Meta-analysis in Stata 

Contact Technical Services at tech-support@stata.com. They should be able to assist you with this request. Another resource is to register to Statalist, The Stata Forum. It is a great place to discuss Stata and statistics. It is free to register at statalist.org.

Hello, I have the same problem. Have you found the solution? Is it possible to add the sample size when using this command?

If you need to compute the effect sizes, you can use *meta esize* . You can type "help meta esize" in Stata and take a look at the examples. If you already have the effect sizes in your dataset, you can use *meta set* to declare the data to be meta-analysis data.

Contact Technical Services at tech-support@stata.com with the commands you are running. We weren't able to reproduce this error message.

After loading any of the example datasets, such as *webuse pupiliq* or *webuse nsaidsset* you can click on the Data Editor icon to view the data in a spreadsheet format in Stata. Or, you can simply type *browse* to bring up the Data Editor. You'll notice that for meta analysis, we have one observation per study with variables recording the effect size, sample size, and other aspects of each study. Additionally, you can type *help import excel* to learn how to import data from Excel into Stata. Please email us at tech-support@stata.com if you have any other questions; we will be happy to help.

If you have the standard errors for Steiger's psi, then you can use *meta set* . Generally, as long as you have the effect sizes and their standard errors you can use *meta set* to declare your meta-analysis data. Since you mentioned that you have three groups, and the official *meta* commands are designed for studies with two groups, you may want to consider the community-contributed network suite for network meta-analysis. Please type 'search network meta analysis' to learn more about these commands.

@@statacorp It is very helpful. Thank you very much.

For the data set shown in the video, the overall log odds-ratio is 1.06 for the IV method and 1.09 for the MH, so the vertical lines are at different positions but it may be hard to see because the 2 values are pretty close.

@@javierzanet thank you so much

To perform meta-analysis in Stata, your data should contain one observation (row) per study. You can type *help input* to learn about entering data directly into Stata. You can also type *help import* to get an overview of bringing data from several formats into Stata. Please email us at tech-support@stata.com if you have any other questions.

Our official suite of meta-analysis commands was introduced in Stata 16, but since you are working with Stata 14 you can take a look at the community-contributed meta-analysis commands that are available here: www.stata.com/support/faqs/statistics/meta-analysis/

@@statacorp Thanks! And in Stata 16 is it possible to run the meta-analysis for the RRR? I'm not sure which will be the appropriate command (I could only do it for the RR).

@@dafnecapelusnik3929 My understanding is that the terms *relative risk ratio* (RRR) and the *relative risk* or *risk ratio* (RR) are used interchangeably in the meta-analysis literature. In Stata 16, you can use the *meta esize* command with option *esize(lnrratio)* to compute the log risk-ratio (which is the same as the log relative risk-ratio), and then use command *meta summarize, rr* (option *rr* will display risk ratios as opposed to log(RR)) to obtain your results. I hope this helps!

@@javierzanet Thanks for answering! However RRR is actually the ratio between 2 RR, so I cannot replace for the RR command since I will have 8 variables to include instead of 4..

@@dafnecapelusnik3929 Please send your dataset to our Technical Services team at tech-support@stata.com and they will be happy to take a look and help you out.

Hi, I was wondering the same? Have you gotten any response other than exp(logOR)?

@@iliasdoulamis8065 Stata 16 only gives the option for lnoratio, lnrratio, rdiff, or lnorpeto and any other resize() specification is invalid.

Whenever you want to present your results for example on a forest plot, you can use option *or* (or option *eform* ) to transform your effect sizes into odds ratios. The computation in meta-analysis happens on log scale for risk ratios and odds ratios. This is why Stata expect you to provide your data on log scale and this why the option name is *esize(lnoratio)* . But this fact will not prevent you from reporting odds ratios or risk ratios.

Hello, which command did you use to combine p-values in Stata 12?

@@javierzanet I want to combine the P-value of Meta-analysis in STATA 16. In Stata 12 it is obvious. Aber Stata 16 has a different interface in Meta-analysis. Could you operate it, Sir?

@@user-wi9zl5ur8o You have probably used a user-written command to combine p-values in Stata 12 (perhaps the *metap* command?). In any case, if you have Stata 16, you can email your dataset to Stata's tech support and let them know about the command you used in Stata 12, then they should be able to tell if you can replicate the analysis in Stata 16. Good luck!

@@javierzanet Thank you very much dear teacher H Assaad

As long as you have the value of log(OR) (stored in a variable called *logor* ) and its standard error (stored in a variable called *se* ), you can use the following command *meta set logor se* and then perform any meta-analysis technique, like constructing a forest plot using command *meta forestplot* .

@@javierzanet Thank you. Just to check, is this the correct approach to use for meta-analysing ordinal data which has been analysed using proportional odds logistic regression?

Thank you for your question. Please contact Technical Services at tech-support@stata.com for details.