5 Relationships between Variables

Many of the critics of the methods we are to describe argue that, for twin studies at least, the so-called traditional methods such as taking the difference between the MZ and DZ correlations and doubling it as a heritability estimate give much the same answer as the more sophisticated methods taught here. In the final analysis, it must be up to history and the consumer to decide, but in our experience there are several reasons for choosing the methods presented here. First, as we have already shown, the puzzle of human variation extends far beyond testing whether genes play any role in variation. The subtleties of the environment and the varieties of gene action call for methods that can integrate many more types of data and test more complex hypotheses than were envisioned fifty or a hundred years ago. Only a model building/model fitting strategy allows us to trace the implications of a theory across all kinds of data and to test systematically for the consistency of theory and observation. But even if the skeptic is left in doubt by the methods proposed for the interpretation of variables considered individually, we believe that the conventional approaches of fifty years ago pale utterly once we want to analyze the genetic and environmental causes of correlation between variables. The genetic analysis of multiple variables will occupy many of the succeeding chapters, so here it is sufficient to preview the main issues. There are three kinds of ``multivariate'' questions which are generic issues in genetic epidemiology, although we shall address them in the context of the twin study. Each is outlined briefly.