A previous multivariate analysis of variance (MANOVA) study of attitudes toward interactive video (ITV) of participants in distance education had indicated that generalized variance was not homogeneous across all groups. Since heterogeneity of variance in MANOVA has previously been shown to change Type I error rates in unequal group size, detected differences were suspect. This study investigated whether the detected differences in attitude toward ITV would be identical if the total participant population with unequal group sizes were used or if random sampling of equal size groups was conducted. The analysis was conducted by using all participants (unequal group size) and by randomly selecting 31 subjects from each of 4 groups formed by site (home or remote site) and grade (high school or college). Random selection of 31 subjects per group and analysis was replicated 4 times. Results support the hypothesis that a larger generalized variance associated with a larger sample size does produce a conservative Type I error rate. These results also indicated that if there was no association between sample size and generalized variance, the effects of differences in variance will be canceled. When, however, the difference in generalized variance was less than two times the smaller, the hypothesis that Type I error was liberal did not occur. This suggests that there may be a minimum difference in variance (and possibly in sample size) to be exceeded before Type I error is seriously affected. (Contains three tables and five references.) (Author/SLD)