Can someone help me interpret those results? I get very high p-value in my probit model but I do not understand why...

enter image description here

Some details about my variables:

confucius = dependent (binary 1=yes, 0=no)

totalagri = continuous (tons)

bribes = percentage

size = continuous (km2)

coast = binary

agrigdp = percentage

impchina, expchina, gdp = continuous (value in $)

lifeexp = continuous (years)

urbpop = percentage

coalprod, nargasprod, oilreserv = continuous

diamond = binary

cropprod = index


  • 2
    $\begingroup$ Well... you have n=39 split up across 16 variables, for one thing. $\endgroup$
    – Alexis
    May 2 '14 at 14:28
  • $\begingroup$ Thank you Alexis for your reply! That indeed seems obvious (I'm quite in this whole regressions world). I've removed/replaced some of the variables in order to loose less observations and I end up with this: imageshack.com/a/img835/4459/a9cp.png Some p-values are still high but would this be enough for my hypothesis: H1: The higher the agricultural production (here squared) the more likely to have confucius H0: No relation between them $\endgroup$
    – Mafalda
    May 2 '14 at 14:29

As Alexis was already hinting, you have too many variables. It's time to throw out some variables, not add new ones. Or aggregate: e.g. all the oil+gas+coal production can be aggregated into the "energy" score.

Also, it seems to me that many variables (like "totalagri") should scale with the population size. I propose logging them all (totalagri, popsize. coalprod etc.). This way, all scaling relationships will be moved into the intercept term.

  • $\begingroup$ Thanks! I have made those changes and it seems to work much better! $\endgroup$
    – Mafalda
    May 4 '14 at 19:14

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.