Affirmative Intervention to Reduce Stereotype Threat Bias: Experimental Evidence from a Community College

ABSTRACT

At all levels of education, the racial achievement gap in performance between Black and Latino students and their White peers stubbornly persists. While the causes of this gap are numerous and interrelated, one theory posits that students from underrepresented racial groups may face stereotype threat, meaning that fear of failing and thereby fulfilling negative group stereotypes leads to anxiety and suboptimal cognitive performance. Though low-cost value affirmation interventions have been shown to reduce achievement gaps in some classroom settings, these findings have not been consistently replicated. In this study, we test the efficacy of this intervention among a new sample of students enrolled in a community college in the Midwest. At the beginning of the fall 2016 semester, students in English courses (N = 1,115 in 59 course sections) were randomly assigned to short writing exercises that were either self-affirming or neutral. Using administrative data collected at the end of the term, we compared treatment and control students on a range of outcomes that included course grade, overall GPA, and course persistence. Overall, we find little evidence of a positive effect of this one-time affirmation of social identity. Moderation analyses, however, show heterogeneous effects across course sections, suggesting that the classroom setting may play a role in the interaction between social identity and student outcomes.

KEYWORDS:

• Community colleges
• stereotype threat
• randomized control trial

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. We note that self-affirmation writing interventions are only one way that institutions can work to reduce stereotype threat. We do not advocate for their use over other methods, but rather wish to evaluate their merits in a new context for the very reason that their scalable, low-cost design makes them attractive to budget-constrained institutions.

2. Authors’ calculations using IPEDS data. We suppress the graduation rates of MWCC and its home state in order to maintain confidentiality. The United States completion rate (150%) in 2014 was 21.84%.

3. It is difficult to posit the role of individual faculty members at community colleges compared to selective institutions due, in part, to the consistent trend for part-time or adjunct faculty members to comprise a significant share of the teaching force (Charlier & Williams, 2011).

4. In order to most closely replicate the environment of the Cohen et al. (2006) study, we chose to conduct our intervention in English classrooms. This decision means that all analyses involving gender effects are exploratory since most prior research that found gender effects used a STEM classroom as the site of the intervention.

5. This day was chosen to make it more likely that students would have already dropped or switched classes before the rosters were compiled and randomization took place.

6. See Appendix B for the full description of the power analysis.

7. Two students had a randomized envelope but instead completed an “extra” envelope. Of these, one received a packet with the correct treatment status; the other the opposite. The third student received the correct envelope, but the packet had an incorrect treatment status.

8. Fewer than 10 corrections were made and all were due to initial transcription errors of the values selected by a student.

9. We considered students who consented to be part of the experiment. All who consented also completed at least the first values selection section (25% of the packet).

10. Though the research team randomized all students in the 92 classrooms, only the instructors of 59 classrooms distributed the packets, thereby giving students the opportunity to consent. Thus, while researchers often estimate ITT based on the full sample, we estimate ITT based on the sample of students who had an opportunity to consent.

11. We also estimated treatment-on-the-treated estimates (Gerber & Green, 2012) using both treatment status and 2SLS. All results were substantively the same as the ITT estimates, likely due to almost complete compliance.

12. Our data do not allow us to distinguish between first- and continuing-generation students so we were not able to investigate the heterogeneity of the treatment effect for these student groups.

13. Only the estimates for interactions with assignment to treatment are shown due to space limitations.

14. We also estimated the effect of assignment to treatment on the number of days it took students to withdraw. As fewer than 100 students withdrew from their course, there was little statistical power to detect a significant relationship (estimates available upon request).

15. Because there were too few Black students in the sample to run a separate analysis, we combine them with other groups.

16. Subpopulation estimates available upon request.

17. It is not clear in the published materials which racial/ethnic groups were included in this category.

18. From conversations with site partners, it appears that non-participating faculty may have been overrepresented by contingent faculty. We note their status only to acknowledge the differential demands on their time, particularly in their ability to attend departmental meetings during which faculty were notified about the study as well as their availability to pick up study materials once randomization was complete.

19. In order to investigate the salience of the prompts, we reviewed the paragraph participants wrote, looking for statements incongruent with assignment. Only approximately 1% of the analytical sample appeared to deviate from the prompt, which supports that the participants generally understood the difference in the two prompts.