[Boliang Guo]

Hi Daniel,
the mimrgns has beening working very well but recently when I run the following code, I got error message, any help please?
mimrgns group@time, contrast

(here are output)

st_numscalar(): 3253 nonreal found where real required
mimrgns_return(): - function returned error
<istmt>: - function returned error

[Richard Williams]

Daniel may very well be able to answer your question as is but if it was me I would want to see the code and output for the command that preceded mimrgns. Use code tags (see pt. 12 of the FAQ). A replicable example, if you have one, would be great. Also make sure that you have the most current version of mimrgns:

. which mimrgns
c:\ado\plus\m\mimrgns.ado
*! version 2.1.5 18aug2015 daniel klein

[daniel klein]

This is a stupid bug that should have shown up while testing. I will send you private message for a quick fix and will send a fixed version to Kit Baum for upload on SSC. Sorry for inconvenience.

Best
Daniel

[daniel klein]

Thanks as always to Kit Baum, an updated version of mimrgns is now available from the SSC. In Stata use the ssc or adoupdate command to install the latest version.

The update fixes some minor and two somewhat serious bugs. The first concerns margins' undocumented saving() option. mimrgns accepted this function but the produced file contained incorrect results. The option is now no longer allowed.

The second bug concerns mi estimate's (documented) saving() option. mimrgns messed up the saved file, replacing it with its own estimation results without notification. This bug has been converted to a feature, as mimrgns can now work with ster-files, previously saved by mi estimate. This results in slightly shorter execution time.

Best
Daniel

[daniel klein]

Thanks once again to Kit Baum, another update for mimrgns is now available from the SSC archives. I trust you know how to do this.

The updated version now supports the atcontrast() suboption, as requested here.

The second change worth mentioning concerns fixing covariates in the at() option at sample depended values (like the mean). Prior versions would not report a single value and refuse to return the matrix, necessary to plot results, when one or more of these values varied across imputations. The updated version now reports the MI point estimate of such values and stores them in r(at) (or e(at), with the post option). A note will be issued below the coefficient table when the fixed values vary across imputations and I leave it up to others to judge the appropriateness of the way mimrgns handles the situation.

Best
Daniel

[daniel klein]

Thanks to Kit Baum, an updated version of mimrgns is now available from the SSC.

The update fixes some minor bugs, such as the one reported here.

Wolf Edler reported a problem when using mimrgns via a remote access that would not allow uploading .class files, such as mimrgns_work.class. The update fixes this problem; mimrgns_work.class is no longer required.

Technical note: To prevent any problems with .class files when installing mimrgns, mimrgns_work.class has been removed from the SSC archives. As a consequence, this file will not be automatically removed from your ado-path when you previously had mimrgns installed and update the files. If you feel the need to remove it, you will have to delete the file manually but this is not necessary for mimrgns to work correctly.

Best
Daniel