Psychological Review 2000, Vol. 107, No. 4, 947-949 Copyright 2000 by the American Psychological Association, Inc. 0033-295X/00/$5.00 DOI: 10.1037//0033-295X.I07.4.947 The Importance of Theory: Response to Brenner (2000) Thomas S. Wallsten University of North Carolina at Chapel Hill Ido Erev The Technion: Israel Institute of Technology David V. Budescu University of Illinois at Urbana-Champaign L. Brenner's (2000) critique of 1. Erev, T. S. Wallsten and D. V. Budescu (1994) focuses on their (a) use of a model to explain the paradox of the same data appearing to suggest over- and underconfidence, depending on how they are analyzed; (b) definitions of true judgment and error; and (c) specific use of judgments transformed to log-odds and a model formulated in those terms. The authors of the present article strongly disagree with the first point and discuss the importance of using models to interpret data. With regard to the second, the authors admit that the constructs of true judgment and error are poorly named but dispute L. Brenner's specific criticisms. Concerning the third, the authors had not claimed that the log-odds metric has any special status in judgment research and thus agree with L. Brenner's basic point. We appreciate the opportunity provided by Brenner's (2000) critique in this issue to clarify some important issues in Erev, Wallsten, and Budescu (1994). Brenner raises three main points. First, he chides us for focusing on "model constructions (true judgments) over observed data" (p. 943). Second, he says that the work "suffers from a basic flaw: the lack of a clear definition of either the true judgment t or the error term e" (p. 944). Third, focusing on our use of the log-odds model, he says that "no compelling rationale was provided by Erev et al. (1994) for the log-odds model, beyond the suggestion that covert confidence should be represented on an unbounded (—»,"») rather than [0, 1] scale" (p. 944). We respond to these criticisms in reverse order so that we may concentrate on the first, which is the most fundamental. With regard to the arbitrariness of the log-odds model, we agree with Brenner. Indeed we wrote, "we make no special claims regarding the log-odds model" (Erev et al., 1994, p. 545). We further stated that its primary virtue is that it is a convenient special case, used for illustrative purposes only, of a much more general model representing a broad class of psychological theories. Our point was not that any particular metric is better than another, but to provide Thomas S. Wallsten, Department of Psychology, University of North Carolina at Chapel Hill; Ido Erev, Faculty of Industrial Engineering and Management, The Technion: Israel Institute of Technology, Haifa, Israel; David V. Budescu, Department of Psychology, University of Illinois at Urbana-Champaign. This research was supported by National Science Foundation Grants SBR-9632448 and SBR-9601281. We thank Lyle Brenner for comments on an earlier draft of this article. Correspondence concerning this article should be addressed to Thomas S. Wallsten, who is now at Department of Psychology, University of Maryland, College Park, Maryland 20742-4411. Electronic mail may be sent to twallsten@psyc.umd.edu. a concrete illustration that a model explicitly incorporating random perturbations can handle the dual results of over- and underconfi- dence. Similar points have been made by Budescu, Erev, and Wallsten (1997), Juslin, Olsson, and Bjorkman (1997), Pfeifer (1994), and Soil (1996). His second criticism concerns our definitions of true judgment t and error e. He considers our definition of true judgment to be circular, in that it invokes the ideal case in which a respondent operates in a "fully repeatable error-free manner" (Erev et al., 1994, p. 524). The use of unattainable ideals (including frictionless surfaces and absolute vacuums) to clarify thinking has a long history in science. Moreover, error has a clear operational defini- tion, that is, identical procedures under identical conditions leading to different results. Thus, its absence (even if impossible to achieve) is clearly defined and circularity is not a problem. Al- though we elaborated on our definitions subsequently (Budescu et al., 1997, pp. 158-159), we agree with Brenner that the term true judgment was poorly chosen. As illustrated by his criticism, it carries excess meaning that interferes with understanding the sci- entific aspects of the construct. Wallsten, Budescu, Erev, and Diederich (1997) did a better job when they restated the general model, assumed that the respondent partitions the infinite covert confidence scale into a finite number of categories (or, equiva- lently, that the scale is fundamentally discrete), and denoted the categories by the respondent's targeted proportion of true events in each one. That proportion plays the role of true judgment in Erev et al. (1994), is defined in psychological terms, and has no excess meaning. Now we turn to the most important criticism, that we treat the model as more fundamental than the data. On this point we strongly disagree with Brenner. Our position is that data cannot be understood properly without a model, not that one component is more important than the other. Over a half-century ago, Hempel and Oppenheim (1948) wrote, "To explain the phenomena in the 947 This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.