Terminating Quest using a confidence interval

1

Hello


I am creating a perimetry instrument and require Quest to terminate at a certain confidence interval, rather than after 40 trials for example, as this number of trials for each stimulus (40 stimuli in total) would simply make the program too long. I have read the Watson and Pelli (1983) paper which describes how to terminate Quest at a certain confidence interval using a likelihood function. 


My problem, however, is in trying to extract and make sense of the data produced by the PTB Quest function. How, for example, would I get the maximum likelihood estimate after each stimulus presentation? Has anyone else had any experience in terminating the Quest function in this manner?


Thank you

Nikita

2
Nikita,

I echo what Denis has said about termination rules.

On a slight tangent, you might take a look at QUEST+, which has generalizations that make it ideal for multidimensional problems such as perimetry. There is a Matlab implementation.

Best
Andrew (Beau) Watson

Watson, A. B. (2017). QUEST+: A general multidimensional Bayesian adaptive psychometric method. Journal of Vision, 17(3), 10-10, http://dx.doi.org/10.1167/17.3.10.

https://github.com/BrainardLab/mQUESTPlus




On Mar 3, 2018, at 12:16 PM, Denis Pelli <denis.pelli@...> wrote:

dear nikita
beau watson and i wrote the article, and i later wrote the MATLAB software. it's my impression that observers don't like running for an unknown number of trials. beau and i also have the impression that the software tends to devote more trials to the bad runs where an early goof by the observer set Quest off in the wrong direction. this results in contaminating more trials with the bad one. so we haven't used that method much. Perhaps it'll work for you.

QuestMode gives you the maximum likelihood estimate if the prior is flat.

You can easily get a confidence interval by making two calls to QuestQuantile to get your low and high bounds.

Please report back and tell us if you come to a conclusion about the utility of a confidence interval as the stopping criterion.
best
denis


Denis Pelli
Professor of Psychology & Neural Science, New York University
+1-646-258-7524 | denis.pelli@...
http://psych.nyu.edu/pelli/ | Skype: denispelli
| http://denispelli.com

On Wed, Nov 29, 2017 at 2:16 PM, nikitajadet@... [PSYCHTOOLBOX] <PSYCHTOOLBOX@yahoogroups.com> wrote:

Hello


I am creating a perimetry instrument and require Quest to terminate at a certain confidence interval, rather than after 40 trials for example, as this number of trials for each stimulus (40 stimuli in total) would simply make the program too long. I have read the Watson and Pelli (1983) paper which describes how to terminate Quest at a certain confidence interval using a likelihood function. 


My problem, however, is in trying to extract and make sense of the data produced by the PTB Quest function. How, for example, would I get the maximum likelihood estimate after each stimulus presentation? Has anyone else had any experience in terminating the Quest function in this manner?


Thank you

Nikita