Shrinking estimates using meta, ebayes

[Aliza Husain]

Hi Everyone,

I have run into something weird.
I want to shrink some value-added estimates I have generated.
I use felsdvregdm and save the effect size and the associated standard errors (prin and prinse respectively).
I then run the command meta prin prinse, ebayes to shrink the estimates.
I get the following error: subcommand meta prin is unrecognized.

The strange part of this is that I ran the very same code five months ago and it ran without a glitch. The only thing that has changed over the last five months is that I upgraded from Stata 14 to Stata 16. Running version 14: before the command, however, does not fix the issue. I have checked to make sure that packages are installed and up to date. Absolutely nothing in the do-file has changed.
Any thoughts on what could be going on?

Thanks in advance!

Best,
Aliza

[Aliza Husain]

Hi All,

I reached out to Stata's tech support, and they helped me address the issue. I am copy-pasting their response here, in case it's of help to other people.

"I think this is because in your older version of Stata, you have downloaded the .ado file for a command called meta.ado. To check that, you could type in this in the old version of Stata 15,

which meta.ado

to see if you got something like this

. which meta.ado
c:\ado\plus\m\meta.ado
*! version 2.06 jacs/sjs March 1998 STB-43 sbe16.2

which indicates the meta.ado file is manually downloaded and placed into the folder c:\ado\plus\m\meta.ado.

If you type the same code in Stata 16, you should get results similar to below,

. which meta.ado
C:\Program Files\Stata16\ado\base\m\meta.ado
*! version 1.0.0 03may2019

This means the command -meta- is officially available in Stata 16. Running -meta- in Stata directly in Stata 16 would error out.

One approach is to rename the old command name. Please go to the folder found by typing

which meta.ado

in Stata 15, and rename the meta.ado. You need to rename both the program name to be a name different from meta, for example, meta_old, and also save the .ado file as meta_old.ado. Then, you should be able to call the command meta_old in Stata 16. You could choose name other than meta_old, be sure to make the name consistent in both the program name (you need to open the .ado file to change the program name) and the name of the .ado file. After that, you could run

meta_old ....

to call the old command now with a new name."

[David Fisher]

Hi Aliza Husain ,

Thanks for posting this. As you discovered, the error you encountered was due to Stata (version 16) releasing its own command named -meta-, but there was already a user-written command with the same name. This is less likely to happen than you might think, and a big part of the reason it happened here is that meta-analysis is a large and important field which Stata had never previously addressed with built-in routines.

However, the user-written -meta- is a *very* old routine, not updated since 1998, and I wonder how many others are still using it. I notice that your use-case is empirical Bayesian shrinkage as described by Carlin (Stat Med 1992), which is not, as far as I am aware, implemented in any other meta-analysis routine. As an interested Stata-using meta-analyst and user-written contributor, may I ask what you are using these shrunk estimates for?

Best wishes,

David.