Re: Interpreting Negative d prime values in simple Yes/No detection task

[yund <yund@xxxxxxxxx>]

A negative d prime means that the false-alarm rate is greater than the hit rate.  It sounds to me that
the experimental design does not let you discriminate between "expected-interval" false alarms (FAs) and
"unexpected-interval" FAs.  If, on the catch trials, the participant generates more FAs based on the
noise-alone stimulus in the expected interval than in the unexpected interval, then the expected-interval
FA rate is higher than the unexpected-interval FA rate.  If you use the FA rate on catch trials as the
FA rate to compute the d prime for both expected and unexpected intervals, without differentiating
between expected- and unexpected-interval FAs (and it may not be possible to differentiate them in the
data), then negative d primes could be found for the unexpected intervals, whenever the combined FA rate
was greater than the unexpected-interval hit rate.  Similarly, the d primes for the expected intervals
may be lower than they should be because the overall FA rate for the catch trials would count both
expected- and unexpected-interval FAs.

One more minor point.  If the participant responded only on the basis of the expected time interval,
that would yield valid expected-interval d primes and chance unexpected-interval d primes.  If the
participant is responding on the basis of the full interval and shifting the response criterion
consistent with target expectation, then the above-chance unexpected-interval negative d prime
becomes possible.

------------------------------------------
E. William Yund, Ph.D.

Hearing Loss Research Laboratory (151/MTZ)
VA Northern California Health Care System
150 Muir Road
Martinez, CA 94553

yund@xxxxxxxxx
(925)372-2296
FAX (925)228-5738
------------------------------------------

On 8/12/2011 2:09 AM, Imran Dhamani wrote:
> Dear List,
>
>
> In one of the experiments that I performed using the yes/no detection task (button press), I have got some negative d prime values which I am  not sure how to interpret.The experiment is actually created in such a way that most of the targets occur at one time interval (expected = 60%) and at other time intervals (unexpected=20%) for other trials, the rest 20% are the catch trials in which the target does not appear in any of the time intervals. Since the frequent presentations of the targets in the expected time intervals may create an internal systematic bias for the observer the d prime are expected to be high at those target intervals and lower in the other intervals (may be due to tuning attention band to the frequently occurring or expected trials, attentional blink, attentional lapses or purposefully not heeding to the targets at the unexpected time intervals due to the lower frequency of their occurrence).The last possibility of lower d primes
>   due to not heeding seems to be very low in this experiment since the participants understood the task well and the instructions given to them were neutral (i.e. press the button as soon as the target is heard and not to press the button on catch trials) and they were aware that the target can occur at
> any time interval. At some of the unexpected trials instead of lower d primes, I am getting negative d prime values.The most common explanation for negative d primes that I could trace in literature was due to sampling error, response confusion or malingering. In this experiment I do not suspect a
> response confusion or malingering. In terms of sampling error, I have read some literature in terms of assuming the confidence interval for d prime to be plus or minus twice the standard error and then if the SE error includes values equal to or close to zero then the negative d prime is actually zero or close to zero.But I am not sure how this calculation exactly works and what is this sampling error that we are calculating in the context of signal detection theory.
> Can someone please throw some light on what sampling error means in the context of a signal detection theory and also a simple way to calculate this ? Do negative prime values have no significance in terms of their sign and thus if the negative d prime values are not too large they can be simply flipped to indicate the discriminability ? Can these negative d primes (pertaining to the current experiment) be interpreted in some logical way (eg: Inattention to unexpected trials) ?
>
> Regards,
> Imran Dhamani