A comparison of two modeling frameworks for two -choice classification tasks

Zhengping Ma, Purdue University

Abstract

The main purpose of this thesis is to investigate the properties of D-SDT to determine whether it is in general better than SDT. Two experiments were conducted to compare the measures derived from these two models. An important finding of this research is the existence of biased U(k) plot. It is the first time that such a plot is found. Due to its nature of unbiased decision rules, D-SDT has no way to account for biased U(k) plot. We found that subjects' performances were not necessarily increased although they took longer time to respond under time manipulation. In D-SDT, longer RT is always predicted to be associated with a better performance. When we applied D-SDT to our empirical data, we got a smaller d under time manipulation, while d' of SDT increased a little at the same time. Empirical results show that subjects usually perform better with more experiences. Further, our results indicated that mean RT decreased with more experiences. It results in a bigger d' of SDT and smaller d of D-SDT. Therefore, in general, D-SDT does not have advantage over SDT in separating sensitivity from response bias. To address these issues, a generalized sequential sampling model, GRJT, is proposed. Based on simulation results, we found that GRJT is able: (1) To predict both unbiased and biased U(k) plot. (2) To account for effect of speed/accuracy tradeoff with an upper limit on subject's performance. (3) To account for the effect of experience, with a stable sensitivity in the model. On sub-optimality, we found that: (1) It's possible that subjects perform suboptimally. (2) Sub-optimal responses tend to have longer RT. (3) These sub-optimal responses can almost be completely eliminated with re-calibration feedbacks. Our GRJT model is able to predict some sub-optimal responses, and to show how they are eliminated. GRJT is also capable to account for the findings on confidence level such as its relationship with RT, and with percentage of correct responses. In addition, the effect of presentation time on performance can be accounted for as well.

Degree

Ph.D.

Advisors

Dzhafarov, Purdue University.

Subject Area

Cognitive therapy

Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server
.

Share

COinS