Can I regress panel data with OSL under this specific circumstance?

[Arren Zhou]

Hi everyone

I want to do regression analysis on a dataset with election outcomes in all countries in the world, say, to explore the relationship between party polarization and voting turnout. Since the elections are held in different years and the gap between two elections are all different by country, am I supposed to treat this dataset like analyzing a cross sectional dataset and just simply run OSL on it? Will I have serious consequences that reduce the validity of my research?

For example, the dataset may look like this:

Argentina	ARG	2007/1/1	Presidential	5	1	0	4
Argentina	ARG	2011/1/1	Presidential	4	1	0	4
Argentina	ARG	2013/1/1	Presidential	1	1	2	4
Argentina	ARG	2017/1/1	Presidential	2	1	2	4
Argentina	ARG	2019/1/1	Presidential	4	1	0	4
Armenia	ARM	1995/1/1	Presidential	5	0	1	-999
Armenia	ARM	1999/1/1	Presidential	1	1	4	4

Thank you.

[Carlo Lazzaro]

Arren:
(pooled) OLS does not shelter you from wide difference in -timevar- across countries.
Therefore, the main issue relates to your research goal and tribal rules in your research field.

[Arren Zhou]

Dear Lazzaro,

Thanks! I will double check it with my advisor!

[Carlo Lazzaro]

Arren:
wise idea indeed!

[Maxence Morlet]

What Carlo said is entirely true. So your dependent variable is voting turnout, and regressor of interest is party polarisation, correct?

I would, like many social scientists, simply throw a two-way fixed-effects (TWFE) model at it to mitigate at best confounding bias. You will not find causal results with pooled OLS, however these results may still be interesting; they will be descriptive, not causal. Maybe you that is just what you want.

Quick question, would you not expect reverse causality here? As in, if I was correct in my first question, could it not be that voter turnout in fact also affects party polarisation?