st: Re: nocons option

[Kit Baum <baum@bc.edu>]

In the model including a constant, is the constant estimate 
distinguishable from zero? If it is, it is absolutely inappropriate to 
respecify the model to omit it, since you are then asserting that the 
response variable is homogeneous in the regressors (that is, the 
expected value of y with all x==0 is 0). On the other hand, if the 
constant is insignificant, you may be justified in omitting it. 
However, this should be checked for every respecification or subset of 
the data you employ: e.g. the constant might be insig. over the whole 
sample, but highly signif. if you included only males, or only females 
in the model.

Kit Baum, Boston College Economics
http://ideas.repec.org/e/pba1.html

On Jan 19, 2005, at 2:33 AM, Omar wrote:


>   Since estimating linear models with and without the constant 
> (specifying
> the "nocons" option in stata) results in coefficient estimates with
> different standard errors and p-values, is it "cheating" to prefer a 
> model
> without a constant if some covariates are statistically significant in 
> that
> model but not in the one which includes the constant? 
*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/