On 6/15/2014 12:39 PM, Annette Taylor wrote:
Sorry this is coming from my microcomputer aka cell phone. Typos
and short sentences may abound.
I have not had a stats class since grad school. I am well into
my 60's now. So here is my confusion. I was taught way back
then that too large a sample was bad. That it would guarantee a
type of error of a false positive effect when in fact nothing was
there. I was taught as a rule of thumb n equals 15. More than
that was not necessary for the rules of probability to kick in.
More would provide a false positive probability level.
Then later I learned to do sample size estimates based on alpha
and effect size estimates. When I did those I'd come up with n
equals twenty - ish.
Now I am reading one of several papers on newer method in stats
and it appears that much greater sample sizes are in order. A
paper in Perspectives on Psychological Sciences is even
advocating samples over 500 if I would want 90 percent power.
Back in the day that would have been denounced as the very
worst kind of stats thinking.
So which view is more correct and how did such massive changes
come about?
Hi Annette:
The concern back then was that increased-N would lead to a small
effect of trivial theoretical importance becoming statistically
significant.
The current concern is that a bunch of articles have reported
theoretically important effects using small N. The statistical
power of these studies often is very low (beta < 50%). So, if a
study is replicated using the same size N and the effect is not
found then a reviewer will say "Dummy, your N is too small and
you don't have the statistical power to detect the effect. So
your replication doesn't count."
Many researchers don't have a feel for the size of N to achieve a
certain level of statistical power. Check out the free software
G*Power to get an appreciation of the relationship between effect
size, sample size, and statistical power.
http://www.gpower.hhu.de/en.html
Ken
--
---------------------------------------------------------------
Kenneth M. Steele, Ph.D. steel...@appstate.edu
Professor
Department of Psychology http://www.psych.appstate.edu
Appalachian State University
Boone, NC 28608
USA
---------------------------------------------------------------
---
You are currently subscribed to tips as: arch...@jab.org.
To unsubscribe click here:
http://fsulist.frostburg.edu/u?id=13090.68da6e6e5325aa33287ff385b70df5d5&n=T&l=tips&o=37153
or send a blank email to
leave-37153-13090.68da6e6e5325aa33287ff385b70df...@fsulist.frostburg.edu
