Hi,
I found my GLM intercept is not significant. Then I explored a bit and came to know that if I center the data (x-mean(x)), it will be significant.
Is it recommended? Should I go for it?
Your answer is highly appreciated.
Many thanks.
Hi Madan,
On the contrary of explanatory variables, it is recommended (except for very specific cases where there is a good (theoritical) reason not to include an intercept) to keep the intercept term, even when not significant. Indeed, excluding an intercept term, even when it is not-significant, can result in biased beta-estimators for explanatory variables.
See for example : http://stats.stackexchange.com/questions/7948/when-is-it-ok-to-remove-the-intercept-in-a-linear-regression-model ou bien Brooks, C. (2008). Introductory Econometrics for Finance. Cambridge University Press, second edition.
So to conclude, there's no need to spend too much time about this issue and I suggest you to keep the intercept in your first regression.
I hope this helps,
Best regards !
When you center a variable the interpretation will change. It is up to you to center or not.
You don't need to make a coefficient significant in order to preserve it in the model. Intercept term generally should remain in the equation regardless of its significance.
There are lots of reasons to center predictor variables---to help a model converge (if you're using something with minimum likelihood estimation rather than OLS regression), or, usually, to make coefficients more interpretable when you have interactions with continuous variables. However, "to make the intercept significant" is not one of them.
In typical applications one often doesn't even care whether the intercept is significant (as usually the hypothesis under investigation is not about the intercept). If you do have a reason to care whether the intercept is significant, that is fine, but it is not advisable to mess around with the data just to try to get things significant (this is called "p-hacking", you can find many articles about it online). The choice of whether or not to center the data should be for independent reasons.
- Badges
- Science topic