Reader Comments

Post a new comment on this article

aparent bimodal distribution may be due to random chance

Posted by perceptionsydney on 13 Mar 2007 at 21:53 GMT


the authors say there is clearly a bimodal distribution, but from what they present it seems this distribution could easily have occurred by chance (for example assuming that each subject fell in any of the categories according to a single-peaked or perhaps even uniform distribution, we suspect the observed "bimodal" distribution would not have been very unlikely, given the low n here). Perhaps the authors mean that they have measured each subject's threshold with a high degree of replicability (small error bar on each subject), so that the differences among people are reliable. However we could not find any presentation of error bars on individual subjects anywhere in the manuscript, so this is also questionable. The authors then divide the subjects into two groups based on the possibly-accidental two groups they fell in and the graph indicates that the subjects who improved on CFF also improved in motion sensitivity (Fig 5b and c). This helps validate the claim that there is something interesting going on with individual differences here (although it would be appropriate to do a interaction test, not just separate ANOVAs on the two groups), it is just that it may just as easily be a simple correlation with a uniform distribution or many other distributions rather than bimodal.

RE: aparent bimodal distribution may be due to random chance

aseitz replied to perceptionsydney on 16 Mar 2007 at 21:50 GMT

While it is conceivable that the visible separation between the flicker learning and flicker stable groups of subjects could be do to chance we believe that this both unlikely and unimportant for the following reasons:

1). As you stated, the Flicker learning group, as selected in this manner, showed significant learning as assessed completely independently in the motion-tests.

2). Our CFFT measurements are reliable in at least two ways. First, the 5 subjects in the Flicker learning group also showed the highest changes in CFFT in a manner that was consistent across sessions. Second, our within session measurements of CFFT are reliable and were based upon the average of two independent assessments of CFF per subject per session.

3). Finally whether or not their is truly a bimodal distribution it is reasonably to break subjects into groups based upon their degree of change in CFFT. We chose a natural breaking point between the groups given that there was a difference of 4% between the lowest changing CFFT score in the Flicker Learning group and the highest changing CFFT score in the Flicker Stable group.