I am myself not very familiar with these concepts, and haven read the
book you mention (which seems quite interesting indeed), but maybe a
suggestion I could give is to look at what is called "minimum
detectable effect" (which effect can be detected for a given sample
size and given confidence level and power), rather than sample size
(which sample size is needed to detect an effect of x at confidence
alpha), look at Bloom 1995, Eval Rev. You might also want to look at
the paper of Andrews 1989, Power in econometric applications.
Hope this helps.
2013/12/10 JOSE FRANCISCO PERLES RIBES <jfperles(a)gmail.com>:
Sorry in advance because I know that possibly it is not the best forum to
ask for this question, due that this is not a very Gretl specific question.
But I would know the expert opinion of Gretl list about this, if possible.
My question is as follows: In my briefly research career in economics, I had
never heard about "effect size" question. Recently, I have studied a very
basic course in statistics at coursera, and when I studied a topic on Null
Hypothesis Significance Testing, I saw this question very interesting. I'm
currently reading the book of Mc.Closkey "The cult of statistical
significance" and I discovered a new world for me.
It appears to be a very important question in psychometric or other
experimental sciences. But not at the moment in econometrics (this is very
criticized in the Mc Closkey book).
Usually my job is related with multiple regression (mostly with temporary
series) and I uses Gretl for this purposes.
Normally showing my results I compute the b (understandardized) regression
coefficient with his corresponding t-value and his p-value. No more. Lastly,
due that I usually works with small samples sizes (normally 30 or 40
subjects/years) I indicates in my regression output a bootstrap confident
interval for b (understandardized) or a bootstrap p-value. Gretl does this
In the peer-review process nobody say me nothing about effect sizes of
My specific question is: somebody at the list knows what is the common
measure of effect size in multiple linear regressions?. In the topic of this
statistical course it is highligted taht beta (standardized regression
coefficient, that GRETL does not provide by default, but there is a GRETL
package to do it) or confidence intervals as measures of effect size, but
looking for internet I have no clear answer to my question. I found some
effect size measures for R2, but not for the regression coefficients.
Are in your opinion bootstrap intervals or p-values enough information for
effects size estimation?
Somebody knows some interesting literature addressed to econometrics in this
sense? GRETL Manual does not any reference to this "effect size" question.
Thanks in advance and really sorry for any inconvenience.
University of Alicante
Gretl-users mailing list