Panel data ttest

[Elisa Giaretta]

Hi,
I'm comparing data (sales) of private equity backed companies (variable pe=1) and non-private equity backed companies (variable pe=0).
I have panel data (846 companies over 6 years).
I would like to run a ttest for equality of means of sales in the two groups of companies.

Is the following command correct? ttest sales, by (pe)
Or should I use another command as I'm dealing with panels?

Thanks

Elisa

[Carlo Lazzaro]

Elisa:
welcome to the list.
Why not considering one of the -xt- regressions, as you're dealing with panel data?

[Rich Goldstein]

maybe a little more detail will help: using ttest will not be correct for panel data as the SE's will be too small; since regression is a ttest (e.g., when the only covariate is a 0/1 indicator (dummy) variable, you will get the exact same result as using the ttest command; further, if your panel data are unbalanced, even the difference in means supplied by the ttest will be incorrect; note also that you show a space after the "by" and before the parentheses and that will cause an error

as Carlo suggest, you could use xtset first and then xtreg:

xtset panelvar
xtreg sales pe

assuming that pe is coded 0/1

if the panel is balanced, or if you want the "raw" difference in means you could also use:

regress sales pe, vce(cluster panelvar)

in the above replace "panelvar" with the actual name of the panel variable