Lincom result when the main var is not sig. but the interaction is sig.

[davidbrown]

Hi members,

Got a quick question about linear combination of coefficients (using lincom command).
I am interested in a policy (binary var.)'s influence on Y (continuous dependent var.) depending on GDP (logged GDP per capita).
When I run a model, the policy var. doesn't have statistical significance, but the interaction with GDP has.

To see, if any, in what range of GDP, the policy var. has significance, I ran multiple sets of lincom with diffferent GDP values with certain interval covering all GDP value ranges.
And all of the lincom result has positive significant coefficient.

I have hard time interpreting this because if the policy has positive effect on all levels of GDP, shouldn't the variable have positive significant effect in the model before postestimations?
I wonder if somebody could put his/her opinion on this.

Thanks!

Best,
Gru

[Carlo Lazzaro]

David/Gru:
as per FAQ, one of the best ways to elicit positive replies from interested listers is to share what you typed and what Stata gave you back via CODE delimiters.
As per FAQ again, you can also provide an excerpt/example of your data via -dataex-. If your data are confidential, you can change the name of the original variables and observations or provide a fake dataset mimicking the same issue you're experiencing with yoiur real one. Thanks.

[davidbrown]

Carlo,
Thank you for the tips! I would be happy to show my commands and results, but I thought my question is one that can be answered without actually seeing the coefficients and significance. I can reformulate the question to a conceptual one that can be answered only by thinking but not by seeing the command and results if needed.
Best,
Gru