Chapter 4. Does ethnic heterogeneity of clubs affect member dropout? 91 drop out, the addition of these predictors has virtually no effect on the observed difference in dropout between members with migrant and Dutch backgrounds. In model 3, member’s ingroup share is included as a third club level predictor. Firstly, we can witness a clear negative association between ingroup share on dropout. This indicates that members of clubs with high ingroup shares are less likely to drop out than members with lower ingroup shares, supporting hypothesis 1 of this study. While the exponentiated association (i.e., -0.39%) may not seem substantial at first glance, this means that a 78 percent point higher ingroup share (the mean difference in ingroup share between Dutch and migrant members) reduces one’s probability of dropping out by more than 25 percent. Secondly, by including ingroup share as a predictor in the model, most of the difference between members with a migrant background and a Dutch background in dropout has been accounted for. Of the 37.17% difference in probability to drop out, only a 3.57% difference remains. This result provides strong support for hypotheses 2 and 3 of this chapter, which stated that members with migrant backgrounds are more likely to drop out than members with Dutch backgrounds but that this difference is mediated by differences in ingroup share. In model 4, I subsequently add the ethnic fractionalization of members’ outgroup as a fourth club level predictor. This addition is important because ingroup share and outgroup fractionalization correlate oppositely for members with migrant backgrounds and members with Dutch backgrounds. When we do not account for this, it may dilute the estimation of the association between ingroup share and dropout (see also Koopmans & Schaeffer, 2015). The results from model 4 show that this indeed is the case. By including outgroup fractionalization in the model, the effect of ingroup share on dropout significantly increases in strength and the difference between Dutch and migrant members remains virtually the same. These results reaffirm hypotheses 1 to 3. Furthermore, the inclusion of outgroup fractionalization as predictor of dropout also allows us to test whether group threat (hypothesis 4) or, instead, social disarray (hypothesis 5) is a (stronger) driver of member dropout. The results show that the association between outgroup fractionalization and dropout is negative. This means that club members with homogeneous outgroups are less likely to drop out than members with ethnically diverse outgroups. Consequently hypothesis 4 on group threat is rejected and hypothesis 5 on social disarray is confirmed. Finally, through the inclusion of two cross-level interaction terms, model 6 allows the effect of ingroup share and outgroup fractionalization on dropout to
RkJQdWJsaXNoZXIy MjY0ODMw