[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: Green's likelihood procedure in Matlab
I have some routines for MML (Method of Maximum Likelihood, Green 1990,
1993) in both Pascal and (at least for simulations) MATLAB. At the risk of
tooting my own horn, I can also offer code for the ZEST adaptive procedure
(that currently we're successfully using for humans and animals).
In a recent paper, (Marvit, et al., 2003, JASA 113(6):3348-3361), we offered
empirical and simulated comparisons of the classic Levitt (2-down, 1-up),
two variations of MML (Green, 1993), and ZEST (King-Smith, et al, 1994). Our
target was a level-discrimination difference limen, using a cued yes-no and
a 2AFC paradigm.
A quick summary, based on our data and analysis: session-to-session subject
performance variance can be much greater than within-procedure variance, so
that it may be much more advantageous to take many (short) estimates) rather
than few (long, but putatively precise) estimates; the efficiency (measured
by sweat factor) of the ZEST and our variation of the up-down was far
greater than the MML; contrary to some published reports (and simulations),
2AFC seemed to produce lower estimate variance than a cued yes-no paradigm.
In our opinion, the ZEST procedure with 2AFC paradigm is an excellent
candidate for short, relatively accurate estimates obtainable in as few as
12-15 trials. An interleaved Up-Down provides different advantages (e.g.,
fewer assumptions about underlying psychometric function) at the expense of
clock time (~85 trials for three interleaved tracks).
I am pleased to provide the MATLAB code I have for ZEST (and MML), with the
usual caveats of "as-is", no warranties, guarantees, etc. Until I can get a
web site working, email me for details.
: Peter Marvit, PhD <email@example.com> :
: Dept. Anatomy and Neurobiology University of Maryland Medical School:
: 685 W. Baltimore Street, HSF 222 Baltimore, MD 21201 :
: phone 410-706-1272 fax 410-706-2512 :