Interpreting overall and individual p-values for categorical variable in regression

[Bree Shi]

Hi all,

I have a question regarding how to interpret the regression output for a categorical variable with more than 2 categories.
As an example, the output for a linear regression that looked at quality of life and included ethnicity (p=0.60). Ethnicity was categorised into white, Asian and other. For Asian vs White, p=0.03. Can I then say that ethnicity is not significantly associated with quality of life, however, Asians ethnity is significantly associated with a better quality of life score compared to whites?

I have always been told at school that if the overall p-value is not significant, then exclude the variable altogether. If overall p-value is significant, when including the result, don't include the individual p-value. But i tried to look at the literature as to why, but I did not find anything useful.

Many thanks for your help in advance!

Kind Regards,
Bree

[George Ford]

The coefficients of a categorical variable (i.cat) represent the differences in the categories with coefficients relative to the one without a coefficient (the base). You can change the base (e.g., b2.cat). The t-stats on the coefficients are a direct test of the difference from the base. To test other relationships, you'd use margins.

You can parmtest all the coefficients related to a categorical variable. Typically, if one coefficient is stat significant, then the joint test is likewise stat significant.

If you really think it ought to be in the model for theoretical reasons, then put it in the model, irrespective of the p-values.

[Carlo Lazzaro]

Bree:
I'm clearly with George about not following p-value to include/exclude a given predictor from the hand-right side of your regression equation.
What matters is giving a true and fair view of the data generating process you're investigating.
As an aside, please read and act on the FAQ when posting: your query would be easier to reply if you shared with interested listers what you typed and what Stata gave you back. Thanks.