1
17 December, 2001 p.
Dr. Marilynn Brewer
Associate Editor Psychological Review
Department of Psychology
Ohio State University, 1885 Neil Avenue
Columbus, OH 43210-1222
U.S.A.
Manuscript: "A Recurrent Connectionist Model of Group Biases" (01-072)
Dear Marilynn,
Enclosed is a revision of our earlier manuscript. We were, of course, excited about your general positive evaluation, as well as that of the three reviewers. We very much appreciated their comments as they showed great care and expert understanding of connectionism. We thank you also very much for your effort in highlighting the most important issues in the reviewers comments.
As you requested, we made a major revision of the manuscript. As you suggested, we dropped the simulations on recency effects in group impression formation and of the supercongruency effect. In addition, because one of the comments of reviewer C on the testing producers (see below) made use realize that the second simulation on accentuation could be explained more parsimoniously in terms of the first accentuation simulation, this simulation was dropped and only discussed as additional finding together with other moderating variables. In sum, only 5 major simulations remain, which limits the overall length and scope of this version.
In response to two additional requests of you, we added an appendix with technical details on the model. We also addressed the manuscript to a greater audience, and also highlighted the contribution of our model to connectionist models in general (see also concluding section).
We also addressed all issues raised by the reviewers. First by addressing all highlighted comments, and then by going through all other comments. In practically all cases, we either agreed with the comments or realized that some more detailed rationale or qualification was needed to motivate our approach. As a result, all comments were certainly helpful in improving the manuscript.
As a consequence of some comments, especially of reviewer C, we redid all simulations except Simulation 1 (size-based illusory correlation) and 3 (accentuation). The major reason was that we agreed with the reviewer's suggestion to be more consistent in our testing procedure. Specifically, in all our simulations, we now begin by cuing the group (or exemplar) nodes and then read off the resulting activation on the attribute node, except for recognition (assignment) measures were we cue the exemplars and read off the activation of the group (see also appendix). This enhanced consistency will probably improve the understanding of the simulations, as they are now very similar.
One of your suggestions was to change the order of the group variability simulation. We decided not to do that because all first four simulations rest predominantly on a measure of central tendency (liking, frequency estimates), while the last variability simulation rests on a different measure of range. Given the simplification in testing procedure that we attempted to achieve (see above), you will probably better appreciate why this seemed a better order to us.
We hope that you find the present version improved, but we realize that perhaps some small issues will remain. However, we are always willing to address such minor comments.
Sincerely,
Frank Van Overwalle
Vrije Universiteit Brussel
Department of Psychologytel: +32 2 629 25 18
Pleinlaan 2fax: +32 2 629 24 89
B - 1050 Brussele-mail:
Belgium