Panel Probit Regression

[Chanik Jo]

In running panel probit regression,

if I use "" xtprobit dependent_variable independent_variable, fe vce(cluster code) "" , the results shows "Fixed-effects model not available".

So, instead, I use "" probit dependent_variable independent_variable i.code, vce(cluster code) "". And, I got the results,

First question : can I call my model fixed effect probit regression? or does my regression have meaning even if it is not fixed effect or random effect model?

Second question : Is it reasonable to use clustered standard error ( vce(cluster code) in above case) in probit regression?


Sincerely,

Chanik Jo

[Elif Krkmz]

Dear Chanik,

If you want to use probit model for panel data you should use "xtprobit". xt makes probit model available for panel data set. And for panel probit there is no option for fixed effects, you have only two options called random effects and population average.

[Maarten Buis]

If you need a fixed effects model for binary data you can use xtlogit with the fe option.

[Chanik Jo]

Dear kind answerers,

Is there any way to obtain pseudo R-square in panel probit Random effect model or logistic fixed effect model?

When I tried, only log likelihood value is reported in STATA.

[Dana Selim]

Dear all,

I am using panel data for 156 countries for the period (1960-2014), where the majority of variables are derived from the World Bank database except for one variable, so I have pretty much missing values. I split the sample into 5 groups to test the following model for 4 groups of countries : (MENA region, Subsaharan countries, OECD countries & Developed countries). My dependent variable is a binary variable.

Pr(Ethnic Inclusion_it) = β₀ + β₁ OilRents_it + β₂ CoalRents_it + β₃ ForestRents_it+ β₄ NatgasRents_it + β₅ GDPPER_it + β₆ oilRents²_it + β₇ UurbanPopulation_it + β₈ MineralRents_it + v_i+ Ø_t+ ɛ_it

The 1st Question is : I ran the Panel Probit model with random effects and the majority of variables were insignificant. However, when I ran the Probit model (Standard probit), the results had logical significant signs for the 4 groups of countries as well as for the general model, which is for the whole sample.
In fact, I know that if I have panel data, I should run the Panel Probit or Panel Logit but my question is : is there anyway to justify relying (even to some extent) on Probit model as it produces very logical results, based on literature ?
I checked the rho for the 5 models for all groups of countries and it was between ( 0 and 0.8) but the LR test of rho=0 at the end of the Panel Probit table : chibar2(01) = 2.4e+04 generates this Prob >= chibar2 = 0.0. So if the test tells me that there is a difference between Probit and Panel Probit, is there any other way to justify relying on the two techniques (Probit and Panel Probit) not only the Panel Probit ?

The 2nd Question is: Am I in need to justify in my paper why I chose the Random effects model rather than Average population ? And if Yes, what will be the explanation of choosing random effects?

Any suggestion and explanation from your part will be extremely useful to me.
Thank you so much in advance.

Sincerely,
Dana

[Carlo Lazzaro]

Dana:
1st Question: statistical significance should not be the main goal of your research. You should give a fair and true view of the data generating process underlying the population from which your sample was (randomly) drawn, instead.
That said, you do not state if you -cluster()- your standard errors (as you should have to) when you switched from -xtprobit, re- to -probit-;
2nd Question: the literature in your research field should inspire your methodologica choices.

[Dana Selim]

Carlo Lazzaro Thank you so much Mr. Carlo for replying. Just another question and excuse me if this sound stupid because am new to panel data:

How to make robust standard errors in Probit model with panel data? Do I have to use clusters standard errors with this command : probit y x, vce(cluster country) or I can use the basic robust command : probit y x, robust in case I assume that individuals are independent within each sample ?

[Carlo Lazzaro]

Dana:

 probit y x, vce(cluster country)

If you have multiple waves of data on the same panel unit, how could your observations be independent within each panel?

[Dana Selim]

Carlo Lazzaro Thank you Sir for being helpful.

[Carlo Lazzaro]

Dana:
Carlo is enough.
Although respectful and appreciated, Sir makes me feel somehow older that I am!

[Juan Miranda]

Good afternoon.
We understand that the theory of panel data has the restriction of the conformation of the sample; this is a follow-up through the years of the same individuals surveyed. Perhaps because of how difficult and expensive, it is to meet this condition, sampling with replacement usually carried out.
However, what happens? Is it academically feasible to use the modelling of fixed effects, random effects, etcetera? In samples that do not meet the theoretical conformation of the panel data?
I think many of us use a panel type conformation without ensuring that the individuals are the same through the years.
Please, I would like your opinion, and if anyone has a bibliography that clarifies my question, I will greatly appreciate your contribution.
Thanks

[Jeff Wooldridge]

Juan: The proper way to think about panel data with individuals is that you sample once from a list of individuals (or families, firms, and so on) and then you follow those units over time. If you take a new sample from a large population every year, that's not panel data. That's repeated cross sections.