Causal interpretations from cross-sectional data : An examination of the stochastic processes involved in the relationship between a personality characteristic and coronary heart disease

Abstract

We have shown that available data from a cross-sectional study is not sufficient to determine what, if any, causal relationship exists between selected personality traits and coronary heart disease. Specificially, we have shown that the observed data could have arisen from either a process where the personality trait has a beneficial effect (i.e. makes one less likely to die from CHD) or a process where it has a harmful effect (i.e. makes one more likely to develop CHD), as well as a process where it has no effect. The assumptions which would be required to choose one interpretation over another have been explicitly pointed out and discussed. It is concluded that more information is required to make a valid choice between the many alternative processes that might be associated with the observed data.Specifically, a conclusive causal analysis will always require a time variable. Mathematically stated: in a causal analysis of this type we are concerned with first derivatives with respect to time, while a cross-sectional study determines only the values of the corresponding variables at one specific time, not their derivatives. To estimate the derivatives from data requires a knowledge of the values at least two distinct times; without this an analysis depends on unverified assumptions about the derivatives.Of course, a cross-sectional study can be used to show that certain factors are unlikely to be causes of a given effect. Positive causal conclusions, however, will require observations over time.