Mortality and Marital Status. Reflections on the Derivation of Etiology From Statistics

Abstract

In seeking to find, from statistical correlations, etiologic significance of environmental factors in causation of disease, it is easy to fall into the practice of selecting particular diseases associated with particular factors that appear plausible, while ignoring similar associations of the same diseases with other factors, and of the same factors with other diseases. The tendency has been flagrantly exhibited in the recently much-publicized correlation between smoking and death rate from lung cancer. The correlations found are general and concern all classes of disease, not only lung cancer. It has been argued, for instance, that the greater death rate from lung cancer found among males than among females is attributable to the fact that men smoke more. Then again it is argued that the death rate from lung cancer is greater in urban than in rural communities because of the greater smoking rates characteristic of urban communities or because of air pollution. But the death rates from virtually all homologous cancers are greater among males than among females, and so are the rates from virtually all non-cancerous diseases. So also the death rate is greater in urban communities for other cancers - cancers that cannot be attributed to either smoking or air pollution - and for noncancerous diseases as well. The sex differential and the urban-rural differential of death rate from diseases generally are among the first laws of vital statistics discovered. They have been known for 300 years - and have not been adequately explained for the same length of time. The example of the association of death rate with marital status is used to illustrate other aspects of this fallacy. Observed differences of death rates may be determined constitutionally, rather than environmentally. The possible implication for the fallaciousness of environmentally oriented theories is of revolutionary import, while the suggestion that constitution can have such exquisitely differentiating effects is itself challenging. Then if we wonder whether the whole may not be due to systematic errors in the statistics, we can find support for this too.