Equity, diversity, and inclusion in sports analytics

3.1 Analysis of demographic variables

Table 1 summarizes personal features. Among all respondents, 72.0 % were younger than 35 years old, 82.0 % identified as male, 89.7 % identified as heterosexual and 69.5 % identified as White (excluding mixed race). One third (33.0 %) of respondents possessed all these features, representing the largest cohort among all combinations of these variables. In addition, 5.2 % of respondents reported having at least one disability (e.g., mental-health, vision, hearing, mobility, etc.). Nearly three quarters of respondents had either a Bachelor’s or Master’s degree (75.6 %) as their highest degree, and nearly 20 % of respondents had a Doctoral degree. Respondents hailed from many different disciplines. Statistics, mathematics, engineering and business management were the primary fields of study for 60.9 % of respondents. Sports-related fields such as kinesiology, sport science and sport medicine were less prevalent (10.3 %). Fields of study in the “Other” category largely consisted of social sciences and humanities.

Table 2 summarizes job-related features of the respondents. The majority of respondents worked for professional sports teams (54.1 %). The salary distribution (Figure 1) ranged from “volunteer”, which we assumed to mean no salary, to >$200 K. The distribution was fairly symmetric between $0 and $200 K, with 53.2 % of all respondents reporting a salary in the range $50 K–124.9 K. However, the overall distribution had a longer right tail, with 6.9 % of respondents reporting a salary above $200 K. Over half of the respondents (51.5 %) reported being a part of an Analytics department, with over 80 % of those individuals working for a professional sports team or sport tech company. Kinesiology/sports departments were more prevalent than analytics departments at the amateur and college sport level. The two most common job levels were “Staff” and “Director”, together comprising 61.3 % of respondents.

Figure 1:

The histogram of salary for all respondents.

To facilitate interpretation of the results with a more homogenous sample, the following analyses focused on respondents from professional sports (n = 126), which was the most prevalent organization type. We opt to limit our data sample here because across organizations, characteristics such as salary and job title may not be standardized. For example, the role and responsibility of a “staff” member may differ significantly between a professional sports organization and a sport entertainment organization. However, for completeness we repeat the following analysis with the entire survey sample in Appendix A, which reveals similar results.

Returning to just respondents from professional sports, Figure 2 shows that as age increases, the proportion of female and racialized employees decreases. The majority of respondents (64.3 %) were in the 25–34 age range. On the other hand, the 40+ category only had 10 respondents (<1 %).

Figure 2:

The percentage of respondents in professional sports, in each age category, broken down by sex and race. The size of each dot represents the relative number of respondents.

Figure 3 shows that there is a consistent imbalance in sex and race, except at the entry-intern level. We note that the x-axis is not a true continuous variable, however, does follow a trend of increasingly more senior positions. We see that at each non-intern job level, males comprise 85–100 % of respondents and White respondents comprise 66–85 % of all respondents. Similar to the previous cross-tabulation with age, some categories had very few observations. For example, four respondents indicated “Intern”, six indicated “Researcher”, and seven indicated “Executive”. Acknowledging that there were very few respondents in the “Intern” category, this was the only job level where racialized individuals were in the majority (75 %) and where there was parity in sex.

Figure 3:

The percentage of respondents in professional sports at each job level category, broken down by sex and race. The size of each dot represents the relative number of respondents.

Next, we examined how salary differed by sex and race. Due to small sample sizes for individual job levels, we grouped respondents into either a management (i.e., manager/supervisor, director or executive) or non-management category (Table 3). Following statistical best practices when working with interval data, we took the midpoint of each salary bucket and computed the mean and standard deviation of the salary distribution for the management and non-management categories, stratified by sex and race. For the salary bucket “>$200 K”, we assumed a value of $250 K. We excluded respondents who selected “prefer not to answer”. Given these modeling choices, Table 3 shows that males earned more than females at both the management (on average approximately $30 K or 27 % more) and non-management levels (on average approximately $15 K or 23 % more). Similarly, White respondents earned more than racialized respondents at the management (on average approximately $17 K or 14 %) and non-management levels (on average approximately $5 K or 7 % more).

We acknowledge that our stated salary gaps may be subject to some level of uncertainty due to the absence of precise salary data. However, Appendix B provides an analysis that demonstrates the robustness of our results in this section, regardless of any distributional assumptions made about the data. Additionally, our analysis shows that it is a zero-probability event that females earn more than males, on average.

The “Management” and “Non-management” categories in Table 3 each consist of three distinct job levels, which may have different salary distributions. To further explore differences in salary by sex and race, we isolated the two most populous job levels within “Management” and “Non-management”, namely, “Directors” and “Staff”, respectively. Table 4 shows analogous results to Table 3 but focusing on these two specific job levels. Conditioned on the same job level, the salary gap persists between male and female respondents, although the gap was smaller. The difference in salary for “Director” was approximately $21 K or 17.8 %, while the difference in salary for “Staff” was approximately $6 K or 7.1 %. When comparing White to racialized respondents, the gap essentially disappeared for “Director”, with a salary difference of less than $2 K on average. Interestingly, for “Staff”, the relationship flipped: racialized respondents earned $10 K more than White respondents on average. This reversal is due to the removal of several racialized “Interns”, “Consultants” and “Researchers” (i.e., non-management personnel who were not “Staff” members), who collectively had a low average salary of approximately $37.5 K.

For every comparison in Tables 3 and 4, we checked if the difference in mean salaries were statistically significant. Our null hypothesis was that the mean salary amongst each of these pairs were equal, and the alternative hypothesis was that the salary of the non-dominant group was lower than that of the dominant group. We performed the Mann-Whitney U test and the permutation test with an alpha value of 0.05, and both tests failed to reject the null hypothesis for all comparisons (even when correcting for multiple testing using the Bonferroni correction method). We believe that these tests were underpowered given the few female and racialized respondents – especially in senior roles.

3.2 Experiences with respect to EDI

For the subsequent analysis in this section, as well as the remainder of the paper, we include all 233 respondents, not just those individuals who worked in professional sports.

3.2.1 Discrimination

Over one in eight (12.9 %) respondents experienced some form of discrimination in the last 12 months (Figure 4). Females experienced discrimination nearly five times more than males (38.2 % vs. 8.1 %). Racialized respondents experienced discrimination at a similar proportion to White respondents (11.3 % vs. 11.5 %). Conditional on experiencing discrimination, the distributions of the number of discriminatory events, the types of discrimination, and the number of different types of discrimination an individual experienced are shown in Figure 5a), b), and c)), b), and c), respectively. A third of these individuals experienced discrimination five or more times. The most common types of discrimination were sex (23.2 %) and racial/ethnic (23.2 %). Eleven out of 13 respondents who indicated sexual discrimination were females (one was male and the last respondent did not disclose their sex). Six out of 13 respondents who indicated racial/ethnic discrimination were racialized respondents (six were White and the last respondent did not disclose their race). Our sample consisted of 35 females, 191 males, 69 racialized individuals and 162 White individuals. Therefore, the percentage of females and racialized individuals experiencing sexual and racial/ethnic discrimination (31.4 % and 8.7 %) was much higher than the corresponding percentage of males and White individuals (0.5 % and 3.7 %), respectively. Respondents who experienced discrimination experienced between one and five types of discrimination, with 60 % of them experiencing two or more types.

Figure 4:

The percentage of respondents who experienced some form of discrimination in the last 12 months, stratified by sex and race. Note, seven respondents did not indicate their sex, and eight respondents did not include their race.

Figure 5:

Breakdown of discriminatory events.

3.2.2 Workplace satisfaction

The majority (58.2 %) of respondents indicated that they were satisfied or very satisfied with the overall climate of EDI in their workplace. However, there were notable differences when stratifying by sex and race. In particular, females were more than four times as likely to report being “very dissatisfied” (11.4 % vs. 2.7 %) and 0.6 times as likely to report being “very satisfied” (11.4 % vs. 20.5 %). In contrast, racialized respondents were three times as likely to report being “very dissatisfied” (7.5 % vs. 2.5 %) but also one and a half times as likely to report being “very satisfied” (26.9 % vs. 16.4 %). In other words, females were consistently less satisfied by the EDI climate at their workplace, while racialized respondents’ satisfaction was more polarized. The results below provide more insight into these findings.

3.2.2.1 Workplace satisfaction by sex

Across all the positive workplace descriptors shown in Figure 6, females indicated “strongly agree” far less often than males. For example, 5.9 % of females strongly agreed that their workplace was “non-homo/trans/queer phobic” compared to 33.7 % of males. Similar but smaller gaps were seen for other descriptors such as “welcoming”, “supportive”, “non-sexist” and “respectful”. Interestingly where males and females responded most similarly was to the descriptor “diverse” – both sexes generally thought that their workplaces were not diverse.

Figure 6:

Responses to the following question, stratified by sex. “How you would rate your current (primary) work environment/experiences on the following items (prior to COVID) (response options: strongly agree – agree – neutral – disagree – strongly disagree)”.

Figure 7 highlights the respondents’ experiences in their particular department, stratified by sex. Interestingly, the first two statements show that males and females have similar levels of agreement regarding fair processes for rewards and compensation. However, females “strongly agreed” less often than males regarding equitable support, equitable workload, having a voice, and being taken seriously.

Figure 7:

Responses to the following question, stratified by sex. “Considering your experiences over the 12 months prior to COVID, please indicate your level of agreement with each of the following statements for your “primary work unit/department.” (response options: strongly agree – agree – neutral – disagree – strongly disagree)”.

Figure 8 highlights the respondents’ experiences in their specific job or role, stratified by sex. Note that some of these statements were written from a “negative” perspective. Females generally reported more negative experiences in their job. For example, females were twice as likely (i.e., “agree” or “strongly agree”) as males to have considered leaving their job due to isolation or feeling unwelcome (35.3 % vs. 17.4 %). When restricted to respondents who “strongly agree” with this statement, females were nearly four times as likely to consider leaving compared to males (14.7 % vs. 3.8 %). For another example, 36.3 % of females strongly agreed that they have to work harder than others to be valued equally, compared to 9.8 % of males. Males and females answered similarly when it came to questions around fairness of promotions, job performance evaluation and compensation. Looking at all questions from their “positive” perspective, the one that both males and females disagreed with the most is that their employer “provides sufficient programs to foster the success of a diverse staff.” Less than 50 % of respondents from both sexes agreed with this statement, although females disagreed more strongly. This result complements the finding from Figure 6, where both sexes generally did not view their workplace as diverse.

Figure 8:

Responses to the following question, stratified by sex. “Considering your experiences over the 12 months prior to COVID, please indicate your level of agreement with each of the following statements for your current (primary) job (response options: strongly agree – agree – neutral – disagree – strongly disagree)”.

3.2.2.2 Workplace satisfaction by race

Across all the positive workplace descriptors shown in Figure 9, racialized and White respondents shared similar sentiments, with racialized respondents being more positive across the board. We find similar trends in Figures 9 and 10, when examining experiences in the respondents’ department and specific job, respectively. As an example, consider the statement “Within the organization, everyone has access to equal employment opportunities” in Figure 11. The percentage of racialized and White respondents who agreed or strongly agreed with this statement was very similar (62.3 % vs. 59.4 %). Likewise, the percentage of those who disagreed or strongly disagreed was 13.1 % versus 14.8 %. The largest difference in responses appeared to be with respect to the workplace descriptor “Diverse” in Figure 9. Over half (51.6 %) of racialized respondents agreed that their workplace was diverse, compared to 27.1 % of White respondents.

Figure 9:

Responses to the following question, stratified by race. “How you would rate your current (primary) work environment/experiences on the following items (prior to COVID) (response options: strongly agree – agree – neutral – disagree – strongly disagree)”.

Figure 10:

Responses to the following question, stratified by race. “Considering your experiences over the 12 months prior to COVID, please indicate your level of agreement with each of the following statements for your “primary work unit/department.” (response options: strongly agree – agree – neutral – disagree – strongly disagree)”.

Figure 11:

Responses to the following question, stratified by race. “Considering your experiences over the 12 months prior to COVID, please indicate your level of agreement with each of the following statements for your current (primary) job (response options: strongly agree – agree – neutral – disagree – strongly disagree)”.

To better understand the findings in Figures 8–10 (i.e., that racialized respondents seem to have slightly higher levels of satisfaction than White respondents), we conducted further statistical tests. We begin with the statement “I have a voice in decision making.” The categories “strongly disagree”, “disagree” and “neutral” were collapesed into a single category called “disagree”, and “agree”, “strongly agree” were collapsed into a single category called “agree”. This results in a binary response outcome for the statement. The odds ratio of disagreement for White versus Racialized is 3.2 (p = 0.02, 95 % CI: 1.2–11.2). Adjusting this odds ratio for sex in a logistic regression model yielded a similar odds ratio for White versus racialized, and sex was also statistically significant in the model (OR = 3.0, p = 0.01, 95 % CI: 1.3–6.9). The odds ratio of disagreement for males and females are 4 (p = 0.02, 95 % CI: 1.1–21.8) and 1.5 (p = 1, 95 % CI: 0.2–18.0) respectively, but only seven of the 34 females in the sample are racialized. The difference between the odds ratios for males and females is not statistically significant (p = 0.37) which indicates that the interaction between sex and race is not statistically significant.

We then repeated the above analysis for other positive statements found in Figures 8–10 (e.g., “Support is provided fairly and equitably”) and the results varied. That is, for certain statements White respondents were more positive, for others they were more negative and lastly for some, no statistically significant difference existed. As such, unlike workplace satisfaction by sex, there is no clear pattern across all experiences of EDI such that White or racialized respondents were more positive or negative.

3.3 Impact of COVID-19

The final two questions in the survey addressed the respondent’s workplace experiences since the onset of the COVID-19 pandemic. With many organizations needing to pivot their operations to the virtual environment, the aim was to discover how pandemic-induced changes impacted the respondents. Interestingly, females reported more positive workplace changes, but lower levels of satisfaction, due to the pandemic (Figure 12).

Figure 12:

The changes in the workplace and satisfaction levels since COVID, stratified by sex.

Figure 13 presents similar results, stratified by race. Both White and racialized respondents similarly reported mostly reported neutral or positive changes due to COVID. However, racialized respondents reported higher satisfaction levels compared to White respondents. For both Figures 12 and 13, more research and granular questions are needed to determine the reasons for these differences in satisfaction level.

Figure 13:

The changes in the workplace and satisfaction levels since COVID, stratified by race.

4.1 Interpreting the results

Firstly, from the individual demographic variables, we recognize that there are clear dominant groups identified within our survey sample – White, heterosexual, male, young (between the ages of 25–34). Given similar anecdotal evidence, we believe our study’s dominant group extends to the sports analytics industry. The proportions of these groups are heavily skewed, even when compared to other STEM-based industries according to a study by the Pew Research Center (Funk and Parker 2018). For example, across STEM-based jobs, females make up roughly 50 % of the population, compared to only 15 % of respondents in our sports analytics survey. Similarly, Black employees hold roughly 9 % of STEM jobs, compared to only 1.7 % in our sample. Thus, our survey suggests that sports analytics is critically lacking in diversity, even in comparison to STEM in general.

In contrast, when it comes to advanced education degree holders, our study suggests that people working in sports analytics had much higher representation compared to general STEM-based jobs. Indeed, across STEM-based jobs, approximately 29 % of workers hold a postgraduate degree (Funk and Parker 2018). In our sample, it is nearly double that percentage (57.5 %). Although diversity within STEM education has been steadily increasing, there is a still a gap when it comes to racialized or marginalized communities pursuing advanced degrees, which may help explain the above findings. Within respondents who held a doctorate, 37.0 % worked in a traditional academic organization and half of them (50.0 %) worked for professional sports or sports tech companies. Within the advanced degree holders in our survey, there is a much greater prevalence of technical degrees such as statistics, computer science, math and engineering (e.g., 73.9 % of doctorate degrees) compared to more traditional sport-related degrees, such as kinesiology or sport medicine (e.g., 15.2 % of doctorate degrees). This reinforces the perception of sports analytics as a field that favors analytics-based personnel with specialized and focused technical skills, over individuals trained in the sport sciences or business/management.

Within professional sports, our data suggests that there are clear imbalances in representation by sex across different sports analytics job levels, with a much higher proportion of males holding senior positions. Indeed, within our survey the only job level in which there is parity between sexes is at the intern level. Studies of the workforce at large find a similar result (Burns et al. 2021), namely, that as positions become more senior, there is less female representation. Our survey responses also indicate a salary gap between male and female respondents. A 2020 study by the Pew Research Center shows that females earn 84 % of what males earn when analyzing median salaries of full-time and part-time workers in the United States (Barrosa and Brown 2021). Our results were similar. In our sample, females in management-level jobs earned 78 % of what males earned, while females in non-management jobs earned 82 % of what males earned. Similar gaps in representation and salary were observed, with White respondents having higher representation and higher salaries, though the gaps were less pronounced. We acknowledge that none of the differences in mean salaries were statistically significant, however, we believe that these tests were likely underpowered given the few female and racialized respondents – especially in senior roles.

Within our analysis of discrimination, we observed that an equal number of White and racialized respondents experienced racial discrimination. However, this translated to 3.7 % and 8.7 % of White and racialized respondents, respectively. We acknowledge that within the literature, it has been shown that these two groups experience racial discrimination differently. Due to the smaller and heterogenous sample of racialized respondents, we cannot make any generalizations from this result. Moreover, the survey relied on each respondent to answer questions regarding discrimination with their own definitions of what constituted as discriminatory. We hope future studies can provide an education or explanatory component to help all respondents arrive at the same definitions of these terms.

We also comment on our finding that when an individual experienced discrimination, they received anywhere between one to five unique types of discrimination, with an average of two. This highlights an important notion of intersectionality. Intersectionality is a way of explaining complexities in the world and people by viewing the characteristics of an individual (i.e., race, class, gender, etc.) as interrelated and mutually shaping one another (Collins and Bilge 2020). Essentially, inequities are not the result of a single factor, but rather the outcome of multiple different factors that are not simply additive in nature. The presence of individuals experiencing multiple forms of discrimination provides evidence of the existence of intersectionality within our study sample. We hope future studies can more explicitly gather insights into how intersectionality is experienced by those in the sports analytics industry.

In terms of experiences with respect to EDI, there are clear positive advantages afforded to male respondents in our survey that female respondents do not receive. For example, females were nearly five times as likely to experience discrimination compared to males (38.2 % vs. 8.1 %). A recent study across all STEM-based jobs reported a similar finding, but focused on gender, instead of sex: 50 % of women and 19 % of men experienced gender discrimination (Funk and Parker 2018). In general, female respondents reported lower workplace satisfaction than males, for example with respect to having a voice in decision-making, having to work harder to be valued equally, and working in a welcoming and supportive workplace. Interestingly, despite the fact that females earned less than males in our sample, they reported similar levels of satisfaction with respect to fair and equitable processes for determining compensation.

In contrast to the stratification by sex, the responses about workplace EDI experience when stratified by race were quite similar. Our results suggest that racialized respondents had slightly more positive experiences than their White colleagues. However, when conducting further statistical tests (i.e., Fisher’s test, logistic regression), we find that the results varied for different statements, sometimes indicating that White respondents were more positive, and other times more negative (and for certain statements, no differences emerged). These results are at times at odds with the general literature which demonstrates clear evidence that racialized communities have more negative experiences in the workplace (Smith and Calasanti 2005). We note a few limitations of our data that may contribute to this finding. First, we have a relatively small and imbalanced dataset. In our sample, White respondents outweighed racialized respondents by a 2.3:1 ratio. We also note that comparisons with respect to race found in the literature are primarily done between White individuals and individuals from a specific racialized community (i.e., Black, Latinx, Indigenous, etc.). We acknowledge that this analysis may be challenging to conduct if there are relatively few racialized persons currently employed in sports analytics. Lastly, our sample of racialized respondents is very heterogeneous, which renders any generalizations about this group’s experiences difficult and subject to added noise. Overall, more research is needed to understand if there is a clear difference in EDI experience across racial communities in sports analytics.

We also note that some changes occurred as a result of the COVID-19 pandemic. For example, females reported more positive workplace changes, but lower satisfaction levels compared to males. We cannot conclusively comment on what is driving this trend, but we conjecture that although widespread work-from-home practices provided greater flexibility (a positive change), it ultimately led to greater amounts of office-work and housework for females, leading to more burnout and lower satisfaction. This has been found in the literature (e.g., Burns et al. 2021; Dunatchik et al. 2021; Lyttelton, Zang, and Musick 2020), but we note that we cannot confirm if this phenomenon was present amongst our female survey respondents.

4.2 Limitations

This study has several limitations. While we made every effort to share our survey as widely as possible, we could not confirm which particular sports, teams or personnel received notice of the study. We suspect the sports analytics population is much larger than our survey sample, and it is unclear if the respondents are representative of the overall population. We hope the preliminary results presented in this paper lead to additional, larger-scale studies of EDI in the field. In particular, future studies should include more qualitative data collection to obtain richer details on individuals’ EDI experiences.

Many of our analyses were affected by small sample sizes. Results for groups with small samples were either omitted or are limited in their robustness. It is unclear if these small numbers are due to a low response rate or point to the fact that these groups are under-represented in the sports analytics population. Moreover, the comparison of responses between groups may also be affected by the underlying differences between the groups, rather than a true difference in their responses. Salary comparisons were restricted to means, rather than medians, because our large salary buckets coupled with a small sample would have made comparisons challenging.

We also acknowledge that completion of the survey was voluntary, and as such, may not have attracted a representative sample. For example, the survey may be biased with respondents who simply had the time to fill out the survey. Alternatively, it may have attracted respondents who felt more strongly about EDI issues than the average individual. Moreover, respondents were expected to interpret the phrasing and terminology of the questions based on their own judgment. Given that the definitions around equity, diversity and inclusion are continuing to be defined, this may have impacted some of the survey results.