1997). This test is appropriate only if the dependent variables can be combined
theoretically which is the case in this study. Combing the four dependent variables
creates a general construct of adverse mental health symptoms.
The statistical assumptions for MANOVA include multivariate normality,
homogeneity of covariance matrices, linearity, and independence of observations
(Tabachnik & Fidell, 1989; Weinfurt, 1997). These assumptions were met with the
exception of the distribution of the alcohol/drug use subscale, which had a positively
skewed distribution. Several transformations were attempted without success. However,
this subscale was retained in the model because MANOVA is robust to skewed
distributions with sample sizes greater than 200 (Tabachnik & Fidell, 1989; Weinfurt,
1997). In addition, the central limit theorem is applicable to studies with at least 20
degrees of freedom in each cell. This theory states that the sampling distribution of
means approach normality even if the sample data does not (Tabachnik & Fidell, 1989).
Two MANOVAs were done; the results are in Table 4-6. The first test compared
those who reported the presence of a mentor and those who did not. Those with a natural
mentor had higher levels of adverse mental health outcomes than those without a mentor.
The observed power was 0.78. The first hypothesis was not supported.
Second, the participants who met all four of the quality factor cutoff points
described in the previous section were categorized as having a quality mentoring
relationship. The second MANOVA compared the adverse mental health outcomes of
those with a quality mentor to the rest of the adolescents. These two groups were not
different on the mental health outcomes. There was no significant difference between
these two groups. However, this analysis had an observed power of only 0.20.