ethnic backgrounds will be important to ensure generalizability. Second, the inherently low prevalence of psychiatric disorders and the less than ideal reliability of psychiatric interviews reduced statistical power to some extent. Set against this concern, the GSMS provides a large sample and a number of our analyses aggregated diagnoses across waves of assessment, increasing the frequency of outcome. We also supplemented our analyses of diagnoses with more powerful analyses based on continuous counts of psychiatric symptoms. Third, during the course of analyses we have conducted a number of statistical tests, taking alpha as .05 for each analysis. This increase the possibility that Type I errors may have been made in statistical inference. This highlights the importance of replication in research of this kind. Fourth, our parenting variables were measured with single items that may not have fully captured the dimensions of risk present in the population. This may limit the conclusions that can be made about relationships with these constructs.
