Reader Comments
Post a new comment on this article
Post Your Discussion Comment
Please follow our guidelines for comments and review our competing interests policy. Comments that do not conform to our guidelines will be promptly removed and the user account disabled. The following must be avoided:
- Remarks that could be interpreted as allegations of misconduct
- Unsupported assertions or statements
- Inflammatory or insulting language
Thank You!
Thank you for taking the time to flag this posting; we review flagged postings on a regular basis.
closeAsymptote per condition
Posted by Potsdam_EM_Group on 28 Sep 2007 at 22:17 GMT
The authors do not clearly state how many experimental runs per condition a participant performed. Moreover, it remains somewhat unclear what exact starting rate was used in the staircase procedure (QUEST) and after how many trials / runs the procedure settled at an 80% performance level. We assume that this number varied between conditions. Given this is the case, wouldn't one expect biases towards the starting value in those conditions differing the most. Can the authors state, whether for the data presented in the paper performance already asymptoted at 80% correct? Also, it would be interesting to know whether (possibly as a consequence) the level of training in the different conditions was different.
RE: Asymptote per condition
DenisPelli replied to Potsdam_EM_Group on 26 Oct 2007 at 22:18 GMT
Yes, we used 15 trials (of 6 word/trial) per run, and by then the threshold estimate is stable and has a small confidence interval. QUEST is a Bayesian algorithm, not a simple up-down rule (Watson & Pelli, 1983). Taking into account the experimenter’s initial threshold guess (and standard deviation), and all the preceding trials in this run, QUEST estimates the most informative rate for the next trial. (We used a slightly enhanced version of the algorithm described in the original paper, maximizing expected information gain instead of likelihood.) Of course, the experimenter’s initial guess is, in a sense a bias, but it is impossible to make a measurement without picking what to measure. However, the initial confidence interval based on the experimenter’s guess is generously large (s.d. of 1.3 log word/min) and after the 15 trials of a run, the confidence interval is small (s.d. of 0.03 log word/min), with negligible residual effect of the guess.
Of the 8 conditions in Figure 1, the only one that participants got extra practice in, relative to the others, was the plain text condition. Across the 7 knockout conditions, each observer had similar exposure. Observer KT, an author, had more exposure to all the conditions than the other participants, but this additional exposure does not seem to have had much effect on the percentage contributions of the three processes (Table 2).
Watson, A. B., & Pelli, D. G. (1983). QUEST: a Bayesian adaptive psychometric method. Percept Psychophys, 33(2), 113-120.
Denis Pelli & Katharine Tillman
http://psych.nyu.edu/pell...