solution for non significant coefficiant

[Hela BEJI]

Dear all,

I want to estimate the impact of agricultural and manufactured (high skilled and low skilled) exports on the average years of schooling (schl) in 49 countries (developed and developing) between 1995 and 2015 with 5 year intervals. My control variables are the death rate (per 1000 people), population, nominal GDP and migrant share. All variables are in log.
I used the OLS method with country and year fixed effect, the command is the following:

regress schl lnagri lnlowtech lndeath lnpop lngdp lnmig i.country i.year, robust

I obtained satifying results (in line with the theorie and previous empirical works) exept for the agricultural exports variable which is negative (the expected sign) BUT non significant.
What can I do to fix this problem?

You will find the output of the OLS regression down below.

Thank you very much in advance.

Hela BEJI

[Rich Goldstein]

my first comment is that you have the "wrong" SE's; using "robust" will correct (in a sense) for unequal variances but will definitely not correct for the clustering effect of having multiple years of data per country; you need to replace "robust" with "vce(cluster countryvar)" (replace countryvar with the name of your country id variable)

you don't show the full output (please show within CODE blocks as advised in the FAQ)

this is not an area I know anything substantive about but would certainly worry about (a) collinearity; (b) linearity (note that these are not given in order of importance);

finally, not sure what you mean by "fix this problem"

[Hela BEJI]

Dear Rich Goldstein
Thank you very much for your reply, i tride the vce option but it gave me this:

[Hela BEJI]

.

[Rich Goldstein]

in general, I would expect increased SE's when you use "vce(cluster clustervar)"; I am confused by your

but it gave me this

I remain confused as to what your question is - maybe someone else can help

[Tiago Pereira]

Hela's message was motivated by the fact that one of several coefficients was not statistically significant.

Hence, as far as I can understand, there is no problem at all to fix.