Impact of broad categorization on statistical results: How underrepresented minority designation can mask the struggles of both Asian American and African American students

Although recent research has reported equity gains in the university-level physics classroom, there remains evidence of significant inequities by race. We conducted a series of multiple linear regressions to examine the magnitude of these inequities in introductory mechanics final exam scores. When we followed a common method of grouping students by underrepresented minority (URM) status in our statistical analysis, we found that when controlling for prior knowledge, based on ACT math and physics placement test scores, there was no evidence of racial inequities. However, when we performed the same analysis but replaced URM status with separate groups for each race or ethnicity reported to the university, we found that African American and Asian American students, and to a lesser extent, Hispanic students, underperformed in their physics course relative to their non-Hispanic White American peers, even when controlling for prior knowledge. We conclude that grouping students by URM status may not be sufficient to detect patterns of academic difficulties experienced by specific racial and ethnic groups of students. Furthermore, these results lead us to believe that there is still work to be done at the university level to provide a learning environment supportive of students of all backgrounds.

Figure 1

Multiple linear regression coefficients of the URM variable in the binary models reported in Table 4. Coefficients represent the standardized difference in final exam performance between non-URMs (Asian and White students) and URMs (Black, Hispanic, AIAN, and NHPI students). The error bars represent the standard error for each coefficient. The URM coefficient had $p<0.01$ for models 1 and 2 (URM alone and $\mathrm{URM}+\text{ACT Math}$, respectively), but $p=0.069$ for model 3, in which the placement test was included.

Figure 2

Multiple linear regression coefficients from the differentiated models reported in Table 5. Coefficients represent the standardized difference in final exam performance between non-Hispanic White American students and the specified group. The error bars represent the standard error for each coefficient. The coefficients had $p<0.001$ for all three models.

Figure 3

Asian American and non-Hispanic White American student performance on the final exam by ACT math score.

Figure 4

Multiple linear regression coefficients from the exam models reported in Table 7 with the coefficient from model 6 of Table 5 included for comparison. Model 6 is represented by the leftmost bar for each racial or ethnic group, and models 7–9 follow consecutively. Coefficients represent the standardized difference in final exam performance between non-Hispanic White American students and the specified group. The error bars represent the standard error for each coefficient. All coefficients are statistically significant with $p<0.05$.

Figure 5

Graph corresponding to question 5.

Figure 6

Answer choices for question 6.

Figure 7

Graph corresponding to questions 7 and 8.