RECALL OF STIMULUS FREQUENCIES FOR THE STUDY OF THE RECALL OF STIMULUS FREQUENCIES FOR THE STUDY OF THE COMPETITION PRINCIPLE IN PREDICTIVE LEARNING TASKS COMPETITION PRINCIPLE IN PREDICTIVE LEARNING TASKS Ramos, Ramos, M.M M.M.* .* and and Catena Catena, A.** , A.** * * University University of of Jaen Jaen –Spain Spain– **University of Granada **University of Granada –Spain Spain– In the predictive learning and causal reasoning areas it has been established that the processing of events is subject to a competitive mechanism. According to this principle, the causal or predictive relationship magnitude assigned to a certain element is engaged in a relative way to other potential agents. The study we introduce is based in a relative validity procedure (see Baker, Vallé-Tourangeau and Murphy, 2000). In the task presented there were two intervening stimuli as potential predictors of a criterion. The design was carried out through two experimental conditions. In both these conditions the contingency of the shadowed stimulus was kept constant, whereas the shadowing stimulus was modified. The competition principle was computed by analysing the differences of the contingency judgment in the shadowed –unaltered- stimulus. Usually, competitive effects such as blocking, relative validity or discounting are ascribed to a coding or learning problem (see Shanks, Holyoak and Medin, 1996). That is to say, in any such procedure we would learn that the validity of the shadowed stimulus is low as there is a good predictor of the consequence. Furthermore, research on conditioning (Matzel, Schachtman, and Miller, 1985) has suggested that the competition principle takes place in the final stages, those responsible for the decision or choice of response. Thus, there seems to be data that suggests that the location of the competition principle is not clear. The present work suggests that measures based on recall frequencies of stimulus combinations could help us in the placement of the competition principle within the processing stages. Y ~Y Probability: X a b d P S =a/a+b ~X c P N =c/c+d Contingency: dP=P dP=P S -P N Predictions Predictions. As the validity computation is based on frequencies such as the ones mentioned in Table 1, we could maintain a prototypical contingency judgment situation and ask an estimation of the frequencies of the four types of information at the end of the experimental session. The logic then would be simple. The two relative validity conditions are the same both for the contingency of the shadowed event and for the frequencies in such table. Consequently, if the competition effect took place in the initial stages of the predictive-causal processing, then the frequencies’ estimations, in the same way as the contingency judgment, would be different in the two experimental conditions. In other words, the shadowing stimulus validity would affect the two types of measures in a convergent way. However, if the competition effect took place in the final processing stages, then the judgment would show competition but at the same time the recall of the different information types would remain the same. Method Method Participants Participants The participants were 36 undergraduates at the University of Jaen (Spain) whose attendance was optional. They participated in an introductory Psychology course. There were two conditions, High and Low group, with 18 subjects randomly assigned to each one. Sex and age distributions were very similar in both groups: median of 20 years and an 88% of women in the High validity group, and a median of 21.1 years and an 83% of women in the low validity group. Design and Procedure Design and Procedure A medical diagnosis task was used (Price and Yates, 1993, 1995). On each trial there appears the photograph of a face representing a specific patient. On this face, information is presented about symptomatology (presence or absence of the symptoms) and, immediately afterwards, about the disease (presence or absence). In this context of predictive learning, two symptoms of a syndrome were presented according to a relative validity design with two conditions. We made the validity for the symptom Regmentacion equal in the two conditions (dP of 0.43), and different regarding the symptom Prurosis: dP of 0.80 in the High condition vs. dP of 0.41 in the Low condition. The contingency judgment was measured as a validity subjective estimation, which is a typical measure in this kind of tasks (Allan, 1993). We also asked, at the end of the experiment, an estimation of the stimulus frequencies for both symptom-illness pairs. Since the competition mechanism is asymptotically evaluated, we will only present the judgment and the frequency estimations at the end of the experimental session and only for the potentially shadowed stimulus (Regmentacion). All the statistical decisions were adopted fixing a level of significance of 0.05. Results Results The most interesting results are summarised in Figure 1. Mean R judgments are different in the two conditions -difference of 18,972-. The t-test indicated that the mean judgment of the “High condition” was lower to that of the “Low condition” [T(34)= -2,448; Standard Error of Difference= 7,750; p<0.05; η 2 = 0,150; Power= 0,662]. Mean frequency estimations of types a, b, c, and d were very similar in the two conditions – differences of -4.55, -5.33, 12.22, and -2.33, respectively-. The t-test indicated that all the mean frequencies of these estimations were equivalent between the two conditions. For type a estimation [T(34)= -0,748; Standard Error of Difference= 6,093; p>0.05; η 2 = 0,016; Power= 0,112], for type b estimation [T(25,14)= -1,910; Standard Error of Difference= 2,793; p>0.05; η 2 = 0,097; Power= 0,459], for type c estimation [T(34)= 1,971; Standard Error of Difference= 6,201; p>0.05; η 2 = 0,103; Power= 0,482], and for type d estimation [T(34)= -0,673; Standard Error of Difference= 3,466; p>0.05; η 2 = 0,013; Power= 0,100]. 0 10 20 30 40 50 a b c d dP Type of information Subjective Estim ation High Low 0 10 20 30 40 50 a b c d dP Type of information Subjective Estim ation High Low Discussion Discussion We found that the contingency judgment on a stimulus that was kept constant varied in connection with the magnitude of the validity of the other event intervening in the learning situation. However, there was no variation in the other type of responses measured, the frequencies estimations, when we manipulated the contingency of the alternative stimulus. Consequently, there was a selective predictive competition principle for the first type of answers but not for the second type. The first result, the competition effect on the contingency judgment, adds to the one found in many studies using a procedure similar to ours. In fact most of the studies on competition are based on the relative validity paradigm (for an empirical review, see Baker and cols., 2000; and for the implications of results, see Shanks and cols., 1996). The second type of results we found, that recall of events was not affected by the competition principle, does not follow most of the predictive learning literature, or causal reasoning studies in general, although they are consistent with those of Price and Yates (1993, 1995). Nevertheless, these researchers did not measure the two types of answers in the same experiment, and in fact the competition effect magnitude clearly varied throughout the different experiments introduced. Our results show, as a whole, the possibility that the competition effect is not a matter of codification but rather, of decision-making. Consequently, these results tend to suggest that the effect takes place in final evaluative answers, although the recall of events is not altered following the competition principle. In fact, recent judgment literature seems to point in this direction (for a more comprehensible review, see Shanks et.al., 1996). Jones, Wills and McLaren (1998) explain what could be attributed to the learning mechanism and to the final stages of decision-making, by thoroughly characterising the decision-making mechanism. In their own words: “If the differences observed in an experiment are not attributable to learning but, rather, stem from the operation of decision processes that operate on what has been learnt so as to convert this information into a response, then trying to account for these differences via any learning algorithm is a mistake”. Table 1 Table 1 . Contingency computations Figure 1 Figure 1 . Subjective estimations for Regmentacion symptom. References References Allan, L.G. (1993). Human Contingency Judgments: Rule based or associative? Psychological Bulletin, 114 (3), 435-448. Baker, A.G.; Vallée-Tourangeau, F.; and Murphy, R.A. (2000). Asymptotic judgment of cause in a relative validity paradigm Memory & Cognition, 28 (3), 466-479. Jones, F.W.; Wills, A.J.; and McLaren, I.P.L. (1998). Perceptual categorization: Connectionist modelling and decision rules. Quarterly Journal of Experimental Psychology, 51B (1), 33-58. Matzel, L.D.; Schachtman, T.R.; and Miller, R.R. (1985). Recovery of an overshadowed association achieved by extinction of the overshadowing stimulus. Learning and Motivation, 16, 398-412. Price, P.C. and Yates, J.F. (1993). Judgmental overshadowing: Further evidence of cue interaction in contingency judgment. Memory & Cognition, 21 (5), 561-572. Price, P.C. and Yates, J.F. (1995). Associative and rule-based accounts of cue interaction in contingency judgment. Journal of Experimental Psychology: Learning, Memory & Cognition, 21 (6), 1639-1655. Shanks, D.R.; Holyoak, K.J.; and Medin, D.L. (Eds.). (1996). The Psychology of Learning and Motivation Vol.34. Causal learning. San Diego: Academic Press. Contingency Contingency Frequencies Frequencies