Is it ok to make mistakes at work? Perceptions of differential treatment of mistakes by men and women in the workplace

Highlights

• The authors explore whether professional women and men have different perceptions of their workplace mistakes.
• The authors draw on a new primary survey data set of financial services professionals.
• The authors find that women and men have different perceptions of mistakes, and this interacts with the share of men in the environment.
• Women perceive mistakes in a more negative light when they are in roles with the highest shares of men.
• Women report the most positive perceptions of mistakes in environments with low shares of men; for men, differences across the share of men are small and not statistically significant.
• The authors find no significant gender difference in expected penalization for mistakes, and this null result does not vary with the share of men in the workplace, suggesting that the gender gap they document is located in the informal, psychological dimensions of error response rather than in formal penalties.

1 Introduction

Mistakes at work are inevitable and can play an important role in an individual's career trajectory. If handled constructively, workplace errors can also serve as important learning opportunities. For example, research has shown that making errors on the job can support learning but only if individuals feel safe acknowledging and reflecting on those errors (van Dyck et al., 2005; Hirak et al., 2012; van Mourik et al., 2023). Whether employees feel that safety, however, may depend in part on who makes the mistake. Prior research suggests that men and women are not judged equally when they err. Women in stereotypically male roles tend to be judged more harshly for mistakes, being viewed as less competent and accorded lower status than equivalent male colleagues. Brescoll et al. (2010) found that a woman in a male-typed leadership role (e.g., a female police chief) was evaluated more harshly for an error than an equivalent male leader, consistent with role congruity theory. Likewise, Swim and Sanna's (1996) meta-analysis and Hamilton and Lordan's (2023) review concluded that observers tend to attribute women's failures to lack of ability, whereas men's failures are more often excused as bad luck or other situational factors.

How a mistake is perceived may depend not only on who makes it but also on the context in which it occurs. Prior research suggests that the gender composition of a workplace can influence employee experiences. Usui (2008) found that women's job satisfaction tends to be lower in occupations with higher proportions of men, and tipping-point theories propose that workplace culture shifts once a demographic majority becomes overwhelming (Pan, 2015; Centola et al., 2018; Taylor et al., 2019). When women are only a small minority in a team or firm, they may become “tokens” (Kanter, 1977) and face heightened visibility and pressure, and in environments of “double male dominance,” where most colleagues and supervisors are male, women encounter a harsher culture, including more harassment (de Haas and Timmerman, 2010). Notably, however, this literature concerns observers' judgments of others' mistakes and the external conditions that women face; little is known about whether women and men differ in their own perceptions of the mistakes they make in the workplace. To address this gap, in this study we ask: Are views about mistakes at work gendered, and does the answer depend on the gender composition of the workplace? In other words, do female and male professionals feel differently about the mistakes they make on the job? Do women themselves feel less comfortable about their own mistakes because they anticipate harsher judgment? And is this shaped by how male-dominated their environment is? This focus on self-perceptions is important because an employee's internal comfort with errors can influence learning behaviors and career decisions, independent of how others actually respond. In taking this approach, we complement prior research that has mostly examined observers' reactions to men's vs. women's mistakes.

This question is particularly relevant in male-dominated sectors, such as financial services. Despite some progress, women remain significantly underrepresented in finance, especially in senior leadership (for instance, less than 20% of executive roles in major European financial firms are held by women; Boston Consulting Group, 2024). The industry also exhibits a substantial gender pay gap, which is largest at the top of the wage distribution, with evidence to this effect from both the United Kingdom (Healy and Ahamed, 2019) and the United States (Bertrand et al., 2010). One possible reason for these persistent disparities is a workplace culture that is less supportive or forgiving for women (Holton and Dent, 2016). Consistent with this, research has found that “pro-male” occupational cultures tend to have significantly lower female representation, suggesting that women self-select out of or are excluded from these environments (Lordan, 2019). Financial organizations often involve high stakes and heavy pressure, where any mistake can be costly. This can foster a punitive or low-tolerance “error culture.” If women perceive (or experience) that their mistakes are handled more harshly or seen as less acceptable, it could erode their confidence, job satisfaction, and willingness to stay in such environments (Lordan and Pischke, 2022) due to lower psychological safety for women (Edmondson, 2004). In turn, this might contribute to the difficulty of attracting and retaining women in finance. Motivated by this, we focus on professionals in financial services and explore whether women's and men's perceptions of their workplace mistakes differ and whether these perceptions are influenced by a workplace's gender composition.

Multiple perspectives have sought to explain gender disparities in workplace experiences, spanning structural to psychological factors. Traditional economic accounts highlight discrimination, pipeline issues, and unequal caregiving burdens (Altonji and Blank, 1999; Goldin, 2014). Subsequent research explored behavioral and attitudinal differences as additional contributors (Croson and Gneezy, 2009; Bertrand et al., 2010). Meanwhile, a rich psychology literature finds that men and women often differ in how they experience the workplace in both how others evaluate them (Igbaria and Shayo, 1997; Ashkanasy, 1994) and how they interpret workplace events (Rosenthal et al., 1996; Basford et al., 2014). Our study integrates these perspectives by examining a new facet of workplace experience—individuals' perceptions of their own mistakes—through a gender lens, thereby linking self-perceptions to prior literature on observers' judgments.

To investigate our research question, we administered a novel survey to UK financial services professionals. Respondents (N = 1,485) were asked to imagine a hypothetical mistake of moderate consequence at work and indicate their expected outcomes. We focused on three key self-perception measures: the individual's comfort in admitting the mistake, their expected penalization for the mistake, and whether the mistake would be seen as a learning opportunity. We then examined how these perceptions differed by gender. Importantly, we also analyzed whether any gender gap varied with the workplace gender composition (the share of men in the respondent's environment). This design allowed us to test whether gender differences in mistake perceptions are amplified in male-dominated contexts or diminished in more gender-balanced contexts.

In summary, our contribution is twofold: first, we provide new evidence on gender differences in how people perceive the treatment of their own mistakes at work (i.e., self-perceptions). Second, we show that these differences are moderated by the workplace context (with gender gaps most pronounced in male-dominated environments). By focusing on self-perceptions, we extend the literature on gender and leadership, error management, and organizational culture to illuminate how internalized experiences of mistakes may contribute to gender disparities. Our findings shed light on a previously underappreciated aspect of workplace gender dynamics and suggest avenues for making workplaces more inclusive and learning-oriented for all employees.

The remainder of the article is structured as follows: Section 2 presents a theoretical framework and develops our hypotheses regarding gender and mistakes. Section 3 describes the data and methods used in the empirical analysis. Section 4 reports the results, including descriptive findings and regression analyses. Section 5 discusses the implications of the findings. Additional descriptive and robustness analyses, including binary-coded outcomes, a reduced-sample replication, and an outlier-truncated robustness check, are reported in the Supplemental material.

2 Theoretical framework

We develop our theoretical framework by drawing on two complementary perspectives about workplace errors: one focusing on organizational factors (external context or “error climate”), and the other on individual factors (internal attitudes or “error orientation”). These perspectives lead to several possible mechanisms of gender differences in mistake perceptions. One possibility is that men and women are treated differently by others when they err, leading to different error climates for each (Baumgartner and Seifried, 2014). A second possibility is that men and women have inherent differences in their attitudes and reactions to mistakes—an error-orientation difference (Rybowiak et al., 1999). A third possibility is that both external and internal factors are at play, in which case gender differences in mistake perceptions might be most pronounced when these factors interact (for instance, when a woman with certain dispositions is placed in a heavily male-dominated culture). In this section, we review relevant literature in each domain and derive our hypotheses.

2.1 Organizational factors: error climate and gender composition

2.1.1 Error climate

An organization's norms and practices around handling mistakes can strongly influence how employees respond to errors. Baumgartner and Seifried (2014) introduced the concept of error climate, describing the shared attitudes within a workplace toward errors (e.g., whether errors are openly discussed or hidden, whether they are punished or treated as learning opportunities). Drawing on questionnaire data from apprentices and trainees in commercial occupations (e.g., the hotel and retail trades), their research indicates that a supportive “learning-from-error” climate fosters constructive responses to mistakes, such as greater reflection and problem-solving, whereas a punitive climate induces fear and error concealment. In other words, factors external to the individual—such as the team or organizational culture—can shape one's comfort with making and admitting mistakes.

2.1.2 Gender stereotypes and incongruent roles

Gender-based expectations in an organization are one important aspect of error climate for women vs. men. Role congruity theory (Eagly and Karau, 2002) posits that prejudice can arise when there is a misalignment between gender stereotypes and job roles. For example, leadership roles are often stereotyped as “masculine” (requiring assertiveness, toughness), so a female leader may be viewed as incongruent with that role and face extra scrutiny or skepticism. Mistakes might especially undermine the status of someone in a gender-incongruent role because observers could see the mistake as evidence that the individual does not fit the role. Experimental evidence supports this idea: in a series of experiments in which participants rated hypothetical leaders occupying gender-incongruent roles, Brescoll et al. (2010) found that people evaluated a manager's mistake more harshly when the manager was in a role not traditionally associated with their gender (e.g., a female police chief or a male nurse). In those scenarios, the individual who violated gender-role expectations was rated as less competent and accorded less status after an error compared to a gender-congruent counterpart. Similarly, in a meta-analysis of observer-attribution studies, Swim and Sanna (1996) showed that on stereotypically “male” tasks, women's failures are more likely to be seen as due to lack of ability, whereas men's failures are excused as bad luck or a momentary lapse. These findings underscore that the error climate can be gendered: in traditionally male domains, a woman's mistakes may be met with harsher judgments, creating an environment where she might reasonably worry more about errors.

2.1.3 Gender composition vs. gender congruence

While historically the gender stereotype of a job often matched its actual gender composition, this is not always true today (Ortiz-Ospina et al., 2018). Many occupations that were once male-dominated have seen increases in female participation, although cultural stereotypes may lag. Therefore, distinguishing between gender composition (the share of males vs. females in a workplace) and gender congruence (how culturally “masculine” or “feminine” a job is perceived to be) is useful. Both factors could independently affect the error climate. A highly male-dominated workplace might have a more “macho” culture that could be less forgiving of mistakes or less psychologically safe for women (regardless of the job's stereotype). Indeed, a laboratory experiment by Born et al. (2018) suggests that simply being in a strong minority as a female can impact one's leadership behavior and outcomes, and Lordan and Pischke (2022), using large-scale survey data, find that women are less satisfied in workplaces with higher shares of men. Furthermore, tokenism theory (Kanter, 1977) argues that when women are a very small minority (say <15%) in a group, they tend to experience isolation, heightened visibility, and pressure to perform, which can make any mistake feel more critical. Empirically, Taylor et al. (2019) document “tipping points” in U.S. workplaces such that when female representation falls below roughly 20%−30%, women's presence in management tends to drop precipitously. In sum, the organizational context, including both the prevalent gender culture and the numerical gender composition, is likely to influence how mistakes are perceived and handled. We expect that women working in environments with very high shares of men will face a less forgiving error climate (or at least perceive it to be so), compared to women in more gender-balanced environments or men in general.

2.1.4 Psychological safety

Psychological safety is a related concept. It refers to an individual's belief that they will not be punished or humiliated for speaking up with ideas or admitting mistakes, a concept Edmondson (2004) developed in field studies of work teams. A climate of psychological safety has been shown to facilitate learning from failure (Carmeli, 2007) and improve performance (Rowold et al., 2008). Women may feel less psychological safety in workplaces where they are heavily outnumbered by men. Double male dominance (de Haas and Timmerman, 2010)—being in a setting that is male-dominated in both workforce and leadership—could exacerbate feelings that one's mistakes will not be given the benefit of the doubt. In such contexts, a woman might fear that a mistake will confirm negative stereotypes about her gender (Eagly and Karau, 2002) and thus be particularly anxious about errors. This leads to our expectation that women in very male-dominated environments will report the least comfort with mistakes and the strongest sense that mistakes carry stigma. We formalize this expectation as our context hypothesis (H2) in Section 2.3.

2.2 Individual factors: error orientation, risk, and emotion

In contrast to external factors, another line of thought is that men and women might inherently react to mistakes differently because of personality, preference, or cognitive differences. Unlike the organizational factors reviewed above, which are properties of the environment and therefore vary with workplace context, the individual factors reviewed here are properties of the person and, in principle, travel with the individual across contexts. This is why the present subsection is not organized around workplace context in the way Section 2.1 is. Rather, gender enters here through average differences between men and women in the relevant traits rather than through gendered environments. The interplay between such traits and the workplace context is taken up in Section 2.3. Rybowiak et al. (1999) introduced the notion of error orientation—stable individual differences in how people think about and cope with errors. They developed an Error Orientation Questionnaire (EOQ) with eight scales capturing attitudes toward errors (e.g., whether one is error-averse or error-accepting). Such individual traits could lead to different perceptions of mistakes regardless of the environment. Here, we consider two well-documented individual-level differences that might be relevant: risk preferences (including cognitive reflection) and rumination.

2.2.1 Risk tolerance and cognitive style

Making a mistake at work often implies that one took a risk or attempted a task beyond one's comfort zone. Thus, an individual's risk tolerance could influence their attitude toward mistakes. A comprehensive meta-analysis by Byrnes et al. (1999) found that men generally take more risks than women, across a variety of domains (with a mean effect size suggesting men take more risks). This implies that women, being more risk-averse on average, might also be more cautious about mistakes and view mistakes as more aversive. Additionally, recent studies have investigated the role of cognitive abilities in risk-taking. Cueva et al. (2016), for example, administered the Cognitive Reflection Test (CRT) to more than 1,000 subjects and found that women scored significantly lower on the CRT than men, indicating a lower tendency to override intuitive errors. They also found that individuals with higher cognitive reflection exhibited different risk-taking behavior in certain tasks. Another study by Amador-Hidalgo et al. (2021) showed that higher cognitive ability is associated not so much with different risk preferences as with more consistent, less noisy decision-making under risk. In other words, people with lower cognitive ability might make more decision errors or be inconsistent, which can appear as greater risk aversion. These findings suggest that some of the observed gender gap in risk-taking may stem from cognitive and decision-making differences. When applied to our context, this could mean that women (who, on average, may have a more cautious decision style) could perceive a given mistake as more serious or be less inclined to view “learning from mistakes” positively, compared to men.

2.2.2 Rumination and stress

Another individual difference that could affect reactions to mistakes is the tendency to ruminate on negative events. There is evidence of a gender difference in rumination: Johnson and Whisman (2013) conducted a meta-analysis and found that women ruminate more than men, with a small-to-medium effect size (d ≈ 0.24). Rumination has also been linked to higher stress and slower emotional recovery after adverse events (Michael et al., 2007). Therefore, if a woman is more likely to brood over a mistake, she might experience more intense or prolonged negative feelings about it. This could manifest as lower comfort (the mistake looms larger in her mind) and a diminished sense of mistakes being a learning opportunity (because the emotional cost is high). In contrast, men's relative tendency to “move on” or not internalize failures as much might make them more at ease with errors.

Notably, these individual traits (risk tolerance, cognitive reflection, and rumination) are not independent of environment. Someone who is naturally risk-averse might gravitate toward certain workplaces, and a punitive workplace could in turn heighten anyone's rumination. But insofar as such traits differ on average by gender, they provide a potential internal mechanism for gendered perceptions of mistakes. We formalize the resulting expectation as our gender hypothesis (H1) in Section 2.3.

We emphasize, however, that our survey was designed to measure perceptions of mistakes and the composition of the respondent's work environment, and it did not include validated measures of the individual-level constructs reviewed in this subsection [e.g., incentivized risk-preference tasks, the CRT, or rumination scales in the tradition of Nolen-Hoeksema (1991)]. The discussed literature therefore serves to motivate the direction of our gender hypothesis and aid the interpretation of our results, rather than to generate mechanism-specific predictions that we can test directly. We return to this asymmetry between the breadth of the theoretical framework and the scope of our empirical measures in Section 5.

2.3 Interaction of gender and context and hypotheses

The preceding discussion highlights that external/organizational factors (such as gender-biased reactions and male-dominated cultures) and internal/individual factors (such as risk aversion and rumination) could both lead to women perceiving workplace mistakes more negatively than men. These are not mutually exclusive explanations. In fact, they may reinforce each other: For example, a woman who is somewhat risk-averse to begin with might feel especially uncomfortable in a macho culture that indeed penalizes mistakes. Our study is not designed to definitively tease apart whether the root cause is differential treatment by others or intrinsic differences in perception. Instead, we capture the combined outcome of these influences. Disentangling the organizational from the individual mechanism would, in principle, require either an experimental design that randomly assigns employees to teams with differing gender compositions or longitudinal data that track the same individuals as their team composition changes; absent such variation, the cross-sectional estimates we report should be read as the joint imprint of context and disposition rather than as a clean decomposition of the two. Given the literature review, we hypothesize the following:

• H1 (Gender hypothesis): Women perceive their own workplace mistakes more negatively than men do (i.e., women report less comfort admitting mistakes, higher expected penalization, and lower perceived learning opportunity).
• H2 (Context hypothesis): Employees in male-dominated environments have more negative perceptions of mistakes than those in more gender-balanced or female-dominated environments.
• H3 (Interaction hypothesis): The gender difference in mistake perceptions is largest in male-dominated environments and diminishes in more gender-balanced settings.

These hypotheses flow from the idea that external biases and internal predispositions likely compound in certain scenarios. If women generally are a bit more concerned about mistakes (H1) and male-heavy workplaces make everyone a bit more negative about mistakes (H2), then women in male-heavy workplaces would be affected on both dimensions at once (H3). The following sections describe how we test these hypotheses using our survey data.

3 Data and methods

3.1 Data collection and sample

Our analysis uses a primary dataset collected through an online survey in September 2020. The survey was developed in collaboration with Women in Banking and Finance (WIBF), a professional network in the United Kingdom. WIBF distributed the survey link via its member email lists and social media. The target population was professionals working in the financial services sector. We chose to focus on this sector because it is traditionally male-dominated and high-pressure, making it an apt setting to explore our research questions. By leveraging WIBF's network, we were able to reach many women in finance, as well as men who are supporters or members of the network.

We deliberately chose this partnership-based recruitment strategy over a probability-based sampling design for three reasons. First, there is no public sampling frame of U.K. financial-services professionals from which a random sample could be drawn, and general-population online panels contain too few women in the most male-dominated corners of finance to support our subgroup comparisons. Second, our hypotheses require observing women across work environments that range from female-majority to overwhelmingly male; the WIBF network, which includes women throughout the sector as well as male members and supporters, provided statistical power for women across the full range of share-of-men environments that a random sample of the sector would not. Third, the partnership lent the survey credibility with a professional population that is difficult to reach and typically reluctant to respond to unsolicited surveys. The cost of this strategy is non-representativeness, which we describe below and return to when interpreting the results.

The survey received a total of 1,692 responses. We applied several screening criteria. First, 171 respondents were removed because they were not currently working in financial services (the survey had an initial filter question for sector). Second, 10 respondents who did not disclose their gender or identified as non-binary were omitted. We made this decision because the non-binary sample was too small to analyze as a separate category. We acknowledge this as a limitation. Third, we dropped 26 respondents who failed to answer the key questions about mistakes. These steps yielded a final baseline sample of 1,485 individuals, of whom 19% are men (n = 280) and 81% are women (n = 1,205). In some extended analyses with additional control variables, the sample size is slightly smaller (falling to between 1,474 and 1,333 depending on the specification) due to item non-response on those controls, as detailed in Section 4.2; importantly, our main results are robust if we reestimate all specifications on the common subsample of complete cases (N = 1,333; see Table C.2 in the Supplemental material).

It should be noted that our sample is not representative of the entire finance workforce. By design, it oversamples women because it was distributed through a women-in-finance organization. In fact, women make up about 81% of our respondents; however, in the U.K. financial-services sector, women are estimated to be less than 50% of employees overall and well under 20% of senior roles (Rogish et al., 2021). The men in our sample may also be a select group (possibly those more engaged with diversity initiatives). The implications of this will be considered when interpreting results. Overall, our respondents worked an average of approximately 45 h per week (suggesting full-time roles), and they spanned a range of positions from junior analysts to senior managers and executives. They were also employed across various subsectors of finance (including banking, investment, insurance, and professional services consulting). Our regression models will include controls for job role and sector to account for this heterogeneity.

A natural question is whether survey weights could be applied to obtain a more representative picture in terms of gender and the share of men in the workplace. We considered this carefully and concluded that credible weights cannot be constructed here for two reasons. First, valid weights require external benchmarks for the weighting margins, and while population benchmarks exist for gender in U.K. financial services, there are no population data on the joint distribution of gender and workplace-level gender composition, the margin that matters most for our research question. Second, our parameters of interest are conditional associations: gender, the share of men, and their interaction enter the regressions directly, so under correct specification reweighting on these margins would leave the coefficients essentially unchanged while inflating sampling variance, particularly given that men are only 19% of respondents (Solon et al., 2015). Reweighting would alter the unconditional descriptive statistics, so the means in Tables 1, 2 should be read as descriptions of our sample rather than as estimates for the sector as a whole.

Table 1. Summary statistics for key variables.

SD, standard deviation.

Mistakes comfort/Penalize/Learn are self-rated perceptions regarding a hypothetical mistake at work. Female is an indicator for gender (81% of the sample are women). Share of Men is derived from a question on the percentage of colleagues who are male. Summary statistics for the categorical role/seniority and sector controls used in the regressions are reported separately in Table B.3 in the Supplemental material.

Table 2. Tests of difference for mistake perceptions variables, separately by gender and by share-of-men environment.

Two-sample t-test results.

SoM, share of men; HSoM, high share of men in workplace (≥75% men); LSoM, low share of men (<75% men).

^*p <0.10. ^**p <0.05. ^***p <0.01.

3.1.1 Response rate

Because the survey link was distributed through network emails and postings without a unique sampling frame, we cannot calculate an exact response rate. Given the size of WIBF's mailing list, the response rate can be roughly estimated to be in the 25% range, but this is only an approximation. We mention this for transparency; despite the non-random nature of the sample, the large number of respondents provides substantial variability and statistical power for our analysis.

Finally, we note that the data set generated and analyzed for this study is not publicly posted due to confidentiality agreements. However, it can be made available by the authors upon reasonable request for replication or extension purposes.

3.2 Measures

3.2.1 Mistake perception measures

The core of our survey asked respondents to imagine a scenario in which they made a mistake of moderate consequence in their job (e.g., an error that required colleagues to do rework, or a missed small deadline (significant but not career-ending). They were then presented with three statements and asked to rate how likely each outcome was on a 7-point Likert scale (1 = Not likely at all to 7 = Very likely). The statements were (1) “I would feel comfortable telling my company about a mistake I made”—we call this Mistakes-Comfort; (2) “I would be penalized for the mistake I made”—Mistakes-Penalize; and (3) “Mistakes would be viewed as learning opportunities in my company”—Mistakes-Learn. An option for “Not applicable” was also provided (those few who chose it were treated as missing). These three variables are our primary outcomes of interest, reflecting the respondent's subjective view of how mistakes are treated in their workplace.

We note that respondents were not asked to describe the specific mistake they had in mind when answering these questions. The scenario deliberately fixed the severity of the mistake at a “moderate consequence” level and offered concrete examples to standardize the frame across respondents, but because we did not collect open-text descriptions, we cannot categorize the types of mistakes respondents imagined, examine whether women and men had different types of mistakes in mind, or test whether mistake type moderates the gender differences we document. We flag this as a limitation in Section 5 and outline how future surveys could address it.

We use the Likert-scale responses (1–7) directly in many analyses. In addition, for ease of interpretation, we also construct binary indicators for each outcome: equal to 1 if the respondent answered 5, 6, or 7 (indicating that the outcome is at least somewhat likely in their view) and 0 if they answered 4 or below (neutral to very unlikely). For example, a binary Comfort variable is 1 for those who would feel likely comfortable admitting the mistake (responses 5–7) and 0 for those neutral or unlikely to feel comfortable (responses 1–4). This dichotomization provides a clear way to talk about “being comfortable or not,” although it does discard some nuance. We chose 5 as the threshold since it is the point at which respondents cross into the “likely” side of the scale. In robustness checks, we verified that using alternative cutoffs (e.g., treating only 6–7 as 1, or treating 4–7 as 1) does not qualitatively change the results (we discuss robustness results later).

3.2.2 Gender

Respondents were asked to indicate their gender with the options: 1 = male, 2 = female, 3 = prefer to self-describe, 4 = prefer not to say. As noted, we dropped those who chose options 3 or 4 due to the very small number. Thus, our analytic sample consists of 0 = male and 1 = female. In regressions, we include a Female dummy (Female = 1 for women).

3.2.3 Share of men in workplace

The survey asked participants about the gender composition of their current work environment, defined in terms of their occupation and function. Specifically, they were asked: “What is the current proportion of your gender in your occupation/function at work?” with response options in ten percentage ranges (1 = “1%−10%,” 2 = “11%−20%,” …, 10 = “91%−100%”). We converted this to a measure of the share of men (SoM) in the workplace as follows: for male respondents, their answer directly corresponds to the SoM. For female respondents, we subtracted their percentage from 100%. For example, if a woman selected “1%−10%” (meaning women are 1%−10%), we code that as 90%−99% men. If a woman said “51%−60%” women, we code it as 40%−49% men, and so on. This yields an approximate percentage of the workforce (in the respondent's context) that is male. We then simplified this variable in two ways. First, we define a dummy variable High SoM equal to 1 if the share of men is ≥75%, and 0 if it is <75%. In other words, “high SoM” indicates a substantially male-dominated environment. About 44% of our sample works in a high-SoM context by this definition. We chose the 75% cutoff to capture clearly male-majority contexts because previous literature on organizational tipping points suggests that somewhere in the 60%−80% range of one group can lead to cultural dominance (Pan, 2015). We tested alternative thresholds (e.g., 80%, 90%), and our results are robust to those (see Table S2 in the Appendix). Second, we created a continuous SoM variable by taking the midpoint of the respondent's selected range (for instance, a response of “71%−80%” is coded as 0.75). This ranges from 0.05 to 0.95 in our data. We use this continuous measure in regression models to capture more granular associations of composition.

We note that our measure is based on the respondent's perception and definition of their “occupation/function,” which could introduce some subjective variation. Because responses are recorded in 10-point bands rather than as exact percentages, the SoM measure is observed only up to a 10-percentage-point interval. The distribution of this variable is relatively uniform across the bands, with a slight concentration at the extremes (e.g., a notable number of women reported being in the 91%−100% male category, reflecting the fact that many work in very male-dominated areas of finance); the full distribution across bands is reported in Supplementary Table B.2.

3.2.4 Control variables

In our regression analyses, we include a set of controls for other factors that might influence perceptions of mistakes. Specifically, we control for the number of hours the person works per week (continuous variable, self-reported). Working extremely long hours might be correlated with workplace culture and stress, potentially affecting attitudes toward mistakes. We also include controls for the respondent's role seniority and sector. Role is captured by dummy variables indicating the person's level/position in their organization. The survey asked respondents to classify their role (e.g., analyst, associate, manager, director, executive, and so on—the exact categories were tailored to finance job ladders). We include separate dummy variables to account for seniority differences. This is important because higher-level individuals might feel more or less secure about mistakes. Sector is captured by dummy variables for the broad area of financial services the respondent works in (we had categories such as banking, asset management, insurance, professional services, fintech, etc.). Different subsectors can have different cultures (e.g., perhaps a fintech start-up vs. a traditional bank). Including these controls helps ensure that any gender or composition associations we find are not simply due to women being concentrated in certain roles or sectors.

Descriptive statistics for all variables defined earlier (including the three mistake-perception outcomes, the SoM measure, hours, and the workplace and attitudinal controls) are reported in Table 1, presented at the end of this section. We summarize the key features here for the reader's convenience: In the full sample of N ≈ 1,485 financial-services professionals, 81% are women and 19% are men; the average SoM in respondents' work environments is 0.64 (SD = 0.21); the mean Comfort score is 5.58 (on the 7-point scale), the mean Penalization score is 4.02, and the mean Learning score is 4.46.

Table 1 provides definitions and summary statistics for all key variables used in our analysis, including the three mistake-perception outcomes and all main regressors. We report each variable's definition along with its mean, standard deviation, minimum, and maximum in the full sample (N ≈ 1,485). From these summary statistics, we note that men comprise about 19% of respondents and women 81%, reflecting the oversampling of women in our survey. The average share of men in respondents' work contexts is 64%, with a standard deviation of 21%, indicating that many respondents work in male-majority settings. The mean Comfort score is 5.58 (out of seven), while the mean Penalize expectation is 4.02. The mean Learn score (viewing mistakes as learning opportunities) is 4.46.

3.3 Empirical approach

Our main empirical strategy is to estimate regression models that relate each mistake perception outcome as a function of the respondent's gender, the share of men in their environment, and the interaction of the two, controlling for other factors. The baseline model can be written as

Yi=a0+Fiδ+SoMiβ′+Xiγ+εi.(1)

In Equation 1, Y_i relates to either one of our three mistakes perception variables, defined as the original 7-item Likert scale or, separately, as a binary variable that indicates whether a person indicated a “likely” response to the specific mistake's perceptions question in the survey. SoM_i refers to the share of men in the respondent's work environment. In Equation 1, F_i = 1 if an individual self-reports as female and zero if they self-report as male. SoM is defined as either the midpoint share of men reported by a respondent or a binary variable that is equal to 1 if the SoM in the respondent's occupation equaled × SoM and zero otherwise. a0 is the usual regression intercept term. Standard errors are calculated using White's (1980) correction to allow for heteroscedasticity with an unknown functional form.

In Equation 1, X_i is a vector of control variables that contains hours, role, and sector. Hours relates to the number of hours per week an employee reports working. Role is a dummy variable that captures the worker's current role within their organization. Sector is a set of three dummy variables that capture the sector the worker is employed in.

We estimate linear models [ordinary least squares (OLS)] for all outcomes, treating the Likert scores as approximately continuous. For the binary versions of the outcomes, we primarily use linear probability models (OLS on the 0/1 outcome) for the ease of interpreting the interaction coefficients. In robustness checks, we also ran logistic regressions for the binary outcomes and found very similar patterns of results (in terms of sign and significance of key coefficients), which is not unusual given the sample size. For simplicity, we present the OLS results throughout. All regressions use heteroscedasticity-robust standard errors (White, 1980). Given our sample structure, we did not cluster standard errors (each observation is an individual; while individuals are nested in firms, we did not collect firm identifiers, and many respondents share large employers; we therefore report heteroscedasticity-robust standard errors).

We first present descriptive statistics (Table 1). We next examine simple differences in means (Men vs. Women, High SoM vs. Low SoM) to get a sense of the raw patterns (these are presented in Table 2). This is followed by the regression results (Table 3) corresponding to Equation 1.

Table 3a. Comfort with mistakes.

White-corrected standard errors in parentheses. Outcome measured on 1–7 Likert scale. The decline in N across columns reflects item non-response on the added attitudinal controls; see Section 4.2.

SoM, share of men, midpoint coding (0.05–0.95).

^**p <0.05. ^***p <0.01.

Before turning to the results, we note that we cannot conclusively determine from this analysis why any gender differences arise (e.g., whether due to bias vs. self-selection). What we can say is whether there are stylized differences and associations. In the Discussion, we speculate on mechanisms and implications, but the regression coefficients themselves should be interpreted as conditional associations. We also emphasize that our outcomes are perceptions, which, while important, may or may not reflect objective reality. Even if women perceive they will be penalized for mistakes more than men do, it will require further evidence to verify if they truly are penalized more. Our data nonetheless provide valuable insight into the subjective climate that employees experience.

4 Results

4.1 Descriptive statistics

Table 2 presents the mean values of the three mistake-perception variables by gender and workplace context, along with t-tests for differences. Several patterns are immediately apparent. First, there are notable gender differences in the raw data. Men report significantly more positive views about mistakes on all three measures. For example, the average Comfort score for men is 5.850 (on the 7-point scale) compared to 5.524 for women, and this difference (0.326) is statistically significant at the 1% level (two-sample t = 3.68, p < 0.01). Likewise, men's average Learn score is higher than women's (4.657 vs. 4.422, difference = 0.235, p < 0.05), and men's average Penalize score is lower (3.775 vs. 4.066, difference = −0.291, p < 0.01). In other words, in our sample, on average, women feel less comfortable admitting mistakes, are more inclined to believe mistakes would be held against them, and are less inclined to say mistakes are treated as learning opportunities compared to men.

Second, workplace gender composition also correlates with perceptions in the expected direction. Respondents working in high-SoM environments (with ≥75% men) have a mean Penalize score of 4.194, vs. 3.869 for those in more gender-balanced (low-SoM) environments. This difference (0.325) is significant (p < 0.01), indicating that people in heavily male workplaces are more likely to feel that mistakes are penalized. Similarly, the mean Learn score is 4.241 in high-SoM settings, significantly lower than 4.641 in low-SoM settings (difference −0.400, p < 0.01), meaning that those in male-dominated teams are less inclined to see mistakes as learning opportunities. For the Comfort measure, the difference by context is smaller (5.522 in high-SoM settings vs. 5.635 in low-SoM settings) and not statistically significant. This suggests that overall, employees in male-heavy environments are a bit less positive about mistakes, although the Comfort dimension by itself does not show a strong raw difference without accounting for gender.

Third, when we break the sample down by both gender and high/low environment, we find that women in high-SoM environments appear to be the most negative in their perceptions. A woman in a high-SoM environment reports an average comfort level of 5.458. In contrast, a man in a high-SoM environment reports a higher comfort (5.942), and a man in a low-SoM environment reports 5.809. Both are significantly higher than the woman/high-SoM reference (differences of +0.484 and +0.351, respectively, p < 0.01 for both). Women in high-SoM roles also feel more penalization than some other groups: For instance, compared to men in low-SoM roles, they have a higher Penalize score (4.221 vs. 3.670, difference +0.551, p < 0.01). Meanwhile, women in more balanced (low-SoM) environments report somewhat lower penalization (3.930) than their high-SoM female counterparts (4.221, difference −0.291, p < 0.01). Regarding the Learn variable, women in high-SoM roles have the lowest mean (4.196). This is significantly lower than the mean for men in low-SoM roles (4.711, p < 0.01) and lower than women in low-SoM roles (4.620, p < 0.01). Men in high-SoM settings have a mean Learn of 4.535, which is higher than 4.196 (the difference +0.339 is marginally significant at the 10% level). In summary, without any controls, the descriptive data indicate a consistent pattern: both being female and being in a male-dominated setting are associated with viewing mistakes more negatively, and these factors in combination correspond to the greatest differences in perceptions across all subgroups.

Furthermore, Table A.1 in the Supplementary material shows the bivariate Pearson correlations among the three mistake-perception outcomes (with significance). Although the three mistake-perception items (Comfort, Penalize, and Learn) are conceptually distinct, they are moderately correlated. For example, feeling comfortable admitting mistakes is positively associated with viewing mistakes as learning opportunities and negatively associated with expecting penalization. If combined into a single scale (after reverse-coding the Penalize item), these items yield a Cronbach's α of approximately.65, indicating modest internal consistency. Given this only moderate reliability and the theoretical differences among the three outcomes, we analyzed each item separately rather than as a unified index.

4.2 Comfort with admitting mistakes

Table 3a presents regression results for the outcome Mistakes-Comfort, illustrating how gender and workplace gender composition jointly relate to an employee's willingness to acknowledge errors. The analysis reveals a nuanced interaction between being female and working in a male-dominated environment. In general, we find no significant overall gender difference in comfort with admitting mistakes when workplace composition is not considered (in column 1, the Female coefficient is negative, but that simple model does not account for composition). However, once we account for the SoM colleagues, a clear pattern emerges: Women's comfort levels are increasingly lower in more male-dominated workplaces, whereas men's comfort remains relatively stable. This interaction is both statistically significant and practically important, indicating that workplace gender composition plays a critical role in shaping psychological safety for admitting mistakes.

Before discussing magnitudes, we clarify the decline in the number of observations across the columns of Tables 3a–c, which reflects item non-response on the attitudinal controls added in the later specifications rather than any loss of observations on the core variables. Columns 1–4 use the full baseline sample (N = 1,485) because gender, SoM, and the hours/role/sector controls are complete for everyone in the analytic sample. Adding the collaborative- and competitive-environment ratings in column 5 drops the 11 respondents who skipped those items (N = 1,474); adding job satisfaction and aspiration to senior leadership in column 6 drops a further 129 respondents (N = 1,345); and adding intention to leave in column 7 drops another 12 (N = 1,333). Table C.2 in the Supplemental material reestimates all specifications on the common sample of complete cases (N = 1,333) and shows that the coefficients of interest are essentially unchanged, so the movement in N across columns does not drive our results.

Table 3b. Penalization for mistakes.

White-corrected standard errors in parentheses. Outcome measured on 1–7 Likert scale. The decline in N across columns reflects item non-response on the added attitudinal controls; see Section 4.2.

SoM, share of men, midpoint coding (0.05–0.95).

^*p <0.10. ^***p <0.01.

Table 3c. Mistakes as learning opportunities.

White-corrected standard errors in parentheses. Outcome measured on 1–7 Likert scale. The decline in N across columns reflects item non-response on the added attitudinal controls; see Section 4.2.

SoM, share of men, midpoint coding (0.05–0.95).

^*p <0.10, ^**p <0.05. ^***p <0.01.

To quantify the effect, using the continuous SoM variable (Table 3a, column 3), the interaction term Female × Share of Men is −0.106 (p < 0.05). The negative and significant interaction is the estimate that bears on H3: It indicates that the female–male difference in comfort becomes more negative as the share of men rises; that is, the gender gap in comfort is larger in more male-dominated environments. Because the model is linear and the coefficients are conditional associations rather than a structural model, we interpret the sign and significance of the interaction and do not extrapolate the fitted line to compositions outside the observed range (roughly 5%−95% male).

What is more relevant is the direction and significance of the estimated coefficients. The pattern—a comfort gap that is increasingly negative for women as the share of men rises—is consistent with the tipping-point and tokenism mechanisms discussed in prior work (Kanter, 1977; Pan, 2015). The overall R-squared for these Comfort models is modest (approximately 0.04 without controls and up to 0.21 with full controls), meaning that gender and composition together explain only a small fraction of the individual variation—there is considerable heterogeneity in responses. Nevertheless, the coefficients of interest are statistically robust and substantively meaningful.

In sum, we find that women's comfort with admitting a mistake is significantly lower than men's in male-dominated workplaces, even though in gender-balanced or female-majority settings women's comfort levels are similar to (or even slightly higher than) men's. Men's comfort is essentially flat across workplace compositions (the Share of Men coefficient for men is small and not statistically significant), whereas women's comfort declines as the environment becomes more male-heavy. Formally, the linear combination β_Female + 0.5 × β_(Female × SoM) (the predicted female–male comfort gap at SoM = 0.5) is small in magnitude and not statistically distinguishable from zero at conventional levels (Wald test, p > 0.10), so we treat the apparent equality of comfort at the SoM sample mean as a tested null rather than an unverified inference. Read against our hypotheses, support for H2 (the Context hypothesis) would come from SoM's main effect, which is not significant for Comfort, whereas H3 is supported by the negative and significant Female × SoM interaction.

4.3 Perceived penalization for mistakes

We turn next to the Mistakes-Penalize outcome, which captures whether respondents think they would be penalized for the hypothetical mistake. The regression results (Table 3b) show a strikingly different pattern from the comfort measure. After accounting for controls, we find essentially no significant gender difference or composition effect on expected penalization. In the fully controlled model (Table 3b, col. 3 and onward), the coefficient on Female is small and not statistically significant, and the interaction Female × SoM is near zero. SoM's main effect is also not significant. In other words, once we control for other factors, women are no more likely than men to say that a mistake would result in formal penalties, and this perception does not vary systematically with the workplace's gender makeup.

It is worth noting that in the raw data (Table 2), women had a slightly higher mean Penalize score than men (4.066 vs. 3.775). That difference was driven mostly by women in male-heavy environments. However, in the multivariate context with controls, that small gap vanishes and the estimates are very close to zero. This suggests that other characteristics (perhaps role or sector) were correlated with both gender and the feeling of being penalized; once those are considered, gender, per se, is not associated with expected penalty.

Overall, H1 is not supported for the penalization outcome. In our data, once we consider comparable roles/environments, women do not report a higher expectation of punishment than men do. Similarly, H2 and H3 are not evident here. The share of men in the environment does not significantly alter perceived penalty for either gender. In other words, the expectation of getting penalized is uniformly moderate (around the midpoint of the scale, indicating “neither likely nor unlikely”) across groups.

4.4 Mistakes as learning opportunities

The third outcome, Mistakes-Learn, asked whether mistakes are viewed as learning opportunities in the respondent's company. The patterns for this variable in many ways mirror those for Comfort. Without the interaction term (Table 3c, column 1), Female is negative and significant (−0.227, p < 0.05) and SoM is also negative and significant (−0.081, p < 0.01), suggesting that overall women are a bit less likely than men to say mistakes are seen as learning opportunities, and people in more male-heavy workplaces are less likely (these are simple associations). Once we include the interaction (column 3), however, a richer story emerges: The Female main effect becomes +0.882 (p < 0.01), which can be interpreted as in a 0%-male (all-female) workplace, women's “learning opportunity” score is about 0.882 points higher than men's. Meanwhile, the Female × SoM interaction is −0.190 (p < 0.01), indicating that as SoM increases by one unit (i.e., moving from an all-female to an all-male composition), the female–male gap in the learning score decreases by 0.190. Equivalently, a 10 percentage-point increase in the male share is associated with about a 0.019 decrease in women's relative “learning” score. What matters for our hypotheses is the sign and significance of these coefficients: The negative and significant SoM main effect bears on H2, and the negative and significant Female × SoM interaction bears on H3. Both indicate that, as the workplace becomes more male-dominated, women become relatively less likely than men to see mistakes as learning opportunities.

These results imply that in gender-balanced or female-majority workplaces, women are equally or even more likely than men to say that mistakes are treated as learning opportunities, whereas in male-dominated workplaces, women become significantly less likely than men to believe that mistakes are seen as learning opportunities. Another way to see the pattern is to examine the SoM effect for each gender: In the model, the coefficient on SoM (−0.008) represents the slope for men (because Female is coded 0/1). It is near zero and not significant, suggesting that men's views on “learning from mistakes” are not very sensitive to the gender mix of the environment. In contrast, the slope for women is the sum of the Share coefficient and the interaction: −0.008 + (−0.190) = −0.198 (approximately). This means that for women, moving to a more male-dominated environment has a noticeable negative association with seeing mistakes as learning opportunities. These slopes' divergence is reflected in the significant interaction term.

Pulling the three outcomes together, the support for the three hypotheses is uneven and outcome-dependent. H1 (women perceive workplace mistakes more negatively) is supported in the raw means for all three outcomes but, after controls and the interaction with workplace composition, holds clearly for Comfort and Learning and is null for Penalization. H2 (employees in male-dominated environments have more negative perceptions) is supported only for Learning, where the main SoM coefficient is consistently negative and significant; it is not supported for Comfort or Penalization. H3 (the gender gap is largest in male-dominated environments) is supported for Comfort and Learning, where the Female × SoM interaction is negative and significant across all specifications, but is not supported for Penalization, where the interaction is essentially zero. The pattern across binary outcomes (Table C.1 in the Supplemental material) is qualitatively identical. Table 4 summarizes this hypothesis-by-outcome pattern.

Table 4. Summary of hypothesis support across outcomes.

Coefficients are from the fully controlled specification (Model 7) of Tables 3a–c. “Supported” indicates that the relevant coefficient (or coefficient combination) is signed in line with the hypothesis and statistically significant at p < 0.05. For H1, support is read from the Female main effect together with the Female × SoM interaction (combined linear test). For H2, support is read from the Share of Men (β_SoM) main effect. For H3, support is read from the Female × SoM interaction.

n.s., not significant at conventional levels.

4.5 Robustness and additional analyses

We performed several checks to assess the robustness of our findings (details can be found in the Supplemental material). First, as mentioned, we varied the cutoff that defines a “high share of men” (Table C.3 in the Supplemental material). The overall conclusion—that comfort and learning perceptions deteriorate with more men in the environment, especially for women—held true at each increment. There was no single arbitrary cutoff driving the results; rather, the relationship appears roughly monotonic (see Table C.3 in the Supplemental material).

Second, we considered whether our results might be influenced by how we dichotomized the Likert scales. We repeated the main regressions using the alternative binary outcomes (likely vs. not likely). The linear probability model results (Table C.1) are consistent with the OLS on Likert: For Comfort and Learn, the Female × SoM interaction is negative and significant, while for Penalize, it is null. For example, women in 75% or higher male environments are about 8 percentage points less likely than men to agree that mistakes are learning opportunities (p < 0.01), whereas in female-majority environments they are slightly more likely than men to agree (the difference crossing over at approximately 50% male). These magnitudes are in line with the Likert-scale associations when considering the 7-point range.

Finally, we performed a robustness check by re-estimating the main interaction models after removing potential outliers (see Table D.1 in the Supplemental material). Specifically, we dropped observations in the top or bottom 1% of the distributions of the following variables: each of the three outcome measures, SoM, Hours, Job Satisfaction, and Intention to Leave. This resulted in a slightly reduced sample (N = 1,330). As shown, the key coefficients of interest—Female, Share of Men, and Female × SoM—are qualitatively unchanged. For Mistakes-Comfort and Mistakes-Learn, the Female × SoM interaction remains negative, of similar magnitude, and statistically significant. For Mistakes-Penalize, the gender and interaction coefficients remain small and non-significant. In sum, the main conclusions are robust to excluding outliers; no single extreme observation is driving the gender-composition interaction coefficients.

In summary, our core findings are robust and general: (1) women are less comfortable admitting mistakes and less likely to view mistakes as learning opportunities compared to men, particularly in male-dominated contexts; (2) the share of men in one's workplace is inversely related to perceptions that mistakes are treated as learning opportunities and, for women, to comfort with admitting mistakes; men's perceptions vary comparatively little with workplace composition; and (3) there is no significant gender difference in expected penalization of mistakes, and workplace composition also does not significantly affect that outcome in our data. In the next section, we discuss the implications of these findings and provide interpretation and context, as well as consider limitations and potential avenues for future research.

5 Discussion

In this study, we set out to explore whether there are gender differences in perceptions of workplace mistakes and whether such differences depend on the gender composition of the work environment. We found clear evidence that women and men perceive workplace mistakes differently and that the context of gender composition plays a crucial role. To summarize, women reported significantly lower comfort admitting mistakes and were less likely to view mistakes as learning opportunities compared to men, and these gender differences were most pronounced in male-dominated workplaces. Women reported the most positive perceptions of mistakes (greater comfort, more learning opportunity) in environments with low shares of men, whereas their perceptions were markedly more negative in highly male-majority teams; men's perceptions, by contrast, varied comparatively little with workplace composition. Notably, we found no gender gap in expected formal penalization for a mistake: On average, men and women anticipated similar (moderate) consequences if an error occurred. In other words, women did not think they would be officially “punished” more than men for an error. However, women nonetheless felt less safe about mistakes and perceived fewer learning benefits in male-heavy environments, indicating a disparity in the informal or psychological domain of error responses.

As our sample is non-representative (81% female) and specific to the finance sector, these findings are not directly generalizable. Nonetheless, the study provides a novel proof of concept that gender differences in attitudes toward mistakes can emerge in a high-stakes, male-dominated environment. Our findings have implications for managers and organizations in male-dominated fields. These findings can be viewed through the lens of psychological safety. Psychological safety (the feeling that one can admit mistakes without fear) appears to be lower for women, particularly those in male-majority teams. In a heavily male environment, a woman might worry that admitting a mistake carries extra stigma or confirms negative stereotypes, leaving her significantly less “comfortable” doing so. This interpretation is supported by role congruity theory (Eagly and Karau, 2002): A woman in a male-typed role may indeed be judged more harshly for errors [as demonstrated by Brescoll et al. (2010)], and merely anticipating such judgment can erode her comfort. It also echoes Kanter's (1977) insights on “token” individuals (when women are a small minority, their slip-ups are highly visible and likely amplified under scrutiny). While our study cannot directly observe others' reactions to mistakes, the women's self-perceptions (feeling less safe and expecting fewer learning opportunities) are consistent with the possibility that they anticipate harsher judgment.

The absence of a gender difference in perceived penalization is noteworthy. At first glance, it may seem contradictory: If women feel more fear and less positivity about mistakes, why would they not also expect greater penalty? A possible explanation lies in how respondents interpreted “being penalized for the mistake.” This phrasing probably evokes formal repercussions (official discipline or career setbacks), which, in regulated professional settings, tend to follow standardized procedures for all employees. Indeed, both women and men may assume that a moderate error would result in similar formal consequences for anyone (e.g., a note in one's record), hence the reported parity on that item. In contrast, the comfort and learning measures tap into informal, psychological, and cultural dimensions of error response. It is in these informal domains that biases often emerge (e.g., subtle differences in support or stigma). Women might be more concerned about indirect penalties—for instance, being seen as less competent or having their mistake held against them in unwritten ways—which would not necessarily be labeled as formal “punishment” but would undermine their comfort and learning. Men, however, could be more confident that admitting a mistake will not alter others' perceptions of their competence, so they remain relatively at ease and open to learning from the error. In short, our female respondents may feel on “thinner ice” psychologically, even if they believe that, on paper, they will face the same official consequences as their male peers.

The results highlight the value of examining interaction effects between individual identity and organizational context. Prior studies often considered gender differences alone or overall workplace culture alone, but our findings demonstrate that the gender gap in mistake attitudes is context-dependent. It widens or narrows based on the environment's gender composition. In settings with a critical mass of women, the gap nearly disappears; where women are very scarce, the gap becomes highly pronounced. This pattern offers support to critical mass theory (Kanter, 1977; Konrad et al., 2008). When women constitute a sizable minority or majority, they likely feel more empowered to speak up and can help foster a safer culture. By contrast, when women are only tokens in an otherwise homogeneously male team, they may feel pressure to conform to a “no mistakes” norm, resulting in much lower psychological safety for them.

These findings carry implications for an organization's learning culture. Learning from failure is known to be vital for innovation (Cannon and Edmondson, 2005; Carmeli, 2007). If one group of employees (women, in this case) feels less able to glean lessons from mistakes due to a lack of psychological safety, the organization risks losing valuable learning opportunities. Women who are uncomfortable with errors might adopt defensive strategies (e.g., double-checking excessively or avoiding risky projects) to prevent mistakes. While understandable, such caution can hinder their learning and career growth, and it deprives the organization of some bold ideas or experiments it might otherwise pursue. Men, by contrast, may be more inclined to treat a mistake as “no big deal,” promptly extracting the lesson and moving on. Over time, these differences in response could quietly contribute to gender disparities in expertise, confidence, and even advancement. Importantly, this would not stem from any innate difference in ability but from differing opportunities to learn: If women have less psychological freedom to fail and learn, they may accumulate experience at a slower pace despite equal competence.

Our findings offer valuable input for organizations facing gender imbalance in their workforce. Promoting psychological safety should be a priority, and leaders need to signal that mistakes are treated as learning opportunities, not blunders to punish. For instance, managers (especially male leaders, who often set the tone in male-heavy teams) can model vulnerability by openly sharing their own mistakes and lessons learned. Such behavior helps normalize admitting errors for everyone. In addition, organizations can bolster mentorship and support networks for women. If female employees have mentors or peer groups (even outside their immediate team) where they can safely discuss mistakes and growth, it can buffer the negative impact of a harsh team culture. Establishing cross-team mentorship programs or forums for women can provide those outlets and reinforce an error-positive mindset.

Organizations should also consider training and policy adjustments. Unconscious bias training could be extended to cover error-related biases (e.g., ensuring managers do not judge a female analyst's mistake more harshly than a male colleague's). Performance evaluation systems might be tweaked to encourage smart risk-taking and normalize occasional mistakes for learning. For instance, self-evaluations can prompt employees to describe a failure and what they learned, signaling that well-intentioned errors are not career-ending. If women observe that learning from a mistake is explicitly valued (and not held against them), their comfort with taking risks and admitting errors may increase accordingly.

Importantly, a more inclusive “error culture” stands to benefit everyone, not only women. In our estimates, the negative compositional gradient is concentrated among women; for men, the only suggestion of such a pattern is on the learning outcome, where raw means are somewhat lower in the most male-dominated environments (Table 2) and the estimated slope is at most marginally significant. This suggests that extremely male-heavy, high-pressure teams may foster a general climate of fear or perfectionism that stifles growth for all. Therefore, improving psychological safety and promoting a growth mindset about mistakes (Edmondson, 2004) can enhance overall performance and innovation. In our data, men in more gender-balanced workplaces saw somewhat more learning potential in mistakes than their counterparts in very male-heavy contexts (4.71 vs. 4.54 on the 7-point scale; Table 2), while men's comfort levels were similar across contexts. Interventions that make the error climate more supportive could thus “raise all boats” while also narrowing the gender gap.

We acknowledge a few limitations. First, our data are cross-sectional self-reports about a hypothetical scenario. We captured perceptions at one time point and did not observe actual behavior or long-term outcomes after making real mistakes. This opens possibilities of response bias (e.g., perhaps some women overstated their discomfort out of frustration, or some men downplayed theirs due to norms against admitting fear). We mitigated framing effects by assuring anonymity and embedding these questions in a broader survey, but self-report bias remains a concern. Future research could address this by observing actual workplace mistakes, for instance, analyzing error logs or manager reactions to real errors, to see if gender differences manifest in practice. Second, our study cannot conclusively establish causality. We interpret the patterns as evidence that being in a male-dominated team causes women to feel less safe about mistakes, but we cannot rule out self-selection. It is possible that women who are particularly uncomfortable with mistakes tend to remain in (or end up in) more male-dominated teams, whereas those more confident about errors might seek out or stay in female-friendlier environments. We did control for job satisfaction and turnover intentions (and still saw the significant associations), which gives us some confidence it's not purely selection, but the causal arrow may well be bidirectional. A longitudinal design (tracking employees as the gender mix around them changes) would help unpack this. For example, if a given woman's perceptions improve after her team becomes more gender-balanced, that would strongly support a causal context effect.

Third, our findings' generalizability beyond the finance sector is uncertain. The financial industry is known for a high-pressure, low-tolerance culture (Madden, 2012), so gender differences in mistake perceptions might be especially pronounced there. Our respondents, drawn from a women-in-finance network, may also have above-average awareness of gender dynamics. We controlled for job role and subsector within finance, but ultimately, this study reflects the finance/professional services context. It remains an open question whether similar patterns occur in other fields. Early evidence suggests that they might: For example, studies in technology teams have found that female members report significantly lower psychological safety when they are outnumbered by men, compared to when they work with other women (Cole et al., 2023). These parallels hint that our results could extend to other male-dominated industries (tech, engineering, and certain medical specialties), but future research should confirm this. Investigating a range of sectors (including those with different gender balances or cultural norms) would strengthen the external validity of our conclusions.

Fourth, while our hypothetical scenario held the severity of the mistake constant, it neither fixed nor recorded the type of mistake respondents had in mind. Different respondents may have imagined different kinds of errors, and we cannot rule out that women and men systematically imagined different types, or that perceptions vary with mistake type. Future surveys could ask respondents to briefly describe the mistake they envisaged, allowing the responses to be coded into categories, compared across genders, and used to test whether mistake type moderates the patterns we document.

Fifth, and relatedly, our empirical design tests the organizational side of the framework in Section 2, the association between workplace gender composition and mistake perceptions, but not the individual-level mechanisms. Because the survey did not include validated measures of risk preferences, cognitive reflection, or rumination, those constructs motivate our hypotheses and inform interpretation rather than enter the analysis, and we cannot quantify how much of the joint imprint of context and disposition (Section 2.3) reflects each channel. Future work could field these instruments, for example, incentivized risk-elicitation tasks, the CRT, or the EOQ (Rybowiak et al., 1999), alongside our perception measures to decompose the gender differences we document.

Even with these caveats, our study offers new quantitative evidence on a nuanced topic and points to rich avenues for future research (such as experiments on manager reactions to mistakes, or longitudinal tracking of error-related behaviors and career outcomes). In sum, our findings show that gender and workplace context jointly shape how people perceive and learn from mistakes. In male-dominated environments, women appear to face a double burden: a generally harsher “error climate” plus an extra layer of identity-based pressure. Together, these factors leave them significantly less comfortable with mistakes and less able to treat errors as learning opportunities. Crucially, this is not just a women's issue. It reflects the broader organizational culture. By acknowledging these dynamics, employers can strive not only for greater gender equity but also for a more learning-oriented, resilient workplace for all. In short, to err is human, and creating a culture that truly embraces that fact will enable both women and men to grow and innovate to their full potential.

Generative AI statement

The author(s) declared that Generative AI was used in the creation of this manuscript. Generative AI was used in the last revisions to check references and proof read the text.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/forgp.2026.1619562/full#supplementary-material

Supplementary material Supplementary material at the source

Footnotes

Edited by: Lara Carminati, University of Twente, Netherlands

Reviewed by: Rizqi Nur'aini A'yuninnisa, Gadjah Mada University, Indonesia

Marcello Sartarelli, Complutense University of Madrid, Spain

Eleonora Clerici, University of Pavia, Italy