Dear Statalisters,
I am planning for a meta analysis. The outcome variable of interest is a ratio level continuous variable, just like body weight. This variable has been reported in multiple studies done in populations of various ethnic origins. Within each ethnic, several studies have been published. The aim of my review is to: 1) obtain a summary value of the outcome variable for each ethnic population; 2) compare the statistical significance of the differences in this mean value across ethnics.
I noticed that the most commonly used effect size indices in meta analysis are weighted and standardized mean difference for continuous outcomes. However, the studies I will include in my meta analysis will be those reporting on the observed Mean (SD) of a single population, and there will be no experiment and control groups. The weighted/standardized mean difference therefore cannnot be imputed from these studies.
Even though -metan- has an option to perform analyses using the effect and its standar error, in which I only need to input the mean and SD of the obseved effect, not the mean and SD of both the experimental and control group, to obtain the summary effect, I am wondering the "effect" option should only be used when the authors reported mean difference (effect) and its SD between the experiment and control group. However, the problem with me is that there is simply no distinction of experiment and control group in the studies I have included.
In page 55 of the book "Introduction to Meta-Analysis" by Borenstein et al. published in 2009, it says "While many meta-analyses use one of the effect sizes presented above (standardised mean difference, OR, RR, RD ect), other options exist. ... .. Some indices simply report the mean, risk, or rate in a single group. For example, we could perform a meta-analysis of studies that had estimated the prevalence of HIV infection in different countries". The example in the book is very similar to my objective of review, and the only difference is that the outcome measure is binary in the example but continuous in my review. Therefore I belive there should be certain way of summarizing the coninuous outcome, I am just trying to find it out.
Thank you very much for your attention and assistance!
Best regards,
Patrick Wen
I am planning for a meta analysis. The outcome variable of interest is a ratio level continuous variable, just like body weight. This variable has been reported in multiple studies done in populations of various ethnic origins. Within each ethnic, several studies have been published. The aim of my review is to: 1) obtain a summary value of the outcome variable for each ethnic population; 2) compare the statistical significance of the differences in this mean value across ethnics.
I noticed that the most commonly used effect size indices in meta analysis are weighted and standardized mean difference for continuous outcomes. However, the studies I will include in my meta analysis will be those reporting on the observed Mean (SD) of a single population, and there will be no experiment and control groups. The weighted/standardized mean difference therefore cannnot be imputed from these studies.
Even though -metan- has an option to perform analyses using the effect and its standar error, in which I only need to input the mean and SD of the obseved effect, not the mean and SD of both the experimental and control group, to obtain the summary effect, I am wondering the "effect" option should only be used when the authors reported mean difference (effect) and its SD between the experiment and control group. However, the problem with me is that there is simply no distinction of experiment and control group in the studies I have included.
In page 55 of the book "Introduction to Meta-Analysis" by Borenstein et al. published in 2009, it says "While many meta-analyses use one of the effect sizes presented above (standardised mean difference, OR, RR, RD ect), other options exist. ... .. Some indices simply report the mean, risk, or rate in a single group. For example, we could perform a meta-analysis of studies that had estimated the prevalence of HIV infection in different countries". The example in the book is very similar to my objective of review, and the only difference is that the outcome measure is binary in the example but continuous in my review. Therefore I belive there should be certain way of summarizing the coninuous outcome, I am just trying to find it out.
Thank you very much for your attention and assistance!
Best regards,
Patrick Wen
Comment