too many signficant variables

[Romuald Landry]

Hello
I am doing a multilevel regression using PASEC2014 data
at the first level I included several variables specific to the student
I find that all the variables are significant
I find it a bit suspicious
Please help me to know if there are problems and how I can solve them

[Felix Bittmann]

Most people complain when their variables are not significant and not the other way round to be serious, why should this be a concern? How many variables? How many cases? What about theoretical assumptions, do you have any why these variables should be related? Maybe look at effect size and not at p-values to get a better understanding what really matters. If your N is large, you can get very low p-values (with minimal effect sizes).

[Maxence Morlet]

Bonjour Romuald,

It all looks good to me, the estimate and std error of var(_cons) confirm the appropriateness of a random intercepts model as it is highly significant, you may want to try a random slopes model and cross-random effects model as well if it is not too computationally difficult.

Subsequently, store your estimates and select the most appropriate model using information criteria.

That is what I would do in your shoes.

The variance partioning coefficient (1789.477/(1789.477+5989.108)) seems adequate.

From a theoretical perspective, are your estimates robust to the addition to new (relevant) explanatory variables? You may perhaps want to conduct a few robustness checks and diagnostic tests.

Someone else may however jump in with a wise recommendation or highlight an issue.

[Maxence Morlet]

Very wise answer from Felix, hadn't seen it when I posted my reply

[Carlo Lazzaro]

Romuald:
others have already commented positively.
I would only add that reporting only a part of what you typed and what Stata gave you back (BTW: the preferred way to share the relevant items of your Stata session is via CODE delimiters; screenshots are deprecated for sound reasons explained in the FAQ) forces interested listers to unnecessary guess-work.

[Romuald Landry]

thank you all for your comments
I understand things better now