interaction process between teacher and students: The labeling of students on the basis of relative characteristics, The

The results of these analyses are in Table 2. In Model I it can be seen that both the achievement score and the behavior of the student have significant negative effects. This means that the higher the achievements and the better the behavior, the lower the chance of being labeled as problematic. The effect of the mean achievement of the class is as expected. The higher the mean achievement the greater is the chance that a student with constant achievement and behavior will be labeled problematic. The conclusion is, therefore, that the labeling is relative and not absolute.

In the second model, relative achievements were used instead of the absolute ones. Both the relative achievements and the behavior have significant negative effects. The mean achievement of the class is no longer significant. This means that the labeling is relative, but that in groups with lower mean achievements, students are not labeled as problematic sooner than in groups with higher mean achievements. From Table 2, it can be seen that both models have the same deviance and degrees of freedom. To predict the labeling of problem students, therefore, Model 2 is used, because after leaving out the variable mean achievements (not significant), this Model becomes the best (deviance 1180.54). On the basis of their relative achievements and their behavior, 80% of the problem students and 81 % of the non-problem students can be predicted correctly.

If the individual characteristics gender, SES and ethnicity, are added as discussed in the introduction to the Model; only the student's SES has a significant negative effect. On the basis of Model 3a, 80% of the problem students can still be predicted correctly, but also 84% of the non-problem students. No better prediction can be made on basis of the variable mean achievement or on basis of interactions between individual characteristics. Also, the teacher characteristics-teaching experience and additional educational training-have no effect.

REFERRAL TO SPECIAL EDUCATION

The possible referral of problem students to special education also was considered. From the 304 problem students, for only 19 was a referral procedure started. In Model 4 (Table 4) the ways in which the relative achievements and behavior of these students differ from those of the other problem students is examined. On the basis of Model 4 it was concluded that this is the case with respect to the achievements-but not with respect to the behavior. If the chance that a problem student will be referred to special education is predicted on basis of his/her relative achievements, then 74% of them are predicted correctly. From the other pupils only 64% are predicted correctly.

To improve this prediction, some other variables were added to the model. First, the individual characteristics: gender, SES and ethnicity were entered. All three variables have no significant effect. Second, some characteristics of the teacher were added. Both the variables teaching experience and additional educational training had no effect. Third, the variable help from non-resident sources was added to the model. This variable had a significant, negative effect (see Table 5). Because of this extra help, it is possible to predict more accurately why some problem students are not referred. Seventy-four percent of the students for whom a referral procedure is started, are still predicted correctly, but now so also are 78% of the others.