Robust standard errors vs Clusterd standard errors

[Joris Berg]

Dear community,

For my research I'm doing a sectoral analysis of buyouts. To research a specific part i'm doing a normal OLS regression for my whole dataset (all the sectors combined) and for every sector separately. Because there is heteroskedasticity, I have to use robust standard errors or clustered standard errors. I can only cluster the standard errors on sector-level. This works fine for my general regression but when I'm doing the sector specific regressions I only have one cluster which causes all my coefficients to have a p-value of 0.000. This seems very odd and illogical to me so I was wondering if this is right or if I just have to use robust standard errors for all my regressions?

Kind regards,
Joris

[Carlo Lazzaro]

Joris:
without serial correlation of the epsilon (I assume that you're using -regress- and not -xtreg-), you should go -robust-, not -vce(cluster clusterid)- if you detected heteroskedastcity.
As an aside, instead of running N sector-specific regression, just add -i.sector- in the right-hand side of your regression equation.