The effect of different-race presenter on bias reduction in online diversity training

Diversity training, as is true of most training and education, has been offered online increasingly throughout the twenty-first century (Anand and Winters, 2008). The current study seeks to inform the effective delivery of online diversity training to reduce racial bias. Much research has been conducted to determine best practices in diversity training. Research has shown that the race of the presenter impacts the effectiveness of diversity training in the reduction of bias in in-person and interaction online (Peck et al., 2013). The current study seeks to determine if the same is true in single-session online diversity training.

The US workforce and student populations are more diverse than ever (Jensen, et al., 2021). Diversity and race-based bias are important educational, political, social and business considerations. Research shows there is a persistent gap in pay and promotion (Fisher and Houseworth, 2017; Webber and Canché, 2015; Green and Ferber, 2005; Collins et al., 2017; Myran and Ivery, 2013) and in educational attainment (Devine-Eller, 2012; Mineo, 2021) between racial groups in the USA. Though the education and pay gap has closed some in recent years, African Americans are still more likely to be affected by recessions and to lag in wealth-building than other races (Dias, 2021; Collins et al., 2017; Robinson, 2011; FastCompany, 2021). The existence of this gap may be unavoidable, but the causes are contested. Among those factors are racism and bias of individuals as well as systemic bias built into colleges and employers (Alon, 2015; Rubio, 2001; Mineo, 2021). Because the problem is so complex, the solution is not likely to be simple. However, there are a few things that are effective in reducing interracial bias, one of the major factors thought to be perpetuating the gap, including affirmative action, diversity and inclusion initiatives (Bagenstos, 2006; Bowen et al., 1999). It is incumbent upon a democratic, egalitarian society such as the USA claims to be to reduce bias among business students in Higher Education and among employees in the workplace to truly afford equal opportunity to all of its citizens and residents. This study aims to determine if one small adjustment might help to close this race-based pay and promotion gap, bringing the laudable goal of equal opportunity closer to fruition.

Business is and has been the most popular undergraduate major in the USA, with nearly 20% of undergraduate degrees granted in business during the 2018–2019 academic year (Department of Education, 2020). Business majors enter the workforce as entry-level employees where many are promoted into management positions. Managers regularly make hiring, promotion and pay decisions. Each of these decisions is prone to bias on the part of managers, whether intentional or unintentional. At some level of advancement, managers also begin to have input into the internal policies of organizations making Diversity Training for managers and potential managers that much more important (Cho et al., 2017). Students have been found to be an acceptable proxy for the study of Diversity Training in employment (Kalinoski et al., 2013; Highhouse and Gillespie, 2009). For these reasons, business students are particularly situated in society to affect the race-based pay and promotion gap and to represent employees as participants in the study.

Many efforts have been made to close this gap. Among those efforts have been Affirmative Action, diversity, and inclusion initiatives within both employers and higher education institutions. In the contemporary educational and employment landscape, the main effort continues to be diversity initiatives, including diversity training for students, employees, hiring managers and faculty. Diversity training, as is true of most training and education, has been offered online more and more throughout the twenty-first century (Anand and Winters, A retrospective view of corporate diversity training from 1964 to the present, 2008). The current study seeks to inform the effective delivery of online diversity training to reduce racial bias. Much research has been conducted to determine best practices in diversity training. Research has shown that the race of the presenter affects the effectiveness of diversity training in the reduction of bias in person and interaction online with different-race others been shown to reduce bias (Peck et al., 2013). The current study seeks to determine if the same is true in online diversity training.

Bias and methods of reducing bias are well researched. Racial bias has been studied for decades and several measures have been developed and used to quantify changes in bias. One of the most-used (Chopik et al., 2015; Henry and Sears, 2002; Lambert et al., 1996; Morrison and Kiss, 2017) and trusted (Aosved et al., 2009) instruments to measure bias against African Americans, the Modern Racism Scale (MRS) (McConahay, 1986), was used in this study. The MRS is a Likert type survey which asks respondents to rate their agreement with several statements about views toward African Americans. Among methods of reducing bias through Diversity Training, Intergroup Social Contact Theory is well-researched (Bezrukova et al., 2016). That research began and has continued to be largely focused on the study of the reduction of bias between racial and ethnic groups, but has also been applied to other intergroup situations. Many researchers have found merit in and called for the further use of Intergroup Social Contact in Diversity Training (Bezrukova et al., 2012; Guy, 2019; Paluck, 2006; Stewart and Peal, 2001). Lessons learned in the study of Intergroup Social Contact should inform the development of a diversity program that minimizes the possibility of negative contact between groups and maximize the bias-reducing benefits of positive intergroup contact.

Diversity training is an $8bn industry in the USA, and most U.S. Fortune 500 firms provide diversity training, though they are not sure it works (Chang et al., 2019). Diversity training has been defined as “a distinct set of instructional programs aimed at facilitating positive intergroup interactions, reducing prejudice and discrimination, and enhancing the skills, knowledge, and motivation of participants to interact with diverse others” (Bezrukova et al., 2016, 1228) in the landmark meta-analysis of 40 years of research. Roosevelt Thomas, a pioneer in Diversity research, said that diversity training “should equip the participants to make quality decisions in the midst of differences, similarities, and the related tensions and complexities” (cited in C. D. Johnson 2008). Diversity training is important to academics and practitioners, but a gap between attention given to particular topics in Higher Education management and in management trade journals and practice by employers has been noted (Gibson and Deadrick, 2004; King et al., 2010). Academic research is generally careful to measure the effectiveness of diversity training while companies often miss this important step (Patricia and Ash, 2022). Firms, especially smaller firms, do not carefully determine the need for diversity training related to specific areas of diversity (Pendry et al., 2007; Roberson et al., 2009) and fail to assess effectiveness (Chrobot-Mason and Quinones, 2002). Firms, instead, tend to apply a less effective one-size-fits-all approach (Homan et al., 2015). Firms often use less effective single-session training instead of a more effective curriculum of multiple training sessions (Kormanik and Rajan, 2010; Shaw, 2019; Snibbe, 2007). Furthermore, firms that use single-session training are more likely to perform that training online, which has proven to be less effective than single-session in-person training (Roberson et al., 2009). The use of the internet to provide diversity training is especially important, as more firms continue to move to online diversity training (Brown, 2021; Shaw, 2019). However, many firms choose to continue single-session and online training even if there is no evidence that it is effective (Nathoo, 2021).

Based on the situation revealed in this literature review, most research is focused on improving multisession training programs, which will most likely affect large companies with significant resources if best practices discovered in academic research are followed. The current research approaches the dilemma of continued use of single-session and online diversity training and its relative ineffectiveness by attempting to improve the method that seems to be more likely to affect small businesses, which employ almost 50% of the US workforce (Small Business Administration 2020). While the majority of large firms offer diversity training, less than 25% of small businesses offer any diversity training (Berk, 2017).

One cautionary point discovered in the literature is the risk of reinforcing race-based social hierarchies while exposing them during diversity education and training. Pointing out differences between groups without sufficiently discussing similarities is a risk that must be considered when designing interactions (Amoroso et al., 2010). Organizations with poorly designed and executed diversity training risk litigation (Weaver, 2021) and other types of backlash (Watson, 2008; Whittenburg et al., 2001). However, the long-term benefits do generally outweigh the short-term consequences for firms (Richard et al., 2007). Discussing differences can result in a reduction of bias (Chien and Atwell Seate, 2017).

Education and the workplace intersect most fully in higher education. Therefore, this seemed to be the most appropriate place to explore the use of Inter Group Social Contact Theory in advancing diversity. Wallace and Allen (2016) found that introductory college level business courses covered Affirmative Action well, noting:

However, management courses tend to offer Affirmative Action, Equal Opportunity and Nondiscrimination as legal Human Resource issues and Diversity as a social imperative, with the bonus of contributing to the bottom line (Bierema, 2010). For these reasons, the researcher chose to use a professionally produced diversity training video that clearly relied on the equality of opportunity view (TrainingABC). Combining the Intergroup Contact Theory discussed in this work with classroom education should provide more effective treatment of these topics in the college business classroom and in corporate training.

While most studies have found diversity training to be effective, some have found it to be ineffective, causing a shift in research to determine situations and methods that are most likely to result in success. Richard et al. (2007) found that Diversity Training that theoretically should be effective was more likely to be effective in certain types of companies, such as companies flush with resources, usually large companies. Ironically, even firms that do not think diversity training works tend to continue to use diversity training out of fear of increased litigation if they discontinue its use (Scorzetti, 2019) Many companies do use a combination of methods, (Patricia and Ash, 2022), but usually fail to accurately assess their effectiveness (Pendry et al., 2007).

Using lessons learned from Intergroup Social Contact theory (Pettigrew and Tropp, 2006), the current study proposed that the use of simulated intergroup (in this case inter-racial) interactions can effectively complement diversity education (Guy, 2019) with single-session diversity training through the use of a different-race presenter avatar. Institutes of higher education seem to be doing a sufficient job of teaching students the motives and methods of promoting diversity and of affirmative action (Wallace and Allen, 2016). This research explores a method of adding a component to their education which allows them to virtually encounter those who are different from themselves in the context of online single-session diversity training. This addition of Inter Group Social Contact should better prepare them to react appropriately in workplace situations by reducing bias toward members of that group.

The research question for the current study is: Does the use of a different-race presenter avatar in single-session online diversity training increase the effectiveness of the training in reducing bias toward African Americans?

H1: There is a difference between the mean MRS score of respondents receiving same-race and different-race presenter avatars.

H2: There is a difference between the mean MRS score of respondents receiving a same-race presenter avatar and the mean MRS score of respondents receiving a different-race presenter avatar within one or more self-reported racial group of respondents.

H3: There is a difference between the mean MRS score of respondents receiving different-race presenter avatar and the mean MRS score of respondents receiving a same-race presenter avatar within one or both gender groups.

H4: There is a difference between the mean MRS score of respondents receiving a different-race presenter avatar and the mean MRS score of respondents receiving a same-race presenter avatar within one or more schools based on the diversity of the school’s student population.

H5: School student body diversity does interact with the relationship between the use of same-race versus different-race presenter avatar and MRS scores.

Undergraduate students are a commonly studied group in this type of research (Highhouse and Gillespie, 2009; Kalinoski et al., 2013). To study this change, nine colleges have allowed their business students to be invited to participate as human subjects. Upon institutional approval, business faculty at these colleges invited their students to participate by sharing a link to the online diversity training and the subsequent MRS questionnaire. The measurement of bias was compared between groups using paired two-tailed two-sample t-tests and an analysis of co-variance (ANCOVA). Statistically significant results would indicate that the use of a different-race presenter in online diversity training impacts bias reduction.

In this study, the independent variables include the use of same-race and different-race presenter avatar, self-reported race and self-reported gender, and the diversity of the individual school attended by the participants. The dependent variable considered was the groups’ mean MRS score.

The design of this research was experimental in nature, as respondents received training under different circumstances and the results of those different groups were compared. The research was conducted via a password-protected website built specifically for the research. The website contained the required disclosure statement with opt-out, age verification which only allowed respondents reporting they were 18 year of age or older to continue, demographic questions including race, gender and school, one of two diversity training videos (one with a light-skinned presenter avatar and one with a dark-skinned presenter avatar), and the Modern Racism Scale (MRS) questionnaire. The researcher received permission from a professional provider of training materials to use their diversity training video (TrainingABC) and then added the presenter avatars throughout the video using Screen-Cast-O-Matic. This video training represents the minimum training a small business might provide through an online vendor.

Institutional Review Board (or equivalent) approval was granted by each school. Among these nine schools, faculty invited 1,092 students via school-owned email to participate. No individually identifiable information was collected from respondents. The participant sample was taken from the business school populations using a modified block design that is appropriate when samples cannot be taken randomly within populations (Hair et al., 2010). Within each business school’s population, it was impossible for the researcher to compel a randomly selected group of students to participate. Instead, the students were invited to participate and all complete responses were used as a convenience sample of the respondents’ constituent groups based on the independent variables.

The scope of the research was limited by operational definitions of several important terms. Race was self-reported by respondents. Racial bias was determined based on the respondents’ scores on the Modern Racism Scale (MRS) (McConahay, 1986), an explicit bias scale which has been used in similar research for decades (Chopik et al., 2015; Henry and Sears, 2002; Lambert et al., 1996; Morrison and Kiss, 2017) and has been confirmed by factor analysis, outperforming most surveys created since (Aosved et al., 2009). The MRS asks respondents to record their agreement or disagreement with six statements of attitudes toward African Americans using a Likert type scale (McConahay, 1986). Five of these questions are positively scored and one is negatively scored. Different-race presenter avatar is defined as a respondent who self-reported as white/Caucasian receiving the dark-skinned presenter avatar and a respondent who self-reported as a race other than white/Caucasian (including African American, Latino/Hispanic or other) receiving the light-skinned presenter avatar. Alternately, same-race presenter avatar is defined as a respondent who reported as white/Caucasian receiving the light-skinned presenter avatar and a respondent who self-reported as a race other than white/Caucasian (including Latino/Hispanic, or other) receiving the dark-skinned presenter avatar. Gender was self-reported by respondents. School of attendance was self-reported by respondents who chose among a list of the nine participating schools. Diversity in each school was defined by the publicly reported percent non-white/minority enrollment at each school (College Factual, 2022). The independent variables of different-race presenter avatar, self-reported race, and self-reported gender were categorical variables. The independent variable of diversity of the individual schools was a continuous variable. The dependent variable of the groups’ mean score on the Modern Racism Scale (MRS) and was a continuous variable.

For each of the hypotheses one through four, the data were evaluated for normality to determine whether the parametric or non-parametric methods should be used (Hair et al., 2010). Once normality of the data was established, the data were tested for equal variances to determine if the t-test assuming equal variance or the t-test assuming unequal variance should be used. The t-test is an appropriate test for convenience samples where the representation of the population is unknown and equal sample sizes cannot be expected (Hair et al., 2010). For the fifth hypothesis, an ANCOVA was used. The ANCOVA is the appropriate test to determine the effect of a continuous independent variable on the relationship between a discrete independent variable and a continuous dependent variable (Hair et al., 2010).

This study has added to the body of knowledge in the areas of Intergroup Social Contact and Diversity Training research, but is limited in a few ways; test subject population, convenience sampling, sample size and scope. The subject group was undergraduate business students studying in institutions of higher learning in the USA and the application of the study is not limited to that group, but is also generalized to employees within organizations working in the USA. Undergraduate students are considered a good proxy for full-time employees and are more accessible to researchers working in higher education. Participants were accessed through their professors within the studied institutions and response was voluntary, so in the modified block methodology, responses represented a convenience sample rather than random sampling. The sample size was fairly small at just under 100 results. However, the statistics used take sample size into account in determining significance, so the researcher is confident in the significance of these results.

Most of the responses were received from students at institutions where the researcher was familiar with the faculty member inviting students to participate. In this study, the research was limited to the effectiveness of the test variable or presenter race on reduction in racial bias. While there are many more independent and dependent variables that could be explored, this seemed to be the most parsimonious combination to result in measurable effect given the sample available to the researcher.

The sample population in this study was composed of undergraduate business students in the USA. Eighty schools invited to participate. Of those, nine agreed to participate by inviting their undergraduate business students to complete the training. Business department faculty in those schools invited a total of 1,092 students. The respondents numbered 203, with 198 complete, useable results. All 198 results were used to test H1, H2, H3 and H5. Due to an insufficient number of results and an imbalance in results among two schools, data from seven schools were used for H4.

For each hypothesis, the subset of results was evaluated for normality and variance. For normality, the data were evaluated using the Ryan-Joiner test, considered the most accurate test for samples of less than 1,000 (Hair et al., 2010). For variance, the data were evaluated for homoscedasticity using the F-test two sample for variances. Based on those results, either the T-test two sample assuming equal variances or the T-test two sample assuming unequal variances was used to compare the mean MRS scores.

The first hypothesis, which compares the mean MRS score of all respondents who received a different-race presenter avatar with the mean MRS score of all respondents who received a same-race presenter avatar was evaluated for normality. The data were found to be normal (RJ = 0.998, M = 19.74, SD = 4.29, p > 0.10) using the Ryan-Joiner test. Variance was found to be equal using the F-Test [F(2, 197) = 1.40, p = 0.069], so the T-Test for comparison of means assuming equal variances was used. The results indicate the mean MRS score of the group receiving a different-race presenter avatar (M = 19.38, SD = 3.93) was not significantly different from the mean MRS score of the group receiving a same-race presenter avatar (M = 20.04, SD = 4.57), t(198) = −1.08, p = 0.283.

The second hypothesis was intended to determine if the impact of use of a different-race presenter avatar is different within each racial group. For the group of black/African American respondents, the data were found to be normal (RJ = 0.950, M = 20.50, SD = 3.68, p > 0.10) using the Ryan-Joiner test. Variance was found to be unequal using the F-Test [F(2, 11) = 0.81, p = 0.39], so the T-test for comparison of means assuming unequal variances was used. The results indicate the mean MRS score of the group of black/African American respondents receiving a different-race presenter avatar (M = 20.50, SD = 3.68) was significantly lower than the mean MRS score of the group receiving a same-race presenter avatar (M = 23.25, SD = 2.09), t(12) = −2.25, p = 0.038.

For the group of white/Caucasian respondents, the data were found to be normal (RJ = 0.997, M = 19.23, SD = 4.84, p > 0.10) using the Ryan-Joiner test. Variance was found to be equal using the F-Test [F(2, 141) = 1.38, p = 0.09], so the T-test for comparison of means assuming equal variances was used. The results indicate the mean MRS score of the group of white/Caucasian respondents receiving a different-race presenter avatar (M = 19.07, SD = 4.11) was not significantly different from the mean MRS score of the group white/Caucasian respondents receiving a same-race presenter avatar (M = 19.36, SD = 4.82), t(142) = −0.39, p = 0.70.

For the group of Hispanic/Latino respondents, the data were found to be normal (RJ = 0.972, M = 21.87, SD = 2.90, p > 0.10) using the Ryan-Joiner test. Variance was found to be equal using the F-Test [F(2, 14) = 1.22, p = 0.39], so the T-test for comparison of means assuming equal variances was used. The results indicate the mean MRS score of the group of Hispanic/Latino respondents receiving a different-race presenter avatar (M = 20.71, SD = 2.93) was not significantly different from the mean MRS score of the group of Hispanic/Latino respondents receiving a same-race presenter avatar (M = 22.88, SD = 2.64), t(15) = −1.50, p = 0.157.

For the group of self-reporting their race as other, the data were found to be normal (RJ = 0.972, M = 19.12, SD = 3.621, p > 0.10) using the Ryan-Joiner test. Variance was found to be equal using the F-test [F(2, 16) = 19.4, p = 0.08], so the T-test for comparison of means assuming equal variances was used. The results indicate the mean MRS score of the group of other respondents receiving a different-race presenter avatar (M = 18.67, SD = 1.15) was not significantly different from the mean MRS score of the group of other respondents receiving a same-race presenter avatar (M = 19.21, SD = 3.98), t(15) = −23, p = 0.821.

Because the number of white/Caucasian respondents far exceeded the number of other self-reporting races, this analysis was repeated, combining all non-white (minority) respondent data. For the group of self-reporting their race as something other than white, indicating minority status, the data were found to be normal (RJ = 0.979, M = 20.32, SD = 3.183, p > 0.10) using the Ryan-Joiner test. Variance was found to be equal using the F-test [F(2, 21) = 1.99, p = 0.28], so the T-test for comparison of means assuming equal variances was used. The results indicate the mean MRS score of the group of minority respondents receiving a different-race presenter avatar (M = 30.32, SD = 3.18) was not significantly different from the mean MRS score of the group of minority students receiving a same-race presenter avatar (M = 22.28, SD = 2.64), t(22) = −2.03, p = 0.052.

This section addresses the objective for research question three, which is to determine if the impact of a different-race presenter avatar is different within each gender group. For the group of female respondents, the data were found to be normal (RJ = 0.998, M = 20.33, SD = 4.56, p > 0.10) using the Ryan-Joiner test. Variance was found to be equal using the F-test [F(2, 94) = 01.23, p = 0.23], so the T-Test for comparison of means assuming equal variances was used. The results indicate the group of female respondents receiving a different-race presenter avatar (M = 19.41, SD = 4.25) had a mean MRS score significantly lower than the group of female respondents receiving a same-race presenter avatar (M = 21.30., SD = 4.72), t(95) = −2.06, p = 0.042.

For the group of male respondents, the data were found to be normal (RJ = 0.991, M = 19.19, SD = 3.968, p > 0.10) using the Ryan-Joiner test. Variance was found to be equal using the F-test [F(2, 102) = 0.1.43, p = 0.11], so the T-Test for comparison of means assuming equal variances was used. The results indicate the mean MRS score of the group of male respondents receiving a different-race presenter avatar (M = 19.34, SD = 3.55) was not significantly different than the mean MRS score of the group of male respondents receiving a same-race presenter avatar (M = 19.10, SD = 4.25), t(103) = 0.30, p = 0.76.

This section addresses the objective for research question four, which is to determine if the impact of use of a different-race presenter avatar is different between schools. For the group of respondents from School 1, the data were found to be normal (RJ = 0.972, M = 23.1, SD = 2.079, p > 0.10) using the Ryan-Joiner test. Variance was found to be equal using the F-test [F(2, 9) = 1.03, p = 0.62], so the T-test for comparison of means assuming equal variances was used. The means were found to be the same. The results indicate mean MRS score of the group of respondents receiving a different-race presenter avatar (M = 23.82, SD = 1.94) was not significantly different than the mean MRS score of the group of respondents receiving a same-race presenter avatar (M = 22.00, SD = 2.00), t(10) = −1.45, p = 0.186.

For the group of respondents from School 2, the data were found to be normal (RJ = 0.994, M = 18.78, SD = 4.085, p > 0.10) using the Ryan-Joiner test. Variance was found to be equal using the F-test [F(2, 78) = 2.54, p = 0.115], so the T-test for comparison of means assuming equal variances was used. The results indicate the mean MRS score of the group of respondents receiving a different-race presenter avatar (M = 18.06, SD = 3.27) was not significantly different from the mean MRS score of the group of respondents receiving a same-race presenter avatar (M = 19.25, SD = 4.51), t(79) = 1.26, p = 0.210.

For the group of respondents from School 3, the data were found to be normal (RJ = 0.990, M = 19.62, SD = 4.228, p > 0.10) using the Ryan-Joiner test. Variance was found to be equal using the F-test [F(2, 33) = 0.11, p = 0.747], so the T-test for comparison of means assuming equal variances was used. The results indicate the mean MRS score of the group of respondents receiving a different-race presenter avatar (M = 20.0, SD = 4.77) was not significantly different from the mean MRS score of the group of respondents receiving a same-race presenter avatar (M = 19.43., SD = 4.04), t(34) = −0.36, p = 0.721.

For the group of respondents from School 4, the data were found to be normal (RJ = 0.976, M = 19.42, SD = 4.078, p > 0.10) using the Ryan-Joiner test. Variance was found to be equal using the F-test [F(2, 32) = 2.58, p = 0.118], so the T-test for comparison of means assuming equal variances was used. The results indicate the mean MRS score of the group of respondents receiving a different-race presenter avatar (M = 18.94, SD = 3.13) was not significantly different from the mean MRS score of the group of respondents receiving a same-race presenter avatar (M = 19.94., SD = 4.95), t(33) = 0.70, p = 0.492.

For the group of respondents from School 5, the data were found to be normal (RJ = 0.960, M = 19.67, SD = 5.01, p > 0.10) using the Ryan-Joiner test. Variance was found to be equal using the F-Test [F(2, 5) = 0.08, p = 0.793], so the T-Test for comparison of means assuming equal variances was used. The results indicate the mean MRS score of the group of respondents receiving a different-race presenter avatar (M = 21.00, SD = 5.66) was not significantly different from the mean MRS score of the group of respondents receiving a same-race presenter avatar (M = 19.00., SD = 5.42), t(6) = 0.42, p = 0.695.

For the group of respondents from School 6, the data were found to be normal (RJ = 0.993, M = 21.17, SD = 4.960, p > 0.10) using the Ryan-Joiner test. Variance was found to be equal using the F-test [F(2, 22) = 0.84, p = 0.369], so the T-test for comparison of means assuming equal variances was used. The results indicate the mean MRS score of the group of respondents receiving a different-race presenter avatar (M = 20.71, SD = 6.24) was not significantly different from the mean MRS score of the group of respondents receiving a same-race presenter avatar (M = 21.38, SD = 4.51), t(23) = −0.29., p = 0.776.

For the group of respondents from School 7, the data were found to be normal (RJ = 0.982, M = 23.29, SD = 1.064, p > 0.10) using the Ryan-Joiner test. Variance was found to be equal using the F-test [F(2, 6) = 0.00, p = 1.00], so the T-test for comparison of means assuming equal variances was used. The results indicate the mean MRS score of the group of respondents receiving a different-race presenter avatar (M = 23.00, SD = 1.73) was not significantly different from the mean MRS score of the group of respondents receiving a same-race presenter avatar (M = 23.50, SD = 1.73), t(7) = 0.38., p = 0.721.

For the group of respondents from School 8 and School 9, there were not sufficient data points to complete this analysis.

This section addresses the objective for research question five, which is related to the effect of the use of different-race presenter avatar depending on the amount of diversity within the student body at each school. To test a relationship between the respondent scores and school body diversity, an ANCOVA was performed with the discrete independent variable of same- versus different-race presenter avatar, the continuous dependent variable of score, and the continuous variable of percent diversity of each respondents’ school as a covariate. Homoscedasticity was tested using a Levene test for equal variance. The variance was found to be the same (Hair et al., 2010, p. 82). The ANCOVA test showed an insignificant effect of same- versus different-race presenter avatar, F(1, 195) = 1.29, p = 0.258 and a significant F(1, 195) = 6.70, p = 0.010 but weak, R² (adj) = 2.91), effect of percent diversity on score (Hair et al., 2010).

The current study found, generally, that the use of Intergroup Social Contact by the appearance of a different-race presenter in a single-session online Diversity Training module was not effective in reducing racial bias and had the undesirable effect of reducing same-race bias among black/African American participants. Despite the general finding of no significance when combining all groups, the significant difference among female respondents shows that this approach did have some effect because females are generally more responsive to bias reduction efforts than men. However, the contraindication of reducing felt bias seemed to cancel out the weak benefit of using the different race presenter in this way.

The use of different-race presenter avatar in single-session online diversity training: reduced perceived bias in black/African American respondents, reduced observed bias in female respondents and weakly reduced observed bias with respect to student body diversity.

Specifically, in regard to research question one pertaining to the overall effectiveness of the use of different-race presenter, the null stating there is no difference between the mean MRS score of respondents receiving same-race and different-race presenter avatars was supported. There was found to be no difference between the participants receiving a same-race versus different-race presenter avatar when all participants’ demographic groups were combined. Concerning research question two, there were mixed results. For the majority (white/Caucasian) demographic group, the null hypothesis was supported. There was found to be no difference between the majority (white/Caucasian) participants receiving same-race versus different-race presenter avatars. Because the majority-minority gap in hiring and promotion exists in management and executive levels of the US workforce, the most desirable effect of diversity training is to reduce racial bias by majority (white/Caucasian) participants against minority applicants or employees and black/African American applicants or employees are the most affected by racial bias. This is a significant finding because professors and Human Resource Development practitioners would be likely to apply the best practice of including Intergroup Social Contact to single-session online diversity training because the method has been found to be effective in-person and online in multisession diversity training.

Among the minority groups of respondents, when all minority groups were combined, and considering the Hispanic/Latino group separately, the null hypothesis was supported. There was found to be no difference in the combined minority group or in the Hispanic/Latino group in reduction of racial bias between the participants who received a same-race versus different-race presenter avatar. However, for the black/African American group of respondents, the null hypothesis was not supported as there was found to be a reduction in perceived racial bias against black/African Americans by the participants who received a different-race presenter avatar when compared to those who received a same-race presenter avatar. This result is not surprising, but is complex and important to the implications of this research. It has been found, in other studies, that black/African American perception of racial bias is reduced against black/African Americans by intergroup contact with majority (white/Caucasian) presenters. This reduction is often considered an undesirable effect of Diversity Training in that it reduces black/African American students’ and employees’ sensitivity to racial bias.

Concerning research question three, the null hypothesis, which stated that there is no difference between the mean MRS score of respondents receiving a different-race presenter avatar and the mean MRS score of respondents receiving a same-race presenter avatar within gender groups, was not supported. The female participants did show a measurable reduction in bias when receiving a different-race presenter avatar versus a same-race presenter avatar. The male participants did not show a difference in measured racial bias between the participants who received a different-race presenter avatar and the participants who received a same-race presenter avatar. These results are not surprising, in that female participants in diversity training are often more likely to show measured reduction in racial bias than make participants. These results are encouraging to the practical application of this study because it seems that the use of different-race presenter avatar was effective in reducing the racial bias in some of the participants even though it was not effective in reducing the racial bias of all groups.

In regard to research question four, the null hypothesis that there is no difference between the mean MRS score of respondents receiving a different-race presenter avatar and the mean MRS score of respondents receiving a same-race presenter avatar within each school based on the diversity of that school’s student population was supported. Within each school, there was no difference between the scores of the participants receiving different-race and same-race presenter avatars. However, to further test this relationship, and because the literature overwhelmingly points to a relationship between student-body diversity and racial bias, the data were combined and analyzed using ANCOVA. The ANCOVA analysis, using the percent diversity of each school as a covariate of the relationship between racial bias and same-versus different-race presenter avatar showed a significant, but weak relationship. The diversity of the student-body at the participants’ school did seem to affect participants’ sensitivity to reduction in racial bias due to the addition of a different-race presenter avatar slightly.

Diversity training has been used as a method to support affirmative action, diversity and inclusion initiatives (three terms which are often conflated and confused) for decades. Intergroup Social Contact has been established as a way to augment these initiatives and to bolster the effectiveness of diversity training in reducing racial bias – among other types of bias – in participants, specifically in the use of a different-race presenter or instructor.

The practical application of this research calls for caution in deploying different-race presenter avatars in single-session online diversity training, though the method is generally found to be beneficial in diversity training efforts presented in-person and with multiple-sessions. Though organization size was not tested in this research, the implications of the effectiveness of shorter duration online training is especially important for small to medium enterprises (SMEs). This is because SMEs are more likely to use single-session online training and are more likely to rely on such single-session online training as their only diversity training method (Wilkie, 2014). Based on the results of this study, the use of different-race presenter avatars in brief online diversity training is more effective in reducing racial bias in female participants and in more diverse populations but can cause a reduction in sensitivity to perceived racial bias in black/African American participants. Human Resource Practitioners are encouraged to read not only trade journals but also academic journals to keep up with research in this area so that they can make research-driven design decisions.

Due to the complexity of diversity and the available literature on best practices in diversity training, the possibilities for future research in diversity training are limitless. Based on the results of the current research, several avenues for future research seem to have great potential. Because previous research using multisession diversity training showed a significant effect of different-race presenter avatar on all groups, repeating this research with modules of varying numbers of sessions would help to determine where the benefit of this innovation is most economical. It has been shown that many organizations will continue single-session online diversity training whether they know it is working or not, so improving the quality of such training would make diversity training more effective for the organizations looking for minimum effort and expense. Research has shown that empathy a significant mediator in the effectiveness of diversity training (Holladay and Quinones, 2008; Madera et al., 2011) and in the effectiveness of Intergroup Social Contact (Pettigrew and Tropp, 2008). Therefore, measuring empathy more directly in addition to measuring racial bias against black/African Americans in general and then comparing those results would help to determine how Human Resource Development practitioners and professors could develop more effective diversity training. As with most research, repeating the study with a larger sample would make it possible to better distinguish between the independent variable groups.

Now that the academe knows diversity training can benefit online from a different race presenter in some situations and not in others, more research experimenting with the number of training sessions, length of sessions, etc. would result in some useful parameters to employ this innovation with confidence in its success.

The purpose of this study has been to evaluate the effectiveness of the use of Intergroup Social Contact in single-session online diversity training in the form of a different-race presenter avatar. This purpose was accomplished by presenting undergraduate business students with such diversity training with alternating same-race and different-race presenter avatars and comparing participant scores on a measure of racial bias, then comparing results based on demographic groups of participants. In summary, the use of Intergroup Social Contact to increase the effectiveness of diversity training was supported by this research, within its limitations and with caution in its deployment. It was found that the use of a different-race presenter avatar was effective in reducing racial bias among female respondents, which is desirable and among black/African American respondents, which is undesirable.

Organizations, Human Resource Development practitioners and professors wanting to increase the effectiveness of their diversity training programs should survey the literature to determine best practices for the development and delivery of diversity training. Among those best practices are multimodular training and the use of Intergroup Social Contact using different-race presenter. Caution is warranted against implementing the use the latter without the former. Intergroup Social Contact using different-race presenter seems to be effective in multi-modular diversity training of significant duration but not as effective in single-session diversity training. The small improvement of adding a different-race presenter avatar to online diversity training could be impactful in further reducing the race gap in employment and promotion by reducing racial bias in participants, given that organizations provide diversity training of adequate design to take advantage of the benefits of Intergroup Social Contact.

This work was supported by the Appalachian College Association with an unnumbered grant.

The author has no financial interest or benefit to disclose regarding this research.

Data may be provided upon request to the author.

There is no supplemental online material for this article.