Re: st: significant to non significant variable

[Nick Cox <njcoxstata@gmail.com>]

You asked this question twice before.

http://www.stata.com/statalist/archive/2014-01/msg00344.html

http://www.stata.com/statalist/archive/2014-01/msg00340.html

Please see advice at
http://www.stata.com/support/faqs/resources/statalist-faq/#noanswer
and in particular the firm statement that you are allowed at most one
repost.

In this case, despite your claim of "no collinearity", "no
mu[lti]collinearity", it does seem likely that relationships among
your predictors need closer scrutiny. Collinearity is not a yes-no
property, but a matter of degree.

Also, I wouldn't make a fetish of significance. Sometimes, it is
reasonable to include a predictor that is not significant at
conventional levels in a model.

Perhaps you should look at

xtmixed y X1 X3 || country: || Microfinance

Perhaps X1, X2, X3 all should be included. Much depends on your
economic grounds for including those predictors. With such anonymous
names, I doubt that even conomists could comment on that choice.

Nick
njcoxstata@gmail.com


On 14 January 2014 11:36, Ayayi, Ayi <Ayi.Ayayi@uqtr.ca> wrote:
> - I have stata IC 10
> - I am conducting three-multilevel  analysis let say
>
>
> xtmixed y X1 X2 || country: || Microfinance . In this regression  X1 and X2 are statistically significant
>
> Now I add  X3  the regression becomes:
>
> xtmixed y X1 X2 X3|| country: || Microfinance
>
>  In this regression X2 is no longer significant. The same problem arised when I add more than one variable
>
> I've checked for mulicollinearity among the variables with  VIF and Collin and all the VIF . There is no  collinearity . All the ViF are less than 2. So there is no mucollinearity
>
> Could you be please let me know what is the problem and how can I handle it
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/