Table 4.
Increasing the number of constructed-response questions on an exam disproportionately benefits middle/high-SES students, but not male students, relative to low-SES and female students, respectivelya
| Parameter | Relative variable importance | Model averaged regression coefficient ± SE | p Valueb |
|---|---|---|---|
| Intercept | NA | 0.515 ± 0.0238 | <0.0001 |
| Cum.GPA | 1.00 | 0.165 ± 0.00311 | <0.0001 |
| Course (reference level: course 1) | |||
| Course 2 | 0.85 | 0.0677 ± 0.0381 | 0.0759 |
| Course 3 | 0.85 | 0.0159 ± 0.0247 | 0.521 |
| Exam (reference level: time 1 (exam 1) | |||
| Time 2 (exam 2) | 1.00 | 0.0116 ± 0.00183 | <0.0001 |
| Time 3 (exam 3) | 1.00 | 0.0257 ± 0.00250 | <0.0001 |
| Time 4 (exam 4) | 1.00 | 0.0745 ± 0.00269 | <0.0001 |
| Student gender (reference level: male) | |||
| Female | 1.00 | −0.0252 ± 0.00341 | <0.0001 |
| Student SES status (reference level: middle/high-SES) | |||
| Low-SES | 1.00 | −0.00503 ± 0.00589 | 0.393 |
| Exam characteristics | |||
| W.Diff | 1.00 | −0.243 ± 0.0114 | <0.0001 |
| Percent CR | 1.00 | 0.0789 ± 0.00668 | <0.0001 |
| Student identity × exam characteristics (reference level: male or middle/high-SES) | |||
| Female × percent CR | 0.29 | −0.000607 ± 0.00273 | 0.824 |
| Low-SES × percent CR | 1.00 | −0.0278 ± 0.00645 | <0.0001 |
aThe outputs were produced via model averaging of all possible models using the MuMIn package in the program R. Although not shown, the models include two random-effects terms: (1|Stu.ID) + (1|Instr).
bBolded p values are significant.