Re: Hypothesis testing vs. effect size

hrubin@xxxxxxxxxxxxxxxxxxxx (Herman Rubin) wrote in news:eravjg
$19ss@xxxxxxxxxxxxxxxxxxxx:

As for meta-analysis, unless one has a good idea of the
probability that a result of a given significance level
will be published, it is likely to have a huge error.
Decades ago, a physicist asked me why, after an effect
was discovered, the already published data gave much too
high an estimate of the effect. Answer: if it wasn't
statistically significants, it did not get in print.


In the meta-analysis toolbox is a procedure called a "funnel plot". Each
study is ranked from highest variance (width of confidence interval) to
lowest and then its point estimate and CI at plotted. If the resulting
"funnel" is asymmetric, one has evidence for the publication bias you are
describing. I suspect you are familiar with the method, but the OP may want
a few terms for furhte searching.

--
David Winsemius
.

Relevant Pages
Loading