Submitted for publication.Please do not quote.Version of April 26, 2001.Comments welcome.Cue competition in function learning: Blocking and highlightingJohn K. KruschkeIndiana University, BloomingtonIn function learning, people learn to predict a continuous outcome from continuous cues. Incategory learning, people learn to predict a nominal outcome. The present research demon-strates that two complementary forms of cue competition, previously found in category learn-ing, also occur in function learning. One form of cue competition is blocking of learningabout a redundant cue (Kamin, 1968). A second form of cue competition is highlighting ofa diagnostic cue (a.k.a. the inverse base rate effect; Medin & Edelson, 1988). For tests withconflicting cues, the results show bimodality of responses, as opposed to averaging, whichimplies exclusive selectivity that cannot be discerned from category learning paradigms. Itis argued that these effects are caused, in both category and function learning, by attentionalshifts. No previously published model of function learning can account for these effects, but amodel by Kalish, Lewandowsky, and Kruschke (2001) is promising.This article reports evidence of two types of strong cue com-petition in function learning. One effect is “blocking” oflearning about a redundant relevant cue (Kamin, 1968). Theother effect is what I call “highlighting,” previously referredto as the inverse base rate effect (Medin & Edelson, 1988).There are two main implications of this work. First, becausethe effects are now evident in both function learning andcategory learning, the results suggest that future theories oflearning in either domain should be designed to address theother domain as well. Second, because theories of these cuecompetition effects in category learning posit a central rolefor selective attention (Kruschke, 1996; Kruschke & Blair,2000), theories of function learning should also incorporateselective attention.The new data also indicate mutual exclusivity in cue com-petition, as opposed to averaging, that cannot be measuredin category learning paradigms. Consider what might hap-pen when an observer is confronted by two conflicting cuessimultaneously. For example, suppose that the person haspreviouslylearned that cue A indicates outcome 1, and cue Cindicates outcome 3, where outcomes 1 and 3 are values ona metric scale. Suppose that the person is then confronted bya test case in which the conflicting cues A and C are paired.The observer might contemplate the responses for both cuesThis research was supported in part by NSF grant BCS 9910720.The author thanks Michael Kalish and Stephan Lewandowsky formotivation to pursue this research. For helpful comments on draftsof this article, the author thanks Jerome Busemeyer and StephanLewandowsky. For assistance administering the experiment, theauthor thanks Twanna Allen, Colin Campbell, Dan Hall, KaraKohnen, Indya Watts and Maxine Weiss.Correspondence can be addressed to John K. Kruschke, De-partment of Psychology, 1101 E. 10th St., Indiana Univer-sity, Bloomington IN 47405-7007, or via electronic mail to [email protected]. The author’s world wide web page is athttp://www.indiana.edu/kruschke/and generate a response that best reflects an average of thetwo, i.e., response 2. Alternatively, the observer might exclu-sively select one or the other cue (or cue-outcome link) andgenerate either response 1 or response 3, perhaps bimodally.Yet another possibility is that the person could just decidethat the conflicting cues imply that all bets are off, and sos/he just chooses randomly (uniformly)from the available re-sponse options. Still other possibilities exist, of course. Mypoint is that these three different response tendencies, whichleave different signatures in the function learning paradigm,cannot be so easily distinguished in the standard categorylearning paradigm. This is because in category learning thereare no intermediate response options between the nominalcategory labels. The results reported below are most consis-tent with mutual exclusivity in cue selection, as opposed toaveraging.BackgroundFunction learning is common in everyday tasks. As justa few examples, consider that people can learn how hard tothrow a ball in response to visual cues about the distance ofthe catcher, paramedics can learn how much medication toadminister in response to cues about the weight of the pa-tient, and investors can learn how much stock to purchasein response to cues about market trends. Category learningis distinct from function learning only in that the outcomesare categorical values instead of metric values. That is, theoutcomes are merely labels (without size or order), insteadof magnitudes. This seemingly small difference suggeststhat similar psychological mechanisms should be involved inboth, and that research about the two types of learning shouldmutually inform each other (see review by Busemeyer, Byun,Delosh, & McDaniel, 1997). One of the central types ofphenomenaobservedin category learning is cue competition,wherein alternative cues apparently compete to gain associa-tive predictivenessof the outcome. It seems reasonable to ex-pect that cue competition should also occur in function learn-12 KRUSCHKEing. Cue competition is fundamental in associative learning,but only a little work has found cue competition in functionlearning (e.g. Birnbaum, 1976; Busemeyer, Myung, & Mc-Daniel, 1993; Mellers, 1986).Perhaps the most well known example of cue competi-tion in category learning is blocking, which has been ob-served in a variety of species and procedures (e.g. Dickin-son, Shanks, & Evenden, 1984; Kamin, 1968; Kruschke &Blair, 2000; Shanks, 1985), and which revolutionized theo-ries of associative learning. In blocking, the participant firstlearns that a cue, denoted here as A, perfectly predicts anoutcome. In subsequent training, the participant experiencescases of cue A paired with another cue, denoted B, still per-fectly predictive of the outcome. Despite the fact that B per-fectly predicts the outcome, people apparently do not learnto strongly associate it with the outcome. That is, learningabout B has been blocked by previous learning about A. Tomy knowledge, blocking has never been sought in the contextof function learning, yet it should exist if blocking is truly afundamental and pervasive phenomenon in learning.Another dramatic case of cue competition in categorylearning is what I call