ivreg2 update (v. 4.1.01)

[Mark Schaffer]

With thanks as usual to Kit Baum, an update to ivreg2 (version 4.1.01) is available from SSC. The bigger changes/additions:

1. Support for factor variables.
2. Sanderson-Windmeijer (2015) first-stage F statistics for multiple endogenous regressors.
3. Option for reporting first-stage regressions as a single system (sfirst option).
4. partial(.) option accepts factor and time-series operators; also added a partial(_all) option.
5. Support for Stata output display options noomitted, vsquish, noemptycells, baselevels, allbaselevels plus an extra option bvclean that strips out all omitteds and base vars from the saved e(b) and e(V) matrices.

Minor changes and updates are documented in the help file and at the bottom of the ado file.

[Sergio Correia]

Thanks again for the update Mark, it's been working really great on my end!

Best,
Sergio

[Mark Schaffer]

With thanks as usual to Kit Baum, an update to ivreg2 (version 4.1.06) is available from SSC. The most noteworthy change is a major speed improvement when using large numbers of regressors, either when included in the regression output or when partialled out (and a big thanks to Martin Watzinger for the feedback that led to these improvements).

As usual, all changes are documented at the bottom of the ado file.

[CARLOS G OSPINO H]

Dear Mark,
I just updated ivreg2 to version 04.1.08. I was getting the "cluster option not supported if a panel spans more than one cluster r(198);" error when running xtivreg2 and clustering on a variable that changed across years in my panel. I skimmed the changes in ivreg2 but could not determined if something in the newer versions of ivreg2 intentionally corrected this error in xtivreg2, or if it was unintentional. Thanks. CO

[Mark Schaffer]

Carlos: can you provide more details about which ivreg2, xtivreg2, version of Stata, etc. you are using? xtivreg2 should still catch this problem. I just tried to recreate this error message and succeeded:

webuse abdata, clear
xtivreg2 n w k, cluster(year) fe

FYI the versions I have are ivreg2 v 04.1.08, xtivreg2 v 1.0.17, and this was under Stata 13.1.

[CARLOS G OSPINO H]

Mark: I just ran this code:
xtivreg2 n w k, cluster(id year) fe
And got some output.
I'm running ivreg2 v 04.1.08, xtivreg2 v 1.0.17, and Stata 13.1
CO

[CARLOS G OSPINO H]

BTW,
To me is just fine. It is exactly what I need in my current research to cluster both by panel id and a variable which changes over time (Firm Size). I'm just curios whether this was intentional or not and if you do decide to correct it, how can I still get the code to run. Best, CO.

[Mark Schaffer]

Carlos: this is indeed just fine. It is just a case of 2-way clustering, and one of the two cluster dimensions is where the fixed effects appear (so in that dimension the panel does not span more than one cluster).

This will generate an error, though (using the same dataset):

webuse abdata, clear
xtset, clear
xtivreg2 n w k, cluster(id year) fe i(ind)

[CARLOS G OSPINO H]

Mark: Thanks for clarifying. CO

[Mark Schaffer]

With thanks as usual to Kit Baum, an update to ivreg2 (version 4.1.09) is available from SSC. The most noteworthy change relates to update in 4.1.06 (from 4 July) with speed improvements when partialling out - a bug in partialling out combined with the use of weights and a constant term has now been fixed. The obscure bug that Theresa Fuchs spotted (thanks Theresa!) has also been fixed (http://www.statalist.org/forums/foru...63#post1307063). Also worth noting is the addition of the center option in 4.1.08 for centering moments. More details are in the notes at the bottom of the ado file and/or in the help file.

[Ariel Karlinsky]

I'm sorry for bumping a somewhat old thread, but I just started using ivreg2 (more specificly: xtivreg2) and though ivreg2 supports factor variables, xtivreg2 still does not. Is there an update planned for xtivreg2 to allow the usage of factor variables there as well?

[Mark Schaffer]

Good question. It's on my to-do list, but it's a non-trivial programming exercise and I'm not sure when I'll get to it, so I guess the answer is "yes, but".