Difference between -areg and -xtreg

[Geert van Hooff]

Hi all,

I ran the command -areg y1 x2 x3 x4 i.year , absorb (clustervar) to test my hypothesis. However, I came to the conclusion that my dependent variable: the net score of CSR (CSR strengths-CSR concerns obtained from WRDS) might not be a true interval variable and OLS regressions aren't appropriate. First question: someone agrees/disagrees with this? I can't find any scientific findings about this.

Second question: What is the difference between my areg command and the command : -xtreg y1 x2 x3 x4 i.year, fe , when I set the same cluster variable before with xtset (clustervar)? I found a paper which paper which uses the fixed effects model to test X on the net score of CSR as well (similair setting as my thesis) and they used a fixed effects model. However, when I run a fixed effect regression I don't find any differences with the -areg command.

Thanks in advance!

[Clyde Schechter]

However, I came to the conclusion that my dependent variable: the net score of CSR (CSR strengths-CSR concerns obtained from WRDS) might not be a true interval variable and OLS regressions aren't appropriate. First question: someone agrees/disagrees with this? I can't find any scientific findings about this.

This is a multi-disciplinary, international forum. Please read the FAQ for good advice on how to post questions clearly, so that your likelihood of getting a timely and helpful response is maximized. Among other considerations, don't use abbreviations that won't be known by all reasonably well-educated people regardless of their line of work or expertise. What on earth are CSR and WRDS? I suppose it is possible that anyone who would be able to answer this question would necessarily recognize those abbreviations--but why limit the range of possible responders. The question might be more generic and answered by somebody who just knew what you're referring to here.

Second question: What is the difference between my areg command and the command : -xtreg y1 x2 x3 x4 i.year, fe , when I set the same cluster variable before with xtset (clustervar)?

The -areg- and -xtreg, fe- commands are essentially the same. There is a minor difference in the calculation of degrees of freedom. See http://www.stata.com/statalist/archi.../msg00596.html for more details.

[Phil Bromiley]

You'll increase your chances of a helpful answer by following the FAQ on asking questions - provide Stata code in code delimiters, Stata output, and sample data using dataex.

Your CSR variables come from something like Likert scales - scales of 1 to 5 or some such. That means they are ordinal, and do not have interval measurement properties. Strictly speaking, taking the difference between ordinal scales makes no sense.

You might also want to test whether strengths and concerns are even on the same continuum. At least some recent CSR work finds that they are not measures of the same single dimension but rather are separate dimensions. If they're not on the same continuum, then you'll probably get different results with a model explaining concerns than a model explaining strengths.

While areg and xtreg are essentially the same, many (most?) prefer xtreg when working with panel data. Xtreg has some options that areg doesn't and also makes it easy to test whether you need fixed or random effects (a hausman test).