1

My panel regression model is as follows:

$$Y_{it}= PS_{it}+PF_{it}+EF_{it}+ e_{it}$$

where

$i$ : country
$t$ : year
$Y_{it}$ : GDP per capita
$PS_{it}$ : Political stability
$PF_{it}$ : Political Freedom
$EF_{it}$ : Economic Freedom
$e_{it}$ : error term

I want to determine whether a country that wants to develop gives political freedom to its people first or economic freedom. Is it better to use "stepwise regression" or "hierarchical regression"?

whuber
  • 281,159
  • 54
  • 637
  • 1,101
malik fahim
  • 73
  • 4
  • Stepwise selection for multiple regression is a bad idea. If that doesn't make sense, or you want to understand why, it may help you to read my answer here: [algorithms-for-automatic-model-selection](http://stats.stackexchange.com/questions/20836//20856#20856). – gung - Reinstate Monica Apr 06 '13 at 16:04

1 Answers1

1

You have an endogeneity problem here. A per-capita increase in wealth could increase economic freedom and/or political stability. Without dealing with this your parameter estimates are going to be uninformative. Your design here needs work before you worry about estimating a model.

Stepwise regression would be a bad choice here. There are other (better) options if you want regularization. Taking account of the structure of your data (by not pooling completely) is probably a good choice though, so some form of hierarchical model is appropriate for panel data.

Zach
  • 335
  • 2
  • 11
  • Thanks @Zach and @ Gung for your comments. There is disagreement among economist about role of political freedom and economic freedom in economic growth. Some scholars have view that economic freedom leads to political freedom. To measure the joint effect of both variable on economic growth i can introduce interaction term of economic and political freedom but i am unable to answer the question weather for less develop country is it give political freedom to its people first or economic freedom on the basis of econometric technique . – malik fahim Apr 07 '13 at 02:05
  • yes I understand the disagreement. As you suggest you can't establish causality here since you have reciprocal causation (endogeneity). The best you can do with this design is measure association. – Zach Apr 07 '13 at 03:07