IVpoisson

[Rosmaiza Ghani]

Hi Professor Joao,
Hi everyone,

This is my very first post in Statalist. Like everyone else, I have so many problems in my research. For my second research paper, my data is a set of panel data(248 countries) as bilateral migration flows will be my dependent variable. My EEV (Endogenous Explanatory Variable) wilI be bilateral trade flows and average import tariff is the exogenous shock for EEV. I need to run these using IVPoisson since the dependent variable has a lot of zero values. But, when i ran them in Stata using 'ivpoisson lnMig i.Year ( lnTrde = AverTariff lnDist lnGdp_o lnGdp_d lnPop_o lnPop_d ln_Dist Adjcc Cc Lcns ),fe first' , stata said that lnMig is not a valid estimator. Prof Joao,is that means I should not log the dependent variable? but when I use only migration flow in the command, the outcome is still the same. I really appreciate if you could help me on this matter.

Thank you Prof.Joao

[Joao Santos Silva]

Dear Rosmaiza Ghani,

I think there are two issues here.

1 - After the first argument of the ivpoisson command should the the estimation method. You need to type gmm after ivpoisson and before the dependent variable to implement the estimator proposed by:

Windmeijer, F. and Santos Silva, J.M.C. (1997), Estimation of Count Data Models with Endogenous regressors; An Application to Demand for Health Care, Journal of Applied Econometrics, 12(3), pp. 281-294.

2 - Indeed your dependent variable should not be in logs.

Best wishes,

Joao

[Rosmaiza Ghani]

Thank you very much for your reply Professor Joao. Really appreciate your time.
I used gmm as the estimator now and the dependent variable isn't in log form anymore, I ran my data again with "ivpoisson gmm Mig i.Year ( lnTrde = AverTariff lnDist lnGdp_o lnGdp_d lnPop_o lnPop_d ln_Dist Adjcc Cc Lcns )" but stata gave me 'missing values encountered in analytic gradient' .
I read your paper Prof. about the demand for health care, it came together with stata command (through the log of gravity page) but I have problem to understand the command (still studying on them).

Thank you Prof.Joao

[Joao Santos Silva]

Dear Rosmaiza Ghani,

Try changing the units om Mig (for example, divide by 1000) and try starting with smaller models and then using the results as starting values for models with mode variables.

Best wishes,

Joao

[Rosmaiza Ghani]

Thank you Prof.Joao. I will try these now

[Rosmaiza Ghani]

Dear Professor Joao,

Thank you for your advice. After several trials, the outcome is obtained. But now I am stuck on how to interpret
coefficient after incidence rate ration (irr).

There are 2 questions that I am hoping you could help me with Prof.Joao.

1.If the coefficient obtained for the main explanatory variable which is trade=0.340, can it be interpret as " US$1,000s( unit used in data) increase in trade flows, is associated with 0.34 unit increase in migration flows"?

2. Will the sign of coefficient be the normal interpretation? because when I read this page "https://stats.idre.ucla.edu/stata/output/poisson-regression/", even though the irr coefficent was .9965,
the interpretation was "....would be expected to decrease by a factor of 0.9965".

Thank you very much Prof.Joao.

[Joao Santos Silva]

Dear Rosmaiza Ghani

I am afraid I do not work with IRR, in trade we usually work directly with the coefficients and they have a clear interpretation.

Best wishes,

Joao

[Rosmaiza Ghani]

Dear Professor Joao,

Thank you for your kind assintance previously. I really appreciate your help in resolving my difficulties.

Now am trying to run my unbalanced panel data using this command "ivpoisson gmm Mig i.Year (lnTrde=IV1 IV2) exogenous variables". will this count as fixed effect if I includes the i.year dummy?
Mig is the dependent, lntrade is the endogenous variable, and IV1 IV2 are the excluded instruments used
This command gave results.

But when I tried with this command "xi:xtpoisson Mig exogenous variables (lnTrde=IV1 IV2) i.year, fe", parentheses unbalanced
r(132); came out. Prof.Joao, I think this is becaue something is not right with my command. Really hope that you can help me with this matter.

Thank you very much for your time Prof.Joao.

[Joao Santos Silva]

Dear Rosmaiza Ghani,

Including the year dummy in the ivpoisson regression is OK. About the error with xtpoisson, I have no idea what can be causing it.

Best wishes,

Joao

[Rosmaiza Ghani]

Dear Prof.Joao Santos Silva,

I read about multilateral resistance and Poisson. A lot of papers included remoteness in their study. Do we take remoteness as multilateral resistance (MRT) in Poisson? Or should we exclude MTR if we are using Poisson?

Thank you very much Prof.Joao

[Joao Santos Silva]

Dear Rosmaiza Ghani,

Remoteness is an imperfect way of capturing MRT; the standard way to deal with MRT is to include origin and destination dummies and that can be done with Poisson regression.

Best wishes,

Joao