General knowledge about regressions

[Jakob Berggreen]

Hi Statalist

I am not quite sure if this is the right place to post it, and if there is any help to get. Note though, I am kinda new to econometrics. But I am writing a thesis regarding acquisitions in the US. The topic concerns whether a CEO makes acquisitions locally or nationally based on a KM distance of the acquisition. So the dependent variable would be a measure of how far is acquisition from their headquarter, based on the KM difference from longitude and latitude. The topic is also written below.
Master topic: The association between CEO narcissism (and overconfidence - controlling for Big Five) and the geographical dispersion of acquisitions inside the US

Though some help regarding the statistics, and what regressions to make is needed. So far I have made some thoughts about using RE and FE (robustness) as estimation method, but does it make more sense to make a simple linear regression instead. I have hard time understanding if there is a big difference between RE and linear regression, beside that the random effects is attractive when we think the unobserved effect is uncorrelated with all the explanatory variables.
The regression looks like this atm.:
xtreg kmdistance narcissism CEO-controlvariables FIRM-controlvariables i.year i.industry, re robust

Can anyone help me explain what the differences is, and if it is a good idea to use RE and FE?

Thanks for the help in advance.

Kind regards
Jakob

[Carlo Lazzaro]

Jakob:
welcome to this forum.
I find always difficult to advise on words rather than on numbers (please see the FAQ on how to post more effectively. Thanks.).
That said:
1) if you have N>T (ie, short) panel dataset, you should start with -xtreg,fe- (assuming that your regressand is continuous);
2) however, please note that -fe- estimator wipes out time-invariant variables (such as -i.industry-, unless a company changes industry during the time horizon panel dataset stretches over);
3) using -fe- as a robustness test for the -re- specification (or the other way round) makes less sense;
4) please also note that the main -re- assumption (zero correlation between the ui panel-wise effect and the vector of regressors) rarely holds.

[Jakob Berggreen]

Thank you very much, I will ofcourse look into the FAQ.

Further thank you for the answer, this makes a lot of sense. But a question comes to me, as I have researched panel data a bit, and I am bit unaware if this even is panel data.
Because my data might have the same independent variables twice, as only the dependent variable changes, due to the acquisition, and they might acquire companies twice a day. I have tried to attach a picture with a snap of the data, where you see that deal number 6 and 7, has been done the same year, and the only difference in the 2 lines could be the dependent variable (the distance measured in km).
So when I try to -xtset FiscalYear DealNumber- i get an unbalanced panel, but that is maybe just how it is?

FiscalYear is the year that the acquisition happenened
DealNumber is just a number assigned from 1 ... N for the number of observations

[Carlo Lazzaro]

Jakob:
1) reading (and acting on) the FAQ implies avoiding screenshots and use CODE delimiters, instead, to post what you typed and what Stata gave you back. Thanks;
2) the main elements of a panel dataset are a cross-sectional (or N) dimension (ie, the very same sample) measured repeatedly across time (time-series or T dimension) on the very same set of viariables.
It is perfectly usual that some -panelid- are not present in all the data waves (unbalanced panel).