The Ironic Costs of Performing Well: Grades Differentially Predict Male and Female Dropout From Engineering

¹The results are more unequivocal in the lab than in real-world contexts (Cullen, Hardison, & Sackett, Citation2004; Cullen, Waters, & Sackett, Citation2006; Good, Aronson, & Harder, Citation2008; Huguet & Régner, Citation2007; Keller, Citation2007; Keller & Dauenheimer, Citation2003; Stricker & Ward, Citation2004).

²Once stereotyped students in the field chronically employ the self-protective strategy of disidentification, they may show reduced reactivity to feedback and report even less self-doubt than majority students. Psychological disidentification implies that, to protect their self-esteem, stereotyped students disengage their self-feelings from their academic achievements (Major, Spencer, Schmader, Wolfe, & Crocker, Citation1998; Steele, Citation1997). This results in weakened correlations between self-esteem and academic outcomes (Morgan & Mehta, Citation2004; Osborne, Citation1997; Verkuyten & Thijs, Citation2004). However, as long as women are identified so that they care for succeeding in the domain, this response should be unlikely.

³In principle, it is also possible that—at poor and/or high performance levels—men suffer lower chances of graduation than women. However, we are not aware of any theoretical reasons that would suggest such a pattern.

⁴This means that some students enrolled before 1993; the average year of enrollment was 1995.

⁵Dropout is defined from the university's perspective rather than from a student's view. If a student stopped paying tuition fees and taking exams for a certain period, the student's status is defined as dropout.

⁶A considerable portion could not be reached because of changes in address or family name.

⁷First-term university GPA has also been found to mediate the relationship between standardized test results and dropout (Brandstätter et al., Citation2006). This is important because in Middle Europe standardized testing is less common than in other countries, and students often cannot report SAT or GRE scores.

Note. The continuous variables in the model were centered at the mean. N = 2,397. GPA = grade point average.

^a 1 = male, 0 = female. ^b 1 = gap, 0 = no gap. ^c 1 = participation, 0 = no participation.

*p < .05. **p < .01. ***p < .001.

⁸For probing interactions, the Modprobe macro for SPSS was used (Hayes & Matthes, Citation2009).

⁹Among nonrespondents to the survey, 78% of the female and 56% of the male trajectories end in dropout; among survey respondents the numbers are 58% for female students and 19% for male students.

¹⁰Simple slope analyses corroborate this conclusion. In separate analyses for the two subsamples, we test for differences at the total sample's mean GPA (M = 3.08), and 1 SD above and below the mean (SD = 1.03), controlling for experience gaps and age. For both survey respondents and nonrespondents, there are no significant gender differences at a performance level −1 SD (no participation: B = .40, p = .29; participation: B = 1.36, p = .09). At the mean and at higher performance levels (+1 SD), the gender differences are significant in both subsamples (M: no participation: B = .92, p < .01; participation: B = 1.31, p < .01; +1 SD: no participation: B = 1.43, p < .01; participation: B = 1.25, p = .04).

Note. The grade point average (GPA) variable was centered at the mean. R ² = .23 for self-doubt and .07 for social discomfort. N = 444.

^a 1 = male, 0 = female. ^b 1 = prospective, 0 = retrospective.

*p < .05. **p < .01. ***p < .001.

¹¹All probings in this section are conducted at the mean and one standard deviation below and above the mean of the total sample (M = 3.08, SD = 1.03). Although the three-way interaction does not reach statistical significance, simple slope analyses indicate that in the prospective subsample there are no statistically significant gender differences at any of the performance levels (Bs between −.07 and .01, all ps > .83). In the retrospective subsample, in contrast, men report less self-doubt than women across all performance levels with the difference being particularly pronounced at poor performance levels (−1 SD: B = −1.36, SE = .22, p < .01; M: B = −.96, SE = .16, p < .01; +1 SD: B = −.57, SE = .21, p < .01).

¹²For both self-doubt and social discomfort, the models are recalculcated including the variables experience gap and age. For neither of the dependent variables the explanatory power of the model increases (self-doubt: R ² change =.00, F = 1.11, p = .33; social discomfort: R ² change = .00, F = 1.03, p = .36).

Note. The continuous variables in the model were centered at the mean. N = 111. GPA = grade point average.

^a 1 = male, 0 = female.

*p < .05. **p < .01.

Note. R ² = .08.

^a 1 = male, 0 = female. ^b 1 = prospective, 0 = retrospective. ^c 1 = graduate, 0 = dropout.

*p < .05. **p < .01. ***p < .001.

¹³Domain importance is related to domain identification but does not include the self-evaluative aspect, which is central to the latter concept. In a separate study (N = 16) we asked engineering and natural sciences students to rate the domain importance items, the six-item academic competence subscale suggested by Crocker et al.(Citation2003), and a two-item domain identification measure used by Keller (Citation2007). The correlations of domain importance with the two measures were r = .57 (p < .05) and r = .68 (p < .01), respectively, indicating considerable construct overlap in the measures of domain importance and domain identification.

¹⁴Respondents indicated for 26 factors whether they played a role in their decision for dropping out. The most frequently chosen factors were unfulfilled expectations with regard to contents (56%), difficulties with exams (56%), long duration (54%), ways of teaching (52%), and did not meet my expectations (48%). Fisher's Exact Tests (two-sided) suggest that there are no gender differences at conventional significance levels for 24 of the 26 factors. Male and female dropouts only differ in reporting on an “attractive alternative to the major” (49% vs. 9%, p = .03) and on “job-related pressure” (41% vs. 0%, p = .01). None of the male and female dropouts felt burdened with childcare.