1

I am conducting research for migration from third countries outside EU to Slovakia, Czechia, Hungary and Poland.

I have $t=11$ and $N=20$. I am using Gretl.

The only good, economic meaningful results are from pooled OLS. I understand that I should not use this method in panel data (and Breusch Pagan test confirmed me that), but all other results from FE and RE are not economically meaningful and they have completely different signs from pooled OLS.

Is there some way that I can use pooled OLS, or what should I do in this case?

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
Radoslav Gasper
  • 11
  • 2
  • Welcome to CV. What do you mean when you say *economically meaningful*? – Thomas Bilach May 16 '20 at 02:49
  • Hi, thank you for welcoming. – Radoslav Gasper May 16 '20 at 03:10
  • Hi, thank you for welcoming. Basically my hypothesis for diplomma is as follows: People migrate due to higher standard of living. Higher standard of living is represented with HDI index (Human development index - includes GNI, education and health). In model I always have HDIj/HDIi. i - origin country, j - final country. In pooled OLS Im getting HDI with plus - it supports my hypothesis. In RE, FE is HDI either insignificant or with minus. I am out of ideas, thats the reason why Im asking if I can use pooled OLS - all tests for pooled OLS are right. Thank you. – Radoslav Gasper May 16 '20 at 03:18
  • 1
    So you want to ‘pool’ your data and ignore some of the panel structure because it more closely aligns with your hypothesis? – Thomas Bilach May 16 '20 at 03:50
  • No, thats not the goal of this question. I dont want to ignore anything, thats the reason why Im asking. I want to get it right. I have "good" results in pooled OLS and opposite in RE or FE, thats why Im looking if it is possible to use pooled OLS, if no, then of course I will use RE or FE. I do not desire to "cheat". – Radoslav Gasper May 16 '20 at 04:09
  • Yes, it's possible to use pooled OLS. In fact, that is the base estimator in pooled or panel data models, the benchmark against which more complex models are to be compared, e.g., see Wooldridge's classic *Econometric Analysis of Cross-Section and Panel Data*. –  May 16 '20 at 14:15
  • Hello, and thank you for the answer. My problem is really weird. In my analysis I use Human Development Index which represents standard of living. For my thesis I need to calculate the ratio between country to which people migrate and country from which people migrate. Formula is as follows: HDI of country to which people migrate / HDI for country from which people migrate. What is weird, that in my analysis I always get very extreme coefficients (once with really high minus, once with really high plus). This ratio is then logged. I dont know what can cause this, all tests seems fine. – Radoslav Gasper May 18 '20 at 04:18
  • Can it be because it is index with values from 0 to 1? For example if I replace HDI ratio with GDP ratio I am getting normal values without any extreme. There is strong collinearity between GDP and HDI, thats the reason why I am not using it together in one model. But I suspect, that the results of these two variables should be similar. I reall dont know what to do, please if it is possible to answer this problem. – Radoslav Gasper May 18 '20 at 04:18

0 Answers0