Hi,
I am currently trying to undertake a meta-analysis that looks at the difference between community-dwelling adults and those residing in various care settings.
Originally I had planned to calculate standardised mean differences (SMD) as the studies I found in my initial search presented the means and standards deviations of the community-dwelling and 'instititutionalised' adults. Unfortunately, this is obvious not the norm as the majority of the studies I have now found (which provide relevant and needed information) solely present the mean and standard deviation for one 'institutionalised' population. From this I have been able to calculate the 95% confidence intervals for each study and have put them into a forest plot to see how homegenous they are.
However I do not know if this is a valid type of meta-analysis, if it is - what is it called? Everytime I find something about meta-analysis of means its about SMD. If it isn't a valid form of meta-analysis is there another method that I could use? Or is there a way to impute a community-dwelling mean for the studies that don't present one so that I can use SMD's?
Many thanks,
Lucy
Yes, you can estimate the mean value of a dependent variable using the mean values (and standard deviations and sample sizes) from a bunch of studies. (See the attached PDF for some screenshots of how I would do it using the Comprehensive Meta-Analysis software. I just made up some numbers for means, standard deviations, and sample sizes.) As you can see in the top screenshot, the Std Err for a study is simply the standard deviation divided by the square root of the sample size. The Variance is the square of the Std Err. The weighting of a study's mean in the calculation of the overall effect size is 1/Variance. (Technically, this weighting is only when the fixed-effect model is used.) If you wanted to compare the value for studies with community-dwelling subjects versus that for studies with institutionalized subjects, you could run a subgroup analysis on a categorical moderator variable that distinguishes between the two types of studies. Be cautious in interpreting this analysis because you are assuming that the only difference between the two groups of studies is the type of subject and that may or may not be true.
You don't need to run a meta-analysis on the study standard deviations and confidence intervals. You get a 95% confidence interval when you calculate the overall effect size. See the Lower limit and Upper limit values on the yellow row in the bottom screenshot.
Thank you for your reply Tahir!
From reading your answer I think I am correct that you are talking about mean differences; my understanding is that to calculate mean differences a study needs to assess either a population at two time points, or assess two different participant groups.
As the studies my search is returning tends to only assess one population at one time point I did not think calculation of mean differences would be appropriate, as I would have to impute both means and standard deviations!
Please correct me if I am wrong.
Thanks again!
this this link may be helpful to beginners too
https://www.erim.eur.nl/fileadmin/erim_content/images/meta-essentials/How_to_interpret_results_of_meta-analysis_1.4.pdf
I am not sure from the description provided if the data is based on observational studies or trials.
mid the data is from trials then you can separately analyze the community and institutional studies as Subsets and then a combined meta-analysis can be performed.
if your data is based on observational studies then you can do a meta-analysis of proportions. Metafor in r is a software I have used for this purpose.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Meta-Analysis