Summary

Abstract

The basic idea of the book is:

• 1. Besides conventional physical risk factors like smoking, pre-cancer diseases or heredity, psychosocial conditions are crucially relevant for predicting morbidity and mortality.

• 2. Multivariate statistical methods are important for determining the statistical relevance of a variable in the context of others, which is a precondition for testing hypotheses of causality. These methods allow to determine the relevance of a variable when any number of other potentially relevant variables are held constant computationally — they need not be held constant physically e. g. by using groups or pairs equalized with respect to certain control variables (whose number is severely limited for practical reasons). A variable which is positively related to some criterion may become insignificant or even change the direction of its relevance when control or competing variables are brought in. Multivariate analysis allows to decide whether the variations in a criterion variable are »really« due to some risk factor when competing variables are taken into consideration, or rather to the latter. Multivariate models also allow to study interactions between variables, by which we mean that the efficacy of a risk factor depends on the level of some other risk factor, and vice versa. More refined conclusions can be drawn when there are several consecutive measurements of the same variables. When there is only one, we have a correlation between x and y which is symmetrical and does not allow do decide between (a) x directly influences y, (b) y directly influences x, or (c) neither. With consecutive measurements, the relations (a) between x and a subsequent change of y, and (b) between y and a subsequent change of x are completely separate facts; (a) fixes the possible causal direction as x→y, (b) as y→x, and both relationships are possible simultaneously and independently of each other. The question of direct or indirect influence again depends on the role of competing variables.

• 3. A crucial test for some hypothesized causal relationship is provided by an experiment. If it is found that (a) certain variables commonly associated with health or morbidity/mortality have been changed by the treatment in a favourable direction, and (b) that in a follow-up, health status is better than in an untreated control group (and is in fact associated with the changed values of these variables), then it seems safe to conclude that the treatment in question is a means for favourably influencing the health status of risk persons or diseased persons.