Xtabond2 with twostep, robust, and small

[Manuel Almodovar]

Dear Stata users.

My question concerns the use of Xtabond 2 in conjunction with the twostep, robust, and small commands.

If twostep is used together with robust, “where the errors are already robust, robust triggers the Windmeijer correction” (Roodman, 2009:123). Three questions then arise for me:

1) if we use twostep + robust+ small at the same time would we be being redundant...would we be telling STATA to use the Windmeijer correction twice?

2) What difference would there be between the following combinations.

2.1 Using twostep + robust
2.2 Using twostep + small
2.3 Use twostep + robust + small

3) Which would be the correct option if the aim is to obtain robust errors and the Windmeijer correction at the same time?

Thank you very much in advance

[David Roodman]

No, they are not redundant. The twostep errors without robust are "robust" in the technical sense that as the width of the panel goes to infinity, the standard errors will correctly account for correlation of errors over time. But in finite samples, they are usually quite biased. The Windmeijer correction fixes this to a wonderful extent. Small introduces classical small-sample corrections. You should find they don't make a big difference. I recommend robust + small or twostep + robust + small.

[Manuel Almodovar]

Thank you very much for your precise answer Dr. Roodman.