You need to either (1) convert them onto the same scale so that they are all measuring the same thing, then combine them, or (2) use standardized mean difference, but then struggle to interpret the unitless estimate.

Thank you for your answer. Is this mean i also have to convert different scale means. lets say one mean is belong to 5-likert point scale and the other is 7-likert point scale. do i have to convert one of them. if yes, how. and if i use CMA software these converting are made otomatically or not. 

Hi Halim,

Here is an easy and powefull software that is very good for those (non statisticians) who want to perform meta-analysis.

http://www.meta-analysis.com/index.php

Tell me what do you think !

I already use it. Thank you

Hi Halim,

I suggest you enter the data as is and then choose the SMD option for analysis. Then all of the means will be converted into SDs allowing you to combine them.

Be aware that if you use SMD, you should not combine means that are changes from baselines with means that are final values. Also you should pay careful attention as to whether all scales have the same direction for example lower=better or if some of them go in the other direction. See the Cochrane Handbook section 9.2.3.2 'The standardized mean difference' for more information.

Cheers,

Britta

Hello Halim,

if you use RevMan by the Cochrane Group, you have to convert all the different means, which probably occur as the authors have used different scales, questionnaires or whatever, into "standardized mean differences= SMD", as Darren's already mentioned. Read the Cochrane Handbook carefully, that helps a lot. For the interpretation, as Darren also mentioned, you have to rely on common sense within your field of research. There are some broader guidelines but I don't know whether they also apply for your research field.

All the best

Markus

thank you for your answers and hi to all you

i read cochrane book carefully, but i dont find answer. okey if all articles use scales and mesaure same thing than we dont need standartize mean. but in some article total scores of items of scale divided by number of items of scale. so mean is between 1-5 and sd is small than mean. but in some article total scores of items were not divided by number of items and mean is 126 and sd is 31. do i have to convert one of them or not.

secondly 

i dont want to compare two grup. i will perform meta analysis for one group at one time point. in some article 5-likert type scale was used and in some 7-likert type scale used. also, for example, in one article mean is 3,4 and sd is 1,66  n is 200 and other article mean is 126 sd is 28 n is 420 (same scale were used in two article). 

thirdly 

when we convert mean for SDM we consider number of sample in the formula. why they say if we measure same thing with different way in the books? i didnt see different way in the formula.

1. Ask the authors. I don't know why www.researchgate.net/publication/49672850 scaled down the 100pt FIQ to a figure between 0&10 either, but a decimal point shift put the figure into total accord with all other studies. Always beware of unclear SE or SD shown however.

2. You have disparate scales, and must Standardise Mean Diff. I've noted Alice Theadom in Cochrane CD001980 claims that a translated (and validated in that language) instrument differs sufficiently to demand SMD also.

Further to Britta's comment, Handbook 9.4.5 is useful, but if 9.4.5.2 indicates Bayesian corrections are required to impute SD for a change in the means (due to mismatched baselines) then Abtrams '05 or Fullman & Cutler '92 may be of use.

Article Treating fibromyalgia with mindfulness-based stress reductio...

You can check your library for a book titled "Applied Meta-Analysis with R" by Ding-Geng (Din) Chen & Karl E. Peace. Download R (it is free) and follow the procedure in the book. They also provided link to some codes.

Hi 

I had this problem previously and by using  the SMD in REVMAN it was solved perfect. 

Using (SMD) is the best way to avoid the range differences in the measured means, however, some argue that this method is not preferable if all your studies have been measured with the same tool, rather than you can just have the WMD for the studies that have the same scale