deltie.net archive

Extracted abstract

How do returns to physical attractiveness in the labour market vary depending on the alignment between an employee's gender and the gender composition of their occupation? Research suggests that attractiveness often invokes gender-typical stereotypes. If these stereotypes imply a mismatch between what are considered to be necessary traits in an occupation and those which are ascribed to a person, this may lead to penalties for attractive individuals with a weak trait-job fit. To test this proposition, we combine longitudinal data from the German Family Panel (pairfam) including interviewer ratings of attractiveness with official statistics on occupational gender segregation. Employing a multiverse analysis across 6,912 model specifications, we consistently find positive associations between individuals' attractiveness and their earnings for both women and men. However, we do not find evidence that attractive individuals face earnings penalties when employed in occupations typically dominated by the opposite gender.

The study of physical attractiveness and its consequences for social inequality and social stratification is receiving increasing attention in the social sciences. The general consensus appears to be that attractive individuals have considerable advantages in many areas of life. Benefits of physical attractiveness have been reported to occur as early as infancy and childhood (Langlois et al., 1995 (Langlois et al., , 2000;; Dunkake et al., 2012) and continue throughout life (Hamermesh, 2011; Jaeger, 2011; Hakim, 2012) . In the labour market, attractiveness is linked to higher earnings and greater likelihood of being invited for a job interview, as shown by extensive research (Hamermesh and Biddle, 1994; Hosoda, Stone-Romero and Coats, 2003; Hakim, 2010; Jaeger, 2011; Wong and Penner, 2016; Kuwabara and Thébaud, 2017) .

Although physical attractiveness is often assumed to be more consequential for women than for men, the evidence is contradictory (Kukkonen et al., 2024) . Physical attractiveness seems to be beneficial for men and women, but not in all contexts, and there is still no conclusive evidence on how gender moderates attractiveness returns in the labour market (Wolbring and Riordan, 2016; Kukkonen et al., 2024) . This mixed evidence might be due to the fact that the occupational context has not been systematically accounted for, particularly the interaction between perceived physical attractiveness, perceived gender-typicality (understood here as being seen as typically feminine or masculine), and the lack of person-job fit (between the gender of the employee and the sex-typed occupation). A theoretical basis for the relevance of the occupational context is the lack-of-fit model (Heilman, 1983) . If gender stereotypes imply a mismatch between traits considered necessary in an occupation and those ascribed to a person based on their physical appearance, this may result in adverse outcomes, such as lower earnings. Thus, highly attractive women (men) in predominantly male (female) occupations may experience attractiveness penalties. This occurs because attractive persons could be perceived as gender-typical and therefore be ascribed gender-typical traits, potentially creating a perceived mismatch between those traits presumably necessary in an occupation and those attributed to the person based on their appearance. Previous empirical evidence on potential beauty penalties for persons in gender-atypical tasks and roles comes mainly from U.S. studies using (quasi-) experimental designs in specific contexts, such as entrepreneurial pitch competitions (Brooks et al., 2014) or peer-to-peer lending (Kuwabara and Thébaud, 2017) . These results raise the question of whether the context-specific nature of gendered beauty penalties can be generalized to other contexts.

To address this gap, this paper examines how physical attractiveness affects earnings for men and women in Germany using representative longitudinal survey data-from the German Family Panel pairfam (Brüderl, Garrett, et al., 2021) -in combination with data from official statistics from the Federal Employment Agency on the level of occupational gender segregation. We examine whether the returns to physical attractiveness for women and men on the German labour market systematically vary depending on the share of opposite-gender employees in their occupation.

The contribution of our study is threefold: First, by using representative survey data we extend existing research, which has primarily relied on field or experimental data in low-information situations where reliance on signals and stereotypes is prevalent (Brooks et al., 2014; Kuwabara and Thébaud, 2017) , to more common, 'real'-world settings. Second, we focus on Germany-a country where gender-specific stereotypes regarding the traits presumed important in an occupation also exist (Lotzkat and Welpe, 2015) with a comparatively high degree of labour market regulation and less employer discretion in wagesetting compared to countries with liberal welfare regimes (Esping-Andersen, 1990; Hall and Soskice, 2001) . We therefore extend previous research, which has mostly been conducted in countries with comparatively less regulated labour markets and a high degree of employer discretion in wage-setting-most often in the U.S. (Hamermesh and Biddle, 1994; Hosoda, Stone-Romero and Coats, 2003; Jaeger, 2011; Kuwabara and Thébaud, 2017) . By doing so, we aim to contribute to the growing body of literature on attractiveness-based social stratification that highlights how reactions to perceived physical attractiveness are context-dependent and interact with other characteristics, such as gender (Hakim, 2010; Johnson et al., 2010; Kuwabara and Thébaud, 2017; Monk, Esposito, and Lee 2021; Kukkonen et al., 2024) . Third, we contribute methodologically by conducting a multiverse analysis (Steegen et al., 2016) with 6,912 model specifications. This approach allows us to run a large set of reasonable analytic models in parallel, providing transparency into how different data analytic decisions influence the results, and to assess the robustness of our conclusions.

Our results consistently show that both women and men rated as highly attractive tend to have higher earnings compared to those who are rated as less attractive. However, our analysis of the interaction between attractiveness and the level of gender segregation in an occupation finds no earnings penalty for highly attractive individuals employed in gender-atypical occupations in Germany.

Previous research and theoretical framework

Gender, occupational gender segregation, and physical attractiveness Cross-sectional (Fletcher, 2009; Wong and Penner, 2016; Anýžová and Matějů, 2018) and longitudinal research (Frieze, Olson and Russell 1991; Biddle and Hamermesh, 1998; Judge, Hurst and Simon, 2009) consistently find positive associations between physical attractiveness and labour market outcomes, such as earnings. A beauty premium is also corroborated in experimental studies (Mobius and Rosenblat, 2006; Deryugina and Shurchkov, 2015) . While some studies report that both men and women benefit from physical attractiveness (Jaeger, 2011; Sala et al., 2013; Wong and Penner, 2016) , others find that the effect of attractiveness varies by gender. In particular, women are more likely to experience attractiveness-related premiums and penalties (Kuwabara and Thébaud, 2017; Monk, Esposito, and Lee, 2021; Eberl, Kühn and Wolbring 2022) . Therefore, it remains unclear how gender affects the economic returns of physical attractiveness in the labour market (Kukkonen et al., 2024) .

One explanation for gendered effects of physical attractiveness is the moderating role of occupational context in shaping attractiveness premiums and penalties in the labour market. For example, Johnson et al. (2010) suggest that perceived attractiveness is only disadvantageous for women applying for masculine sextyped jobs. Heilman's (1983) lack-of-fit model serves as an explanation for this finding: Disadvantages arise when a person's characteristics do not fit the job requirements. In fact, several scholars find that attractive individuals tend to be ascribed more gendertypical attributes (Gillen, 1981; Heilman and Stopeck, 1985; Drogosz and Levy, 1996) , which are perceived as incompatible with the characteristics believed to be necessary for jobs stereotypically associated with the opposite sex. Much of this research is based on experimental studies conducted in the U.S., including the pioneering work by Heilman and Saruwatari (1979) , who first suggested that attractive women encounter disadvantages when applying for managerial positions.

Recently, researchers have used experiments and data from online markets to study the relationships between gender, task gender-typing, and the impact of physical appearance on economic outcomes, yielding mixed results for males and females (Brooks et al., 2014; Kuwabara and Thébaud, 2017) . Brooks et al. (2014) find that attractive men are particularly persuasive in entrepreneurial venture pitches in terms of their ability to secure investment, whereas physical attractiveness did not matter for female entrepreneurs. Conversely, Kuwabara and Thébaud (2017) observe that women seeking online peer-to-peer business loans are less likely to receive funding if they are attractive, suggesting a beauty penalty for women in male-typed domains, like business ownership. However, existing survey evidence on gender-specific attractiveness outcomes has not taken into account the person-job fit concerning the gender of the employee and the occupational gender segregation, or in other words the sex-typing of occupations.

Theoretical framework

Expectation states theory (Berger, Fisek and Norman, 1977; Webster and Driskell, 1983; Ridgeway, 1991; Correll and Ridgeway, 2006) can explain how physical attractiveness can lead to labour market advantages. Various elaborations of the theory (Correll and Ridgeway, 2006; Berger and Wagner, 2007) agree that cultural beliefs shape expectations, impacting evaluations and interactions, thereby perpetuating existing status hierarchies 1 . When shared expectations associate competence and worthiness with categorical distinctions such as gender, these become diffuse status characteristics. Physical attractiveness is considered such a diffuse status characteristic (Webster and Driskell, 1983) because it is easily observable and differentiates individuals (attractive vs. unattractive). This characteristic carries expectations about general and specific competencies (Webster, and Driskell, 1983) : Attractiveness influences perceptions of skills and competence, with attractive individuals often seen as more intelligent, healthier (Thornhill and Gangestad, 1999) , and socially skilled (Eagly et al., 1991; Jackson, Hunter and Hodge, 1995; Langlois et al., 2000) , which can result in preferential treatment.

While the general theoretical expectation has been that perceived attractiveness has positive consequences for its 'bearer', a growing body of research has provided evidence that reactions to perceived physical attractiveness are context-dependent and interact with other characteristics (Heilman and Saruwatari, 1979; Heilman and Stopeck, 1985; Eagly et al., 1991; Johnson et al., 2010; Lee et al., 2015; Wolbring and Riordan, 2016; Kuwabara and Thébaud, 2017) . Specifically, perceived attractiveness may interact with gender, making gender beliefs (Ridgeway and Correll, 2004; Ridgeway, 2011) or gender stereotypes (Eagly and Karau, 2002; Heilman et al., 2015) more salient. It is argued that people who are perceived as being very attractive, in particular women, are also perceived as being very gender-prototypical, such that attractive women are perceived as very feminine (Heilman and Saruwatari, 1979; Heilman and Stopeck, 1985; Eagly and Karau, 2002; Heilman et al., 2015; Lee et al., 2015; Kuwabara and Thébaud, 2017) 2 . This can be disadvantageous for the attractive person if the salience of the gender belief creates an incongruity or mismatch (Heilman and Saruwatari, 1979; Heilman and Stopeck, 1985) between the skills that are presumed to be important in a situation or task and the gendertypical skills attributed to a person. If this proposition holds, attractive women (men) can be expected to face disadvantages in contexts which are culturally typed as masculine (feminine) (Ridgeway and Correll, 2004) . Male-typed tasks are seen to require agentic traits-for example, being aggressive, ambitious, dominant, and self-confident-while female-typed include communal, supportive traits-for example, being affectionate, helpful, kind, and nurturing (Heilman, 1980; Heilman and Stopeck, 1985; Eagly and Karau, 2002; Kuwabara and Thébaud, 2017) . Such gender-specific stereotypes about the traits that are presumed to be important in an occupation are prevalent in the German labour market (Lotzkat and Welpe, 2015) . Consequently, highly attractive women may face disadvantages in predominantly male occupations, such as engineering, or finance, if their perceived attractiveness is associated with ascriptions of female traits, which are culturally deemed unsuitable for these occupations. Conversely, in predominantly female occupations, like nursing or care-giving, where feminine traits are seen as more suitable, attractive men perceived as prototypically masculine may face disadvantages due to these stereotypes. Building on these theoretical foundations, we test the following two hypotheses in this paper: First, we expect that both women and men who are perceived as attractive will have higher earnings compared to those perceived as less attractive (hypothesis 1). Second, we expect that women and men perceived as attractive, but working in gender-atypical occupations, will have lower earnings than their counterparts perceived as less attractive in the same occupations (hypothesis 2).

Data and methods

Data

The analyses are based on longitudinal data from the German Family Panel pairfam (v12.0) (Brüderl, Drobnič, et al., 2021) . Starting in 2008 with more than 12,000 randomly selected individuals from the birth cohorts 1971-73, 1981-83, 1991-93, pairfam annually collects survey data with a focus on partnership and family living arrangements in Germany. A fourth birth cohort, 2001-2003, was added in 2018. We used all 12 waves from the main sample (Brüderl, Garrett, et al., 2021) 3 .

Multiverse analysis

We do not estimate a single set of models but rather conduct a multiverse analysis (Steegen et al., 2016) and present the results as specification curves (Simonsohn, Simmons and Nelson, 2020) . All analyses are based on numerous and, at times, arbitrary data analytic decisions, including study-centric (e.g. exclusion criteria), variable-centric (e.g. variable coding or transformations), or model-centric decisions (e.g. model choice, inclusion of random or fixed effects, covariates in the model) (Rijnhart et al., 2022) . Researchers face a multitude of decisions, often compared to navigating a 'garden of forking paths' (Gelman and Loken, 2014) , where a single analysis represents just one of many possible paths. A multiverse analysis offers insight into how robust (or sensitive) the results are to these decisions, addressing two fundamental problems of scientific research: the lack of transparency and model uncertainty (Young, 2018) .

However, multiverse analysis does not offer a remedy if the data are unsuited or the models misspecified, or more generally if the study design is unsuited. In such cases, relying on multiverse analysis may mistakenly suggest the robustness of results, when in fact, they are inherently biased or incorrect. Moreover, a multiverse analysis should not be used to 'outsource' data-analytic decisions. These decisions must be grounded in theoretical rationale, carefully reasoned, and transparently reported, regardless of whether a single analysis or a multiverse analysis is employed. Nevertheless, multiverse analysis presents a useful tool to examine the robustness of research findings and communicate them transparently, particularly when theory does not offer specific predictions.

In the following, we first describe the selection and coding of the variable sets for the analysis and then define the model space (Young, 2018) , as detailed in Table 1 . To make our decisions transparent, we also include those for which we do not specify alternatives. Our initial specifications are displayed in italics, these represent the choices we would have made if we had not conducted a multiverse analysis. However, this does not imply that we consider these associated data-analytic decisions as superior; rather, we aim to demonstrate that the various specifications in the multiverse analysis are equally plausible.

Variables

Outcome

The outcome variable in this study is the respondents' earnings, which we analyse using six different versions.

We estimate a Mincer-type wage equation (Mincer, 1974) , assuming that the input factors, such as physical attractiveness and education, are multiplicatively linked to earnings. Therefore, and given the fact that earnings are strictly positive (Gelman and Hill, 2006, p. 56) , we use the natural logarithm of gross hourly earnings as the dependent variable. Gross hourly earnings are calculated from respondents' reported gross monthly earnings in the month prior to the interview and average work hours per week (monthly earnings / (hours worked per week * 4.3)). Since information on gross earnings is collected only every two years in pairfam, we also use the natural logarithm of net hourly earnings to increase sample size, as net earnings are collected annually. However, we prefer gross earnings as they are indicative of earnings before taxes and transfers and are thus a more accurate measure for labour market rewards. The reported earnings contain some implausible values, that is values which are unreasonably low or unreasonably high given other information about the respondents' work-potentially because those respondents reported annual instead of monthly earnings. Because of this, we created additional versions of the hourly earnings, either winsorized or trimmed at the 5 per cent and 95 per cent percentiles.

Physical attractiveness

The main independent variable, the treatment, is the interviewer's rating of the respondents' attractiveness in wave 1. In the first wave of pairfam, every interviewer was asked to rate the main respondents' attractiveness on a scale ranging from 1 'very attractive' to 7 'very unattractive' at the beginning of the interview. This measure shows a skewed distribution in pairfam, leaning towards the right, with few respondents being rated as very unattractive (see Figure A1 in the appendix). This may be due to interviewers' reluctance to categorize respondents as being unattractive in a face-to-face situation (see Tables A2 and A3 in the appendix). Given this skewed distribution, we estimate, among others, interviewer fixed effects models (see below), where effect estimation is based only on within-interviewer (co-)variance, making the measure relative within each interviewer thereby mitigating this issue.

Following different operationalizations in the literature, we used two versions of physical attractiveness; in its original scale as a continuous variable and as a dichotomized variable distinguishing between very attractive respondents (1) and not very attractive respondents (2 through 7). The binary variable is our initial choice as we assume that gender-prototypical beliefs will be salient particularly for very attractive people and given the skewed distribution of the measure, we believe that the binary measure best captures the correct contrast.

There are some interviewers with no or very little variance in their ratings of respondent attractiveness (overall s.d. in attractiveness ratings in wave 1 was 1.4). Since it is unlikely these interviewers encountered only respondents of the same level of physical attractiveness, we created three specifications for the multiverse analysis: (1) exclude data from interviewers with no variation (s.d. = 0), (2) exclude data from interviewers with little variation (s.d. <= 0.4), (3) use all data.

While interviewer ratings of physical attractiveness are an established measure (Kukkonen et al., 2024) , they are not perfect as they rely on the interviewer's subjective judgment. Ideally, physical attractiveness ratings would be constructed through the truth-ofconsensus method (Berscheid and Walster, 1974) , where each individual is rated by a relatively large number of independent raters to increase reliability. However, this approach is not feasible in a CATI setting, where interviewers are tasked with rating respondents' attractiveness. This drawback presents a limitation of our study. Nevertheless, previous research has shown that studies using interviewer ratings of physical attractiveness produce comparable results to those that use attractiveness measures based on the truth-of-consensus method (Schneickert, Steckermeier and Brand, 2020) . With each interviewer conducting 25 interviews on average, we can estimate models with interviewer random and fixed effects to account for constant differences between interviewers (see below). This allows us to rule out that constant interviewer characteristics affect the attractiveness ratings.

Gender

Until the 14th wave in 2021/2022, the pairfam survey only distinguished between female and male respondents. The information was based on the interviewers' assessment and not provided by respondents themselves, which may be problematic as it relies on external judgment. Starting in the 14th wave, pairfam began asking respondents to self-report their gender, distinguishing between female, male, and diversef. 32 respondents reported a different gender than recorded in previous waves by the interviewers, indicating that for the majority of respondents (> 99 per cent in wave 14), there is congruence between interviewer-ascribed (binary) gender and self-attributed gender. Given these data limitations and the fact that official labour market statistics only distinguish between female and male employees, our analysis is constrained by a binary classification of gender.

Occupational gender-typicality

Lacking a direct measure of the gender stereotypes associated with specific occupations, we operationalize the gender atypicality of the occupations through a proxy by differentiating between 'male', 'female', and 'integrated' occupations, based on the respective proportion of women or men employed in said occupations. While a direct measure would be preferable, gender composition is theoretically and empirically related to gender stereotypes (Bielby and Baron, 1986; Charles and Grusky, 2018; Damelang and Ebensperger, 2020) .

To obtain the occupations' gender composition, we used official statistics from the Federal Employment Agency indicating the percentage of female employees in 36 main occupational groups based on the German Classification of Occupations 2010 ('Klassifikation der Berufe' -KldB 2010) 4 . The KldB 2010 uses job titles to group jobs according to their content. It is hierarchically structured, with ten main fields at the top (one-digit level) and 1,286 job classes at the bottom (five digits) (Bundesagentur für Arbeit, 2011a, 2011b). The two-digit occupational main groups consist of 37 groups, but since the official data did not include information on category 01 (armed forces personnel), we had to exclude 37 persons from the analysis (32 men, 5 women). Since the Federal Employment Agency updated their classification in 2011, information on the percentage of female and male employees based on the current classification (KldB 2010) is only available starting in 2012. Thus, to test our second hypothesis, we can only use pairfam data starting with wave 4. The distribution of female employees in each occupational group as well as the distribution of the pairfam sample of the initial specification across the occupational groups is displayed in Table A4 in the appendix.

For our initial specification (1), we use Hakim's (1993) classification of occupations as 'male', 'female', or 'integrated'. Following the policy-oriented definition (Hakim, 1993) , occupational groups were created based on the 40 per cent average female share in the workforce-which also holds in the KldB data. Integrated or mixed occupations are defined as those within a 30 per cent band around the midpoint (40 per cent ± 15 per cent), male occupations as those with <25 per cent female workers, and female occupations as those with >55 per cent female workers. In two additional specifications, we use (2) a broader bandwidth (40 per cent ± 25 per cent), and (3) a continuous variable measuring the percentage of opposite-gender employees in the 36 main occupational groups.

Control variables

Building on theoretical considerations and previous studies, we control for relevant confounding variables that may jointly determine a person's perceived attractiveness and earnings. These variables include gender, age, immigrant status, height, BMI, health, and parental socio-economic background. To clarify the assumed causal relationships, we provide a directed acyclic graph (DAG) and a table listing the covariates and their causal roles in the appendix (Figure A2 and Table A2 in the appendix).

We can control all specified confounders, except for family background since the only information available is parental education. Considering the myriad ways (e.g. Bourdieu, 2018) in which (parental) background may affect the socially perceived physical attractiveness of persons and their earnings, it is very likely that there is an open backdoor path introducing bias in our estimates. The estimates should thus be interpreted as associational rather than causal. For some covariates, e.g. respondent personality, it is unclear whether they are a confounder or a mediator. Respondent personality-measured in pairfam via the Big Five-may impact how the interviewers rate respondent attractiveness (Dunkel et al., 2017) . However, physical attractiveness may also, through the reaction it elicits from the social environment, impact personality. Controlling for a mediator may result in overcontrol or overadjustment bias (Schisterman, Cole and Platt, 2009) , whereas not controlling for a confounder would lead to confounding bias. For this reason, we systematically vary the inclusion of covariates with an unclear status in the multiverse analysis (see below and Table A1 in the appendix). Since the specifications differ in case numbers, we present descriptive statistics only for the two initial model specifications (see Table A2 and A3 in the appendix).

Missing values

Missing values were handled through listwise deletion. This is a suboptimal strategy as it requires the strong assumption that missingness is completely at random (Rubin, 1976) . However, the use of multiple imputation was not feasible due to excessive computational time 5 .

Methods

Model space

The pairfam data have a three-level structure: Individual observations are nested within respondents and interviewers. We restrict the model space (Young, 2018) to a linear regression framework. A linear model which distinguishes between these three levels can be characterized by the following equation:

y ijt = β 0 + β 1 x ijt + β 2 x ij + β 3 t + u i + v j + ε ijt (1)

with i indicating respondent, j indicating interviewer, and t indicating period. y ijt is the outcome (earnings) and x ijt and x ij are time-varying and time-constant covariates. The model contains a respondent effect (ui), an interviewer effect (v j ), and the idiosyncratic error (ε ijt ). Unbiased estimates would require E(u i |x ijt , x ij ) = 0, E(v j |x ijt , x ij ) = 0, and E(ε ijt |x ijt , x ij ) = 0. In this context, we have to decide on how to model the dependencies in the data, which arise through multilevel structure.

We estimate three models: (1) A linear model with interviewer fixed effects, which adjusts the standard errors for two-way clustering in interviewers and respondents (Correia, Guimarães and Zylkin, 2020) . This model also excludes so-called singleton observations to avoid overstated statistical significance (Correia, 2015) . (2) A linear model with respondent random effects and cluster-robust standard errors and (iii) a linear model with interviewer random effects and cluster-robust standard errors. We cannot estimate models with respondent fixed effects, as our measure of physical attractiveness is time constant 6 . All models use calibrated design weights (Brüderl, Garrett, et al., 2021) .

Statistical inference for the multiverse analysis

To answer the question of whether the results of the multiverse analysis are consistent with the null hypothesis of no effect, Simonsohn et al. Simonsohn et al. (2020) suggest using a resampling method since it does not seem possible to generate the distributions for any test statistics under the null hypothesis analytically. The basic idea is to use a resampling method to estimate the distribution under the null hypothesis empirically. Unfortunately, this strategy is not feasible for our purpose as a reasonable set of resampled specification curves (e.g. 500 resamples) (Simonsohn, Simmons and Nelson, 2020) would be computationally excessive (see Footnote 5).

Table 1 gives an overview on all decisions and their alternatives. Overall, we estimate 6,912 unique specifications. The analyses were carried out using Stata 17.0. Code for reproducing the analysis has been archived on the Open Science Framework. 7 The analysis partly relies on code provided by Simonsohn et al. (2020) .

Results

In total, we estimated 6,912 specifications, including 864 specifications each for men and women for the first hypothesis and 2,592 specifications each for men and women for the second hypothesis. We present three sets of results. Summary statistics for the multiverse analysis are presented in Table 2 . Figures 1 and 2 graphically depict the results of the influence analysis (Young and Holsteen, 2017) and Figures 3 through 6 present the specification curves (Simonsohn, Simmons and Nelson, 2020) .

Out of the 864 specifications for female respondents, we obtained a median effect size of 0.02 (Table 2 ). Since the outcome is the natural logarithm of the respondents' hourly earnings, this can be interpreted approximately as a percentage change: On average, attractive women earn around 2 per cent more than less attractive women. In our initial specification, this is estimated as 0.05 (s.e. = 0.021, see Figure 3 )-the larger size being mainly due to the fact that the initial specification uses the binary operationalization of interviewer rated attractiveness-which would correspond to predicted hourly gross earnings of 12.42 € for very attractive female respondents and 11.85 € for those not rated as being very attractive. The median p-value for all the specifications is 0.02 and about 68 per cent of the specifications return statistically For male respondents, the median effect size across the 864 specifications is 0.03 (Table 2 ), indicating approximately 3 per cent higher earnings for attractive men compared to their less attractive counterparts. The initial specification returns a somewhat larger estimate (b = 0.05, s.e. = 0.02, see Figure 4 ), predicting a difference in gross hourly earnings between very attractive male respondents and those not rated as being very attractive of 14.15 € vs. 13.63 €. The median p-value across all specifications is (smaller than) 0.00, 91 per cent of the specifications return statistically significant estimates (p-values <= 0.05), and all specifications return positive effect estimates.

For the influence analysis we regressed the estimated effects ( β) on the dimensions of the multiverse analysis. Importantly, the influence analysis does not tell which data-analytic decisions are correct. Rather it illustrates which decisions have the largest impact on the results of the multiverse analysis (Young and Holsteen, 2017) . Figures 1 and 2 show the results of the influence analysis for the association between physical attractiveness and earnings for women and men respectively. For both women and men, the estimated associations are smaller in those specifications which use interviewer or respondent random effect models rather than interviewer fixed effect models. Data-analytic decisions with regard to exclusion criteria-respondent age or interviewers with no or little variance in their ratings of respondent attractiveness-have little impact on the estimated effects. For women and men using a smaller set of covariates results in larger effect estimates and using a continuous measure of attractiveness results in smaller estimates. For women, using the initial version of the outcome-log hourly earnings trimmed at the 5 and 95 percentile-leads to the smallest estimators in the multiverse analysis. All other versions of earningswinsorized or unadjusted-result in larger estimates of the association between physical attractiveness and hourly earnings. For men this holds too with the curious exception of trimmed net hourly earnings, which results in estimates which are smaller than those obtained when using trimmed log hourly earnings. It is advisable to be cautious with interpretations at this point, as we do not know whether the seemingly implausible earning values were actually wrong or not.

Figures 3 and 4 present the specification curves, which provide an additional graphical illustration of how the data-analytic decisions affect the estimated effects. The advantage of the specification curves is that, unlike an influence regression, it does not assume purely additive effects of the data analytic decisions. The bottom panel of the figures provides information about the specifications. The top panel depicts estimated effect sizes. The specifications are sorted by effect size in ascending order.

Figure 3 -the specification curve for female respondents-shows that small and statistically insignificant effect estimates come from specifications where a continuous indicator of attractiveness was combined with using all controls and respondent random effects models. Whereas there is no clear pattern observable for the vast majority of statistically significant estimatesexcept for the unsurprising fact that using a binary indicator of interviewer rated attractiveness is associated with larger effect estimates. Figure 4 , the corresponding figure for male respondents, does not reveal any telling pattern. Overall, across all specifications there is robust evidence for an association between interviewer rated physical attractiveness and earnings for both female and male respondents, lending support for hypothesis 1.

Regarding hypothesis 2, we observed robust null findings for the interaction between interviewer-rated attractiveness and the gender-atypicality of the respondent's occupation across 5,184 specifications (Table 1 and Figure 5 and 6 ). The median effect size of the interaction for female respondents is 0.01 and male respondents 0.00, the median p-values are 0.37 and 0.49, and the vast majority of specifications return statistically insignificant effect estimates (97 per cent and 98 per cent). Moreover, the estimated effects are not stable with regard to their sign (see Table 2 and Figure 5 and 6 ). Figure 5 and 6 show that it would have been possible to 'find' a specification which returned a statistically significant effect estimate supporting the hypothesis but the multiverse analysis clearly speaks against such a conclusion. This strongly suggests that within our set of reasonable specifications, there is no support for hypothesis 2, that there is an attractiveness penalty if attractive persons are employed in gender-atypical occupations. Therefore, we do not discuss the influence analyses in detail (see Figure A4 and A5 in the appendix).

Discussion

This study examines gendered attractiveness inequalities in the German labour market, focussing on how mismatches between an employee's gender and the level of occupational gender segregation in their field may result in a beauty penalty using nationally representative data. Using longitudinal data from the German Family Panel and conducting a multiverse analysis (Steegen et al., 2016; Simonsohn, Simmons and Nelson, 2020) , we found a stable positive association between perceived physical attractiveness and earnings for both women and men. Based on 864 specifications, we estimated the median magnitude of the earnings disparity between men who are perceived as being attractive and those who are not as 3 per cent. For women, the same number of specifications returned smaller effect estimates-a median difference of approximately 2 per cent in hourly earnings. It is an interesting finding that both the effect size for men is larger than for women and that the effects for men are more robust, understood as the proportion of specifications that are statistically significant. These results suggest that attractiveness is linked to higher earnings for both genders, with a stronger and more robust effect for men. It is noteworthy that the most recent and most comprehensive review on the relationship between physical attractiveness and labour market outcomes (Kukkonen et al., 2024) finds a similar number of studies suggesting that attractiveness benefits women more than men, and vice versa.

Furthermore, contrary to our theoretical expectations based on the lack-of-fit model (Heilman, 1983) , we find no evidence that attractive individuals employed in gender-atypical occupations have lower earnings than their less attractive counterparts in the same occupation. The vast majority of the 5,184 specifications, that is about 97 per cent for female and 98 per cent for male employees, yielded statistically insignificant results with estimates of negligible magnitudes.

This study makes several important contributions to a growing body of empirical research on social stratification and labour market inequalities by highlighting how gender and attractiveness intersect in shaping differences in earnings. First, we find a beauty wage gap in the German labour market. While similar gaps have been documented before in other national contexts (Hamermesh and Biddle, 1994; Hosoda, Stone-Romero and Coats, 2003; Jaeger, 2011; Kuwabara and Thébaud, 2017) , ours is the first study using representative longitudinal data from Germany to investigate this question systematically. It is noteworthy that we found evidence for the association between physical attractiveness and earnings in the German context, where the labour market is more regulated and offers less employer discretion in wage-setting compared to the United States (Esping-Andersen, 1990; Hall and Soskice, 2001) . But, as with the gender-wage gap (Auspurg, Hinz and Sauer, 2017) , our results clearly indicate that there is apparently enough room for beauty premiums in the German labour market.

Second, research on gendered labour market outcomes has largely neglected how gender stereotypes and role expectations with regard to the occupational context may moderate the relationship between gender, physical attractiveness, and earnings. Theoretical models, like the lack-of-fit model (Heilman, 1983) and expectation states theory (Berger, Fisek and Norman 1977; Webster and Driskell, 1983; Ridgeway, 1991; Correll and Ridgeway, 2006) , suggest that a person-job mismatch in terms of gender-typing of the task and the gender of the employee may result in a beauty penalty. This study is the first to test this proposition using largescale longitudinal survey data combined with official statistics on the share of opposite-gender employees in one's occupation. Contrary to previous findings suggesting that attractiveness can be a disadvantage for women in male-typed tasks (Brooks et al., 2014; Kuwabara and Thébaud, 2017) , our results show no evidence of a beauty penalty in gender-atypical occupations.

Lastly, by conducting a multiverse analysis (Steegen et al., 2016; Simonsohn, Simmons and Nelson, 2020) , we address two fundamental problems of scientific research-the lack of transparency and model uncertainty (Young, 2018) -in investigating how perceived physical attractiveness is related to earnings.

Given that gender-specific stereotypes about the traits deemed important in occupations also exist in the German context (Lotzkat and Welpe, 2015) , the question arises as to how the findings of this study can be reconciled with previous research. One possible explanation could be that gender-attractiveness stereotypes are less influential in repeated interactions, as in the workplace, compared to one-time interactions or low-information situations (e.g. entrepreneurial pitch competitions), where decision makers rely more on limited or imperfect information about candidates' competence, which might increase the probability to evaluate a (potential) employee's competence based on other available information (Phelps, 1972; Arrow, 1973) , such as attractiveness (Lee et al., 2015) . However, if this were the case it would contradict the overall positive association between perceived attractiveness and earnings. Alternatively, occupational (self-)selection could counteract adverse consequences of a lack-offit. The vocational specificity of the German education system and its rather tight coupling with the labour market (Allmendinger, 1989) , could result in a positive selection of persons who work in a gender-atypical occupation-although exploratory analyses did not reveal clear signs of such sorting.

The discrepancy between our results and those of previous studies could also be due to the fact that we use gender composition in occupations as a proxy for occupational gender-typicality. While this approach is common and while gender composition is both theoretically and empirically related to gender stereotypes (Bielby and Baron 1986; Charles and Grusky 2018; Damelang and Ebensperger 2020) , the proxy may not be accurate enough. Furthermore, it is possible that people working in gender-atypical occupations take on or are assigned gender-typical tasks: women in male-dominated occupations could take on tasks traditionally associated with femininity (e.g. emotional labour, support), while men in female-dominated occupations might perform tasks traditionally associated with masculinity (e.g. lifting heavy objects, repairing equipment). In addition, 'female' occupations, understood here as those with more than 55 per cent female employees, may not be female occupations at all, but rather gender-neutral ones (Hakim 1998) . This could explain why other studies which could more directly observe gender-typicality (Brooks et al., 2014; Kuwabara and Thébaud, 2017) found penalties for attractive persons, while we do not.

This brings us to the study's additional shortcomings. First, although interviewer ratings of physical attractiveness are commonly used (Kukkonen et al., 2024) , they are not without flaws. Existing literature demonstrates that high-status individuals are generally perceived as more attractive (Schunck, 2016; Bourdieu, 2018; Connell and Mears, 2018) . Thus, attractiveness may be confounded with (unobserved) characteristics, which impact both how attractive a person is perceived and their labour market outcomes. This problem might be particularly relevant when using interviewer ratings, where the attractiveness rating might be confounded by the respondents' habitus and surroundings. We addressed this concern by adjusting for relevant confounders and by estimating models accounting for rating subjectivity (interviewer random and fixed effects models). Despite these potential biases, it is worth noting that studies in which physical attractiveness is assessed by interviewers yield comparable results to studies in which physical attractiveness is measured using the consensus method (Schneickert, Steckermeier and Brand, 2020) .

Second, while longitudinal survey data enables us to test for beauty advantages and penalties in cases of job-person mismatches regarding gender stereotypes across occupations in the German labour market, our study is constrained by its reliance on observational data. Moreover, we could not estimate models with respondent fixed effects, since we had to treat attractiveness as a time-constant characteristic. Consequently, we cannot establish causal relationships, which would require an experimental design.

Third, importantly, this limitation of not being able to establish causal relationships is not resolved by employing a multiverse analysis. Although the multiverse analysis makes our data-analytic decisions transparent and provides information on how robust our findings are, it does not address potential misspecifications in the estimated models. Moreover, a multiverse analysis should not be used to 'outsource' analytical decisions. This would render the method ad absurdum. It could result in a large number of possibly nonsensical specifications, giving the impression that the result is a robust social fact, while in fact, it is just a robust artefact. When applied carefully, however, a multiverse analysis can enhance transparency and be a useful tool to assess the uncertainty of data analytic decisions.

Our study suggests several areas for future research. First, further investigation is required to examine the intersectionality of gender and physical attractiveness with other individual-level and context-level characteristics. Such research will provide a better understanding of the mechanisms driving social stratification and shed light on how gendered beauty stereotypes and beliefs contribute to labour market (dis)advantages. Second, it remains uncertain whether our findings on attractiveness premiums and penalties in the German labour market, as well as the lack of significant results on sex-typing of occupations and role congruency, can be generalized to other economies with greater employer discretion, lower dismissal costs, and less institutional regulation. Third, while we found no evidence of occupational sorting with respect to physical attractiveness and the gender-atypicality of an occupation, a more systematic examination of this phenomenon seems promising, in particular with more direct measures of gender-typicality in occupations and tasks. Fourth, it could also be promising to move beyond the binary framework of femininity and masculinity and take a closer look at how diverse gender identities intersect with specific stereotypes and biases. Understanding these associations could provide insights into their labour market implications and broader mechanisms of inequality.

Despite these open questions and the aforementioned limitations, this paper makes important contributions to the literature on physical attractiveness and labour market inequalities. It is the first study to examine how attractiveness and gender-atypicality interact in the German labour market and to use a multiverse analysis. Our findings reveal that physical attractiveness is robustly associated with higher earnings for both women and men in the German labour market. However, we did not find evidence that attractive individuals face earning penalties when employed in gender-atypical occupations. This challenges assumptions about the role of physical attractiveness in contexts of occupational gender incongruence and highlights the need for further research into the interplay between physical appearance, gender norms, and labour market outcomes.

Notes

1. These models align with (social) psychological models, such as social expectancy theories (Langlois et al., 2000) , which propose that cultural stereotypes can create their own self-fulfilling reality because those stereotyping and those stereotyped act according to these expectations. However, stereotypes and status characteristics are distinct concepts (Correll and Ridgeway, 2006, p. 32 ). 2. The direction of the association does not matter for the theoretical argument presented here. What matters is the assumption that these two attributes-attractiveness and gender-prototypicality-are cognitively connected. 3. We had to exclude the so-called DemoDiff subsample, as it does not contain an attractiveness rating in the first wave. 4. Data were retrieved through https://statistik.arbeitsagen- tur.de/Statistikdaten/Detail/201603/iiia6/beschaeftigung- sozbe-bo-heft/bo-heft-d-0-201603-xlsx.xlsx on September 22nd 2021. 5. Running the multiverse analysis took around 48 hours (Stata 17 MP on an Intel(R) Core(TM) i7-9700 CPU @ 3.00GHz with 32,0 GB). 6. The interviewer rating of physical attractiveness was only collected in waves 1 and 8 (and for the refreshment sample only in wave 11). Using it as a time-varying characteristic would first dramatically reduce the sample size to those respondents observed from wave 1 to 8, making the sample potentially rather selective. Second, the seven-year gap in between the first and second measurement is arguably too long for it to be used as a time-varying characteristic. Third, it could only be combined with net earnings since gross earnings are collected only every two years. Fourth, data on the percentage of opposite-gender employees in the occupations is available only from 2012, limiting us to use waves 8 and 12 to test hypothesis 2. Using it as a timevarying characteristic therefore appears infeasible. 7. https://osf.io/ug4xj/

References

1. Educational systems and labor market outcomes J Allmendinger European Sociological Review 5 1989
2. Beauty still matters: The role of attractiveness in labour market outcomes Petra Anýžová Petr Matějů 10.1177/0268580918760431 International Sociology International Sociology 0268-5809 1461-7242 33 3 2018 SAGE Publications
3. Information and Economic Behavior Kenneth J Arrow 10.21236/ad0768446 1973 Defense Technical Information Center Stockholm
4. Why Should Women Get Less? Evidence on the Gender Pay Gap from Multifactorial Survey Experiments Katrin Auspurg Thomas Hinz Carsten Sauer 10.1177/0003122416683393 American Sociological Review Am Sociol Rev 0003-1224 1939-8271 82 1 2017 SAGE Publications
5. Expectation states theory J Berger D G Wagner 10.1002/9781405165518.wbeose084.pub2 The Blackwell Encyclopedia of Sociology G Ritzer Hoboken John Wiley & Sons 2007
6. Status characteristics and social interaction: An expectation-states approach J Berger M H Fisek R Z Norman 1977 Elsevier New York
7. Physical Attractiveness E Berscheid E Walster Advances in Experimental Social Psychology L Berkowitz Academic Press 1974 7
8. Beauty, productivity, and discrimination: Lawyers' looks and lucre J E Biddle D S Hamermesh Journal of Labor Economics 16 1998
9. Men and women at work: Sex segregation and statistical discrimination W T Bielby J N Baron American Journal of Sociology 91 1986
10. Distinction: A social critique of the judgement of taste P Bourdieu Inequality Classic Readings in Race, Class, and Gender D Grusky New York Routledge 2018
11. Investors prefer entrepreneurial ventures pitched by attractive men A W Brooks Proceedings of the National Academy of Sciences of the United States of America 111 2014
12. GESIS Data Archive, Cologne. ZA5678 Data file Version J Brüderl 12.0.0 (2021a) The German Family Panel (pairfam)
13. Pairfam Data Manual, Release 12.0. LMU Munich: Technical Report. GESIS Data Archive, Cologne. ZA5678 Data File Version J Brüderl 12.0.0 2021b
14. Klassifikation der Berufe 2010-Band 1: Systematischer und alphabetischer Teil mit Erläuterungen Arbeit Bundesagentur Für Klassifikation der Berufe Definitorischer und beschreibender Teil 2011b. 2010 2
15. Egalitarianism and gender inequality M Charles D B Grusky The Inequality Reader D B Grusky S Szelényi New York Routledge 2018 nd ed.
16. The Oxford Handbook of Pierre Bourdieu C Connell A Mears Medvetz T. and Sallaz J. J. 2018 Oxford University Press New York Bourdieu and the body
17. Singletons, cluster-robust standard errors and fixed effects: A bad mix S Correia Duke University 2015 Technical Note
18. Fast Poisson estimation with high-dimensional fixed effects S Correia P Guimarães T Zylkin The Stata Journal: Promoting communications on statistics and Stata 20 2020
19. Expectation states theory S J Correll C L Ridgeway Delamater J. 2006 Springer
20. Gender composition of occupations and occupational characteristics: Explaining their true relationship by using longitudinal data Andreas Damelang Sabine Ebensperger 10.1016/j.ssresearch.2019.102394 Social Science Research Social Science Research 0049-089X 86 102394 2020 Elsevier BV
21. Now you see it, now you don’t: The vanishing beauty premium Tatyana Deryugina Olga Shurchkov 10.1016/j.jebo.2015.05.007 Journal of Economic Behavior & Organization Journal of Economic Behavior & Organization 0167-2681 116 2015 Elsevier BV
22. Another look at the effects of appearance, gender, and job type on performance-based decisions L M Drogosz P E Levy Psychology of Women Quarterly 20 1996
23. Schöne Schüler, schöne Noten?: Eine empirische Untersuchung zum Einfluss der physischen Attraktivität von Schülern auf die Notenvergabe durch das Lehrpersonal [Good Looks, Good Grades?: An Empirical Analysis of the Influence of Students' Physical Attractiveness on Grading by Teachers] I Dunkake Zeitschrift für Soziologie 41 2012
24. Physical attractiveness and the general factor of personality C S Dunkel Adaptive Human Behavior and Physiology 3 2017
25. What is beautiful is good, but…: A meta-analytic review of research on the physical attractiveness stereotype A H Eagly Psychological Bulletin 110 1991
26. Role congruity theory of prejudice toward female leaders A H Eagly S J Karau Psychological Review 109 2002
27. Using deepfakes for experiments in the social sciences-A pilot study A Eberl J Kühn T Wolbring Frontiers in Sociology 7 907199 2022
28. The Three Worlds of Welfare Capitalism G Esping-Andersen 1990 Princeton University Press Princeton
29. Beauty vs. brains: Early labor market outcomes of high school graduates J M Fletcher Economics Letters 105 2009
30. Attractiveness and income for men and women in management I H Frieze J E Olson J Russell Journal of Applied Social Psychology 21 1991
31. A Gelman J Hill Data Analysis Using Regression and Multilevel/Hierarchical Models Cambridge Cambridge University Press 2006
32. The statistical crisis in science: Data-dependent analysis-a 'garden of forking paths'explains why many statistically significant comparisons don't hold up A Gelman E Loken American Scientist 102 2014
33. Physical Attractiveness Barry Gillen 10.1177/014616728172015 Personality and Social Psychology Bulletin Pers Soc Psychol Bull 0146-1672 1552-7433 7 2 1981 SAGE Publications
34. Segregated and integrated occupations: A new approach to analysing social change C Hakim European Sociological Review 9 1993
35. Erotic Capital C Hakim 10.1093/esr/jcq014 European Sociological Review European Sociological Review 0266-7215 1468-2672 26 5 2010 Oxford University Press (OUP)
36. Honey Money: Why Attractiveness is the Key to Success C Hakim 2012 Penguin Books London
37. An introduction to varieties of capitalism P A Hall D Soskice Varieties of Capitalism. The Institutional Foundations of Comparative Advantage P A Hall D Soskice Oxford Oxford University Press 2001
38. Beauty Pays: Why Attractive People Are More Successful D S Hamermesh 2011 Princeton University Press Princeton
39. Beauty and the Labor market D S Hamermesh J E Biddle The American Economic Review 84 1994
40. The impact of situational factors on personnel decisions concerning women: Varying the sex composition of the applicant pool. Organizational Behavior and Human Performance M E Heilman 1980 26
41. Sex bias in work settings: The lack of fit model M E Heilman Research in Organizational Behavior 5 1983
42. When beauty is beastly: The effects of appearance and sex on evaluations of job applicants for managerial and nonmanagerial jobs. Organizational Behavior and Human Performance M E Heilman L R Saruwatari 1979 23
43. Attractiveness and corporate success: Different causal attributions for males and females M E Heilman M H Stopeck Journal of Applied Psychology 70 1985
44. Presumed incompetent: Perceived lack of fit and gender bias in recruitment and selection M E Heilman F Manzi S Braun Handbook of gendered careers in management: Getting in, getting on, getting out A M Broadbridge S L Fielden Cheltenham Edward Elgar Publishing 2015
45. The effects of physical attractiveness on job-related outcomes: a meta-analysis of experimental studies M Hosoda E F Stone-Romero G Coats Personnel Psychology 56 2003
46. Physical attractiveness and intellectual competence: a meta-analytic review L A Jackson J E Hunter C N Hodge Social Psychology Quarterly 58 1995
47. A thing of beauty is a joy forever?' Returns to physical attractiveness over the life course M M Jaeger Social Forces 89 2011
48. Physical Attractiveness Biases in Ratings of Employment Suitability: Tracking Down the “Beauty is Beastly” Effect Stefanie K Johnson Kenneth E Podratz Robert L Dipboye Ellie Gibbons 10.1080/00224540903365414 The Journal of Social Psychology The Journal of Social Psychology 0022-4545 1940-1183 150 3 2010 Informa UK Limited
49. Does it pay to be smart, attractive, or confident (or all three)? Relationships among general mental ability, physical attractiveness, core self-evaluations, and income T A Judge C Hurst L S Simon The Journal of Applied Psychology 94 2009
50. Is beauty-based inequality gendered? A systematic review of gender differences in socioeconomic outcomes of physical attractiveness in labor markets I Kukkonen 2024 European Societies 26
51. When Beauty Doesn't Pay: Gender and Beauty Biases in a Peer-to-Peer Loan Market Ko Kuwabara Sarah Thébaud 10.1093/sf/sox020 Social Forces 0037-7732 1534-7605 95 4 2017 Oxford University Press (OUP)
52. Infant attractiveness predicts maternal behaviors and attitudes J H Langlois Developmental Psychology 31 1995
53. Maxims or myths of beauty? A meta-analytic and theoretical review J H Langlois Psychological Bulletin 126 2000
54. When beauty helps and when it hurts: An organizational context model of attractiveness discrimination in selection decisions S Lee Organizational Behavior and Human Decision Processes 128 2015
55. Auswahl von Männern und Frauen als Führungskräfte. Perspektiven aus WIrtschaft, Wissenschaft, Medien und Politik G Lotzkat I M Welpe Welpe, I. M., Brosi, P., Ritzenhöfer, L. and Schwarzmüller, T. 2015 Springer Gibt es Geschlechtsstereotype in der Wahrnehmung von Berufsgruppen?
56. Schooling, Experience, and Earnings J Mincer Human Behavior & Social Institutions 2 1974 National Bureau of Economic Research, Inc New York
57. Why beauty matters M M Mobius T S Rosenblat American Economic Review 96 2006
58. Beholding Inequality: Race, Gender, and Returns to Physical Attractiveness in the United States Ellis P Monk Michael H Esposito Hedwig Lee 10.1086/715141 American Journal of Sociology American Journal of Sociology 0002-9602 1537-5390 127 1 2021 University of Chicago Press
59. The statistical theory of racism and sexism E S Phelps The American Economic Review 62 1972
60. The social construction of status value: gender and other nominal characteristics C Ridgeway Social Forces 70 1991
61. Framed by gender: how gender inequality persists in the modern world. Framed by Gender: How Gender Inequality Persists in the Modern World C Ridgeway 2011
62. Unpacking the gender system: a theoretical perspective on gender beliefs and social relations C Ridgeway S J Correll Gender & Society 18 2004
63. Assessing the robustness of mediation analysis results using multiverse analysis J J M Rijnhart Prevention Science : The Official Journal of the Society for Prevention Research 23 2022
64. Inference and missing data D B Rubin Biometrika 63 1976
65. Exploring the impact of male and female facial attractiveness on occupational prestige E Sala Research in Social Stratification and Mobility 31 2013
66. Overadjustment bias and unnecessary adjustment in epidemiologic studies E F Schisterman S R Cole R W Platt Epidemiology 20 2009 Cambridge, Mass
67. Lonely, poor, and ugly? How cultural practices and forms of capital relate to physical unattractiveness C Schneickert L C Steckermeier L. -M Brand Cultural Sociology 14 2020
68. Reich und schön? Eine Untersuchung zur ungleichen Verteilung physischer Attraktivität Reinhard Schunck 10.1007/s11577-015-0354-1 KZfSS Kölner Zeitschrift für Soziologie und Sozialpsychologie Köln Z Soziol 0023-2653 1861-891X 68 1 2016 Springer Science and Business Media LLC
69. Specification curve analysis U Simonsohn J P Simmons L D Nelson Nature Human Behaviour 4 2020
70. Increasing Transparency Through a Multiverse Analysis Sara Steegen Francis Tuerlinckx Andrew Gelman Wolf Vanpaemel 10.1177/1745691616658637 Perspectives on Psychological Science Perspect Psychol Sci 1745-6916 1745-6924 11 5 2016 SAGE Publications
71. Facial attractiveness R Thornhill S W Gangestad Trends in Cognitive Sciences 3 1999
72. Beauty as status M Webster J E Driskell American Journal of Sociology 89 1983
73. How beauty works. Theoretical mechanisms and two empirical applications on students' evaluation of teaching T Wolbring P Riordan Social Science Research 57 2016
74. Gender and the returns to attractiveness J S Wong A M Penner Research in Social Stratification and Mobility 44 2016
75. Model uncertainty and the crisis in science C Young Socius 4 2378023117737206 2018
76. Model Uncertainty and Robustness Cristobal Young Katherine Holsteen 10.1177/0049124115610347 Sociological Methods & Research Sociological Methods & Research 0049-1241 1552-8294 46 1 2017 SAGE Publications