Cognitive choice processes for sequentially or simultaneously presented alternatives

Although decision makers use dimensional processing strategies when choosing between simultaneously presented and multidimensionally described alternatives, Tversky (1969) has conjectured that alternative based processing would occur when the same alternatives are presented one after the other. This hypothesis was elaborated upon with effort—quality considerations. The first experiment which was a replication of a previous study only partially supported the above hypothesis. As predicted, longer latencies indicated more judgment based processing for the sequential presentation than for the simultaneous presentation. However, memory measures did not indicate any differences. Think aloud protocols of a second experiment revealed that for sequentially presented alternatives dimensional comparisons are applied in addition to judgment based processing. The application of the two processing components can be described by a Markov model which specifies a hybrid decision strategy. In his 1969 paper, Tversky demonstrated that decision makers use dimensional processing strni î ies when choosing between multidimensional alternatives. Although the choice alternatives were simultaneously presented in Tversky's experiments, he conjectured that alternative based processing would occur for the same alternatives when they are presented one after the other. Tversky's conjecture is intuitively appealing. For example, when shopping for a product, the products which are offered by one store are simultaneously available and can thus be processed by dimensional strategies. The products of different stores must be inspected in sequence, and therefore it is quite likely that alternative based processing occurs. While Tversky elaborated upon dimensional processing strategies in considerable detail, he referred quite generally to the additive model with respect to alternative based processing. The additive model, however, does not capture the underlying cognitive processes, i.e., the procedural character of judgment formation (Lopes, 1982). Cognitive choice processes for choosing between two alternatives when they are sequentially or simultaneously presented to a subject were described by Schmalhofer and Gertzen (1986). Their description was 1) This research was supported by a grant from DFG (Al 205/1). We gratefully acknowledge the comments of Dietrich Albert, Michael Aschenbrenner, and Otto Kuhn. based upon the framework of criterion dependent choice models, whose empirical validity has been shown in several studies. Aschenbrenner, Albert and Schmalhofer (1984) have shown that such models can account for the information processing of dimensionally described and simultaneously presented choice alternatives: The models not only predicted choices better than the additive model, there was also a positive relation between choice latencies and the number of processing steps predicted by the models. Using a process — tracing method, Schmalhofer, Albeit, Aschenbrenner and Gertzen (1986) have shown that the criterion dependent choice models adequately describe the decision maker's selective and adaptive information utilization. Since the current research builds upon these models, we will first describe the generic criterion dependent choice model as well as the two specific models for dimensional and alternative based processing. Next, we will present an experiment which replicated a previous investigation in order to test the stability of the results of that experiment. As a second experiment, a think aloud study will be reported. The verbal protocols will be used for a detailed exploratory analysis of the information processes for sequentially presented alternatives revealing a hybrid decision strategy which can be described as a Markov model. Criterion dependent choice models The basic assumption of criterion dependent choice models is that the information processing of an alternative or choice pair continues until some evidence criterion is surpassed. This evidence criterion is specified by a number k, which is the only free parameter in a criterion dependent model. Thus, decision making is assumed to be a selective sequential process. The criterion dependent choice models postulate that the availability (and importance) of the choice alternatives' features determines both which features will be processed and the order in which they will be processed. The processing of a feature yields an attractiveness value for that feature. For attractive and unattractive features, positive or negative attractiveness values are obtained, respectively. The attractiveness values are combined according to some rule. This rule may specify dimensional comparisons or the formation of overall judgments as component processes of choices. Dimensional comparisons as component processes. It is assumed that at the beginning of the choice process neither alternative is favored. Therefore, at the beginning the evidence value is assumed to be zero. In the first processing step, the features on the most important dimension are evaluated, and the difference of the two attractiveness values is calculated. This calculated value represents the evidence value after the processing of the first dimension. Then the second most important dimension is processed. After the processing of the second dimension, the evidence value is updated by adding the attractiveness difference determined for the second dimension. This process continues until all dimensions have been processed or one of two criteria is surpassed, i.e., the evidence value is larger than k or the evidence value is smaller than — k. A positive evidence value determines the choice of one alternative, and a negative evidence value determines the choice of the other alternative. For a more detailed description see Albert, Aschenbrenner and Schmalhofer (in press). This dimension based processing strategy may be difficult to apply for sequentially presented alternatives, because the features of the first alternative would have to be stored in memory until the next alternative becomes available. To decrease the demands upon working memory, subjects could however use judgment as a component decision process in this case. Thus, subjects would make an overall judgment of the first alternative and store it in memory rather than a list of its features. A second overall judgment is then made for the second alternative, and the decision would be based upon a comparison between the two overall judgments. For such a strategy, the criterion dependent processing occurs for the formation of the judgments of the two choice alternatives. Judgments as component processes. The present conception about the formation of judgments assumes that the feature evaluations, which may be positive or negative, are summed up. In particular, it is assumed that at the beginning of the judgment process the evidence value for an alternative is zero, i.e., there is no bias toward a positive or negative judgment. In the first processing step, the feature of the most important dimension is evaluated and represents the evidence value after the processing of the first dimension. Then the second most important dimension is processed. After the processing of the second dimension, the evidence value is updated by adding the new attractiveness value. This process continues until one of two criteria is surpassed or all features have been processed. If the boundary k is surpassed the alternative is considered to be attractive. If the boundary — k is surpassed the alternative is unattractive. A judgment about the choice alternative is obtained by dividing the evidence value, i.e., the sum of the attractiveness values, by the number of processed features. A judgment of the second alternative presented is derived in the same way. The alternative which received the better judgment will be chosen. Effort—Quality Relations In order to assess the advantages of applying dimension versus judgment based processing, Schmalhofer and Gertzen (1986) performed an effort — quality analysis for the two choice situations of simultaneous or sequential presentation of the alternatives. This analysis was based upon the two described versions of the criterion dependent choice models. For the two strategies, the effort and quality of a choice have been analyzed under the conditions of simultaneous or sequential presentation of the choice alternatives. These effort — quality analyses showed that dimension based processing requires less computational effort, but that because of memory load, judgment based processing should be more economical for sequentially available alternatives. Since judgment based processing requires more accumulation operations than dimensional processing and these accumulations are probably more difficult and therefore more time consuming than dimensional comparisons (Klayman, 1982; Russo & Dosher, 1983), longer choice latencies should be observed for sequentially presented alternatives than for simultaneously presented alternatives. Furthermore, an experimental investigation of Tversky's hypothesis indicated that for sequentially presented choice alternatives decision makers are more likely to apply judgment based processing, but not necessarily to the complete exclusion of any dimension based processing. In particular, overall judgments obtained from the subjects after their choices were in moderately better agreement with the observed choices for sequentially presented alternatives than for simultaneously presented alternatives. In addition, choices between sequentially available alternatives, which supposedly require more computational effort due to judgment based processing, actually required more time than choices between simultaneously available alternatives. While these results are consistent with the effort — quality analysis, they also showed that the specific criterion dependent choice models may be too simple to account for the experimental findings in