Dear All,
I have a simple question related to GLM: Suppose I categorize a continuous
predictor (e.g. Age) into levels (eg. <30, 30-40, 40-50, and >50). Suppose
the p-value suggest that this variable is statistically significant in the
model. However, some levels are not statistically significant.
In the past, I've only look at the overall significance of the variable
rather than that for the specific levels. However, I found a paper that
suggests the following: "sometimes some levels of a categorical factor may
be clearly significant, while other levels may be less so. Although the
factor as a whole may be statistically significant, this may indicate that
it is appropriate to re-categorize this factor, grouping the less
significant levels with other levels" Do you agree with this? Is this a
common practice?
Thanks in advance for your help.
Regards,
Lars.
|