Generate data for meta-analysis

Accepted Solution Solved
Reply
Occasional Contributor
Posts: 8
Accepted Solution

Generate data for meta-analysis


I want to generate raw data for a study that has multiple treatment groups (2 treatment groups and 1 control group). Every participant in each of the groups are measured on two related outcomes ( r=.80). Then I need to repeat the same process to generate 10 studies, which I  plan to meta-analyze. Each group has equal sample size (n=10). Any help with the simulation code will be appreciated. Outcomes are generated from a multivariate normal distribution.

StudyGroupsOutcomes
1Treatment Group 1Y1
Y2
Treatment Group 2Y1
Y2
Control GroupY1
Y2
.
.
.
10Treatment Group 1Y1
Y2
Treatment Group 2Y1
Y2
Control GroupY1
Y2

Accepted Solutions
Solution
‎07-03-2014 03:17 PM
SAS Super FREQ
Posts: 3,752

Re: Generate data for meta-analysis

Look at the ANOVA and GLM examples in Chatper 11 of Simulating Data with SAS.  Because you want to simulate correlated responses, also read the section in Chapter 8 that deals with mixtures of MVN data. (Your example is a little simpler because your example does not contain correlations between treatment groups.)

View solution in original post


All Replies
Solution
‎07-03-2014 03:17 PM
SAS Super FREQ
Posts: 3,752

Re: Generate data for meta-analysis

Look at the ANOVA and GLM examples in Chatper 11 of Simulating Data with SAS.  Because you want to simulate correlated responses, also read the section in Chapter 8 that deals with mixtures of MVN data. (Your example is a little simpler because your example does not contain correlations between treatment groups.)

Occasional Contributor
Posts: 8

Re: Generate data for meta-analysis

I will Rick. I really appreciate your help!

Occasional Contributor
Posts: 8

Re: Generate data for meta-analysis

Hi Rick! To extend my question, how then do you simulate correlated response between groups. Using the table above with slight modification, how do you simulate an outcome where each group (Control, Treat1 and Treat2) is only measured on one outcome (Y), but the outcome (Y) for each group is correlated. Again, thank you!

SAS Super FREQ
Posts: 3,752

Re: Generate data for meta-analysis

Could you explain more or give an example?  Each patient is an observation, right? Each patient only belongs to one group. So what do you mean that the outcomes are correlated?

Perhaps this will help: Sometimes researchers assume that

Trt1 ~ N( mu1, sd1)

Trt2 ~ N( mu2, sd2)

Control ~ N( mu3, sd3)

That makes each sample a draw from a different population, where the population parameters depend on the group.  Mathematically, this is a draw from a finite mixture distribution. See p 120-121 of Simulating Data with SAS.

Occasional Contributor
Posts: 8

Re: Generate data for meta-analysis

Thank you for the response, Rick. I think I may have to re-think my question, but thanks for your suggestion.

🔒 This topic is solved and locked.

Need further help from the community? Please ask a new question.

Discussion stats
  • 5 replies
  • 2305 views
  • 1 like
  • 2 in conversation