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

st: iweight and heckman

From   Marie-Benoit MAGRINI <>
Subject   st: iweight and heckman
Date   Thu, 5 Nov 2009 16:11:19 +0100


I am using weighted data, but my weights are not integer.

So I cannot use the “fweight’ option.

Therefore I have used the ‘iweight’ option with the estimation of different regression functions,

notably OLS estimation and Heckman estimation.

And I observe this difference concerning the number of observations:

When using ‘iweight’ within OLS regression, the number of observations displayed in the ‘Results’ window corresponds to the whole population, that is the sum of the weights.

When using ‘iweight’ with Heckman procedure, the number of observations in the ‘Results’ window corresponds to the number of observations of the unweighted data sample. But the estimated coefficients in the regression are not the same with or without ‘iweight’ option. So, something is going on when using ‘iweight’ option, but I would like to know why the number of observations displayed in the window is not the weighted one with Heckman. Although when using ‘fweight’ (if the weights are integer) with Heckman, the number of observations corresponds to weighted data.

Perhaps it is this reason: as the weight coefficient I use is not an integer, when the regression is done over all the observations, the sum of this weight coefficient is well an integer. Otherwise if we select a part of the data (what it is done with the Heckman second step), the sum of weights is no more an integer. Therefore it appears a little strange for STATA that does not displayed the weighted observations in the Results window ???

Could anyone give me some details concerning iweight option with the heckman estimation ?

best regards,

mb magrini

using Stata 10

Marie-Benoît MAGRINI
PhD in Economics
INRA - French National Institute for Agricultural Research
BP 52627
31326 Castanet Tolosan
Phone: 33 (0)5 61 28 54 22
Fax: 33 (0)5 61 73 20 77

*   For searches and help try:

© Copyright 1996–2015 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index