[Clyde Schechter]

That is correct. You do not have to include country indicators in your -xtpoisson- command. In fact, if you try to add them, Stata will throw them out anyway because they would be colinear with the automatically represented country effects.

[Peniana Patrick]

Thank you very much Mr Schechter

Since im dealing with gravity model for trade, I know that distance is an important predictor variable however, the xtpoisson..fe output stated that
"ln_dist dropped because it is constant within group".

Any suggestions on how to deal with the distance variable?

[Clyde Schechter]

I can't help you with this. While I have heard of gravity models, I am an epidemiologist, not an economist, and I know next to nothing about how they work.

But let me make a few comments that are generic about fixed-effects models:

In your #90 you referred to -xtset country year-. If this is what you did, Stata is telling you that your variable ln_dist is constant within country. That would indicate that your data set is wildly incorrect: no country lies at the same distance from every other country! So you need to get the correct data before proceeding.

Alternatively, perhaps you did something equivalent to -xtset country_pair year-. In that case, ln_dist would, indeed, be constant within group: the distance between any two countries does not change over time (unless we are talking geologic time scales). If that is what you have done, you cannot include ln_dist in a fixed effects regression. This is not some peculiarity or limitation of Stata: it is a general property of fixed effects estimators that you cannot estimate the effect of anything that is constant within those groups, and any effect that they have on the outcome is already accounted for by those fixed effects. So if the gravity model is crucially dependent on ln_dist, I can only conclude that it should not be fitted with a fixed-effects estimator with country-pairs as the group variable. I do not, however, know what you should do instead. There are people on the Forum who are experts in gravity models, and perhaps one will chime in.

[Joao Santos Silva]

Hello again,

Adding to Clyde's comments, the fact that distance is dropped is not a problem, it just means that once you control for fixed effects the distance variable contains no additional information. So, tat is fine. Of course, you do not get to see a coefficient for it, but that is a result of the nature of your data.

Best wishes,

Joao

[Peniana Patrick]

Mr Schechter & Mr Santos, thank you very much for the valuable comments you've made.

Kind regards

[Peniana Patrick]

Also, How would I interpret the coefficients estimated by xtpoisson..fe?

[Said Jafar]

Dear Peniana,

You can find answer to your last question and many other questions about the gravity model in this new book. It is free of charge. You can also download all stata files from here and replicate the results.

Hope this helps
Dias

[Peniana Patrick]

Hello Dias,

I am going through the book now.

Thanks you

[Clara Leanderson]

Hi, I have a question regarding the xtpoisson command. I a trying to run a gravity model according to the following equation
xtpoisson importvalue gdp_imp gdp_exp pop_imp pop_exp dist contig colony language time_i
where time_i represents time to import. When I do the estimation with the fe command, stata just keeps on doing iterations and (backed up). What does this mean?

Thanks in advance!

[Joao Santos Silva]

Clara,

You probably have "perfect predictors". One option is to use the -ppml- command (available form SSC) and explicitly include the dummies for the fixed effects.

Best wishes,

Joao

[Clara Leanderson]

Hi,

thanks for your answer! I tried doing the ppml approach instead, this time using fixed effects dummies for importers, exporters and years. When I include the importer effect, the time to import variable becomes insignificant. Only running fixed effects for exporters and years give significance, which I believe is correct. Is it possible that the effect is captured by the importer fixed effect? The one only running with exporter effects gives the following results:

I tried running the xtpoisson with fixed effects for country pairs and found small significant effects for time to import. Are my interpretations right for the variable? A one day increase in time to import results in exp(-0,002)=0,9973818233, a decrease of 0,26% in trade flows.

Thanks for your help!

[Joao Santos Silva]

Dear Clara,

It looks as if you are not using logs of the GDP variables; you should do that before anything else.

Best wishes,

Joao

[Peniana Patrick]

Dear forum members,

I used the Xtpoisson, fe model to estimate a gravity model of trade which dropped the time-invariant variables such as distance and common colony. But I am also interested in determining whether historical ties has any impact on current exports. Literature suggest estimating a second regression where the individual effects will be the dependent variable are the time in-variables will be the independent variable. My question is how do I get the individual effects? Does the individual effects equal the residuals from the fixed effects estimation. In the pooled effects model, the constant term is the individual effects however there is no constant term when using the FEM.

I appreciate any comments or suggestions on how to proceed with this

[Joao Santos Silva]

Dear Peniana,

Please see this thread, especially the second post.

Best wishes,

Joao

[Peniana Patrick]

Noted with many thanks Mr. Santos.