I am conducting cox proportional hazard regression modelling. Two predictors in the model, for example, the age and height, have significant interaction effects on the death outcome when both using as continuous variables (age * height). But when I categorized age into <30, 30-50, >50 groups, and added the interaction terms as age_group * height into the model, the interaction effect was not significant any more. I don't know how to interpret this result.
$\begingroup$
$\endgroup$
1
-
2$\begingroup$ Does this answer your question? How to best model interaction effect of two continuous predictor variables? $\endgroup$– kjetil b halvorsen ♦Aug 21, 2020 at 16:39
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
1
It is already hard to get sufficient statistical power for an interaction effect. By categorizing a continuous variable you are loosing statistical power, so it is not that surprising that interaction effects for the categorized continuous variable is no longer significant.
-
$\begingroup$ Try instead to model the
agewith a spline, see stats.stackexchange.com/questions/193313/…, stats.stackexchange.com/questions/77819/… $\endgroup$ Aug 21, 2020 at 16:38