Effect size in meta-analysis

#1
Dear all,

I am an animal nutritionist currently working on fish nutrition in the
Czech Republic. I am busy with an analysis on the effect of the
replacement of fishmeal with plant proteins in fish diets. I am
calculating effect size according to the conventional method of {Mean
(Control)-Mean(TRT)}/{[SD(Control)+(SD(TRT)]/2}. Unfortunately several studies did not report SD's or SE's (that I can convert to SD) for individual
means, but a "SE calculated from the RMS in the ANOVA" or a "pooled SEM"
over treatments. My problem is as follows:

TRT Growth
Control 1.65
1 1.45
2 1.34
3 1.87
4 1.61
5 1.54
SEM 0.54

I want to calculate an effect size between Control and TRT 3. TRT's 2, 4
and 5 are not suitable for my analysis. But the SEM was calculated over
ALL means. How do I eliminate the variance due to the latter treatments
that I do not include?

I really would appreciate any help. I am in a country where people cannot
speak English, and I cannot speak their language. Furthermore, the
knowledge on statistics is very limited.

James Sales
 
#2
Not sure how you eliminate the variance (there must be a way to do it but can't think just now) - but you can calculate cohen's d from the F value if you don't have the MSE, but only if it has 1 df. See link below - think it's page 8.

http://www.work-learning.com/white_papers/effect_sizes/Effect_Sizes_pdf5.pdf

If you don't have the F value you could always use a correlational measure of effect size and then convert it to cohen's d, or even better Hedge's g as this takes account of your sample size which is preferable when doing meta-analysis.

Sorry can't answer your question but hope this helps!