Calculating sensitivity for a single interval 5AFC syllable discrimination task (Imran Dhamani )


Subject: Calculating sensitivity for a single interval 5AFC syllable discrimination task
From:    Imran Dhamani  <imrandhamani@xxxxxxxx>
Date:    Fri, 2 Dec 2011 19:55:46 +0530
List-Archive:<http://lists.mcgill.ca/scripts/wa.exe?LIST=AUDITORY>

--513538877-1474428231-1322835946=:92929 Content-Type: text/plain; charset=iso-8859-1 Content-Transfer-Encoding: quoted-printable Dear List, =A0=A0=A0=A0=A0=A0=A0=A0=A0=A0=A0=A0=A0 I have currently collected some dat= a using a single interval 5 alternative forced choice syllable discriminati= on task. The task is to detect a target syllable amongst other syllables. T= he target can occur as any of the 5 alternatives, but occurs more frequentl= y (60% of the time) as the first alternative and less frequently as others.= Thus there is an induced response bias that we have generated in the task = using unequal distribution of trials at each of the the alternatives. The p= roblem that i am facing is how to analyze this data using sensitivity as th= e index of discriminability. I understand that the traditional signal detec= tion theory cannot handle induced response bias such as in the current task= as it assumes equal distribution/probability of target occurrence at each = of the 5 alternatives. Moreover as I understand d primes itself is relative= ly independent of response bias in the signal detection theory which is counterproductive in the current task as there is a specific reason of ind= ucing this bias in our task. The other option I had was calculating d prime= s using the High threshold theory by correcting the hit rates for guesses/f= alse alarms but that too is difficult since the guess rate for a multiple a= lternative forced choice is usually calculated as guess/lambda=3D1/number o= f alternatives but in the current task we also have additional catch trials= (20%) in which the target syllable does not occur at all which allows us t= o calculate individual false alarm and hit rates for the target occurrence = at each of the 5 alternatives within an interval by using the participants = reaction times (using the method described by Watson and Nichols (1976) for= signals presented without defined observation intervals). So in a way it i= s a combination of a mAFC and Yes/No task, thus making it complex for me to= analyze. One other way I explored was using the correction for z scores for response bias instead of correcting the hit rates as described = by Klein (2001) but the method described in the research article does not a= llow me to calculate 5 separate d primes for the target occurrence at each = of the alternatives. One point to note in addition in this task is that sin= ce the probability of target occurring as the first alternative is much hig= her than others, along with higher hit rates for target presented at that a= lternative there is also a higher tendency of participants to false alarm o= n the catch trials on that alternative as well (i.e unequal distribution of= false alarms).=20 Can someone please suggest me a way of analyzing this data ?=20 Regards, Imran Dhamani $$$$$ monty@xxxxxxxx@xxxxxxxx@xxxxxxxx@xxxxxxxx --513538877-1474428231-1322835946=:92929 Content-Type: text/html; charset=iso-8859-1 Content-Transfer-Encoding: quoted-printable <table cellspacing=3D"0" cellpadding=3D"0" border=3D"0" ><tr><td valign=3D"= top" style=3D"font: inherit;">Dear List,<br>&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&= nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; I have currently collected = some data using a single interval 5 alternative forced choice syllable disc= rimination task. The task is to detect a target syllable amongst other syll= ables. The target can occur as any of the 5 alternatives, but occurs more f= requently (60% of the time) as the first alternative and less frequently as= others. Thus there is an induced response bias that we have generated in t= he task using unequal distribution of trials at each of the the alternative= s. The problem that i am facing is how to analyze this data using sensitivi= ty as the index of discriminability. I understand that the traditional sign= al detection theory cannot handle induced response bias such as in the curr= ent task as it assumes equal distribution/probability of target occurrence = at each of the 5 alternatives. Moreover as I understand d primes itself is re= latively independent of response bias in the signal detection theory which = is counterproductive in the current task as there is a specific reason of i= nducing this bias in our task. The other option I had was calculating d pri= mes using the High threshold theory by correcting the hit rates for guesses= /false alarms but that too is difficult since the guess rate for a multiple= alternative forced choice is usually calculated as guess/lambda=3D1/number= of alternatives but in the current task we also have additional catch tria= ls (20%) in which the target syllable does not occur at all which allows us= to calculate individual false alarm and hit rates for the target occurrenc= e at each of the 5 alternatives within an interval by using the participant= s reaction times (using the method described by Watson and Nichols (1976) f= or signals presented without defined observation intervals)<span style=3D"font-weight: bold;">. </span>So in a way it is a combination of a= mAFC and Yes/No task, thus making it complex for me to analyze. One other = way I explored was using the correction for z scores for response bias inst= ead of correcting the hit rates as described by Klein (2001) but the method= described in the research article does not allow me to calculate 5 separat= e d primes for the target occurrence at each of the alternatives. One point= to note in addition in this task is that since the probability of target o= ccurring as the first alternative is much higher than others, along with hi= gher hit rates for target presented at that alternative there is also a hig= her tendency of participants to false alarm on the catch trials on that alt= ernative as well (i.e unequal distribution of false alarms). <br><br>Can so= meone please suggest me a way of analyzing this data ? <br><br>Regards,<br>= Imran Dhamani<br><br><br><br>$$$$$ monty@xxxxxxxx@xxxxxxxx@xxxxxxxx@xxxxxxxx</td></tr></table> --513538877-1474428231-1322835946=:92929--


This message came from the mail archive
/var/www/postings/2011/
maintained by:
DAn Ellis <dpwe@ee.columbia.edu>
Electrical Engineering Dept., Columbia University