2000

Integration of Multiple Cues in Learning
Manuel Miguel Ramos Alvarez
My research deals with predictive processing in binary situations when these situations present more than one potential predictor of consequences. The main result obtained in our laboratory up to now is that predictors are processed independently in some situations such as the blocking one. These results do not agree with the competitive mechanism as shown by related literature. Psychologically speaking, the discounting or competition principle cannot explain our results. In our previous research we proposed an Information Integration Model based on Norman Anderson's theory and also related it to the Brunswik Lens Model. Our theoretical proposal takes its roots from two parallel regression equations, as in the lens model: One for the ecological structure, and the other for the subjective system. Such equations allow the estimation of multiple regression parameters for each of the predictors included in the situation. These estimations are carried out through the computation of the focal cue validity (or pairwise correlation) relative to the validities of the other potential cues. The linking function between the ecologicalenvironmental structure and the cognitivesubjective systems incorporates the subject's beliefs or assumptions about the causal texture. For instance, in the blocking context the person may have the belief that the cues are independent of each other, and this fact would lead to a simplified subjective estimation only based on the crude validity (the cuecriterion or cueconsequence correlation). In addition, relevant information to estimate the regression index of each predictor could be integrated according to different rules. We propose a formal model with two types of Information Integration Strategies, including every possibility up to date. Either the person uses a regression rule based on absolute frequencies (suboptimal heuristic strategy) in which the weight and the information sign may differ, or the person uses a rule in which the information is processed in a relative or probabilistic way. At the same time, each rule can be subdivided according to the person's assumptions. For instance, if one kind of causal link is assumed, the model is slightly different than if the assumption is the opposite one. Within this framework, we have carried out new research related to the processing strategy and research related to complex stimulus situations: A) In the former research area, we have manipulated the contingency, the expected associations between cues and outcomes (high positive expectancy or Null expectancy), and the type of information (Symmetric or Asymmetric). We have shown the usefulness of all the regression strategies in our model. We found that a high percentage of the people used heuristic rules in all the experimental conditions. With respect to the Normative Strategies, a high percentage of the subjects assumed a direction of regression opposite to the conventional, XtoY. There was also evidence of a very low percentage of strategies that did not take the appropriate information sign, according to the association established beforehand in the predictive situation. In addition, our research work helped us establish the subjective weight pattern of the different information types. Only some of the experimental conditions were adjusted to a uniform weight distribution. Some other conditions followed a nonuniform pattern proposed in the investigation of causalpredictive learning. Still some other conditions deviated from the two patterns above. B) Regarding our research about complex stimuli, we have extensively explored all those conditions that have an influence on the competition principle. The competition is only present when the situation allows the direct comparison of diverse types of experiences, i.e., a relative validity paradigm, and when the context is a causal one. In the blocking paradigm, contingency judgments do not follow the competition principle unless the experiment induces the belief that the cues are associated with each other.
Contact Manuel Miguel Ramos Alvarez 
