial data characteristics like mean, variance, skewness, and kurtosis. They are calculated as the ratio of a score’s deviation from the minimum to the range between maximum and minimum possible scores. No further weighting, for instance based on questionnaire length, was applied to the personality scores, as that variable was intended for use as a moderator later and should not be incorporated twice.

Voting measures that were not already dichotomized were transformed into dichotomous variables, given that most measures in the dataset were dichotomous. Each non-dichotomous voting variable was assessed for its content and distribution to decide which scale points should be used as the poles of the new dichotomous variable. This dichotomization was carried out in five datasets. In The American National Election Studies (ANES; Wang, 2014) and in study samples 1 (Gerber et al., 2013) and 5 (Smith et al., 2019), where voting intention was initially measured on equidistant five-, ten-, or one-hundred-point scales, all scores exceeding 75% voting intention were recoded as one (“yes”) and all scores at 75% or below were recoded as zero (“no”). In the Cooperative Campaign Analysis Project (CCAP; Gerber et al., 2011), voting behavior was aggregated over elections in 2000, 2002, 2004, and 2005, resulting in a five-point scale ranging from 0 (individuals who never voted) to 4 (individuals who voted in every election). A score of one (“yes”) was assigned to those with a total of one or higher (indicating participation in at least one election), while a score of zero (“no”) was assigned to those with a total of zero (indicating no participation). In the Minnesota Twins Political Survey (MTPS; Verhulst, 2012), voting intention was measured on a four-point scale (0 = “definitely will not vote”, 1 = “probably will not vote”, 2 = “probably will vote”, 3 = “definitely will vote”), and the scores were split so that an inclination toward voting was coded as one (“yes”) and not voting as zero (“no”).

Additionally, sociodemographic variables were standardized. The gender variable was harmonized to include only male and female categories, and the nationality variable was adjusted to distinguish between samples from the U.S. and those from other countries for consistency with the meta-analysis. Finally, data from the various studies were pooled into a single, combined dataset. In this process, any potential sampling weights, such as those found in the Cooperative Campaign Analysis Project (CCAP; Gerber et al., 2011) or the AmericasBarometer (AB; Weinschenk, 2013) were disregarded due to the impossibility of extending them to other samples. Hence, in these instances, the entirety of the available data was used.

Summary of the literature

Table 2 summarizes key statistics for the 13 studies included in the mega-analysis. These studies included 14 datasets (e.g., Gerber et al., 2011; Gerber et al., 2013; Mattila et al., 2011).

The average sample size across individual studies in the mega-analysis was 3,158 participants (SD = 2,192, Range = 800–9,790), nearly double the average sample size observed in the meta-analysis. Gender distribution remained balanced in most studies, with 54.8% women on average, and a maximum of 61.6% in Brandt et al. (2022). Participants were again mostly middle-aged adults, with an average age of 46.2 years, and a minimum age of 23.0 years in one study (Weinschenk et al., 2019). Similar to the meta-analysis, the majority of studies used scales with one or two items per trait (i.e., the FIPI or TIPI; Gosling et al., 2003) or the BFI (eight to ten items per trait; John et al., 1991) to assess the Big Five personality traits. The distributions of voting conceptualization (intention/behavior), operationalization (dichotomous/continuous), and geographic location (U.S./other) were relatively balanced in the mega-analysis. Outside the United States, datasets were sourced from various countries, including Germany (Brandt et al., 2022; Schoen & Steinbrecher, 2013; Sindermann et al., 2021; Sindermann & Montag, 2023; Weinschenk et al., 2019), Great Britain (Furnham & Cheng, 2019), and Spain (Gallego & Oberski, 2012). However, the data in the mega-analysis were somewhat less culturally diverse compared to the meta-analysis, which included a broader range of East Asian countries.

Analysis plan

In order to mirror the three-level structure employed for the meta-analysis as closely as possible, we tested hierarchical linear models (Burke et al., 2017; Curran & Hussong, 2009). Our approach entailed the inclusion of random intercepts to capture the inherent variability at the individual and study levels, acknowledging that individual responses are nested within broader study contexts. Additionally, to address the variability associated with the Big Five personality traits across studies, we incorporated random intercepts for each trait within studies. This modeling decision allowed us to account for the fact that both the baseline levels of voting outcomes and the effects of personality traits on these outcomes could exhibit variability across different study settings. The basic form of the model was as follows; personality scores were centered at zero:

Level 1 ( within participants ) : v o t i n g S c o r e i j k = β 0 j + β 1 j × p e r s o n a l i t y S c o r e i j k + β 2 j k × p e r s o n a l i t y T r a i t k + β 3 j k × p e r s o n a l i t y S c o r e i j k × p e r s o n a l i t y T r a i t k + ε i j k

Level 2 ( within studies ) : β 0 j = γ 00 + u 0 j ; β 1 j = γ 10 ; β 2 j k = γ 20 k + u 2 j k

Level 3 ( across studies ) : u 0 j ∼ N ( 0 , σ u 0 2 ) ; u 2 j k ∼ N ( 0 , σ u 2 k 2 )

Then, we fitted a series of mixed-effects binomial logistic regression models to the data, closely replicating the meta-analytic procedure. The initial model was designed to examine how the Big Five personality traits correlate with voting outcomes. Each of the Big Five traits was represented by a dummy variable (0 or 1), indicating the presence of that trait, which, when interacted with the trait’s continuous score, served as the independent variable in predicting the dependent variable of voting outcomes. We then progressively included additional moderators in the analysis: (1) the number of items used for personality measurement (brief/longer), (2) the conceptualization and operationalization of voting (intention/behavior, and dichotomous versus continuous scaling), and (3) sociodemographic variables (nationality, gender, age). Ultimately, we also converted the model statistics into correlations and reported point estimates, along with 95% confidence intervals (Borenstein et al., 2021).

As with the meta-analyses, our mega-analytic procedure was preregistered and based on a limited set of a priori hypotheses, and thus we did not apply formal multiple-comparison corrections (e.g., Bonferroni adjustments).

Results

Table 3 shows the main effects of the mega-analytic correlations between the Big Five personality traits and voting, without considering the influence of moderators. Figure 4 illustrates the mega-analytic correlations between the Big Five personality traits and voting across various moderator models. 95% confidence intervals for these correlations are accessible on OSF (https://osf.io/57kgd).OPEN IN VIEWER

Importance of integrative data analysis, longer personality measures, and modest effects

Our integrative data analysis suggests that previous studies have lacked the necessary statistical power to consistently capture the modest yet significant associations between personality traits and voting outcomes. In the meta-analysis, the range of prediction intervals often spanned from negative to null to positive effect sizes, reflecting the variability and inconsistencies found within the literature. In the mega-analysis, the addition of multiple moderators rapidly resulted in increased standard errors, an issue that is likely to be even more pronounced in individual studies with limited statistical power. Only the negative correlation between neuroticism and ratio-based measures of voting intentions and turnout proved strong and robust enough to potentially be detected in individual studies with substantial power. In other cases, our analysis cautions against overreliance on single-study outcomes, as they may mislead rather than illuminate the nuanced links between personality traits and civic actions like voting (Hedges & Tipton, 2010; Hunter & Schmidt, 1982).

To find robust effects, future research might consider consistently incorporating comprehensive personality measures in extensive data collection efforts. Our results indicated that the use of longer personality measures did not necessarily lead to more consistent results. However, this observation may stem from a limited dataset, where the prevalent approach in existing literature typically matches large samples with concise personality scales, or smaller samples with in-depth measures (Credé et al., 2012; Sleep et al., 2021; Soto & John, 2019). In our meta-analysis, studies utilizing longer measures were assigned lesser weights due to their smaller sample sizes, compared to those with larger samples but brief measures. Overall, the systematic review did not yield any studies that included highly comprehensive personality measures, except for the study by Mattila et al. (2011), which could not be considered in the mega-analysis. To evaluate the role of different personality trait scales, a systematic incorporation of longer measures into large-scale studies is necessary (e.g., Haehner, Krämer, et al., 2025). Beyond the potential for more consistent results, longer measures can offer superior construct coverage, greater variance, reduced measurement error (via latent variable modeling), and more finely grained descriptions (e.g., via facet-level analysis), as will be discussed further (Anglim & O’Connor, 2019; Sleep et al., 2021; Smith et al., 2000).

Despite their modest size and the challenge of robustly estimating them in single studies, it is important to highlight the significant influence that correlations between personality traits and voting can wield when aggregated across populations (Funder & Ozer, 2019; Ozer & Benet-Martínez, 2006). Previously, we illustrated how even odds ratios of 1.1 could sway tightly contested elections if leveraged to their potential. However, in light of our findings, these numbers were underestimated. If we rerun our calculation with assumed odds ratios between 1.2 and 1.4, depending on whether we look at the meta-analysis or the mega-analysis, we can estimate that a political campaign with 5% success rate would yield an estimated 692,000 additional votes when targeting individuals high in traits like extraversion, conscientiousness, agreeableness, and openness. ⁶ Conversely, targeting a group low in these traits might only yield 527,000 additional votes. ⁷ This means that, when targeting people with personalities more inclined to vote instead of those less inclined to vote, up to 165,000 additional votes may be secured, marking a significant difference of 31.3% in favor of effective personality targeting. This calculation demonstrates the importance of not underestimating or disregarding these seemingly modest effects in political campaigning or similar endeavors that could leverage Big Five profiling.

Differences in domain-coverage and between lower-order traits

We found no moderation effects of sociodemographic factors such as gender and age, and minimal evidence for moderation effects of the sampling region and the operationalization of voting measures as either dichotomous or continuous measure. The only notable sociodemographic effect was that associations were somewhat more pronounced outside the U.S. This may reflect contextual variation in how personality translates into political behavior. For instance, personality–participation links may differ across countries due to variation in political norms, institutional rigidity, and the openness of civic spaces (Weinschenk, 2017). Personality may play a larger role where political engagement is less tightly structured by external rules or expectations.

Other than that, the persistent heterogeneity observed both within and across the studies in the meta-analysis suggests the possibility of other moderators not examined in this research. These other moderators, either alone or in combination, might explain additional differences in the correlation between personality traits and voting. Some of these differences may be partly attributed to the inherent variability in how traits are conceptualized and measured across questionnaires and test families (Miller et al., 2011). Such variability arises because lower-order components, including aspects, facets, or nuances, are represented and emphasized differently in each questionnaire (Crowe et al., 2018; Mõttus et al., 2017; Schwaba et al., 2020). For instance, some measures of neuroticism might capture more content related to heightened anxiety and withdrawal tendencies, while others might capture more content related to anger and volatility. This discrepancy poses challenges, particularly when hypotheses formulated based on one model of the Big Five are tested using other questionnaires with slightly different conceptualizations of these traits. Therefore, alongside questionnaire length, the conceptual distinctions within the Big Five, such as between questionnaires like the TIPI and BFI, warrant further exploration as potential moderators in future research.

Moreover, while facets within broader personality trait dimensions are highly correlated, their predictive validity may differ across outcomes and situations. For example, although a high level of neuroticism typically heightens stress sensitivity, the related lower-level facets of anxiety and anger may produce contrasting effects on voting. Anxiety may discourage voter participation due to the stress associated with political debates (Gerber et al., 2013; Schoen & Steinbrecher, 2013), whereas anger may motivate individuals to participate in voting as a means of protest or alleviation (Brader, 2005; Marcus & MacKuen, 1993). There is a growing body of research supporting the idea that considering both domain-level traits and their sub-components can resolve inconsistencies and improve the accuracy of predictions (Anglim et al., 2019; Mõttus, 2016; Stahlmann et al., 2023). However, the literature in our systematic review did not use such detailed personality assessments. Therefore, employing more comprehensive personality questionnaires, which provide scores for more domains and specific facets, could elucidate effects that appear contradictory at the domain level but are coherent at the facet level.

Finally, we approached this project with the idea that overarching moderators might govern the correlations between Big Five traits and voting outcomes. For example, our initial prediction was that these associations would be strongest for voting attitudes measured on continuous scales among middle-aged adults (35–64 years), with uniform effects across all traits. However, some moderators may affect only specific traits or even counteract one another, leading to null average effects. Thus, more sensitive predictions that focus on specific trait-moderator interactions are warranted. One promising strategy is specification curve analysis—a method that aggregates and disaggregates results across various moderators (Simonsohn et al., 2020). In the context civic engagement, this approach revealed that extraverted younger adults were less likely to register as organ donors, whereas extraverted older adults were more likely (Stahlmann et al., 2023). However, given the reduced sample sizes in specific personality–moderator subgroups when multiple moderators are considered, these findings should be interpreted with caution. Therefore, integrating specification curve analysis with meta-analysis would thus provide a rigorous means to assess the stability of such nuanced effects (Voracek et al., 2019).

Neuroticism’s role in voting behavior and its broader societal effects

The personality trait that stood out the most in our analysis was neuroticism, which not only diminished the intent to vote but also actual participation in voting. This finding complements previous research demonstrating the personal and societal costs of higher neuroticism. For instance, neuroticism is associated with increased healthcare costs (Cuijpers et al., 2010; Lahey, 2009; Widiger & Oltmanns, 2017) and a wide range of adverse health outcomes, including unhealthy habits (e.g., smoking and excessive alcohol consumption), chronic physical conditions, mental health issues, and higher mortality rates (Bleidorn, Stahlmann, & Hopwood, 2025; Charles et al., 2008; Jokela et al., 2014; Kotov et al., 2010; Malouff et al., 2007; Terracciano & Costa, 2004). Altogether, the societal costs of neuroticism are estimated as more than twice as high as the costs of common mental disorders (Cuijpers et al., 2010).

In light of the significant repercussions associated with neuroticism, the strategy of diminishing its impact through specific interventions has attracted growing interest. Altering a personality trait like neuroticism may seem unconventional, yet the desire to improve personal traits is common throughout the general population (Hudson & Fraley, 2015; Hudson & Roberts, 2014; Thielmann & de Vries, 2021). Evidence suggests that psychotherapy can effectively reduce neuroticism levels (Armstrong & Rimes, 2016; Nyklíček & Irrmischer, 2017; Roberts et al., 2017; Wright et al., 2024). However, it is still a matter of some debate whether such changes are sustained over the long term, as opposed to being mere temporary alleviations of symptoms or other difficulties. Furthermore, research indicates that purposeful personality change is possible within the general population, particularly for those who are driven to change, believe in the possibility of change, and engage in new practices that lead to personality adjustments (Allemand & Flückiger, 2022; Haehner et al., 2024; Olaru et al., 2022; Stieger et al., 2018). These insights collectively suggest the potential for volitional personality change beyond natural maturation, although the precise drivers, incentives, and moderating factors of such transformations are yet to be comprehensively understood.

Addressing the 10.68% drop in U.S. voter turnout over the past six decades (Coppedge et al., 2023) may well lie in the realm of personality psychology, particularly through interventions aimed at reducing neuroticism. Our findings show that people scoring one standard deviation above the mean on neuroticism are less likely to vote, with odds ratios between 1.4 and 4.9, depending on whether we look at the meta-analysis or the mega-analysis. With about 80 million of the 240 million eligible voters in the U.S. motivated to increase their emotional stability, the implications of such personal growth are significant (see Thielmann & de Vries, 2021). If we assume that 1% of these motivated individuals, or 800,000 people, successfully decrease their neuroticism by one standard deviation, the adjusted voting probabilities for this group could settle in between 68.65% ⁸ and 88.46%, ⁹ compared to the current average of 61%. These heightened probabilities would lead to an additional 61,200 ¹⁰ to 219,6800 ¹¹ votes. Such findings not only highlight the significance of neuroticism on electoral participation but also suggest that interventions aimed at reducing it could play a pivotal role in mobilizing voters, especially in closely contested elections.

Future research should test whether personality interventions lead to lasting changes in political behavior. Longitudinal and cross-national studies are needed to assess how robust and generalizable such effects are. Large panel studies, such as the Longitudinal Internet Studies for the Social Sciences (LISS; Scherpenzeel & Das, 2011) and the Understanding America Study (UAS; Alattar et al., 2018), provide infrastructure for tracking personality and civic engagement over time. In contrast to these observational designs, the CHILL study (Haehner, Wright, et al., 2025) targets neuroticism through structured training and includes multi-informant assessments and repeated follow-ups across German-speaking countries. While it does not measure political behavior, CHILL and similar studies can offer a model for studying sustained personality change and may serve as a blueprint for future research linking such change to political participation.

Implications for the role of personality psychology in politics

While the potential to enhance voter participation by addressing neuroticism is promising, our findings also highlight risks, particularly the possibility of misuse by unscrupulous actors aiming to manipulate political outcomes. The Cambridge Analytica controversy, for example, has brought to light the potentially impactful role of personality psychology in the political arena. Although the precise effect of Cambridge Analytica’s strategies on the 2016 U.S. presidential election remains debated (Gibney, 2018; Grassegger & Krogerus, 2017), there is an accumulating body of evidence showing that strategies employing personality psychology can indeed foster more favorable attitudes toward specific political parties or candidates (Kosinski et al., 2015; Matz et al., 2017).

The underlying process of these strategies involves the initial assessment of personality traits, followed by the alignment of messages that resonate with the identified personality profiles (Matz et al., 2023). These messages are often combined with imagery or details about a party’s program or a candidate, reminiscent of classical conditioning techniques. For instance, individuals characterized by high levels of neuroticism might encounter political ads that highlight threats or fears, intertwined with topics such as warfare, criminal activity, or immigration issues, which a party aims to address (Caprara & Zimbardo, 2004; Krotzek, 2019). This strategy of personality assessment followed by targeted political messaging has proven effective within social media platforms, where personality data can be gleaned from biographical texts and political ads are integrated into users’ timelines (Zarouali et al., 2022).

Understanding how the results of our analyses tie into this line of research could help in developing strategies to guard against undue influence. For example, our findings suggest that focusing on individuals with high neuroticism for political targeting may be ineffective, as their likelihood to vote is lower despite potential support for a political party. Instead, directing efforts toward individuals with lower neuroticism and customizing messages to their other traits, like extraversion or agreeableness, may become a preferred strategy for organizations seeking to influence voter allegiances. Consequently, it might be prudent to adjust preventive efforts toward individuals who are emotionally stable, as they are more likely to vote and thus may be more exposed to persuasive political content.

This illustrative example underscores the ethical dilemma of our research findings: Psychological targeting, particularly without individuals’ awareness, may threaten autonomy and informed choice (Matz et al., 2017). Yet such strategies are already used by political actors and tech firms, often without ethical oversight (Matz & Netzer, 2017). For this reason, some have argued that psychology has an ethical obligation to engage with these developments in order to help guide responsible use (Matz et al., 2023). Rather than dismissing this line of research as inherently unethical, it may be more prudent to stay one step ahead—so that potential misuse can be identified and prevented early on.

Strengths and limitations

This study leverages both meta-analysis and mega-analysis to offer a detailed and robust examination of the link between the Big Five personality traits and voting outcomes. By systematically reviewing data from both psychological and broader academic sources via Scopus and the grey literature database BASE, we compiled information from 65,036 individuals across 17 studies for the meta-analysis, and from 44,206 individuals across 13 studies for the mega-analysis. Our integrative analysis encompassed zero-order correlations and explored how nationality, gender, age, and the specific measures of personality and voting outcomes might moderate these correlations. Despite these strengths, the study faces certain limitations.

Conclusion

Our combined meta-analytic and mega-analytic approach has clarified the links between the Big Five personality traits and voting intentions and behavior, hitherto obscured by a fragmented and inconsistent literature. Our findings suggest that the Big Five are modest but significant predictors of voting intentions, with neuroticism additionally predicting voting behavior. To obtain reliable results, large sample sizes and extensive personality assessments are important, enabling a deeper understanding of the connection between personality and civic participation like voting. These insights may prove valuable for developing strategies to increase voter participation and protect against manipulative influences on the democratic process. Questions remain about the processes by which the observed links between personality traits and intentions may manifest into actual voting behaviors. A crucial next step for personality theory and research would thus be to document how personality traits may shape voting intentions and ultimately cause actual voter behavior.