Treatment of the fourfold table by partial association and partial correlation as it relates to public health problems.

Many public health studies require treatment of plural variables and the bulk of data which is available falls into a dichotomous classification. The following are examples of studies which have required analysis by a technique applicable to alternative categories: (a) An estimate of the risk of operative mortality for cancer patients was desired [1]. Obesity, malnutrition, old age, hypertension, cardiac history and length of operation were the variables studied. Many of these were not quantitatively measurable, and for consistency the few that were measurable were consolidated into alternative categories. I(b) In another study aimed to determine the amount of cancer knowledge in relation to age and economic factors, the age could have been measured quantitatively, but the economic condition could not [2]. (c) At present a study is being conducted in which approximately 80 variables dealing with heredity and environment as possible causative factors of cancer are being considered. Over three-quarters of these variables cannot be measured quantitatively and the solution of the many problems inherent in this study depends on the treatment of the 2 x 2 table. If all variables in a study were independent, the examination of the data in the zero order of association would be sufficient. In the presence of data requiring the analysis of plural dependent variables, the two methods to be considered are partial association and partial correlation. The one is the effort to eliminate the fallacy of mixed classification by using partial universes; the other is the effort to eliminate the effect of certain variables in the whole universe, so as to be free of the fallacy of mixed classification. With a satisfactory coefficient of correlation the method of partial correlation could be used. This would be of great advantage in those studies in which the data are not sufficiently numerous to enable subdivision of all the variables. If the data are numerous enough, partial association could be used. This is the association between x and y