The findings of the exploratory regression analysis of
the main effects and the one of main and interaction effects
were used to select a fewer number of variables for use as
independent variables in four additional regression analyses.
The variables selected were age, stage, histology, initial
treatment, subsequent treatment, and the interaction of stage
and initial treatment. At this point in the analysis, the
dependent variable was transformed by taking its natural log;
that is, the natural log of survival. This procedure was
carried out in order to correct for the skewed distribution
of this variable.
Four regression models were studied in this series of
analyses. In the first age, stage, initial treatment, and
subsequent treatment were used as independent variables. The
regression analysis was based on 214 of the sample of 266.
Cases excluded were those with missing values for stage and
cases with zero months survival. A comparison of the group
of 214 cases to the sample of 266 is given in Tables 37 41,
Appendix F.
1
Because the natural log of zero cannot be computed,
all cases with less than one month of survival, which were
coded as zero months, were excluded from the regression
analysis. Since the model used for prediction was not used
to predict any cases with less than one month survival, this
approach was deemed acceptable.