Skip to content

[R-meta] Binomial Effect Size Display?

3 messages · Mark White, Viechtbauer Wolfgang (STAT), Michael Dewey

#
Hello all,

Many meta-analyses will take their smaller-than-they-would-have-hoped
summary effect size and make it look bigger by using Rosenthal and Rubin's
(1982) binomial effect size display:
http://www.cognadev.com/publications/BESD_Rosenthal_and_Rubin_1982.pdf. I
have always thought this is a misleading metric. Someone asked a question 5
years ago on CrossValidated about it, and I tried to answer it with a
little simulation:
https://stats.stackexchange.com/questions/24067/is-the-binomial-effect-size-display-besd-a-misleading-representation-of-effect
(this post provides a good, short summary of the original paper and the
metric, if you are unfamiliar with it).

I'm more of a simulate-and-see-if-it-works type of person?is there anyone
who understands more of the math behind *why* this metric might be
misleading?

Best,
Mark
3 days later
#
I don't have the time to really dig into this and since I hardly see any use of the BESD in practice, it seems like a non-issue to me. But this article seems highly pertinent:

Hsu, L. M. (2004). Biases of success rate differences shown in binomial effect size displays. Psychological Methods, 9(2), 183-197.

Some criticisms are also discussed in:

Randolph, J. J., & Edmondson, R. S. (2005). Using the Binomial Effect Size Display (BESD) to present the magnitude of effect sizes to the evaluation audience. Practical Assessment, Research & Evaluation, 10(14).

Best,
Wolfgang
1 day later
#
Dear Mark

There are some examples in the Cooper and Hedges handbook in a chapter 
by Rosentahl where clinical trials which were stopped early on ethical 
grounds because of the large size of effect are shown to have very small 
values of r. The BESD was, I imagine, a response to the use of r which 
gave a misleading result here. But, as Wolfgang said in another post, 
the BESD is hardly ever seen in the wild because nobody running a 
clinical study would usually use anything other than odds/risk ratios or 
mean differences.

Michael
On 15/07/2017 14:35, Mark White wrote: