The Diversity Challenge: Social Identity and Intergroup Relations on the College Campus

Appendix C. Attrition Analysis

Appendix C: Attrition Analyses T his appendix contains the results of the attrition analyses we conducted to examine whether respondents who participated in all waves of the study differed from those who dropped out of the study at earlier stages. Our main objective was to determine whether attrition was selective with respect to our core variables. Differential attrition might have introduced various kinds of biases into our findings. We were particularly concerned about three potential biases. One was the possibility that the most prejudiced or most conservative students might have dropped out of the study out of reluctance to expose attitudes that were counternormative on the campus. This would have led us to overestimate the liberalizing effects of the college experience. A second potential bias was the possibility that the ethnic minority students who least identified with their own ethnic groups might have selectively dropped out because they were reluctant to expose attitudes that were counternormative within those groups. A third potential bias was the possibility that students whose attitudes about ethnic diversity were the least well formulated and least crystallized might have selectively dropped out to avoid exposing their poorly thought-out attitudes. This would have left us generalizing from a subset of students who were unusually interested in and knowledgeable and passionate about issues concerning ethnic diversity. In the following analyses, we compared four groups of respondents. The “persisters” were those who participated throughout the study (n = 831). This group included those who participated in waves 1 to 5 inclusive (as well as those who spent an additional year at UCLA and par- ticipated in wave 6 in addition to waves 1 to 5). It also included the black and Latino oversample added at wave 2 if they participated in waves 2 to 5 inclusive (or waves 2 to 6 inclusive). These persisters were compared to three attrition groups. The “partway persisters” were those who participated in only the first few waves, completing wave 1 and wave 2 only or waves 1 to 3 only (n = 317). This group also included those in the black and Latino oversample added at wave 2 who participated in waves 2 and 3 only. The “voluntary early dropouts” were those who dropped out early in the study (n = 346), having completed wave 1 only, or wave 2 only if they were in the black and Latino oversample added at wave 2. Finally, the “involuntary early dropouts” (n = 175) consisted of the white or Asian students we decided not to recontact after wave 1 either because they had failed to complete much of the wave 1 questionnaire (having at least thirty missing values) or because they had not provided contact information . Since this last group was composed only of whites and Asians, and since most of our key variables were strongly correlated with ethnicity, we decided to conduct the attrition analyses within ethnicity , to unconfound attrition from ethnicity. We compared the attrition groups in terms of five categories of variables: demographic indicators , background variables (such as SAT scores and ethnicity of high school friends), major sociopolitical attitudes at wave 1, consistency of such attitudes at wave 1, and stability of such attitudes from wave 1 to wave 2. A limitation of performing analyses within ethnic groups was a reduction in sample size. The number of cases in the persister, partway persister , voluntary early dropout, and involuntary early dropout groups seemed adequate among whites (n = 203, 112, 129, and 84, respectively) and Asians (n = 284, 89, 91, and 64, respectively). Among Latinos, the sample sizes were smaller, but they seemed adequate if our threshold for a significant effect was not too stringent (n = 182, 57, and 56, respectively , with no involuntary dropouts, by design). Among blacks, the sample sizes were small, at best (n = 44, 16, and 20, respectively, with no involuntary dropouts). Given that there were also some missing values on each variable, these numbers occasionally became even smaller. In the case of African Americans, the sample sizes were often too small to analyze . As a result, we tried to set lenient standards for detecting attrition effects. Appendix C 354 Demographic Indicators The first set of analyses tested for differential attrition on the eleven key demographic variables reported in chapter 3: age, gender, ethnicity-race, birthplace of student, religion, birthplace of parents and grandparents, language spoken at home, father’s and mother’s education, and family social class. For...