Regression with year and country fixed effects

[Maria Sousa]

Hi there,

First of all I am new to this forum and I've been trying to find a relatable post but wasn't able to find it. I'm running a regression to study the potential relation between accruals earnings management and corporate social responsibility. I included several control variables but I followed previous studies and included the variables year and country to control for specific year and country effects.

So, this is what I ran on Stata:

xtset id year
egen year_country = group(year country)

xtreg REM CSR AEM GROWTH SIZE LEV ROA BIG4 IND, fe vce(robust)
xtreg REM CSR AEM GROWTH SIZE LEV ROA BIG4 IND i.year_country, re vce(robust)
xtreg REM CSR AEM GROWTH SIZE LEV ROA BIG4 IND i.year i.country, re vce(robust)

My question is: does this fulfil the propose of my research? And if I want to export the coefficients should it be from the first regression with fixed effects and cluster robust?



Thank you in advance for your help.

[Phil Bromiley]

You will increase your chances of useful answer by following the FAQ on asking questions.

I suggest you look very carefully and broadly for related work. I know I've seen a paper similar to what you propose in the last year. That doesn't mean yours won't be legitimate and valuable – there are many ways to measure accruals and corporate social responsibility and it appears you're doing it on an international data set.

The issue is what do you want to hold constant and what do you want to be the definition of the within condition. You are automatically getting ID fixed effects so the fe analysis is only looking at variation in REM and CSM over time within firms. This may be problematic if a major part of the action is stable differences in REM and CSM – you might look at a hybrid model.

Including year as a control is quite standard. Including year by country is also reasonable but certainly not standard in the literatures I read. Since most firms are only in one country, the firm fixed effects in the last model would fully take care of any country effects making them unnecessary. Why exactly you would estimate one model with fixed effects and the other two with random effects is unclear to me.

I would look at how people in the journal you want to publish in as set up these models. There are different norms for what needs to be controlled for in different research domains.