On Sun, 30 Aug 2015 11:15:29 -0700, William Scott wrote:
Mike Palij writes:
"it is not always clear what a Bayesian approach buys one
though there may [be] situations when it is the method of choice."
-------------------------
Well, it seems to help identify those studies that REALLY
REALLY didn't replicate and those that REALLY REALLY did.
To me, this is better than simply categorizing those that did vs.
those that didn't based on significant NHST. Yes, in the end
it remains a similarly dismal state of affairs but one might do
some correlations of various factors with the calculated BF's
to see what might be predictive of replicability.
If you REALLY REALLY believe that a Bayesian analysis allows
you to determine which results did or did not replicate after
a single replication, well, you are entitled to your BELIEF.
Let's test your belief in the power of Bayesian analysis by
examining what it tells us in the following:
http://papers.ssrn.com/sol3/Papers.cfm?abstract_id=2423692
and see also
http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3114207/
and see also
http://www.jstor.org/stable/2245728?seq=1#page_scan_tab_contents
So, do you now believe in parapsychology too? ;-)
Those of use with a more empirical bent are left pondering the
following question:
How many replications does one need in order to determine that
an effect exists or does not exist?
The answer to that question should also take into account:
(1) The "decline effect". See previous posts on Tips for this.
(2) The results from the first corpus callosotomies performed by
van Wagenen that produced the first "split brain" patients;
see:
Mathews, M. S., Linskey, M. E., & Binder, D. K. (2008). William P.
van Wagenen and the first corpus callosotomies for epilepsy.
Journal Of Neurosurgery, 108(3), 608-613.
doi:10.3171/JNS/2008/108/3/0608
and also see
VAN WAGENEN, W. P., & HERREN, R. Y. (1940). Surgical division
of commissural pathways in the corpus callosum: relation to spread
of an epileptic attack. Archives of Neurology and Psychiatry, 44(4),
740.
Note #1: For some reason Wagenen's name gets misspelled in the
psychology literature. I first encountered it in Springer & Deutsch's
"Left Brain, Right Brain" (I believe in the first edition) and it
appears
in Kolb & Whishaw's "Fundamentals of Human Neuropsychology";
see page 283 in the edition on books.google.com:
https://books.google.com/books?id=z0DThNQqdL4C&pg=PA283&lpg=PA283&dq=%22William+Van+Wagnen%22&source=bl&ots=V3YNZKlJfr&sig=9APGsgFqCpHs_3JzxzLkLMLum5Y&hl=en&sa=X&ved=0CD0Q6AEwCGoVChMInYXK78LRxwIVSo4NCh0peg46#v=onepage&q=%22William%20Van%20Wagnen%22&f=false
Note #2: Van Wagenen found a small positive effect of cutting the
corpus callosum in epileptics but the results were often interpreted
negatively or of no effect (I believe Springer & Deutsch present it
this way). It would take Sperry, Gazzaniga, Bogen and Vogel to
resurrect the technique and gain fame. Sperry won the Nobel Prize
in 1981 for this work (sharing the Prize with Hubel & Wiesel) and
Van Wagenen getting no recognition.
(3) The results from Leo Dicara research on operant conditioning
of the autonomic nervous system; see:
Dworkin, B. R., & Miller, N. E. (1986). Failure to replicate visceral
learning in the acute curarized rat preparation. Behavioral
neuroscience,
100(3), 299.
or
http://psycnet.apa.org/journals/bne/100/3/299/
Like I said, there are probably situations that Bayesian analysis is
to be preferred, similarly, a Neyman-Pearson approach may be
appropriate in other situations (a number of people really don't
understand what is involved in this), and classical Fisherian NHST
which is probably the default analysis when one has no idea what
the populations/probability distributions are for their data (and
the use of the t-test and ANOVA F is legitimized because their
results are similar to permutation tests).
But what do I know, right?
-Mike Palij
New York University
m...@nyu.edu
---
You are currently subscribed to tips as: arch...@mail-archive.com.
To unsubscribe click here:
http://fsulist.frostburg.edu/u?id=13090.68da6e6e5325aa33287ff385b70df5d5&n=T&l=tips&o=46498
or send a blank email to
leave-46498-13090.68da6e6e5325aa33287ff385b70df...@fsulist.frostburg.edu
