Demographic “Stickiness”: The Demographic Identity of Departing Group Members Influences Who Is Chosen to Replace Them

Introduction

In 2020, former President Trump nominated Amy Coney Barrett to the U.S. Supreme Court (Kim and Itkowitz 2020). Prior to her nomination, it was reported that the entire short list of candidates under consideration for the seat were women; in fact, Donald Trump committed publicly to nominating a woman for the position (Wagner et al. 2020). His commitment to choosing a woman may have been surprising given that his first two Supreme Court nominees, Neil Gorsuch and Brett Kavanaugh, were White men (Blake 2020). Trump was also not known for embracing diversity: Concurrently with Barrett’s nomination, Trump issued an executive order banning federal contractors from administering many sorts of diversity training, and the White House issued a memo asking federal agencies to halt diversity trainings as they were “un-American” (Torbati et al. 2020).

Given this apparent distaste for diversity and a track record of nominating White men to the Supreme Court, what might explain Trump’s decision to nominate Amy Coney Barrett? We suggest that the demographic identities of the members of the Supreme Court being replaced may help elucidate Trump’s decisions about whom to nominate. Amy Coney Barrett was nominated to fill the seat vacated by Ruth Bader Ginsburg, a woman, whereas Brett Kavanaugh and Neil Gorsuch were both nominated to replace White men (Anthony Kennedy and Antonin Scalia, respectively).

We propose that the decision about whom to nominate for an open seat may be influenced by the demographic identities (i.e., the gender or race) of their predecessors. There are two primary reasons why we believe this may be the case, both stemming from loss aversion. First, past research has documented that people tend to maintain the existing state of the world through inaction or perpetuation of previously made decisions (Samuelson and Zeckhauser 1988, Kahneman et al. 1991). Although people cannot perfectly maintain the existing state of the world when replacing a departing group member, they may seek to avoid changes (i.e., they may be “impact averse”) because the potential losses from changing a group’s demographic composition loom larger than potential gains. Decision makers may thus be disproportionately likely to choose new group members who demographically resemble the group members they are replacing to maximize the perceived similarity between departing group members and their replacements and preserve the previous demographic composition of the group.

Second, people may exhibit a form of “diversity loss aversion”: People may view losing ground on demographic diversity to be more aversive than equivalent gains on diversity are positive. This focus on demographic diversity as a source of gains or losses may stem from beliefs that diversity can be valuable (e.g., by yielding reputational benefits or by increasing creativity; Phillips 2014, Chang et al. 2019a). As a result, people may be more motivated to avoid losses on demographic diversity than they are motivated to make progress on diversity. For example, when a non-White man (compared with a White man) leaves a group of mostly White men, decision makers may be especially likely to try to select a non-White man to fill that seat so as not to incur a diversity loss.¹

These two accounts make very similar predictions—that the demographic composition of a group or organization will be “sticky” when making replacement hiring decisions—and they are generally not mutually exclusive. However, they do make some divergent predictions. The impact aversion account predicts that regardless of who is departing a group—including when majority group members leave (i.e., White men in many U.S. contexts)—people should be more likely to select someone of that same demographic identity to preserve the previous demographic composition of the group. The diversity loss aversion account does not make this prediction when majority group members leave a group, as there would not be a diversity “loss.” In addition, the diversity loss aversion account does not make a strong prediction that decision makers will choose someone of the exact same demographic identity when a minority group member leaves a group. For example, whereas the impact aversion account predicts that a decision maker would be especially likely to select an Asian person—and not a Black or Hispanic person—to replace an Asian person leaving a group of mostly White men, the diversity loss aversion account might predict that a decision maker might also be more likely to select a Black or Hispanic person when replacing an Asian person, as a Black or Hispanic person would also help mitigate the “loss” of diversity when the Asian person left the group.

We examine demographic stickiness in two field studies and four experiments. In Study 1, we consider appointments of U.S. federal judges from 1945 to 2020. We show that when filling open seats on a court, presidents are disproportionately likely to appoint a judge whose demographic identity matches that of the judge who previously held the seat. For example, presidents are much more likely to nominate a woman to a seat last held by a woman as opposed to one last held by a man. Similar effects hold when examining racial identity. In Study 2, we replicate this finding in another organizational context: directors added to S&P 1500 corporate boards from 2014 to 2019. We find that the demographic identities of departing directors significantly influence the demographic identities of newly added directors, such that a new addition is disproportionately likely to share the demographic identity of a director who recently left the board. In both field studies, the demographic stickiness effect is specific to demographic categories, which is more consistent with an impact aversion account than a diversity loss aversion account.

In Studies 3 and 4, we provide experimental evidence of these effects. We randomize either the race or gender of a departing group member and find that the demographic identity of the departing group member has a significant effect on whom participants choose to replace that group member. Study 4 also shows a significant indirect effect of desires to minimize changes relative to the previous world—in other words, impact aversion—on these decisions. In Study 5, we do not find evidence of demographic stickiness in participants’ willingness to select a White man to replace a White man relative to a departing group member whose identity is unknown. In Study 6, we highlight the myopic nature of demographic stickiness. Studies 5 and 6 provide evidence more consistent with a diversity loss aversion account than an impact aversion account, suggesting that the psychology underlying demographic stickiness may differ between participants making hypothetical choices in online experiments and decision makers making real selections in the world.

This research builds our understanding of diversity-related replacement hiring and selection decisions in organizations, elucidating a factor that influences people’s willingness to select candidates with historically marginalized identities to join a group. We show that demographic composition in groups is “sticky” because the demographic identities of departing members influence the demographic identities of their chosen replacements. To be clear, we do not mean to suggest that this is the only factor that influences diversity-related hiring decisions—copious past research shows that factors like bias, discrimination, and homophily all contribute to the underrepresentation of women and racial minorities in many contexts—and our results are limited to replacement hiring decisions. However, our results suggest that human decision-making tendencies, separate from the biases or stereotypes people may apply to individuals from different demographic groups, can also influence diversity-related selection decisions in organizations.

Our work also has implications for interventions that aim to reduce inequality in organizations. Our results suggest that progressive leaders or one-time interventions that increase diversity in the short-term may have longer lasting effects than might otherwise be expected. Of course, we do not mean to suggest that organizational leaders should focus only on increasing diversity in the short term without focusing on issues around the inclusion, retention, and promotion of women and racial minorities. However, past research indicates that combating discrimination is difficult because changing people’s hearts and minds or creating systems that will maintain equality in perpetuity is challenging (Acker 2006, Lai et al. 2016, Chang et al. 2019b, Ray 2019). Meanwhile, one-time actions that change the demographic composition of a group or organization might produce progress toward demographic diversity–and our work suggests that there will be less backsliding on this progress than might otherwise be expected.

Theory and Hypotheses

Loss aversion refers to people’s tendency to view losses as more aversive than equivalent gains are positive (Kahneman and Tversky 1979). In general, loss aversion means that people are more motivated to avoid losses than they are to achieve equivalent gains. For example, people may work harder to avoid a $5 loss than to earn a $5 gain. How might loss aversion play out in replacement hiring? We propose two primary ways in which loss aversion might lead to demographic stickiness in replacement hiring decisions.

First, loss aversion has been posited to help explain people’s desires to maintain the status quo or existing state of the world, as the potential downsides of making a change feel more salient and severe than the potential upsides (Kahneman et al. 1991). However, what happens when people cannot actually maintain the status quo because they are forced to make a decision that inherently involves change (e.g., when they must hire someone to replace a departing group member)? We suggest that even when people cannot maintain the status quo, they may still prefer to make decisions that minimize changes relative to the previous state of the world. In other words, they may be “impact averse.”

If people have a tendency to minimize changes relative to the previous state of the world, then decision makers may be particularly attracted to replacement hires that allow them to maintain core or salient features of the previous state, as decisions that change core dimensions are more likely to feel like significant deviations (Masatlioglu and Ok 2014). Given that race and gender are often primary attributes for social categorization (Stangor et al. 1992), race and gender may be features that decision makers prioritize when seeking to minimize changes. Furthermore, people rely on demographic characteristics like gender or race as a way to make inferences about others’ internal characteristics, like personality or skills (Fiske and Neuberg 1990, Cuddy et al. 2008, Czopp et al. 2015). Because people tend to hold essentialist beliefs that people who look the same (i.e., resemble each other demographically) have the same internal qualities (Rothbart and Taylor 1992, Bastian and Haslam 2006), demographic identity may be used as a proxy for underlying similarity between departing group members and potential replacements. By choosing someone who demographically resembles a departing group member, people may implicitly be assuming that they have chosen someone who will fulfill a similar function in the group in terms of skills, perspectives, or behavior, thus minimizing changes to the group.

Second, loss aversion may influence replacement hiring decisions more directly if people are “diversity loss averse.” Diversity is an inherent property of groups (Harrison and Klein 2007), and past research has shown that people can and do spontaneously evaluate the demographic diversity of a group (Phillips et al. 2018). Decision makers may want to maintain a particular level of diversity in a group because they believe that diversity can aid in group functioning or performance (Tsui and O’Reilly 1989, Williams and O’Reilly 1998, Phillips and Loyd 2006). They may also be worried about how it will look to others if they lose ground on diversity (Chang et al. 2019a). Because diversity is a primary way in which people perceive and understand groups and can be viewed as a valuable group characteristic, people may attend to potential diversity “losses” and “gains” created by replacement hiring decisions. If losses on diversity are more aversive than corresponding diversity gains are desirable (Tversky and Kahneman 1991), then decision makers may be more motivated to choose people whose identity is underrepresented in the group when replacing group members with underrepresented identities (to avoid losses on diversity) than when replacing group members with overrepresented identities (to achieve gains on diversity).

In general, impact aversion and diversity loss aversion make similar predictions. However, impact aversion suggests decision makers will be narrowly focused on maintaining similarity between departing group members and their replacements, while diversity loss aversion suggests they will be broadly focused on maintaining ground on diversity. In environments dominated by White men, decision makers primarily worried about avoiding diversity losses when making replacement decisions may be satisfied to select any non-White man to replace other non-White men. For example, if an Asian person were to leave a majority-White male group, a decision maker focused solely on avoiding diversity losses may prefer to select an Asian person as a replacement, but they may also be satisfied to choose a White woman, Black person, or Hispanic person as a replacement, given that people from any of these demographic groups would also add to the diversity of the group. Thus, a diversity loss aversion account might predict that a woman leaving a group would lead to an increased likelihood of both a woman or a racial minority being added to the group, and a racial minority leaving a group would lead to an increased likelihood of both a racial minority or a woman being added to the group; or that when an Asian person leaves a group, decision makers might be more likely to select a Black or Hispanic person as a replacement, in addition to being more likely to select an Asian person.²

On the other hand, people trying to maximize similarity to a departing group member to avoid change-related losses or risks should only be increasingly likely to choose racial minorities (and not women) to replace racial minorities and should only be increasingly likely to choose women (and not racial minorities) to replace women. Past research has shown that people perceive and stereotype members of different demographic groups differently. For example, the stereotypes that White women face are not the same stereotypes that racial minorities face (Rudman and Glick 2001, Cuddy et al. 2008, Rosette et al. 2008), and people view White women and racial minorities as making distinct contributions to groups (Unzueta and Binning 2012, Geerling and Chen 2020). Similarly, people from different racial categories are perceived and stereotyped differently (Rosette et al. 2008, Berdahl and Min 2012, Czopp et al. 2015), and people’s judgments of how individuals contribute to teams vary for members of different racial groups (Unzueta and Binning 2010, 2012; Danbold and Unzueta 2020). As a result, if people were trying to reduce differences between new group members and their predecessors following an impact aversion account, people would not find replacing a White woman with a racial minority (and vice versa) to be an attractive option; similarly, people would not want to replace an Asian person with a Black or Hispanic person; and so on for other racial categories.

In addition, an impact aversion account predicts that decision makers should be more likely to select a White man when replacing a White man than when replacing someone who is not a White man. Indeed, an impact aversion account predicts that the increase in likelihood of selecting a White man to replace a White man vs. a non-White man should be roughly equal to the increase in likelihood of selecting a Black person to replace a Black person versus a non-Black person or the increased likelihood of selecting a woman to replace a woman versus man. A diversity loss aversion account would predict that the increase in likelihood should be higher when replacing a minority group member as compared with when replacing a White man, given that a departing minority represents a potential diversity loss, whereas a departing White man does not.

Finally, it is important to note that both accounts reflect a myopic approach to diversity. Although most extant theorizing on diversity preferences suggests that decision makers focus on the end state (current group-level diversity), our theorizing suggests that they instead attend myopically to the demographic identity of the departing group member, even though this person no longer affects group composition. For example, imagine two groups with identical preferences for gender diversity: one that was composed of four men and one woman until a man left the group, and the other that was composed of three men and two women until a woman left the group. Both groups now include three men and one woman, so extant theories of diversity would predict that they should be equally likely to select a woman as their next hire given their equivalent gender diversity preferences. However, demographic stickiness predicts that the latter group will be more likely to select a woman as their next hire because decision makers will attend myopically to the fact that the group just lost a woman, whereas decision makers for the former group will be more likely to select a man because they will attend to the fact that the group just lost a man. In other words, rather than focusing on achieving a target level of diversity, people focus narrowly on the demographic identity of departing group members when making replacement hires.

Overview of Studies

We examine these predictions in two field settings and four experiments. In Study 1, we present analyses of all U.S. federal judges confirmed between 1945 and 2020 to test whether the demographic identity of a judge who previously held a seat on a court influences who is chosen to replace them. We also test whether these effects are specific to demographic category (as would be predicted by an impact aversion account) or whether we find cross-category effects (which would be more consistent with a diversity loss aversion account). In Study 2, we replicate these analyses in another field setting: S&P 1500 corporate board directors appointed between 2014 and 2019. In Study 3, we present results from an experiment where we manipulate the race of a departing group member and show that people are more likely to choose Black men to replace departing Black men relative to departing White men. In Study 4, we find a significant indirect effect of desires to minimize changes relative to the previous state of the group on these decisions. In Study 5, we present an experiment that shows evidence more consistent with a diversity loss aversion mechanism than an impact aversion mechanism, given that participants are equally likely to hire a White man when told either a White man or someone whose identity is not revealed left a group. Finally, in Study 6, we highlight the myopic nature of demographic stickiness effects.

Study 1: U.S. Federal Judges

In Study 1, we examined demographic stickiness in U.S. federal judicial appointments from 1945 to 2020. Federal judicial appointments are an ideal organizational context to test our hypotheses because judges are nominated and confirmed to specific seats on courts, so we know exactly who was selected to replace whom.

Methods

Data and Sample.

Our data come from the Federal Judicial Center (FJC). The FJC is the research and education agency of the judicial branch of the U.S. government. The FJC maintains data on all federal judges, including their gender and race. Federal judges in the United States are nominated by the president and confirmed by the Senate. Given our interest in demographic diversity, we restricted the data to all judges confirmed starting in 1945, the year of the first confirmation of a non-White federal judge (1928 was the year of the first confirmation of a female federal judge, and all results related to gender are robust to including data starting from 1928).

Judges are confirmed to specific seats on courts. As a result, we know exactly which seat each judge is selected for, and we know the demographic identity of the judge’s predecessor (i.e., the person who previously held the judge’s seat on the court). For the most part, the number of seats on each court is fixed, though the number of seats on a given court can change based on acts of Congress. Because we were interested in whether the demographic identity of a departing judge influenced the demographic identity of their replacement, we only analyzed appointments where there was a predecessor in the seat. Thus, we did not include judges who were appointed to newly created seats by Congress. This provided a sample of 2,163 judicial confirmations over 75 years of U.S. history. See Table 1 for summary statistics about the data set and Table 2 for a correlation matrix of variables in the data set.

Table 1. Summary Statistics of U.S. Federal Judiciary Data

Table 2. Correlation Matrix for U.S. Federal Judiciary Data

Dependent Variables.

The primary dependent variables of interest were indicators for whether a newly confirmed judge was a woman (one if a woman, zero if not a woman) or a racial minority (one if not White, zero if White; anyone who was not designated as only “White” by the FJC was coded as not White). In additional analyses, we considered racial categories separately (e.g., one if Black, zero if not Black; one if Hispanic, zero if not Hispanic) and the intersection of gender and race (e.g., one if a Black woman, zero if not a Black woman).

Independent Variables.

Our primary independent variables were indicators for whether the judge who previously held the seat was a woman (one if a woman, zero if not a woman) or a racial minority (one if not White, zero if White; anyone who was not designated as only “White” by the FJC was coded as not White). In additional analyses, we considered racial categories separately (e.g., one if Black, zero if not Black; one if Hispanic, zero if not Hispanic; one if Asian, zero if not Asian).

Control Variables.

Because political parties have different coalitions and because the judiciary has become more diverse over time, we included fixed effects for nominating president. Presidents serve at different points in time, so fixed effects for president inherently control for time, and fixed effects for president allow us to partial out variation due to individual differences in presidential nominating patterns (e.g., due to political party; due to individual levels of bias). We also included fixed effects for the political party of the predecessor judge’s nominating president, the party in control of the Senate, interactions between the political party of the nominating president and the party in control of the Senate, and continuous controls for the racial demographics of the court’s jurisdiction (e.g., the percent of the population that was White, Black, Hispanic, and Asian) taken from decennial U.S. Census data (we matched the confirmation year to the closest decade’s Census data; e.g., we used 2010 Census data for confirmations from 2006 to 2015). These additional controls allow us to account for the effects of political coalitions and pressures to create courts that demographically resemble the population they serve. We added fixed effects for court type (Supreme Court, courts of appeals, district courts, and other) given different levels of scrutiny and visibility of different courts (e.g., the Supreme Court is more visible and scrutinized than district courts) and that past research has suggested that scrutiny and visibility can influence diversity-related selection decisions (Chang et al. 2019a). We included fixed effects for the number of women and number of racial minorities already on the court to show that our effects persist even when controlling for the diversity of the group and to highlight the myopic nature of the effects we document.

Analysis Strategy.

We ran ordinary least squares (OLS) regressions with robust standard errors and errors clustered by nominating U.S. president to predict the demographic identities of confirmed judges. Our primary predictor variable was an indicator for whether the confirmed judge’s predecessor was a woman (for our gender analyses) or was non-White (for our racial analyses). In additional analyses, we included both predictor variables or indicators for whether the confirmed judge’s predecessor was a member of each racial category subgroup in the sample (i.e., whether the predecessor was Black, Hispanic, or Asian).

Using robust standard errors in our regressions corrects for the heteroskedasticity resulting from using OLS regressions to predict binary outcomes (Angrist and Pischke 2008) and clustering standard errors by president accounts for serial correlation of decisions. Analyses using logistic regression instead of OLS are identical in terms of statistical significance and are reported in the online supplement. We also include specification curves in the online supplement to show the robustness of our findings to alternative specifications and analysis decisions. We find statistically significant evidence of demographic stickiness for nearly all regression specifications for race and most specifications for gender, demonstrating the robustness of the effects we document.

Results

Do New Judges Demographically Resemble the Judges They Replace?

Descriptively, when a man previously held a seat, new judges were women 16.15% of the time; when a woman previously held a seat, new judges were women 36.60% of the time, z = 7.08, p < 0.001. When we predicted whether a new judge was a woman using our OLS regression analysis strategy, the coefficient on the predictor—an indicator for whether the judge’s predecessor was a woman—was positive and significant, b = 0.089, p = 0.031, 95% confidence interval (CI): [0.00940, 0.169], see Table 3, Model 1. These results suggest that the gender of the preceding judge had a significant effect on the gender of the judge selected to replace them. In other words, male judges are more likely to succeed male predecessors than female predecessors, and female judges are more likely to succeed female predecessors than male predecessors.

Table 3. U.S. Federal Judge Appointments Show Evidence of Demographic Stickiness

Turning our attention to the likelihood of selecting racial minority judges, we found similar results. Descriptively, when a White person previously held a seat, new judges were racial minorities 12.92% of the time; when a racial minority previously held a seat, new judges were racial minorities 35.45% of the time, z = 8.31, p < 0.001. When we predicted whether a new judge was a racial minority using our OLS regression strategy, the coefficient on the predictor—an indicator for whether the judge’s predecessor was a racial minority—was positive and significant, b = 0.117, p = 0.011, 95% CI: [0.0316, 0.202], see Table 3, Model 3. These results suggest that the racial identity of the preceding judge had a significant effect on the racial identity of the judge that was selected to replace them. In other words, White judges are more likely to succeed White predecessors than racial minority predecessors, and racial minority judges are more likely to succeed racial minority predecessors than White predecessors.

Is This About Demographic Resemblance to Departing Group Members or Just Choosing Judges Who Are Not White Men?

To test this, we ran our OLS regressions with fixed effects for court type and nominating president and included two separate predictors: an indicator for whether the judge’s predecessor was a woman and an indicator for whether the judge’s predecessor was a racial minority. Using both indicators to predict whether a newly confirmed judge was a woman, we found that only the coefficient for the predecessor being a woman was significant, b = 0.089, p = 0.030, 95% CI: [0.0101, 0.169], while the coefficient for the predecessor being a racial minority was not, b = 0.018, p = 0.602, 95% CI: [−0.0542, 0.0898], see Table 3, Model 2. On the other hand, when we used both indicators to predict whether a new judge was a racial minority, we found that only the coefficient for the predecessor being a racial minority was significant, b = 0.117, p = 0.011, 95% CI: [0.0316, 0.202], whereas the coefficient for the predecessor being a woman was not, b = −0.0094, p = 0.737, 95% CI: [−0.0683, 0.0496], see Table 3, Model 4.

We also examined racial minority groups separately. We ran four separate regressions predicting whether new judges were (1) Black, (2) Hispanic, (3) Asian, or (4) women. In each regression, we included as predictors an indicator for whether the judge’s predecessor was Black, an indicator for whether the predecessor was Hispanic, an indicator for whether the predecessor was Asian, and an indicator for whether the predecessor was a woman. These results are depicted in Figure 1. We found that the demographic identities of new judges were typically only positively predicted by whether the preceding judge had the same demographic identity. For example, when predicting whether a new judge was Black, the only significant positive predictor was whether the preceding judge was Black, b = 0.119, p = 0.003, 95% CI: [0.0488, 0.190], while the coefficients on the preceding judge being Hispanic, b = −0.025, p = 0.030, 95% CI: [−0.0473, 0.00290]; being Asian, b = −0.065, p = 0.018, 95% CI: [−0.117, −0.0133]; or being a woman, b = 0.027, p = 0.094, 95% CI: [−0.00523, 0.0583], were not significant positive predictors, see Table 4, Model 1.

Table 4. Examining Race-Specific Demographic Stickiness in U.S. Federal Judiciary Data

Finally, although we did not have ex ante theorizing about women of color, we also examined how demographic characteristics of predecessor judges influenced whether women of color were selected as new judges.³ These results are depicted in Table 5. Overall, we found that people’s behavior was consistent with theorizing on intersectionality, suggesting that women of color are not seen as prototypical of women, as women of color appeared to be categorized and treated primarily as people of color rather than as women (Purdie-Vaughns and Eibach 2008, Sanchez-Hucles and Davis 2010, Rosette et al. 2018). For example, when predicting whether a new judge was a Hispanic woman, the only significant positive predictor was whether the preceding judge was Hispanic, b = 0.060, p = 0.018, 95% CI: [0.0123, 0.107], while whether the preceding judge was Black, b = −0.017, p = 0.059, 95% CI: [−0.0345, 0.000748]; was Asian, b = −0.0037, p = 0.734, 95% CI: [−0.0269, 0.0194]; or was a woman, b = −0.014, p = 0.074, 95% CI: [−0.0287, 0.00153], were not significant positive predictors, see Table 5, Model 3.

Table 5. Examining Demographic Stickiness for Women of Color in U.S. Federal Judiciary Data

Discussion

Across 2,163 U.S. federal judge appointments over 75 years of data, Study 1 shows that there are significant effects of a departing judge’s demographic identity on who is subsequently chosen to replace them. New judges are disproportionately likely to resemble the judges who preceded them both in terms of gender identity and racial identity. This effect holds for majority group members as well as minority group members. Although we coded the data such that White men were the baseline in the analyses, we could have chosen any demographic group to be the baseline, which would show that presidents are more likely to select men when replacing men and more likely to select White people when replacing White people. In addition, these effects are specific to demographic categories: A departing woman does not predict a higher likelihood of selecting a racial minority and vice versa, and a departing Black judge does not predict a higher likelihood of selecting a Hispanic or Asian judge and vice versa. This suggests that our effects are not simply due to decision makers striving to maintain a particular ratio of White males to non-White males, as a diversity loss aversion mechanism might predict, because any non-White male would also contribute to the diversity of the group. Rather, decision makers appear focused on maintaining similarity between departing judges and their replacements, consistent with an impact aversion account. We also do not find evidence that our effects are moderated by the visibility of the court (see online supplement).

Decision makers selecting federal judges do not seem to be more motivated to replace minority group members with other minority group members than they are to replace White men with other White men, as would be predicted by a diversity loss aversion account. For example, a newly appointed judge is a White man 76.6% of the time when the judge being replaced is a White man and 45.0% of the time when the judge being replaced is not a White man, so replacing a White man is linked to a 31.6-percentage-point increase in the likelihood of selecting a White man. On the other hand, a newly appointed judge is Black 24.8% of the time when a Black judge is being replaced and 7.4% of the time otherwise, so replacing a Black judge is tied to a 17.4-percentage-point increase in the likelihood of selecting a Black judge. Thus, decision makers do not seem to exhibit demographic stickiness more in absolute terms when replacing minority group members than when replacing majority group members.

Interestingly, we found that presidents were more likely to select women of color to replace racial minority judges than White judges, but they were not more likely to select women of color to replace women than men. In other words, women of color tended to be treated as representatives of their racial identity group rather than as representatives of their gender identity. We also found that even when replacing women and racial minorities, presidents were still most likely to appoint men and White people to the federal judiciary: presidents selected men 63.4% of the time when replacing women, and they selected White people 64.6% of the time when replacing racial minorities. These results underscore that despite demographic stickiness, progress toward diversification is likely to remain slow in the absence of other interventions.

Study 1 provides empirical evidence of our theoretical claims in an organizational context of importance. However, these results are ultimately correlational, so we cannot make causal claims, and they represent only a single organizational context. In the subsequent studies, we aimed to replicate these findings in another organizational context and test our theorizing in experiments to provide more confidence that there is a causal relationship between the demographic identities of departing group members and their chosen replacements.

Study 2: S&P 1500 Corporate Boards

In Study 2, we examined our predictions in the context of directors added to S&P 1500 corporate boards between 2014 and 2019. This study allows us to see whether our demographic stickiness effects replicate in a different organizational context where all of the data are more recent.

Methods

Data and Sample.

Our data come from Institutional Shareholder Services (ISS). ISS has data on the board members of S&P 1500 companies, including each director’s name, gender, and ethnicity. We examined S&P 1500 boards from 2013 to 2019, so we analyzed new director additions to S&P 1500 boards for the years 2014, 2015, 2016, 2017, 2018, and 2019.

Because boards are not of fixed size and because directors do not fill specific seats on boards in the same way as U.S. federal judges fill specific seats on courts, our data are less granular than in Study 1. In particular, we can only observe whether a board lost directors of a given demographic identity immediately prior to adding new directors, as opposed to knowing precisely who replaced which director. Thus, we examined whether the demographic identities of the directors who leave a board in a given year influence the demographic identities of directors added in the subsequent year. Because of our interest in examining how recently departed group members influence the demographic identity of group members chosen to replace them, we only analyzed new director additions when the board lost directors in the previous year (i.e., we excluded selection decisions made when the board had not lost anyone in the year immediately prior). This provided us a sample of 5,616 additions to S&P 1500 corporate boards in the time period we analyzed. See Table 6 for summary statistics about the data and Table 7 for a correlation matrix of variables in the data.

Table 6. Summary Statistics for S&P 1500 Corporate Board Data

Table 7. Correlation Matrix for S&P 1500 Corporate Board Data

Dependent Variables.

The primary dependent variables of interest were indicators for whether a newly added director was a woman (one if a woman, zero if not a woman) or a racial minority (one if not White, zero if White; any director not classified as “White” or “Caucasian” was considered not White). In additional analyses, we considered racial categories separately (e.g., one if Black, zero if not Black; one if Hispanic, zero if not Hispanic), as well as the intersection of gender and race (e.g., one if a Black woman, zero if not a Black woman).

Independent Variables.

Our primary independent variables were indicators for whether a board lost any directors in the preceding year who were women (one if the board lost women, zero if the board did not lose women) or racial minorities (one if the board lost racial minorities, zero if the board did not lose racial minorities). In additional analyses, we also considered racial categories separately (e.g., one if the board lost Black directors, zero if the board did not lose Black directors).

Control Variables.

We included fixed effects for the firm to partial out variation at the firm level (e.g., industry, firm size, profitability). We also included fixed effects for year to partial out variation due to time trends in diversification of corporate boards. We included fixed effects for the size of the board to account for the fact that larger boards have more opportunities to add directors. Finally, we included fixed effects for the number of women already on the board and the number of racial minorities on the board to account for concerns that effects could be driven by diversity thresholds (Chang et al. 2019b) to show that our effects persist even when controlling for the existing diversity of the group, and to highlight the myopic nature of the effects we document.

Analysis Strategy.

We ran OLS regressions with robust standard errors and errors clustered by firm to predict the demographic identities of new directors. Our primary predictor variable was an indicator for whether the board lost women in the preceding year (for our gender analyses) or lost non-White directors in the preceding year (for our racial analyses). In additional analyses, we include both the aforementioned predictor variables. In analyses focusing on specific racial groups, we included multiple predictor variables: indicators for whether the board lost Black directors, Hispanic directors, Asian directors, and women in the preceding year.

Using robust standard errors in our regressions corrects for the heteroskedasticity resulting from using OLS regressions to predict binary outcomes (Angrist and Pischke 2008) and clustering standard errors by firm accounts for serial correlation of decisions. Analyses using logistic regression instead of OLS are identical in terms of statistical significance and are reported in the Online Supplement for robustness. We also include specification curves in the Online Supplement to show the robustness of our findings to alternative specifications and analysis decisions. We find statistically significant evidence of demographic stickiness for all specifications for gender and for nearly all specifications for race, suggesting that the results we document are robust.

Results

Do New Directors Demographically Resemble Directors Who Recently Left the Board?

Descriptively, when a board had not lost any women in the previous year, new directors were women 25.98% of the time; when a board had lost at least one woman in the previous year, new directors were women 35.31% of the time, z = 6.79, p < 0.001. When we predicted whether a new director was a woman using our OLS regression analysis strategy, we estimated that the coefficient on the predictor of whether the board had lost women was positive and significant, b = 0.198, p < 0.001, 95% CI: [0.158, 0.239], see Table 8, Model 1. These results suggest that the gender of departing directors had a significant effect on the gender of newly added directors.

Table 8. S&P 1500 Director Selections Show Evidence of Demographic Stickiness

When examining the likelihood of selecting racial minority directors, we found similar results. Descriptively, when a board had not lost any racial minorities in the previous year, new directors were racial minorities 16.85% of the time; when a board had lost at least one racial minority in the previous year, new directors were racial minorities 24.58% of the time, z = 5.88, p < 0.001. When we predicted whether a new director was a racial minority using our OLS regression strategy, we estimated that the coefficient on the predictor of whether the board had lost racial minorities was positive and significant, b = 0.119, p < 0.001, 95% CI: [0.0748, 0.164], see Table 8, Model 3. These results suggest that the racial identity of departing directors had a significant effect on the racial identity of newly added directors.

Is This About Demographic Resemblance to Departing Group Members or Just Choosing Directors Who Are Not White Men?

To test this, we included both whether the board had lost women and whether the board had lost racial minorities as two separate predictors in our OLS regressions. When we predicted whether new directors were women using both predictors, we found that only the coefficient for whether the board had lost women was significant, b = 0.201, p < 0.001, 95% CI: [0.160, 0.242], whereas the coefficient for whether the board had lost racial minorities was not, b = −0.019, p = 0.397, 95% CI: [−0.0638, 0.0253], see Table 8, Model 2. On the other hand, when we predicted whether new directors were racial minorities, we found that only the coefficient for whether the board had lost racial minorities was significant, b = 0.125, p < 0.001, 95% CI: [0.0800, 0.170], whereas the coefficient for whether the board had lost women was not, b = −0.029, p = 0.128, 95% CI: [−0.0656, 0.00827], see Table 8, Model 4.

We also examined different racial minority groups separately and ran four regressions predicting whether new directors were (1) Black, (2) Hispanic, (3) Asian, or (4) women. In these regressions, we included four predictors: indicators for whether the board had lost Black directors, Hispanic directors, Asian directors, and women in the preceding year. These results are depicted in Figure 2 and Table 9 and, although noisier, share the same general pattern as the results from Study 1: The largest positive predictors of the demographic identities of newly added directors were indicators for whether the board had lost directors of the same demographic identity in the year prior. In the online supplement, we also report analyses for women of color.

Table 9. Examining Race-Specific Demographic Stickiness in S&P 1500 Data

Discussion

Across 5,616 additions of directors to S&P 1500 corporate boards from 2014 to 2019, Study 2 provides additional evidence that the demographic identities of departing group members have a significant influence on who is subsequently chosen to join that group. New directors are disproportionately likely to demographically resemble directors who recently left the board. In addition, although the data and results are noisier because of a lack of precision in knowing exactly who replaced whom, the general pattern of results replicates Study 1 and suggests that these effects are specific to demographic categories. Similar to Study 1, the demographic specificity of the stickiness results is more in line with an impact aversion account than a diversity loss aversion account. Additionally, these effects do not seem to be driven primarily by diversity thresholds identified in past research (Chang et al. 2019b) as these effects persist for companies both above and below the social norm for diversity and are not moderated by firm visibility (see online supplement).

As in Study 1, we do not find evidence that decision makers are more motivated to replace minority group members with other minority group members than they are to replace White men with other White men, as would be predicted by a diversity loss aversion account. For example, White men were 10.5 percentage points more likely to be appointed to a board after a White male director left than after a female or racial minority director left. Meanwhile, Black directors were 8.9 percentage points more likely to be appointed after a Black board member left than after a non-Black board member left. Thus, decision makers do not seem to exhibit demographic stickiness more in absolute terms when replacing minority group members than when replacing majority group members in this context.

Study 2 thus provides additional empirical evidence of our theoretical claims in a second organizational context. Of note, even when boards were replacing departing women or racial minorities, they still primarily chose to add men (65% of the time) and White people (75% of the time). Thus, although boards were more likely to choose women and racial minorities when replacing women and racial minorities (and equivalently, they were more likely to choose men and White people when replacing men and White people), they still had an overwhelming tendency to choose men and White people overall. These results suggest that diversification may remain slow in the absence of other interventions. Of course, both Studies 1 and 2 are ultimately correlational, so in our remaining studies, we aimed to identify causal evidence of these effects.

Study 3: Experimental Evidence of Demographic Stickiness

In Study 3, we wanted to test whether people were more likely to select candidates who shared the demographic identity of a departing group member in an experimental context. Participants were randomly assigned to learn that either a White man or a Black man had recently left a group in a scenario study. After learning about the departing group member, participants selected a new candidate to join the group, allowing us to examine the effects of predecessor race on hiring decisions.

Method

Participants.

We recruited 600 participants from Amazon’s Mechanical Turk (42.5% identified as men). Participants were paid $0.40 for a survey they were told would take approximately three minutes of their time. This survey was preregistered on AsPredicted.org (https://aspredicted.org/6ek3a.pdf). Anonymized data and analysis code from this study and all experiments included in this paper can be found at this OSF folder: https://osf.io/2ht5v/?view_only=4ca0da0c473f4838bc651adee154b21c.

Procedure.

Participants were asked to imagine that they managed the Strategic Management Group, a select group of promising employees, at the management consulting firm Banks & West Consulting Co. They were randomly assigned to one of two conditions, either the departing White man or departing Black man condition. Participants in the departing White man condition viewed a photo of a White man and were told that he was a member of the Strategic Management Group but had recently decided to leave the firm, so they would need to select a candidate to fill his spot in the group. Participants in the departing Black man condition were instead shown a photo of a Black man and informed that he was a member of the Strategic Management Group but had decided to leave the firm, so they would need to select a candidate to fill his spot in the group. In addition to the departing group member’s photo, participants also saw his years of experience, billable hours for the past two years, and MBA program name and rank. All photos and names used in the survey were stimulus sampled to ensure that results were not driven by particular photographs or names used.

Participants were then informed that all other members of the Strategic Management Group would be remaining at the consulting firm. They were shown photos of the five employees remaining in the group: two White men, two White women, and a Black man. Participants were reminded that they would be selecting an employee to join this group. The composition of the remaining employees was held constant across conditions, so the decision that participants faced regarding whom to select to join the group was identical across conditions, except for the demographic identity of the employee who had just left. In other words, if people merely cared about the overall diversity of the group and were not myopically attending to the demographic identity of the departing group member, decisions should not have varied between conditions, given that the demographic composition of the remaining group members was the same between conditions.

Next, participants made their selection decision. They were shown a set of three different candidates: a White man, a Black man, and a White woman. They were asked to select one of the three candidates to promote to their group. Participants saw a photograph of each candidate, their name, their years of experience, their billable hours for the past two years, and the name (and ranking) of their MBA program. Qualifications were randomly assigned to the three candidates, and they were also designed so that each candidate was the top performer on one of the three objective dimensions, the second performer on another, and the worst performer on the third, where the objective dimensions were years of experience, billable hours, and ranking of their MBA program.

We then collected participants’ gender and race. Complete study materials are reported in the online supplement.

Results and Discussion

Our dependent variable of interest was a binary indicator for whether a Black man was selected. We ran a proportions test to compare the proportion of participants selecting the Black man in the departing White man condition to the proportion of participants selecting the Black man in the departing Black man condition. Participants in the departing White man condition selected a Black man to join the group 39.4% of the time, whereas participants in the departing Black man condition selected a Black man 51.3% of the time, z = 2.94, p = 0.003. Thus, participants were significantly more likely to select a Black person when the departing group member was Black as opposed to White, providing experimental evidence of demographic stickiness. Participants were also more likely to select the White man when replacing a White man as opposed to a Black man: participants in the departing White man condition selected a White man to join the group 17.5% of the time, while participants in the departing Black man condition selected a White man 10.4% of the time, z = 2.52, p = 0.012. Rates of selecting a White woman to join the group did not vary significantly across conditions (43.0% versus 38.3%, z = 1.19, p = 0.232), suggesting that the demographic stickiness effect was specific to demographic categories.

Study 3 establishes further support for the demographic stickiness effect identified in Studies 1 and 2. Specifically, we present an experiment where participants were randomly assigned to replace someone of a given demographic identity in a group while all other aspects of the group and the position were held constant. Participants were more likely to hire Black men when replacing Black men than when replacing White men, and they were more likely to hire White men when replacing White men than when replacing Black men. A diversity loss aversion account would predict that the demographic stickiness effect should be larger when replacing the Black man than the White man, given that the departing Black man represents a diversity “loss” for the group whereas the White man does not. Although the demographic stickiness effect was directionally higher when replacing the Black man than the White man (contrary to the results from Studies 1 and 2), this difference was not statistically significant.

Unlike the results from Studies 1 and 2, however, participants in this study were generally less likely to choose the White man than would be expected by chance and appeared to show an overall preference for diversifying the group (i.e., by choosing either the Black man or the White woman). We return to this difference in preferences for diversification in the General Discussion.

Study 4: Experimental Evidence of a Mechanism for Demographic Stickiness

Study 3 provided causal evidence of a demographic stickiness effect. In Study 4, we used mediation to examine whether desires to minimize changes relative to the previous state of the world can help explain the effects of a departing group member’s identity on the demographic identity of selected candidates.

Method

Participants.

We recruited 800 participants on Amazon’s Mechanical Turk (47.2% identified as men). Participants were paid $0.40 to complete a survey they were told would take approximately three minutes of their time. This survey was preregistered on AsPredicted.org (https://aspredicted.org/5g9hi.pdf).

Procedure.

The study design was almost identical to that of Study 3, with one exception: Rather than varying the race of the departing group member, we instead varied whether information about the departing group member’s gender was known. Specifically, participants were randomly assigned to either the departing White woman condition or the departing member’s identity unknown condition. Participants in the departing White woman condition viewed a photograph of a White woman and were told that she had recently decided to leave the firm, so they would need to select a candidate to fill her spot in the group. Participants in the departing member’s identity unknown condition were instead told that a group member had decided to leave the firm, so they would need to select a candidate to fill the departing member’s spot in the group: Participants in this condition were not provided with any other details about the departing group member.

As in Study 3, participants were shown the photos and names of the remaining group members (two White men, two White women, and a Black man), so the composition of the remaining employees was held constant across conditions. If people merely cared about the overall diversity of the group and were not myopic, decisions should not have varied between conditions, given that the demographic composition of the remaining group members was the same between conditions. Participants were then asked to choose one out of three candidates to replace the departing group member. The three candidates were a White man, a White woman, and a Black man with equal (randomly assigned) qualifications, and all photographs and names used in the survey were stimulus sampled.

Participants then rated their agreement with three statements designed to test the degree to which desires to minimize changes relative to the previous state of the world (i.e., impact aversion) affected their decision making on a scale from “1: Strongly disagree” to “7: Strongly agree.” These three statements were as follows: (1) “I prioritized promoting someone who would retain the original group dynamic”; (2) “I attempted to make a promotion decision that wouldn’t change the group too drastically”; and (3) “I tried to choose someone who would fulfill a similar role as the departing group member.” We standardized each item and took their average to create a measure of desires to minimize changes (Cronbach’s α = 0.79).

Finally, participants reported their gender. All study materials are shared in the online supplement.

Results and Discussion

Our dependent variable of interest was a binary indicator for whether a woman was selected. We ran a proportions test to compare the proportion of women selected in the departing White woman condition with the proportion of women selected in the departing member’s identity unknown condition. Participants in the departing White woman condition selected a woman to join the Strategic Management Group 57.2% of the time, whereas participants in the departing member’s identity unknown condition selected a woman 43.0% of the time, z = 4.031, p < 0.001. Thus, participants were significantly more likely to select a woman when the departing group member was a woman than when they did not know the departing group member’s identity. Participants’ increased likelihood of selecting women in the departing White woman condition was accompanied with a decreased likelihood of selecting White men relative to the departing member’s identity unknown condition (9.2% versus 15.2%, z = 2.588, p = 0.010), as well as a decreased likelihood of selecting Black men relative to the departing member’s identity unknown condition (33.5% versus 41.7%, z = 2.408, p = 0.016). In other words, the desire to replace departing women with female candidates significantly decreased the selection rates of both male candidates (and equivalently, the rate of selecting male candidates was significantly higher when replacing someone of unknown identity as compared with replacing a White woman), again providing support for the demographic specificity of the stickiness effect.

Next, we tested whether there was a significant indirect effect of desires to minimize changes relative to the previous state of the world on participants’ likelihood of hiring a woman. Participants indicated that a desire to minimize changes influenced their decision to a greater extent in the departing White woman condition (mean = 4.825, SD = 1.355) than in the departing member’s identity unknown condition (mean = 4.505, SD = 1.277, t(798) = 3.437, p < 0.001). Following our preregistration, we used a 10,000-sample bootstrapped mediation model to test whether desires to minimize changes significantly mediated the effect of knowing a departing group member’s gender on hiring decisions. We found that the 95% bias-corrected confidence interval for the size of the indirect effect excluded zero (95% CI: [0.009, 0.040]), suggesting significant mediation.

In exploratory analyses, we also conducted our mediation analyses using the average causal mediation effect approach of Imai et al. (2010). This general approach to mediation analysis allows us to both test for mediation without relying on OLS (i.e., without assuming a linear model is appropriate) and to conduct sensitivity analyses to determine the robustness of the mediation effect to deviations from the sequential ignorability assumption (the assumption that there are no omitted covariates that are correlated with both the mediator and the outcome). The sensitivity analyses generate a parameter, rho, which varies between −1 and 1 and indicates how large the correlation between the errors of the mediation and outcome models would need to be for the indirect effect of our mediator to be null or reverse in direction. We confirmed that the 95% confidence interval for the indirect effect of desires to minimize changes on willingness to hire the female candidate excluded zero: [0.008, 0.040]. In particular, desires to minimize changes were estimated to account for 15.6% of the effect of the departing group member’s identity on willingness to hire the female candidate. Finally, the mediation results were robust to moderate deviations from the sequential ignorability assumption, as the indirect effect was positive and nonzero for any rho < 0.19.

Study 4 offers suggestive evidence that demographic stickiness arises, at least in part, because people seek to minimize changes relative to the previous state of the world. However, our analyses suggest that this accounts for only a part of the phenomenon, suggesting that the effect is multiply determined. For example, demographic stickiness could also be driven in part by diversity loss aversion. In addition, similar to Study 3, participants in this study were generally less likely to choose the White man than would be expected by chance and appeared to show an overall preference for diversifying the group (i.e., by choosing either the Black man or the White woman). We discuss the difference in preferences for diversification across the online experiments and the field studies in the General Discussion.

Study 5: Comparing White Man Leaving to Identity Unknown Leaving

To more directly test the divergent predictions of an impact aversion account versus a diversity loss aversion account, we ran a study where participants were randomly assigned to learn that either a White man or someone of unspecified demographic identity had recently left a group in a scenario study. After learning about the departing group member, participants selected a new candidate to join the group. An impact aversion account would predict that participants would be more likely to select a White man when replacing a White man as opposed to someone of an unspecified demographic identity, whereas a diversity loss aversion account would not.

Method

Participants.

We recruited 600 participants from Amazon’s Mechanical Turk (52.8% identified as men). Participants were paid $0.40 for a survey they were told would take approximately three minutes of their time. This survey was preregistered on AsPredicted.org (https://aspredicted.org/p86wg.pdf).

Procedure.

The study design was similar to that of Study 4, but in this study, participants were randomly assigned to either the departing White man condition or the departing member’s identity unknown condition. Participants in the departing White man condition viewed a photograph of a White man and were told that he had recently decided to leave the firm, so they would need to select a candidate to fill his spot in the group. Participants in the departing member’s identity unknown condition were instead told that a group member had decided to leave the firm, so they would need to select a candidate to fill the departing member’s spot in the group—participants in this condition were not provided with any other details about the departing group member. We did not provide participants with any information about the remaining members of the group, contrary to Studies 3 and 4.

Participants were then asked to choose one out of three candidates to replace the departing group member. The three candidates were a White man, a White woman, and a Black man with equal (randomly assigned) qualifications, and all photographs and names used in the survey were stimulus sampled. Finally, participants reported their gender. All study materials are shared in the Online Supplement.

Results and Discussion

Our dependent variable of interest was a binary indicator for whether a White man was selected. We ran a proportions test to compare the proportion of participants selecting a White man in the departing White man condition with the proportion of participants selecting a White man in the departing member’s identity unknown condition. Participants in the departing White man condition selected a White man to join the Strategic Management Group 18.7% of the time, while participants in the departing member’s identity unknown condition selected a White man 17.3% of the time, z = 0.444, p = 0.657. In this case, participants were not more likely to select a White man when replacing a White man than when replacing someone of unknown identity. The lack of demographic stickiness when replacing a White man in this study is more consistent with a diversity loss aversion account than an impact aversion account. Overwhelmingly, participants preferred to not select a White man, contrary to the overall patterns in the field data presented in Studies 1 and 2.

One potential alternative explanation could be that participants in the departing member’s identity unknown condition assumed that a White man was leaving. If that were the case, the departing member’s identity unknown condition would be an uninformative baseline against which to compare the departing White man condition, as the two conditions would be functionally equivalent. In the United States, people typically consider White men to be the default (Merritt and Harrison 2006, Purdie-Vaughns and Eibach 2008, Bailey et al. 2019). This assumption could also help explain why Study 4 participants selected White men significantly more often when the departing group member’s identity was not specified than when they were told that the departing group member was a White woman.

To assess participants’ assumptions about the unknown departing candidate, we conducted an additional study. Participants (n = 151) saw the stimuli from the departing member’s identity unknown condition and responded to the question: “If you had to guess, what do you think the demographic identity of the person who is leaving is?” Participants were given the option of selecting “White Man,” “White Woman,” “Black Man,” “Black Woman,” “Hispanic/Latinx Man,” “Hispanic/Latinx Woman,” “Asian Man,” “Asian Woman,” or “Another identity not listed.” A total of 81.5% of participants guessed that a White man had left. Thus, when forced to make a choice, the vast majority of participants assumed that a White man was departing whether or not we specified the demographic identity of the departing group member. However, we acknowledge that the forced choice paradigm does not necessarily reflect participants’ overall belief distribution and that we sampled different participants from those included in the original study, limiting our ability to test this alternative explanation.

Study 6: Highlighting the Myopic Nature of Demographic Stickiness

In our final study, we wanted to more directly highlight the myopic nature of demographic stickiness. Participants were randomly assigned to learn that they were either replacing a White man, a White woman, or no one (i.e., just selecting a new member to add to a group), but we kept constant the demographic composition of the remaining group members. If people were not myopic and just cared about the overall diversity of the group, the rates of selecting a White woman should be equivalent across conditions, given that the demographic composition of the remaining group members was constant.

Method

Participants.

We recruited 900 participants from Amazon’s Mechanical Turk (55.0% identified as men). Participants were paid $0.40 for a survey they were told would take approximately three minutes of their time. This survey was preregistered on AsPredicted.org (https://aspredicted.org/cn7pb.pdf).

Procedure.

Participants were asked to imagine that they handled personnel selection for the Head Investment Panel at Cardinal Asset Management. They were randomly assigned to one of three conditions: the departing White man condition; the departing White woman condition; and the no one departing condition. To keep the demographic composition of the remaining members of the Head Investment Panel the same, participants in the departing White man condition saw a group of four White men, one White woman, and one Black man and were told that one of the White men was leaving. Participants in the departing White woman condition saw a group of three White men, two White women, and one Black man and were told that one of the White women was leaving. Participants in the no one departing condition saw a group of three White men, one White woman, and one Black man (and were not told about anyone leaving the group).

Participants were then told they needed to promote someone to the Head Investment Panel. For each of the three candidates—two White men and one White woman—participants saw the candidate’s photo, years of experience, and highest degree earned. After making their selection decision, participants reported their gender. All study materials are shared in the online supplement.

Results and Discussion

Our dependent variable of interest was a binary indicator for whether a White woman was selected. We ran a proportions test to compare the proportion of participants selecting the White woman in the departing White man condition to the proportion of participants selecting the White woman in the departing White woman condition. Participants in the departing White man condition selected a White woman to join the group 36.6% of the time, while participants in the departing White woman condition selected a White woman 45.5% of the time, z = 2.22, p = 0.026. Participants in the departing White woman condition were also significantly more likely to select a White woman than participants in the no one departing condition, who selected a White woman 37.5% of the time, z = 1.96, p = 0.047. The rates of selecting a White woman did not significantly differ between the departing White man and no one departing conditions, z = 0.244, p = 0.807. In other words, rather than attending to the demographic composition of the remaining group, which was held constant across conditions, people seemed to myopically attend to the demographic identity of the departing group member. However, the fact that the rates of selecting a White woman did not significantly differ between the departing White man and no one departing conditions also means that the rates of selecting a White man did not significantly differ between these conditions, again suggesting that online participants are not particularly driven to replace White men with White men. This result parallels the findings from Study 5 and suggests that the psychology of our experimental participants is more consistent with a diversity loss aversion account than an impact aversion account.

General Discussion

Across various field studies and experiments, we find evidence that the demographic identities of former group members have a lingering influence on the demographic identities of those selected to succeed them in the group, as people tasked with filling vacated positions are disproportionately likely to select candidates who share the demographic identity of departing group members. In other words, the demographic composition of groups tends to be sticky in replacement hiring decisions. We find evidence of demographic stickiness in the appointment of U.S. federal judges, in the selection of S&P 1500 corporate board directors, and in hiring decisions in experiments where we manipulate the demographic identity of a departing group member. We find these effects when the decision maker is not a member of the group (as is the case in the judiciary data) and when the group itself helps select the next member (as is the case in the corporate board data), for both gender and race, and across time. In addition, these effects seem to be specific to demographic categories, as women are disproportionately likely to be selected to replace women—but not racial minorities—and racial minorities are disproportionately likely to be selected to replace racial minorities—but not women.

Theoretical and Practical Contributions

We integrate insights about human judgment and decision making into our understanding of hiring decisions to provide a novel perspective on diversity-related hiring decisions in organizations. Our work is not meant to minimize the important roles that bias, discrimination, and homophily can play in diversity-related selection decisions. Instead, our work augments our understanding of the factors that influence diversity-related decision making by providing insight into when decision makers are likely to select people from historically marginalized groups as well as potential guidance about how we can improve diversity in organizations.

Specifically, our work highlights the impact that loss aversion can have on replacement hiring decisions, suggesting that individuals may be myopically influenced by the demographic identity of departing group members when selecting their replacements. People may be motivated to maximize similarity between departing group members and their replacements to preserve group dynamics (an impact aversion account), and people may place more weight on avoiding diversity losses rather than seeking diversity gains (a diversity loss aversion account). Across studies, we document evidence consistent with both accounts, suggesting that demographic stickiness effects may be multiply determined. These findings highlight that fundamental decision-making tendencies (i.e., loss aversion) can play a role in driving diversity-related decisions.

Considering demographic stickiness can shift the way we think about addressing inequality in organizations. One interpretation of our findings is that they contribute to our understanding of how inequality propagates in organizations. However, our results also suggest that progress toward diversification, once it has happened, will be relatively “sticky.” This suggests that one-time interventions to improve diversity may have more enduring effects on group composition than might otherwise be expected. Returning to our opening example of Trump’s nominees to the Supreme Court, although Trump was not known to be a champion of diversity (and in many ways actively campaigned against diversity), he did not have a significant negative effect on the gender or racial diversity of the federal judiciary as he largely maintained the demographic diversity of those he replaced. Indeed, the overall pattern of diversification in the federal judiciary data suggests that one-time diversification efforts can have enduring effects on group composition, as drastic increases in diversity under some presidents are followed by periods of little-to-no backsliding under presidents who did not increase diversity (Figure 3). A runs-test further confirms the federal judiciary data are characterized by “runs” of diversification, as would be predicted by our theorizing (z = 10.64, p < 0.001; see online supplement for full results). Thus, demographic stickiness could be seen not only as a force that hinders diversification of homogeneous organizations but also as a force that helps maintain progress on diversity once it is achieved.

Some extant research provides support for the idea that organizations can exhibit inertia in diversification (Stainback et al. 2010, Tinsley et al. 2017), but also that one-time interventions that change group composition can have lasting effects. For example, temporary affirmative action programs and quotas have been shown to have enduring effects on the racial composition of workplaces, even once the programs are removed (Miller 2017). Mergers and acquisitions have also been shown to be key opportunities for companies to improve diversity, as they disrupt the status quo and force organizational demographic changes (Zhang 2019).

These results suggest that interventions that can generate one-time behavior change in hiring decisions—even if the behavior change does not persist—may be more desirable than previously appreciated. Rather than focusing on trying to change people’s attitudes in the long term, which is the goal of many diversity-related interventions like diversity training (Bezrukova et al. 2016), it may instead be more fruitful to focus efforts on changing hiring decisions in the short term through, for example, nudge-based interventions or incentives. Our findings suggest that such one-time interventions to hiring processes that successfully increase diversity may continue to pay dividends even once the interventions are removed.

Although we draw from the judgment and decision-making literature, which often documents evidence of biased decision making, we are not claiming that the effects we document are evidence of a bias; we are agnostic as to whether demographic stickiness is evidence of biased or rational decision making. We also are not suggesting that organizations should avoid investing in long-term strategies. Organizations must invest in efforts to increase inclusion, retention, and promotion for women and racial minorities in order to make advances toward equity. Indeed, our data suggest that diversity will not increase in the absence of sustained effort: The replacement rates of men and White people were well above 50% in both of our field settings, whereas this was not true of women and racial minorities. However, an optimistic perspective is that backsliding on diversity may be less likely to occur than would be predicted in the absence of demographic stickiness.

Limitations and Future Directions

Although the totality of our evidence suggests that loss aversion can drive demographic stickiness effects, we find evidence consistent with two distinct psychological mechanisms stemming from loss-averse preferences. Our data support both an impact aversion account (people desire to minimize changes to group composition and dynamics because change-related losses loom larger than change-related gains, a psychology that would kick in both when replacing majority and minority group members) and a diversity loss aversion account (people experience outsized concerns with losing ground on demographic diversity relative to interests in gaining ground, a psychology that would primarily activate when replacing minority group members).

In our field studies, the evidence seems more consistent with an impact aversion account than a diversity loss aversion account. People are disproportionately likely to replace departing group members with someone who shares their specific demographic identity rather than anyone who belongs to a similarly underrepresented group. This pattern suggests that decision makers are not just trying to avoid losing ground on diversity but instead are looking to maximize perceived similarity between departing group members and their replacements. Moreover, we find no evidence that the stickiness effect is greater (in absolute terms) when decision makers replace minority group members than majority group members, which is again inconsistent with diversity loss aversion. However, in the experiments, our evidence is mixed and, if anything, provides more support for diversity loss aversion as a mechanism. Although we do find evidence consistent with partial mediation for an impact aversion mechanism in Study 4, with the exception of Study 3, our results largely suggest that there are no demographic stickiness effects when replacing White men in our online experiments, consistent with the predictions made by a diversity loss aversion account.

In the online supplement, we report two studies that further attempt to disentangle these two accounts by examining whether performance moderates the effects of demographic stickiness. If the team or the departing group member was performing poorly, it should reduce people’s desire to minimize changes to the (relatively bad) previous state of the world. Thus, under an impact aversion account, we would expect that poor performance should attenuate demographic stickiness effects. On the other hand, if diversity loss aversion is the primary mechanism, then we would not expect poor prior performance to weaken the effect, given that losing a minority group member would constitute a diversity loss irrespective of the performance of the team or the departing group member.

In Supplemental Study 1, we varied whether the person departing the team had voluntarily left the team or had been fired because they were underperforming at work. We replicated the demographic stickiness effect when the person had voluntarily left the team, but not when the person had been fired, though the interaction between the identity of the departing team member and the reason for their departure was not statistically significant. In Supplemental Study 2, we varied whether the team had performed well or poorly over the past year. Again, we replicated the demographic stickiness effect when the team had performed well, but not when the team had performed poorly, though again the interaction between the identity of the departing team member and team performance was not statistically significant. Thus, we find evidence that poor prior performance can act as a boundary condition for demographic stickiness, which seems more consistent with an impact aversion account than a diversity loss aversion account. However, we caution against reading too much into these findings given that we do not document significant interaction effects. Combined with the results of our mediation study (Study 4), these studies offer some suggestive evidence that demographic stickiness, at least in experimental settings, arises in part due to conscious decision-making processes that account for past performance.

There are several possible explanations for the conflicting evidence we document across decisions made in the real world and decisions made in hypothetical experiments—in particular, for the mixed evidence about whether demographic stickiness effects hold when replacing White men. First, the likelihood of diversifying at baseline differs vastly across the two contexts. In the field studies, men and White people were chosen most often, even when replacing women and racial minorities, suggesting that on average, decision makers preferred to select candidates who were already well represented. In the experiments, however, participants chose White male candidates less often than would be expected by chance, suggesting that on average, participants in scenario experiments preferred to select underrepresented candidates. The greater preferences for diversity in our experiments may be unsurprising, given past research showing that people typically exhibit more positive attitudes and behaviors toward diversity when they know they are being observed than they actually exhibit in the world (Kawakami et al. 2009, Chang et al. 2021, Kellar and Hall 2022) and that people are unwilling to express bias against minority groups when directly asked, regardless of their underlying attitudes (Petsko and Rosette 2022). Because these participants may value and attend to diversity more than they would in the real world, the experimental context itself may be more likely to generate diversity loss aversion relative to the field. If true, these results suggest that in contexts in which diversity is highly valued and salient, diversity loss aversion may dominate impact aversion as a motivation for demographic stickiness.

Second, in the field studies, there are likely other considerations guiding decisions that are hard to replicate in experimental studies, which may alter the psychology of replacement hiring processes. For example, in our study of U.S. federal judiciary nominations, there are political considerations that we cannot control for, and it would be challenging to recreate these considerations in our experiments. In addition, we randomly assigned qualifications to candidates of different demographic identities in our experiments and limited the choice set to only three candidates, but in the real world, qualifications are not randomly assigned, and the choice sets are much larger. Differences like these may lead to different underlying mechanisms of demographic stickiness in different contexts.

Future research could further investigate when and why each account—impact aversion and diversity loss aversion—operates as a motivation for demographic stickiness, as well as explore additional boundary conditions of the effect. The organizational contexts we examine—federal judges and corporate board members—are historically dominated by men and White people, and they are high-status positions. This makes these contexts interesting to study from the perspective of understanding what drives diversification, but we do not know if we would find these same effects in contexts dominated by women or racial minorities or for low-status positions. Our field data are also limited to U.S. contexts, and our experiments were conducted with U.S. participants, so demographic stickiness may play out differently in other cultural contexts.

Although we did not find evidence that the effects we observe were moderated by the visibility of the court or company (see online supplement), as might be predicted if the phenomenon was driven primarily by impression management concerns (Chang et al. 2019b), it is possible that decision makers may exhibit demographic stickiness in part due to impression management. Decision makers may be worried that it may reflect poorly on them if they were to replace a minority with a majority group member. The phenomenon may thus be driven in part by decision makers catering to the preferences that they expect constituents or observers to have (McGraw et al. 2011, Dorison and Heller 2022).

We also cannot make strong claims about intersectional identities and the experience of racial minority women. Although we find some evidence that racial minority women are treated primarily as representatives of their race as opposed to their gender, these results merit further exploration and research. It would also be valuable to understand whether these effects do or do not extend to other sorts of identities beyond gender identity and racial identity. For example, future research could examine whether demographic stickiness influences selection decisions when considering nonvisible identities.

Finally, our research focuses solely on replacement selection and hiring decisions, and it does not illuminate what happens once candidates are selected or shed light on nonreplacement decisions. The degree to which a hiring decision is explicitly labeled as a replacement hire could moderate these effects. Past research has also identified that beneficiaries of affirmative action programs can face stigma (Heilman 1994, Leslie et al. 2014). It would be interesting to see if women and racial minorities who are selected to replace same-identity others are perceived to be beneficiaries of some sort of affirmative action and, as a result, must contend with similar negative consequences. Similarly, it would be interesting to understand the experiences of the trailblazer women and racial minorities who replace White men for the first time in groups.

Conclusion

Our work highlights the role that demographic stickiness plays in diversity-related hiring decisions. A pessimistic take on our results suggests that demographic stickiness inhibits diversification or slows it to a sluggish pace. However, there is also an optimistic view of our findings: Demographic stickiness may reduce backsliding once progress has been made. Thus, our work suggests that renewed attention and focus should be paid to one-time interventions that change group or organizational composition, as successful one-time interventions can have longer lasting effects than may previously have been appreciated.