Confidence intervals for d primes based on Maximum likelihood estimation

Hi everyone, I am searching for a code/script for calculating the 95% confidence intervals for d primes using the Maximum likelihood estimation procedure. Does someone have it and be willing to share.

Thanks in anticipation,

Cheers, Imran

[Walter Roberson]

What are "d primes" in this context??

[Imran Dhamani]

Hi,

I am evaluating the sensitivity of listeners for a sound. The d primes are calculated using the hit and false alarm rates (d'=z(hit rate)-z(false alarm rate)). The z scores are calculated using the norminv function in matlab.

Cheers,

Imran

[Daniel Shub]

d' is the common name given to the "index of sensitivity": http://en.wikipedia.org/wiki/D%27

[Daniel Shub]

If someone is willing to write code for you (and I doubt they are), you would need to provide more information: How many measurements of d' do you have? How many signal and no signal presentations is each measurement based on? Your equation suggests 1I-2AFC, is this correct?

Answer 1

[Daniel Shub]

The easiest way is to probably bootstrap the confidence intervals. You can create N experiments where you flip M coins. On each flip, you flip a coin to decide if you will use either your signal (heads) or no signal (tails) coin. The probability of heads with this coin is based on the probability of a signal trial. The probability of heads (yes response) of the signal and no signal coins is based on the hit rate (signal coin) and false alarm rates (no signal coin).

Once you make N*M flips (ignoring the extra flip to decide the trial type), you can generate N estimates of d'. Sort these and then the 95% CI is the values for the top and bottom 2.5%.