Hi All,

I am an acoustics researcher, conducting an experiment about the effect of certain sounds on restoration of ability to pay attention. I am using the Sustained Attention to Response Test (SART).

In similar studies, after which I am modeling my experiment, the setup is as follows. Two stimuli are being compared for restorative effects. A participant is asked to take the SART, then he/she is presented with a stimulus, then they take the SART again. The difference in performance on the two tests indicates whether the stimulus was restorative or not. If 2 stimuli are being compared - let's say stimulus A and B - then there is a distinct group of participants for each stimulus and a third Control group with no stimulus. The means of the scores of the three groups are then analyzed using the MANOVA technique.

My setup varies slightly from above. In my case, the groups for conditions A, B, and Control, are not distinct. I am using one group. Each participant is exposed to both of the stimuli as well as the control. Each participant is exposed to either of the stimuli or the control on separate days, and the order of exposure is quasi-randomized across all the participants. In other words, on the first day participant 1 will get stimulus A, on second B, on third control, whereas participant 2 will get B on first day, control on second day and A on the third, etc (with 3 conditions there are 6 permutations).

I realize this is the psychology board, but I am not sure how familiar most member are with the SART. This is a test that asks the participant to respond with a key press to a random sequence of digits 1-9 appearing on a screen. On a certain 'target' digit the participant is asked to withhold. Performance is judged on both the accuracy of detecting and withholding the target, as well the response time.

My question to the group is the following. Is it still appropriate to use the MANOVA method, since the groups for each condition are not truly distinct? Does randomizing the order of conditions make it valid again? If not what is a better method of analyzing the means of performance under the different conditions.

Thanks for the help!


Super Moderator
I am a little confused by the application of MANOVA here, but maybe that just reflects a lack of familiarity with the area (I had not heard of the SART before). MANOVA is usually used when you have two or more groups, and two or more continuous response variables. I'm not sure if see multiple continuous response variables here - there's response time, which is continuous, but then accuracy seems to be dichotomous (unless accuracy is aggregated over multiple trials to become a very roughly continuous variable?) If you use MANOVA, the key will be to make sure you specify your grouping variables as within-subject rather than between-subjects variables.

Maybe also worth thinking about distributional assumptions here - response times are rarely normally distributed.