Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

st: SUEST vs IV

From   John Antonakis <>
Subject   st: SUEST vs IV
Date   Thu, 07 Apr 2011 17:40:32 +0200


I am examining the effect of an endocrinological variable (X) as a predictor in a regression model; X correlates very strongly with being male.

Suppose the basic model is:

y = a0 + a1X + control variables + u

One way to account for the difference in X for males and females is to include the dummy variable male in the regression:

y = b0 + b1X + b2Male + control variables + e

I could also interact Male with X and all the predictor variables to remove any heterogeneity in the model due to being Male. Or a simpler way to do this is with seemingly unrelated estimation (suest). That is, I estimate the model separately for women and men and then stack the models with suest. In this way, I can clearly see the effect of X on y in both groups and do cross-equation tests as needed.

However, to see the general effect of X on y, while accounting for what causes X, yet another specification might be to instrument X with Male (because X is actually endogenous to Male):

ivreg2 y (X= male + other instruments) controls

If the Hansen J test is non-significant, it means that the effect of being Male on Y is mediated fully via X.

Of course, the suest and ivreg2 models are looking at different things; however, which is more defensible do you think?

I would be interested in your thoughts (and alternative modeling procedures if relevant).



Prof. John Antonakis
Faculty of Business and Economics
Department of Organizational Behavior
University of Lausanne
Internef #618
CH-1015 Lausanne-Dorigny
Tel ++41 (0)21 692-3438
Fax ++41 (0)21 692-3305

Associate Editor
The Leadership Quarterly

*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index