Voting like a boss: the effect of implicit leadership prototypes within the political context Open Access

Voters’ perceptions of political leaders can significantly impact election outcomes (Foti et al., 1982). Unlike managerial contexts, voters often interact with political candidates only superficially, relying on ILTs for categorization. Research suggests that when followers have limited individuating information about a leader, they depend more on category-based prototypes (Lord & Maher, 1991; Pitsi et al., 2025; Sy & van Knippenberg, 2021). Political leaders, who operate at the highest levels of visibility yet with minimal personal interaction, exemplify this dynamic. This study seeks to identify which individual differences matter in political ILT perceptions and how they relate to voting behavior.

While leadership research has traditionally focused on leaders, followers play a crucial role in shaping perceptions of leader behavior and effectiveness (Wang, VanIddekinge, Zhang, & Bishoff, 2019). ILTs develop over time based on experiences with leaders in different contexts, leading to variations across individuals (Epitropaki & Martin, 2004; Offermann et al., 1994). Studies have found differences in ILTs based on several individual difference factors, including gender (Epitropaki & Martin, 2004; Nye & Forsyth, 1991; Deal & Stevenson, 1998), professional status (Singer, 1990; Offermann et al., 1994), culture (e.g. Gerstner & Day, 1994), managerial level and industry type (Epitropaki & Martin, 2004).

Gender in particular has been extensively studied in ILT research. Women tend to associate ideal leadership with traits such as understanding, sincerity and participativeness (Epitropaki & Martin, 2004; Paris, Howell, Dorfman, & Hanges, 2009), while men are more likely to emphasize competitiveness, aggressiveness and intelligence (Deal & Stevenson, 1998). Recent longitudinal work confirms that gender continues to shape ILTs, though patterns may be evolving. For example, Offermann and Coats (2018) found that while sensitivity remains the strongest ILT factor for women and dedication for men, the overall ILT structure is stable and “think leader, think male” persists despite increased gender diversity in leadership roles. This underscores the enduring salience of gender in leadership schemas even amid societal change.

Although demographic variables such as age or tenure have not shown consistent effects on ILTs, other factors have. For example, managers are more likely than non-manager employees to value dynamism (Epitropaki & Martin, 2004). Similarly, manufacturing employees rate tyranny as more important and sensitivity as less important compared to service employees (Epitropaki & Martin, 2004). Students are more likely to believe that effective leadership is determined by external uncontrollable factors (Singer, 1990), compared to professionals. Given the variation among these findings, we sought to examine how rater attributes might impact perceptions of ILT in the context of politics.

There is limited scholarship on ILT in the context of political leadership. Political analysis often ties leader ratings to party identification. The 2016, 2020 and 2024 US presidential elections highlighted increased polarization between parties (MacWilliams, 2016). For a primer on the US political context, please see  Appendix. Studies have shown that political polarization has been rising since the 1970s, affecting both public and private sectors (Hare & Poole, 2014; Bonica, 2013; Shor & McCarty, 2011; Swigart, Anantharaman, Williamson, & Grandey, 2020).

Political ideology plays a significant role in shaping voters’ perceptions of leaders. Ideology helps voters categorize political leaders and make informed choices (Jacoby, 2010; Converse, 2006). Research has shown that ideology influences vote choice and party identification, with Democrats aligning with liberal ideals and Republicans with conservative values (Jost, 2006, 2009; Stimson, 2015; Mason, 2015). Seminal work by Downs (1957) argued that simply knowing where a candidate fell on the liberal-conservative spectrum provided enough information for voters to choose the candidate that best fit where they were situated on the ideological spectrum Although there has been debate in the field of political science over how deeply voters understand ideology, there is a general consensus that ideological categorization facilitates voter choice (Jacoby, 2010). While the debate on the utility of political ideology among the electorate continues in political science, both ideology and party identification are included in countless voter studies to understand how both may shape electoral outcomes.

The increasing political polarization combined with the influence of political ideology on party identification has contributed to closely contested presidential and congressional elections throughout the 21st century. This dynamic was especially evident in the 2016 presidential election and the emergence of Donald Trump as a presidential candidate, which coincided with a measurable shift in American public preference toward more dominant and masculine leadership traits (Federico & Malka, 2022; MacWilliams, 2016).

Political science research indicates that rising polarization is associated with divergent preferences for leadership traits across ideological and party lines, and these preferences map onto ILT’s prototypical and anti-prototypical dimensions. Because ideology and party identification consistently predict voting behavior, we expect ILT to serve as a cognitive mechanism linking these variables to electoral choices. Specifically, differences in ILT perceptions, such as liberals and Democrats favoring ILT prototypical traits and conservatives and Republicans favoring anti-prototypical traits, should partially explain voting patterns:

Research on implicit leadership theories supports the mediating role of prototypes in linking individual differences to leader evaluations. For example, Sy and van Knippenberg (2021) demonstrated that implicit theories of leadership emotions (ITLEs), which is the emotional counterpart to the more cognitive ILT, mediated the relationship between leader gender and leadership perceptions. This paralleled earlier findings that ILTs mediate the link between leader race and leadership evaluations (Sy, 2010). These findings reinforce our proposition that ILTs can serve as a cognitive mechanism through which political ideology influences voting behavior:

Taken together, these hypotheses reflect the purpose of this study, which is to examine how ILT explains voters’ perceptions of leadership characteristics and serves as a cognitive link between political ideology and voting behavior.

The primary purpose of this study is three-fold. First, it seeks to extend ILT research into the political context by examining whether the factor structure of ILT, developed for business leaders, holds in politics. Second, it explores the generalizability of ILTs across political ideology and party affiliation. Third, the study tests a model (Figure 1) where ILT mediates the relationship between political ideology and voting behavior.

Beyond its relevance to political science, extending ILT into the political domain offers significant implications for organizational behavior (OB) research. Early ILT theory emphasized that leaders are embedded in context (Foti et al., 1982; Lord et al., 1984; Offermann et al., 1994) and political leaders represent a highly visible and impactful leadership context. Examining the robustness of the ILT construct beyond organizational settings will offer insights into whether leadership perceptions are contextually contingent and broaden the understanding of how implicit prototypes operate across domains. In addition, political leaders often influence organizational environments through policy and regulation, making it critical for management researchers to understand how societal-level leadership perceptions shape expectations of organizational leaders.

Data were collected from 577 participants via Amazon Mechanical Turk during a two-week period before and up to the 2020 US presidential election. Participation was limited to those 18 years of age or older located in the USA. After data screening, 18 cases were removed for missing data, failed attention checks or response bias, resulting in 559 cases used in the analyses. Of the participants, 63.1% identified as male, 74.2% as White, 8.9% as Black/African American, 6.6% as Hispanic/Latino and 6.8% as Asian/Pacific Islander. The mean age was 38.13 years; 69.2% reported working full-time and 2.90% had previously served in the military.

Amazon’s Mechanical Turk (MTurk) is the most commonly used online data collection method (Porter, Outlaw, Gale, & Cho, 2019), and has been used to conduct research in political spheres for over a decade (Clifford, Jewell, & Waggoner, 2015; Huff & Tingley, 2015). It boasts a large and diverse participant pool, convenient access to participants and data, rapid data collection and relatively low data-gathering costs (Aguinis, Villamor, & Ramani, 2021). In addition, research has demonstrated that data provided by MTurk participants in political contexts is generalizable to the broader population (Clifford et al., 2015; Coppock, 2019). However, using MTurk for data collection does present some challenges and concerns, such as inattention and high attrition rates (Aguinis et al., 2021) participants failing response validity indicators and inability to replicate well-established findings (Chmielewski & Kucker, 2020). Although the present data was collected prior to the publication of Aguinis et al. (2021), its methodology aligns with the best practice recommendations subsequently outlined in their work. For example, we evaluated the appropriateness of MTurk to test our theory, screened MTurk respondents based on preset qualifications (e.g. age, location, etc.), established a required sample size, formulated compensation rules, designed data collection tools to gather responses (e.g. used an informed consent form, included CAPTCHA verification, used attention checks and qualitative open-ended questions), launched the study via pilot tests, screened the data for incomplete data and response times and approved or denied compensation for completed responses. Consequently, we believe the data collection methodology is appropriate for the research questions.

Implicit leadership was measured using 21 items developed by Epitropaki and Martin (2004). Participants were asked to think about their image of a typical leader and rate the degree to which each trait is characteristic of a typical leader on a five-point Likert scale (1 = Not at all characteristic, 5 = Extremely characteristic). The scale included prototypical facets of sensitivity (three items; Cronbach’s alpha = 0.721), intelligence (four items; Cronbach’s alpha = 0.774), dedication (three items; Cronbach’s alpha = 0.768) and dynamism (three items; Cronbach’s alpha = 0.625) and anti-prototypical facets of tyranny (six items; Cronbach’s alpha = 0.920) and masculinity (two items; Cronbach’s alpha = 0.760).

Political affiliation and ideology were measured using items from the general social survey (GSS; Davern, Bautista, Freese, Herd, & Morgan, 2023). One item asked participants to report their political affiliation on a seven-point Likert scale (1 = strong Democrat, 7 = strong Republican), with options for “other party” and “don’t know.”

Political ideology was measured by one item on a seven-point Likert scale (1 = Extremely liberal, 7 = Extremely conservative), with an additional option of “I don’t know.”

Participants reported their intended vote for the 2020 election and their vote in the 2016 election. Options included the major party candidates, “Other,” “I did not vote” and “I don’t know.”

Participants provided demographic information, including gender, race, national origin, age, employment status and military service.

Before testing the model, we addressed ILT measurement and generalizability. To examine content and factor structure of ILT in a political context, principal component analysis with Varimax rotation tested the initial survey items’ loading on six dimensions of ILT: sensitivity, intelligence, dedication, dynamism, tyranny and masculinity. The criterion used in the analysis was a factor loading greater than 0.5, and Eigen values greater than 1.0 (Tabachnick & Fidell, 2019). Most items loaded on their respective theorized constructs, but one item in Intelligence and one item in Dynamism were dropped since the factor loadings were less than 0.5 or loaded in multiple constructs. The results of the factor analysis are shown in Table 1.

Internal consistency was evaluated using composite reliability (CR) and Cronbach’s alpha, with all the reliability measures except for dynamism above the cutoff level for both measures (CR = 0.7; Cronbach’s alpha = 0.7; Bagozzi & Yi, 1988; see Table 2). Cronbach’s alpha for dynamism did not meet the cutoff, likely due to the removal of one item, as Cronbach’s alpha tends to underestimate internal consistency for scales with few items. In this case, composite reliability can be used as an alternative (Peterson & Kim, 2013). Composite reliability for dynamism was 0.75, which is above the cutoff level, so we kept the construct in our analysis. Convergent validity is demonstrated when average variance extracted (AVE) is 0.5 or more (Fornell & Larcker, 1981). All of the items meet this requirement.

Discriminant validity was checked by examining whether the correlations between the variables are lower than the square root of their AVEs. As seen in Table 3, all the square roots of AVEs (the main diagonal) are greater than the pairwise correlations between the constructs (the off diagonal), indicating discriminant validity.

We performed factor structure analyses, comparing the goodness of fit of the six-factor ILT model with competing models. The competing models include a null model, one-factor model and two-correlated factor model (one leadership prototype and one leadership anti-prototype). On the basis of prior literature (Epitropaki & Martin, 2004; Lord et al., 1984), we also tested a second-order model with two higher-order constructs to see whether the four dimensions of sensitivity, intelligence, dedication and dynamism represent the higher latent construct of a leadership prototype and whether the two dimensions of tyranny and masculinity measure the higher latent construct of a leadership anti-prototype. For the model comparison, chi square (χ²), comparative fit index (CFI), normed fit index (NFI) and root mean square error of approximation (RMSEA) fit indices were calculated (Mulaik et al., 1989).

As shown in Table 4, the null model, one-factor and two-correlated factors did not meet the minimum criteria (CFI ≥ 0.90, NFI ≥ 0.95, RMSEA ≤ 0.08). The first-order six-correlated factor model provides the best fit to the data, χ² (137) = 253.70, p ≤ 0.001 (χ²/df = 1.85, CFI = 0.98, NNFI = 0.97, RMSEA = 0.04), with the lowest chi-square value among all the compared models. While the second-order factor model had a reasonable fit to the data, χ² (155) = 311.99, p ≤ 0.001 (χ²/df = 2.15, CFI = 0.97, NNFI = 0.95, RMSEA = 0.05), it was significantly different from the first-order model, Δ χ² (8) = 58.30, p ≤ 0.001, indicating that the ILT construct is best represented by the first-order six-factor model, supporting HH1. In the context of politics and political decision making, the six-factor ILT model, which was developed in and widely used in work contexts, was the best fit for the data in the context of political leadership.

To examine the generalizability of ILT across individual difference variables, t-tests and ANOVAs for mean differences and multiple group analyses for factorial invariance were conducted (Meredith, 1993). We primarily considered individual differences in political party affiliation (Democrat = 296, Republican = 201) and political ideology (liberal = 300, conservative = 257), though five additional groups were included in the analyses: gender (male = 353, female = 203), age (younger = 285, older = 272), income ($0–$49,999 = 240, $50,000–$99,999 = 256, above $100,000 = 63), race (white = 415, non-white = 142) and education (high school = 44, college = 403, graduate school = 112). Groups for age were formed by median splits that yielded groups of similar size (age median = 35 years). These factors were included due to their relationship to outcome variables of interest, such as voting behavior.

T-tests and ANOVA. Independent sample t-tests or ANOVAs were performed to investigate mean differences between groups on the six dimensions of sensitivity, intelligence, dedication, dynamism, tyranny and masculinity. No statistically significant differences were found for the groups of age, income and race on the six dimensions of ILT. Significant differences were found between male and female respondents on the dimensions of tyranny and masculinity. consistent with previous studies (Deal & Stevenson, 1998), males rated tyranny and masculinity (M = 2.78 and 3.44, respectively) higher than females (M = 2.57 and 2.93, respectively).

One-way ANOVA revealed that there was a statistically significant difference in dedication (F(2, 556) = 3.985, p = 0.019) and tyranny (F(2, 556) = 8.937, p = 0.000) among education groups. Tukey’s honestly significant difference (HSD) test for multiple comparisons found that dedication was significantly different between high school and college (p = 0.014, 95%) and between high school and graduate school (p = 0.038, 95%). Tyranny score was also significantly different between high school and college (p = 0.009, 95%), between high school and graduate school (p = 0.000, 95%) and between college and graduate school (p = 0.031, 95%). This suggests that individuals with high school degrees perceive their ideal leader as more motivated, dedicated and hardworking while people with graduate degrees view their ideal leader as more domineering, pushy and manipulative. However, this finding is affected by small and different sample sizes across groups, so caution in its interpretation is warranted.

In support of H2, significant differences were found between political party and ideology groups on the dimensions of tyranny and masculinity. As hypothesized, Republicans rated tyranny and masculinity (M = 2.99 and 3.55, respectively) higher than Democrats (M = 2.52 and 3.06). Similarly, conservatives rated tyranny and masculinity (M = 2.90 and 3.47, respectively) higher than liberals (M = 2.52 and 3.06).

Factorial invariance. Factorial invariance refers to whether a construct is measured in the same way across different groups. Our confirmatory factor analysis suggests a first-order six-factor ILT model, but the factor structure may not be equivalent across groups or settings. To verify the generalizability, additional multiple-group analyses are required (Brown, Harris, O'Quin, & Lane, 2017). Previous literature indicates that a large sample size (more than 200) is necessary for the accuracy of estimates (Liu et al., 2017). We ran factor invariance testing for the primary variables of interest (political party and ideology) as well as gender and age; due to insufficient sample size in some groups, we excluded income, race and education.

Prior to any invariance constraint, we tested a six-factor model for each of the eight groups separately. This step determines whether the six-correlated ILT factor model should be adopted for each group. If the baseline model for each group is the same, more restrictive constraints can then be applied on the model (Karim, Weisz, & Rehman, 2011; Byrne, 2016). The subgroup analysis results (Table 5) show that the six-correlated factor model provides a very good fit in all groups (χ²/df between 1.00 and 3.00, CFI and NNFI ≥ 0.90, RMSEA ≤ 0.06; Epitropaki & Martin, 2004). Therefore, the six-factor model was used as a baseline model for subsequent multi-group analysis.

Next, we carried out a comparison of three nested models as part of multi-group analysis (see Table 6). Model 1 (baseline) hypothesized the six-factor structure with the unconstrained model allowing all factor loadings and variances to be estimated freely. In Model 2, factor loadings were constrained to be equal across the groups. In Model 3, factor covariances were also constrained to be equal (Byrne, 1994). A chi-square difference indicates a statistically significant reduction in fit as a result of adding equality constraints.

Total invariance (number of factors, factor loading pattern, factor covariance) is supported for the gender and age suggesting that ILT are generalizable for both males and females and for younger and older individuals. However, partial invariance is supported for the groups of party and ideology. The comparison of Model 1 (baseline) and Model 2 (invariant factor loadings) were statistically insignificant. That is, the measurement model is invariant across the groups (Democrat vs Republican and liberal vs conservative). However, we observed the significant chi-square differences in factor covariances between Democrat and Republican, Δχ² (15) = 34.089, p ≤ 0.01 as well as between liberal and conservative, Δχ² (15) = 41.267, p ≤ 0.01. Therefore, despite the differences in factor covariances, we can conclude that the overall six-factor structure of ILT holds across party and ideology groups.

The mediating role of ILTs on the relationship between ideology and voting behavior (Figure 1) was tested using SPSS 25.0 and PROCESS macro by Hayes (2022). Both indirect and direct effects of ideology via ILT on voting behavior were tested using Model 4 of the PROCESS macro using 5000 bootstrap samples. Ideology was an independent variable, ILT was the mediating variable and voting behavior was the dependent variable. Gender and age were controlled as covariate variables. The results (Table 7) revealed a significant indirect effect of ideology on voting behavior through ILT (b = 0.026, t = 2.250) and a significant direct effect of ideology on voting behavior in presence of the mediator (b = 0.423, p ≤ 0.001). Hence, in support of H3, we conclude that ILT partially mediated the relationship between ideology and voting behavior. Age was a significant covariate affecting voting behavior but had an insignificant effect on ILT.

This study revisited the notion that context matters for leadership prototypes by examining whether the factor structure for ILT, created with reference to business leaders, holds in the political context. We also explored the generalizability of ILTs across political ideology and party affiliation and tested a model where ILT mediates the relationship between political ideology and voting behavior. The results are consistent with prior research on ILT measurement within a business context, showing that the first-order six-factor model accurately represents the ILT construct in the context of political leadership. The overall six-factor structure of ILT remains consistent across various groups, suggesting its robustness and applicability across contexts. Statistically significant differences in the dimensions of tyranny and masculinity when considering party affiliation and ideology were also discovered. In addition, ILT partially mediated the relationship between political ideology and voting behavior. Our findings suggest that while core ILT dimensions remain robust across contexts, individual differences can amplify certain traits; this parallels the adaptability observed in organizational settings by Offermann and Coats (2018).

The current study extends ILT research into the political context, highlighting its importance for understanding voter behavior. Although early research discussed leader prototypes and ILT in various contexts (e.g. Foti et al., 1982; Lord et al., 1984), subsequent ILT research has predominantly focused on management and organizational contexts (e.g. Epitropaki & Martin, 2005; Lord, Epitropaki, Foti, & Hansbrough, 2020; Riggs & Porter, 2017). Extending ILT research into the political space is crucial, as voters’ perceptions of political leaders have practical implications for election outcomes. The results indicate that ILT explains why people with different political ideologies endorse and vote for different candidates; they have different prototypes of ideal leaders and thus endorse different leadership characteristics. As studies have argued that some implicit measures of attitudes help predict voting behavior (Friese, Smith, Koever, & Bluemke, 2016; Friese, Bluemke, & Wänke, 2007), the current study suggests that ILT should be considered when predicting voting behavior. Accounting for voter ILT may provide incremental validity to polling estimates.

For organizational behavior and management scholars, this study underscores the importance of context in leadership perceptions. Demonstrating that ILT factor structures, which were developed in an organizational context, hold in politics suggests that implicit leadership prototypes may be stable across domains, supporting cross-contextual generalizability. This supports the early foundations of leadership cognition literature (e.g. Foti et al., 1982; Lord et al., 1984) and highlights that follower cognition operates similarly in organizational and societal spheres.

Practically, managers and human resources professionals can draw on these findings to anticipate how broader societal prototypes may spill over into organizational settings, influencing employee expectations of leaders. For example, periods of heightened political polarization may amplify preferences for certain traits (e.g. dominance or sensitivity), affecting leader selection, development and communication strategies. Understanding these dynamics, and that politically oriented individual differences affect employee leadership perceptions, may help organizations manage leader-follower fit more effectively and to design leadership development programs that account for these different prototypes.

The study used self-report data about voting behavior, which may not accurately reflect actual voting behavior due to social desirability or influence bias (Abelson, Loftus, & Greenwald, 1992; Anderson & Silver, 1986; Stout & Kline, 2011). Future studies should examine the relation between ideology, ILT and voting behavior using actual voting data. In addition, the convenience sample from Amazon’s MTurk may not reflect the national voting population, so results should be generalized cautiously. Future studies should replicate the findings with a larger, more representative sample. Finally, the data were collected during a small window around the 2020 presidential election, which may limit the generalizability of the findings to other elections or contexts.

The current study contributes to the literature on ILT by examining its factor structure and outcomes in a political context, as opposed to the business or organizational context in which ILT theory has developed. The findings have important implications for understanding voting behavior and other political outcomes.

Future research should explore the application of ILT in other areas of political science and political involvement, such as whether individual differences such as gender or ideology, working through ILT, affect self-selection into politics. The field of political science has shown that women tend to identify as democrats and be more liberal than men (Chaney, Alvarez, & Nagler, 1998; Kaufman, 2006; Gillion, Ladd, & Meredith, 2020). However, many women, particularly white suburban women, voted for Donald Trump. The ILT literature has begun to study the effects of social roles on men versus women and how they evaluate their leaders (Eagly & Karau, 2002; Junker & VanDick, 2014; Epitropaki & Martin, 2004).

Beyond cognitive prototypes, emerging research highlights the role of emotional schemas in leadership perception. Sy and van Knippenberg (2021) introduced ITLEs and found that not only do people hold schemas for the types of emotions that leaders display or should display, but also that ITLEs also predict leadership evaluations. Given that political leaders are often evaluated based on emotional displays (e.g. calm in crises, enthusiasm in campaigns), future research should explore the role of emotional schemas and how they may or may not interact with ILTs in shaping voter perceptions.

In addition, researchers should examine whether follower ILT drives changes in leader behavior. Public perception affects the effectiveness of political careers (Foti et al., 1982), so understanding whether follower ILT predicts the likelihood of political candidates trying to “play up” desired attributes is warranted. Finally, ILT should be examined in contexts other than business and politics, such as friend groups and the military.

One of the co-authors has chosen to publish anonymously in line with Emerald’s policy on author anonymization. The identity of the author is known to the co-authors, the journal and the publisher.

The USA operates under a two-party system: the Democratic Party and the Republican Party. On a left-to-right ideological spectrum, Democrats are generally associated with liberal positions, while Republicans align with conservative positions. Although smaller third parties exist, structural factors, such as the electoral system, make it very difficult for these parties to win national offices, including seats in Congress or the presidency.

Democratic Party positions typically include support for abortion rights (pro-choice), stricter gun regulations, stronger environmental protections, expanded social welfare programs (e.g. food and housing assistance), protections for LGBTQ rights and a greater role for the federal government in addressing social issues.

Republican Party positions generally include opposition to abortion (pro-life), increased military and defense spending, reduced government welfare programs (favoring individual responsibility), looser gun regulations and fewer environmental restrictions to promote business interests. Republicans often advocate for stronger state authority relative to federal power.

Recent presidential elections illustrate these dynamics:

• 2016: Republican Donald Trump defeated Democratic nominee Hillary Clinton.

• 2020: Trump lost to Democratic nominee Joseph Biden, former Vice President under Barack Obama.

• 2024: Trump returned to win against Democratic nominee and sitting Vice President Kamala Harris.