pape

https://doi.org/10.1177/0956797620977518

Psychological Science 2021, Vol. 32(6) 934 –943 © The Author(s) 2021 Article reuse guidelines: sagepub.com/journals-permissions DOI: 10.1177/0956797620977518 www.psychologicalscience.org/PS

ASSOCIATION FOR PSYCHOLOGICAL SCIENCEShort Report

Research has demonstrated that voters are influenced by split-second judgments on the basis of candidates’ faces (Todorov et  al., 2005). However, even though gender and age are automatically inferred from faces (Brewer & Lui, 1989; North & Fiske, 2015), to our knowledge no research to date has systematically exam- ined the role of candidates’ perceived age and gender on voting intention.

A recent study of 28 national legislatures in the Euro- pean Union (Stockemer & Sundström, 2018) found that although female legislators were generally underrepre- sented, they were most underrepresented among older legislators. Could it be because people are more inclined to vote for female candidates only when the candidates are younger? From Susan Sontag’s classic essay “The Double Standard of Aging” (Sontag, 1972) to reports of gender-based age discrimination ( Dockterman, 2016), anecdotal evidence suggests that age has a greater influ- ence on how women are perceived than on how men are perceived. In the studies reported below, we sought to isolate the roles of candidates’ age and gender (sur- mised only from their photos) and examine how those

two factors interact to predict voting intention in hypo- thetical elections.

The present research also examined how basic char- acteristics of candidates inferred from their photos, such as perceived competence (Olivola & Todorov, 2010; Todorov et al., 2005), warmth (Sutherland et al., 2013), and attractiveness (Berggren et al., 2010; Chiao et al., 2008; McLellan & McKelvie, 1993), account for the rela- tions between candidates’ age and people’s intentions to vote for them.

Study 1

Study 1 was preregistered (https://aspredicted.org/ ng5vg.pdf ) and examined how voting intentions were related to candidates’ apparent age and gender using a hypothetical election paradigm.

977518PSSXXX10.1177/0956797620977518Shen, ShodaCandidates’ Age and Gender Predict Voter Preference research-article2021

Corresponding Author: Yiqin Alicia Shen, University of Washington, Department of Psychology E-mail: [email protected]

How Candidates’ Age and Gender Predict Voter Preference in a Hypothetical Election

Yiqin Alicia Shen and Yuichi Shoda Department of Psychology, University of Washington

Abstract Are preferences for political candidates influenced by how old they appear to be? Amazon Mechanical Turk workers and undergraduate students were shown photos of 93 state legislators as candidates in hypothetical elections. Other information about the candidates (e.g., party affiliation) was held constant, randomized, or not presented. For very young candidates (< 35 years old), participants favored women over men. However, participants’ intention to vote for male candidates increased with age until candidates were about 45 years old and then slightly decreased. In contrast, participants’ intention to vote for female candidates consistently decreased with candidates’ age. Perceived attractiveness and warmth accounted for some of the gender differences in the effect of candidates’ perceived age.

Keywords age, gender, voting, preregistered

Received 9/1/17; Revision accepted 9/29/20

Candidates’ Age and Gender Predict Voter Preference 935

Method

Stimulus selection and preparation. Color images of 365 state legislators from the United States were obtained in 2014 from 14 state government websites (California, Nevada, Idaho, Wyoming, New Mexico, Texas, North Dakota, Iowa, Ohio, Michigan, Florida, Virginia, Pennsylvania, and New York). Those states were selected because in the 2012 U.S. presidential elec- tion, roughly half of them voted for Barack Obama and the other half voted for Mitt Romney. Stimuli were sam- pled to cover a wide age range for both genders and were selected without regard to their perceived attrac- tiveness, competence, or electability (see the Supple- mental Material available online for more information). All images were cropped so that only the head, neck, and shoulders of the person were visible. Backgrounds, often of patriotic themes (e.g., national flag, state capi- tol), were digitally removed and replaced with a uniform color background. Example stimuli are shown in Figure S18 in the Supplemental Material.

Fifteen Amazon Mechanical Turk (MTurk) workers rated the perceived age of these 365 legislators. Those legislators were grouped into five age groups (30–39 years, 40–49 years, 50–59 years, 60–69 years, 70 years and above) on the basis of the average of workers’ responses. Images of approximately 10 female and 10 male legislators were randomly selected from the first four age groups. Because of the small number of legisla- tors in the highest age group (70 years and above), all images in that group were selected. On their web pages, approximately half of the selected legislators self- identified as Republican, and the other half self- identified as Democrat. The resulting stimulus set contained a total of 93 legislators (51 males, 42 females).

Participants. Participants were 300 MTurk workers. The sample size was based on the preregistered plan, which in turn was based on samples sizes used in previ- ous studies using hypothetical election paradigms (see Studies S1–S6 in the Supplemental Material). In accor- dance with the preregistration plan, we excluded four participants because they took an excessively long time (> 60 min) to complete the task, and seven participants were excluded because of an extremely low variation in their responses throughout the study (i.e., SD < 3 on a 100-point slider). This resulted in a final sample size of 289 participants (158 women, 131 men; age: M = 49.1 years, SD = 14.5 years; 237 self-reported as White, 14 as Hispanic or Latinx, 17 as Black or African American, 10 as East Asian, four as “other,” two as American Indian/ Alaska Native, two as South Asian, and three as more than one race).

Prior to the experiment, MTurk collected “premium- qualification” information from the potential partici- pants, including participants’ self-reported gender and age. Using that data, we were able to recruit a sample with a gender and age distribution that approximated that of voters in the 2016 presidential election (for the sampling scheme and sample size in each age and gender bin, see Section III.10 in the Supplemental Mate- rial). Data were collected in early 2019. All participants had Internet protocol (IP) addresses within the United States. No analyses were conducted before data collec- tion was completed.

Procedure. Study 1 employed a highly repeated within- person design (Whitsett & Shoda, 2014; for a more recent implementation of this paradigm, see Zayas et al., 2019), in which each participant responds to a large number of stimuli, and the effect of the stimulus characteristics of interest (e.g., the age and gender of candidates) on behavior (e.g., voting intention) is assessed separately for each participant.

Participants were invited to an experiment on “poli- tics and impression formation.” Using the Inquisit Web platform (Millisecond, 2015), we presented the 93 pho- tos, described above, one at a time in random order, and participants were prompted to estimate each

Statement of Relevance

People make split-second judgments about others only on the basis of their faces. Does this ten- dency extend even to highly consequential deci- sions, such as whom to vote for in an election? To understand how such judgments are influenced by apparent gender and age, we showed partici- pants photographs of a number of state legislators and asked them to indicate the likelihood of vot- ing for each candidate, while controlling other information. For candidates judged to be under 35 years old, participants favored women over men. However, as the perceived age of the can- didate increased, the likelihood of voting for female candidates steadily decreased. In contrast, the likelihood of voting for male candidates first increased until the candidates were about 45 years old and then slightly decreased. Although we can- not directly extrapolate beyond the present studies, these findings suggest that candidates’ gender and age, surmised only from photos, can influence vot- ing intentions, and they underscore the importance of raising voter’s awareness of such influences.

936 Shen, Shoda

candidate’s age in years by entering a two-digit number in a text box below each image. After estimating the age of all candidates, participants were given the fol- lowing instructions and then shown each of the 93 candidates again in randomized order:

In the following task, you will view pictures of the same 93 candidates. Imagine they are all candidates for the next United States of America Senate or House of Representatives. Please indicate how likely you are to vote for each person.

After rating all the candidates, participants reported their age, gender, and race/ethnicity.

Results

To observe the relationship between perceived age and voting intention without imposing assumptions about the shape (e.g., linear, quadratic) of the relationship, we first used locally weighted scatterplot smoothing (LOWESS) with 95% bootstrapped confidence intervals (CIs) to plot participants’ voting intentions for each candidate as a function of the perceived age of that candidate. This process was done in R (Version 3.4.0; R Core Team, 2017). Means were plotted separately for younger and older participants (Figs. 1a and 1b) and for male and female participants (Figs. 1c and 1d). Figure 1 also shows the 95% CI at each age point (shaded areas), which were based on bootstrapping with 5,000 resamples. As shown in the figure, younger female candidates were preferred to younger male can- didates, but voting intentions toward female candidates steadily decreased with candidates’ age. In contrast, for male candidates, the relationship between voting intention and candidates’ age appeared to follow an inverted-U pattern: Voting intentions increased with age up to approximately the perceived age of 45 years and then decreased. This pattern of relationship remained qualita- tively the same regardless of participants’ age or gender.

Using the HLM package (Version 7; Bryk et al., 2010), we applied a two-level random-slope-and-intercept model to formally model the relationship between the perceived age of each candidate and participants’ vot- ing intention toward them. Specifically, the voting intention of participant j for candidate i was modeled as a function of candidate i’s age as perceived by par- ticipant j (ageij), its quadratic term (ageij

2), candidate i’s apparent gender (genderi), and the interaction of candidate gender and candidate age (Ageij × Genderi, Ageij

2 × Genderi). This model is formally expressed as follows:

Voting intention Age

Gender

1 1 2 2ij j ij j

i

g g u g u= + + × + +

× + 00 0 0( ) ( )

(gg u

g u

g u

j i

ij j ij ij

j

3 3

4 4

5 5

Gender

Age Age Age

Ge

0

0

0

+ ×

× + + × ×

+ + ×

)

( )

( ) nnder Age

Age

i ij

ij ijr

×

× +

Variables starting with g are estimates of global slopes and intercept, variables starting with u are estimates of participant-to-participant variations, and variables start- ing with r are model residuals.

Consistent with the trends visible in Figure 1, the quadratic effects of age were qualified by an interaction with candidate gender, β = −1.85, t(288) = −6.04, p < .001 (see Table S4 in the Supplemental Material for other coefficients).

A random-effects test indicated that the standard devi- ations for the regression coefficients were significantly greater than zero (see Table S4 in the Supplemental Material), indicating reliable participant-to-participant variation in the effects of candidate gender and age. Participants’ gender accounted for some of that vari- ability. Participant gender interacted with the effect of candidate age (p < .005), the effect of candidate gen- der (p < .001), and their interaction terms (p < .001) to predict voting intention: Intentions to vote for female candidates decreased with age faster for male participants than for female participants (see Figs. 1c–1d; for statistical details, see Table S6 in the Sup- plemental Material). Comparing younger with older participants, we found that voting intention toward female candidates decreased faster with perceived age than for male candidates, but this two-way interaction was stronger for younger, compared with older, par- ticipants (Figs. 1a and 1b). The Participant Age × Can- didate Age × Candidate Gender interaction was statistically significant (p = .04) in predicting voting intention (for statistical details, see Table S10 in the Supplemental Material).

Nevertheless, even after participant gender and participant age were taken into account, there were still statistically significant individual differences in slopes, suggesting that individual differences in the effects of perceived candidate age and gender could not be accounted for solely by participants’ gender and age.

We conducted six additional studies with a combined sample size of 555 participants. As shown in Section I in the Supplemental Material, the Candidate Gender × Perceived Age interactions were statistically significant in all six studies.

Candidates’ Age and Gender Predict Voter Preference 937

Study 2

Study 2 was preregistered (https://aspredicted.org/ t2yc2.pdf ) and examined the role of candidate gender and age on voting intentions using an expanded stimu- lus set and a forced-choice paradigm that more closely mimics actual voting.

Method

Stimulus selection and preparation. We selected pairs of gender-matched candidates who clearly differed in age. From the initial set of 365 stimuli, we first identified younger candidates by selecting those whose mean esti- mated age was below 40. Then we selected candidates who were perceived as being clearly older than them (operationally defined as being rated as older than the oldest of the younger candidates by at least 14 of the 15 raters in a pilot test).

Participants. The preregistration called for collecting data from 300 MTurk workers. We wanted to let all partici- pants who started the experiment finish, so we collected data from 303 participants. Five participants were excluded in accordance with our preregistration plan. This left a final sample of 298 participants (171 women, 126 men, 1 who did not report gender; age: M = 37.66 years, SD = 12.74 years; 241 self-reported as White, 2 as South Asian, 27 as Black or African American, one as American Indian/ Native, 16 as East Asian, and seven as mixed race; one did not report race/ethnicity). All participants had IP addresses within the United States. No analyses were conducted before data collection was completed.

Procedure. Each participant saw 16 pairs of candidates presented in random sequence: six pairs of younger female versus older female candidates, six pairs of younger male versus older male candidates, and four same-age different-gender pairs featuring middle-age

Female Candidates

Male Candidates

Female Candidates

Male Candidates

Female Candidates

Male Candidates

Female Candidates

Male Candidates

Study 1 (Female Participants)Study 1 (Male Participants)

Study 1 (Older Participants)Study 1 (Younger Participants) 100

80

60

40

20

0

100

80

60

40

20

0

30 7060504030 70605040

30 70605040

100

80

60

40

Vo tin

g In

te nt

io n

Vo tin

g In

te nt

io n

Vo tin

g In

te nt

io n

Perceived Age of Candidate Perceived Age of Candidate

Perceived Age of Candidate

100

80

60

40

20

0

30 70605040

Perceived Age of Candidate

Vo tin

g In

te nt

io n

20

0

ba

dc

Fig. 1. (continued on next page)

938 Shen, Shoda

female versus middle-age male candidates, which were included to make sure that not all trials showed candi- dates of the same gender. Following the preregistration plan, we did not present any pairs containing younger female versus younger male candidates or older female versus older male candidates. The four trials featuring middle-age female versus middle-age male candidates were not included in the analysis.

On any given trial, participants viewed pairs of can- didates presented side by side on the screen on the Inquisit Web platform (Millisecond, 2015). One candi- date was presented on the left and another candidate

on the right. In addition to each candidate’s photo, the following information was given: (a) a name (sampled from a pool of 100 common female and male names), (b) a randomly generated post office box number, (c) a randomly generated phone number, (d) an education attainment (sampled from four different pairs of edu- cational attainments), (e) an occupation (sampled from four different pairs of occupations), and (f) a statement adapted from actual voter pamphlets (sampled from 16 different pairs of statements). Pairs of statements were pilot tested to be about equally likely to receive votes. From the set of matched pairs of statements, one pair

Female Candidates

Male Candidates

Female Candidates

Male Candidates

Study 3 (Female Participants)Study 3 (Male Participants)

Study 2 (Female Participants)Study 2 (Male Participants)

Younger Candidate

Older Candidate

Younger Candidate

Older Candidate

100

80

60

40

20

0

30 70605040

Perceived Age of Candidate

Vo tin

g In

te nt

io n

100

80

60

40

20

0

30 70605040

Perceived Age of Candidate

Vo tin

g In

te nt

io n

Vo tin

g In

te nt

io n

Vo tin

g In

te nt

io n

fe

hg

Candidate Gender Female Candidates Male Candidates

Candidate Gender

1.00

0.75

0.50

0.25

0.00

1.00

0.75

0.50

0.25

0.00

Female Candidates Male Candidates

Fig. 1. Voting intention in Studies 1 through 3. For Study 1 (a–d) and Study 3 (g, h), voting intention on a 100-point scale is shown as a function of participants’ estimates of candidates’ ages. Results are broken down in Study 1 by par- ticipant age (a, b) and gender (c, d) and in Study 3 by participant gender. Shaded areas represent bootstrapped 95% confidence intervals (CIs). For Study 2, the graphs show the proportion of (e) male and (f) female participants who voted for candidates in each of four age–gender categories (i.e., younger female, older female, younger male, or older male). Error bars represent 95% CIs.

Candidates’ Age and Gender Predict Voter Preference 939

was randomly selected for each trial. Within each selected pair, one was randomly assigned to be pre- sented on the left and the other on the right (for stimuli, see Section III in the Supplemental Material). After see- ing the candidate photos and reading their profiles, participants were asked to vote for one of the two candidates by checking the box below that candidate’s profile.

Results

Multilevel logistic regression was used to examine whether participants were more likely to vote for younger candidates than for older candidates and whether those intentions were moderated by candidate gender. For candidate pair i and participant j, the dependent variable ln(Pij /(1 − Pij)) represents the log odds of the younger candidate receiving the vote. The dependent variable was regressed on the gender of candidate pairs (either both females or both males), with a random intercept and random slope that were fitted using the following formula:

ln 1 PairGender1 1( /( )) ( ) ( )P P g u g uij ij j j i− = +00 0 0+ + ×

The parameters were estimated using the generalized linear mixed model (glmer) module in lme4 (Version 1.1-14; Bates, 2005) and lmerTest (Version 2.0-33; Kuznetsova et al., 2014) in R. The results indicated that participants were on average more likely to vote for the younger candidate over the older candidate (g00 = 0.46, SE = 0.07, p < .001). Most importantly, the negative effect of age was stronger for female than for male candidates (g10 = −0.23, SE = 0.08, p = .004).

The effect of candidate gender and age on voting intention differed significantly across participants, χ2(2, N = 298) = 11.99, p = .002. Through exploratory analy- sis, we found that this effect was moderated by partici- pant gender: Male participants were significantly more likely to exhibit a preference for younger candidate when both candidates were female than when both candidates were male (p < .001; see Figs. 1e–1f; for statistical details, see Section II.3 in the Supplemental Material). The Participant Age × Candidate Age × Can- didate Gender interaction was not statistically signifi- cant (p > .1) in predicting voting intention.

Study 3

Study 3 replicated the findings of Study 1 and examined the extent to which the results could be accounted for by the influence of perceived competence, warmth, and attractiveness.

Method

Participants. One hundred eighty undergraduate stu- dents at the University of Washington participated in the study in exchange for course credit. Five participants were excluded because they failed to complete the study. This resulted in a final sample of 175 participants (49% male, 51% female; age: M = 19.38 years, SD = 2.86 years; 76 participants self-identified as White, 33 as East Asian, 28 as South Asian, 11 as Black or African American, 10 as mixed race, five as Native Hawaiian, two as American Indian/Alaska Native, and 10 as “other/unspecified”). Data were collected in early 2016. No analyses were con- ducted before data collection was completed.

Procedure. Participants came to the lab to participate in a study called “Impression Formation and Politics.” They were first asked to estimate the age of the same 93 can- didates used in Study 1. Next, they were asked to rate the same 93 candidates on competence, warmth, and attrac- tiveness. Participants rated all 93 candidates on one dimen- sion (e.g., competence) before moving to the next dimen - sion. The order of the rating tasks and the sequence of candidates were randomized. Finally, participants indi- cated their voting intention toward the same 93 candi- dates on a 100-point slider.

Results

Consistent with the findings from Studies 1 and 2, results showed that participants favored the very young female candidates over the very young male candidates. How- ever, voting intentions toward female candidates decreased with candidate age, whereas voting intentions for male candidates first increased and then decreased with age (see Figs. 1g–1h; for statistical details, see Table S12 in the Supplemental Material). Also consistent with the findings from Study 1, results showed a significant Candidate Age × Candidate Gender interaction, but the interaction was stronger for male participants (Fig. 1g) than for female participants (Fig. 1h; p < .01 for the Candidate Age × Candidate Gender × Participant Gen- der interaction; for other statistical details, see Table S8 in the Supplemental Material).

Next, we examined the role of perceived warmth, competence, and attractiveness in accounting for the relationship between candidate age and voting inten- tion. Specifically, we conducted separate mediation analyses for female and male candidates that assumed a causal sequence in which perceived age influenced perceived competence, attractiveness, and warmth of candidates, which in turn influenced participants’ vot- ing intention. Although we believe this sequence is reasonable, the results should be interpreted as

940 Shen, Shoda

conditional on this specific causal-sequence assumption (Baron & Kenny, 1986, p. 1177).

In this within-subject mediation analysis, first the outcome variable was simultaneously regressed on all the mediators and the independent variable (i.e., can- didate age). Then each mediator was regressed on the independent variable. The analysis was done separately for younger candidates (age < 45 years) and older can- didates (age ≥ 45 years) because the assumption of linear relations made by the mediation analysis would not hold when the entire age range was included (see Fig. 1), whereas the assumption was met reasonably well when the data for younger candidates and older candidates were analyzed separately. The results are shown in Figure 2. The within-subject regression coef- ficients for female candidates (averaged across all par- ticipants) are shown in red, and the coefficients for male candidates are shown in blue.

Most notably, among younger candidates, candidate age had a greater negative effect on perceived attrac- tiveness for female candidates compared with male candidates. In addition, again among younger candi- dates, candidate age had a negative effect on perceived warmth for female candidates but a positive effect for male candidates. Candidate age had virtually no effect on perceived competence for female candidates but a positive effect for male candidates. For candidates per- ceived to be older than 45, gender differences in the relations between candidate age and the mediators were much smaller in magnitude compared with those for candidates younger than 45.

Following the method for assessing the indirect effects through mediators (Preacher & Hayes, 2008), we computed the product of the regression coefficient pre- dicting a given mediator (e.g., perceived attractiveness) from the independent variable (i.e., candidate age) and the coefficient predicting the outcome variable (e.g., voting intention) from the mediator. For example, the indirect effect of perceived age on voting intention through perceived attractiveness was 5.59 (−1.03 × 5.43) for younger female candidates (see Fig. 2). The standard errors of the product of these coefficients were esti- mated by bootstrapping, providing the basis for statisti- cal significance testing.

There were significant differences (p < .001) between younger female candidates and younger male candi- dates in all three indirect effects. Specifically, the effect of candidate age on voting intention toward younger female candidates was substantially mediated by per- ceived attractiveness (mean indirect effect = −5.60, 95% CI = [−6.74, −4.91], p < .001), but for younger male candi- dates, the same indirect path was notably weaker, although still statistically significant (mean indirect effect = −1.72,

95% CI = [−2.28, −1.35], p < .001). Similarly, for younger female candidates, perceived warmth significantly medi- ated the relationship between perceived age and voting intentions (mean indirect effect = −2.71, 95% CI = [−3.39, −2.22], p < .001), whereas the same indirect path was much weaker for younger male candidates, although still statistically significant (mean indirect effect = 0.54, 95% CI = [0.33, 0.92], p  < .001). The indirect effect through perceived competence was not statistically sig- nificant (mean indirect effect = −0.25, 95% CI = [−1.10, 0.57]) for younger female candidates but highly significant for younger male candidates (mean indirect effect = 1.69, 95% CI = [1.06, 2.32], p < .001). Voting intentions toward older candidates were also mediated by perceived attractiveness, warmth, and competence (all ps < .01), but the magnitude of indirect effects did not differ between older female and older male candidates as much as they did for younger candidates (for more statistical details on the candidate gender differences in indirect effects, see Table S5 in the Supplemental Material).

We also applied the approach proposed by Hayes and Preacher (2010) to estimate the indirect effects in the presence of quadratic relationships (see Section II.7 and Fig. S15 in the Supplemental Material). The results from these analyses were entirely consistent with those from the Preacher and Hayes (2008) method.

General Discussion

When averaged across participants, preferences for the candidates in the present studies were systematically related to the candidates’ age and gender, gleaned from their photos even when all factors beyond appearance were held constant or randomized. Specifically, for very young candidates (< 35 years old), participants favored women over men. However, their intention to vote for female candidates steadily decreased with candidates’ age, whereas their intention to vote for male candidates increased until the candidates were around 45 years old and then slightly decreased. When the effects were exam- ined separately for each participant, which was possible because we used the highly repeated within-person para- digm, we found that the effects of candidate age and gender on voting intention varied reliably across partici- pants. Even after participant gender and participant age were taken into account, there were still reliable indi- vidual differences in these effects.

Our findings were highly consistent across the study paradigms, stimulus samples, and participant groups represented in the present studies. We also believe our findings likely generalize to our initial set of 365 candi- dates (see Study S4 in the Supplemental Material).

Candidates’ Age and Gender Predict Voter Preference 941

However, generalizability beyond that set is unknown. Our stimuli came from individuals who had already been elected, and thus the stimulus set may already reflect biases present in the U.S. electorate. The partici- pants of the present studies were undergraduate stu- dents and MTurk workers, who are often younger and

more liberal than the general population (Berinsky et al., 2012). Thus unlike opinion polls, lab studies such as this do not provide direct predictions of election outcomes.

A surprising finding from the current studies was that MTurk workers and undergraduate students consistently

Perceived Age

Perceived Warmth

Voting Intention

Perceived Competence

Perceived Attractiveness

−0.55

0.12

5.43

6.02

−1.03

−0.28

8.52

7.12

−0.02

0.23

∗

Perceived Age

Perceived Warmth

Voting Intention

Perceived Competence

Perceived Attractiveness

−0.10

−0.14

3.95

4.03

6.85

6.47

−0.38

−0.31

6.92

6.45

−0.20

−0.06

∗

4.42

4.86∗

∗

Younger Candidates (< 45 Years Old)

Older Candidates (≥ 45 Years Old) b

a

Fig. 2. Within-subject mediation models showing the effects of perceived age on vot- ing intention, as mediated by perceived attractiveness, perceived warmth, and perceived competence (Study 3). Models are shown separately for (a) candidates perceived to be younger than 45 years old and (b) candidates perceived to be 45 years or older. All variables except for voting intention were standardized. Values in red are within-subjects regression coefficients for female candidates (averaged across all participants), and values in blue are within-subjects regression coefficients for male candidates. The solid lines represent indirect effects, and dashed lines represent the remaining effect not accounted for by the mediators examined in this study. Asterisks indicate paths for which the coefficients for female and male candidates differ significantly (p < .05). The p values were obtained on the basis of 5,000 bootstrapped samples.

942 Shen, Shoda

favored the younger female candidates in our stimulus set. One potential mechanism for this finding is that participants in Studies 2 and 3, who were generally young, preferred younger and presumably more liberal (e.g., female) candidates. However, Study 1 had par- ticipants who were older. They also preferred younger female candidates, albeit less strongly than younger participants. This suggests that there were other mecha- nisms at play. For example, female and male candidates may differ in how their perceived age is related to their perceived attractiveness, warmth, and competence, which in turn predict voting intentions toward them. Study 3 supports such a possibility, but more research is needed in which candidates’ perceived age and gen- der are experimentally manipulated in order to address the causal role of these factors.

Transparency

Action Editor: John Jonides Editor: D. Stephen Lindsay Author Contributions

Y. A. Shen and Y. Shoda developed the hypothesis and designed the studies. Data were collected by Y. A. Shen. Y. A. Shen analyzed the data under the guidance of Y. Shoda. Y. A. Shen and Y. Shoda drafted and revised the manuscript. Both authors approved the final manuscript for submission.

Declaration of Conflicting Interests The author(s) declared that there were no conflicts of interest with respect to the authorship or the publication of this article.

Funding This research was partly supported by the Bolles Fund from the University of Washington Department of Psychology.

Open Practices The design and analysis plans for Studies 1 and 2 were preregistered at AsPredicted (https://aspredicted.org/ ng5vg.pdf and https://aspredicted.org/t2yc2.pdf, respec- tively). Data and materials for Studies 1 through 3 have not been made publicly available. This article has received the badge for Preregistration. More information about the Open Practices badges can be found at http://www.psy chologicalscience.org/publications/badges.

ORCID iD

Yiqin Alicia Shen https://orcid.org/0000-0002-7138-9290

Acknowledgments

We thank Jason Webster for his help through every phase of preparing this manuscript.

Supplemental Material

Additional supporting information can be found at http:// journals.sagepub.com/doi/suppl/10.1177/0956797620977518

References

Baron, R. M., & Kenny, D. A. (1986). The moderator–media- tor variable distinction in social psychological research: Conceptual, strategic, and statistical considerations. Journal of Personality and Social Psychology, 51, 1173– 1182. http://doi.org/10.1037//0022-3514.51.6.1173

Bates, D. (2005). Fitting linear mixed models in R: Using the lme4 package. R News, 5(1), 27–30. https://www .r-project.org/doc/Rnews/Rnews_2005-1.pdf

Berggren, N., Jordahl, H., & Poutvaara, P. (2010). The looks of a winner: Beauty and electoral success. Journal of Public Economics, 94(1–2), 8–15. https://doi.org/10.1016/j .jpubeco.2009.11.002

Berinsky, A. J., Huber, G. A., & Lenz, G. S. (2012). Evaluating online labor markets for experimental research: Amazon. com’s Mechanical Turk. Political Analysis, 20, 351–368. https://doi.org/10.1093/pan/mpr057

Brewer, M. B., & Lui, L. N. (1989). The primacy of age and sex in the structure of person categories. Social Cognition, 7(3), 262–274. https://doi.org/10.1521/soco.1989.7.3.262

Bryk, A. S., Raudenbush, S. W., & Congdon, R. (2010). HLM 7 for Windows [Computer software]. Scientific Software International.

Chiao, J. Y., Bowman, N. E., & Gill, H. (2008). The political gender gap: Gender bias in facial inferences that predict voting behavior. PLOS ONE, 3(10), Article e3666. https:// doi.org/10.1371/journal.pone.0003666

Dockterman, E. (2016, May 11). ACLU: The federal government is investigating sexism against directors in Hollywood. Time. http://time.com/4326519/aclu-government-sexism- female-directors-hollywood

Hayes, A. F., & Preacher, K. J. (2010). Quantifying and test- ing indirect effects in simple mediation models when the constituent paths are nonlinear. Multivariate Behavioral Research, 45(4), 627–660. https://doi.org/10.1080/00273 171.2010.498290

Kuznetsova, A., Brockhoff, P. B., & Christensen, R. H. B. (2014). lmerTest: Tests in linear mixed effect models (R Package Version) [Computer software]. http://CRAN.R- project.org/package=lmerTest

McLellan, B., & McKelvie, S. J. (1993). Effects of age and gender on perceived facial attractiveness. Canadian Journal of Behavioural Science, 25(1), 135–142. https:// doi.org/10.1037/h0078790

Millisecond. (2015). Inquisit 4 [Computer software]. https:// www.millisecond.com/

North, M. S., & Fiske, S. T. (2015). Modern attitudes toward older adults in the aging world: A cross-cultural meta- analysis. Psychological Bulletin, 141(5), 993–1021. https:// doi.org/10.1037/a0039469

Olivola, C. Y., & Todorov, A. (2010). Elected in 100 milli- seconds: Appearance-based trait inferences and voting. Journal of Nonverbal Behavior, 34, 83–110. https://doi .org/10.1007/s10919-009-0082-1

Preacher, K. J., & Hayes, A. F. (2008). Asymptotic and resa- mpling strategies for assessing and comparing indirect effects in multiple mediator models. Behavior Research Methods, 40, 879–891. https://doi.org/10.3758/BRM.40 .3.879

Candidates’ Age and Gender Predict Voter Preference 943

R Core Team. (2017). R: A language and environment for statistical computing (Version 3.4.0) [Computer software]. Retrieved from https://www.r-project.org/

Sontag, S. (1972, September 23). The double standard of aging. The Saturday Review, pp. 29–38. http://www.unz .com/print/SaturdayRev-1972sep23-00029

Stockemer, D., & Sundström, A. (2018). Age representation in parliaments: Can institutions pave the way for the young? European Political Science Review, 10(3), 467–490. https://doi.org/10.1017/S1755773918000048

Sutherland, C. A. M., Oldmeadow, J. A., Santos, I. M., Towler, J., Burt, D. M., & Young, A. W. (2013). Social inferences from faces: Ambient images generate a three-dimensional model. Cognition, 127(1), 105–118. https://doi.org/10.1016/j .cognition.2012.12.001

Todorov, A., Mandisodza, A. N., Goren, A., & Hall, C. C. (2005). Inferences of competence from faces predict elec- tion outcomes. Science, 308(5728), 1623–1626. https:// doi.org/308/5728/1623

Whitsett, D. D., & Shoda, Y. (2014). An approach to test for individual differences in the effects of situations with- out using moderator variables. Journal of Experimental Social Psychology, 50, 94–104. https://doi.org/10.1016/j .jesp.2013.08.008

Zayas, V., Sridharan, V., Lee, R. T., & Shoda, Y. (2019). Addressing two blind spots of commonly used experi- mental designs: The highly-repeated within-person approach. Social and Personality Psychology Com pass, 13(9), Article e12487. https://doi.org/10.1111/spc3. 12487