value of R2 over the earlier study. The standard error of
estimate for the lung cancer equation in the present study
was slightly less than that from the earlier investigation.
The regression equation for breast cancer in the present
study included 2 predictor variables as opposed to 23. These
2 were selected from a total of 13 independent variables
initially evaluated. The equation resulted in a value of R2
that was considerably larger than the value from the earlier
study, as well as in a smaller standard error of estimate.
The validity of the comparison of these two sets of
results is less for the rectocolon group because the earlier
study included no cases of rectocolon cancer. However,
again, the present study used an equation with 2 as compared
to 24 variables that had a slightly lower value of R2 and a
lower standard error of estimate.
The improved efficiency of the regression equations is
the result of the selection of a small number of relatively
powerful variables. In the present study, the identification
of these began with the methods used to code information for
important treatment variables and included the elimination of
variables with little predictive power from the regression
equations.
In the 1974 investigation, each single treatment
modality and all interactions between treatment modalities
and other variables were used as independent variables. For