Correct use of StopInterval in QuestHandler

Coding
1

Hi everybody,

I’m trying to use the questHandler with the stopInterval termination rules. I read the doc of the QuestHandler class but I can’t understand what does it mean :

stopInterval: None or a number
The minimum 5-95% confidence interval required in the threshold estimate before stopping. If both this and nTrials is specified, whichever happens first will determine when Quest will stop.

In my current test I have images with a level of noise between 0 and 1 and I use the QuestHandler object like this :

staircase = data.QuestHandler(0.5, 0.5, pThreshold=0.51, stopInterval=0.01, delta=0.01, beta=3.5, gamma=0.1, minVal=0.0025, maxVal=1, nTrials=30)

In this line I choose stopInterval randomly and it stop in 30 trials. I don’t now how to calculate the correct value in order to have a 95% confidence level.

I also checked the code of the questHandler class and I saw that they compare the result of the confInterval() function and the value of stopInterval. The experience stop if confInterval() < stopInterval.
So I print my values of confInterval() during my experience and the values are between 0.22 at the beginning and 0.13 after 30 trials.

Do you have any ideas how to calculate stopInterval in order to have a 95% confidence level?

2

Apologies for resurrecting an old post - I’m also trying to figure this out. Am I correct in believing that stopInterval = 0.05 is requiring a confidence interval of at least 95%?

As I do my trials (spatial 2AFC contrast discrimination), the figure reported by stairs.confInterval(True) reduces. Is this the expected behaviour?

Kind regards,

Martin

3

I am in the same boat. I cannot find a clarification regarding the proper usage of stopInterval. I am assuming it is related to the certainty regarding QUEST’s estimate: a large value of stopInterval would allow QUEST to end sooner, whereas a small value would force QUEST to continue until the CI is very small.

However, it is not clear what units stopInterval should be specified in:

  • Is the value specified by stopInterval in units of the staircased variable (e.g., intensity)? For example, if the confidence interval surrounding the estimate ranges from 2 to 3, then a stopInterval of 1 would terminate the staircase.

  • Alternatively, is stopInterval specified in internal grain (i.e., arbitrary) units? For example, if the confidence interval surrounding the estimate ranges from 0.2 to 0.25 on the internal grain, then a stopInterval of 0.05 would terminate the staircase.

Martin and Cynthia-- the level of significance associated with the estimated confidence interval (e.g., 95%) is separate from the size of the confidence interval (e.g., in intensity or arbitrary units). I could have a confidence interval ranging between 3.50 and 9.00 or between 10.00 and 10.01, both could have a 95% level significance. That just means that there is a 95% chance that the estimate lies between 3.5 and 9 or between 10 and 10.01 for both CI’s. The size of each CI is 5.50 and 0.01, respectively. I think this is what stopInterval is defining…

Clarification would be much appreciated.