This study aimed to investigate the correlation of speed and accuracy in cognitive testing. The response data was recorded into three kinds of data: time data, accuracy data and time-accuracy data. Each kind of data was estimated with 2PL model.
Two main research questions were investigated: (1) How much the latent traits of speed and accuracy is measured, (2) how the posterior time limits will affect the reliability of the measurements? An empirical data of verbal analogies test was used. It was found that the speed latent trait was basically measured by the time data, while the accuracy latent trait was basically measured by the accuracy data, no matter which posterior time limit was used. The stringent posterior time limit will not affect the reliability of accuracy measurement.
Questions and Comments:
1. The method is similar to factorial ANOVA. Speed and accuracy were two factors in the data. Hence, the main effect of speed and accuracy and their interactions can be examined.
2. It was found that speed and power (accuracy) were uncorrelated. I think, if speed and accuracy were treated as factors (and also found as two factors in PCA), it is deemed that they are independent.
3. If one uses multidimensional IRT approach, as van der Linden (2007) has done, the correlation between the speed and accuracy may be estimated.
4. The effect of posterior time limits on difficulties and discrimination really made me confusing.
5. What’s the rational for the authors to choose 2PL in this study?
6. The speed coefficients was developed within the framework of CCT. Can it be used directly to the estimation of IRT?