Information selection and belief updating in hypothesis evaluation
This thesis is concerned with the factors underlying both selection and use of evidence in the testing of hypotheses. The work it describes examines the role played in hypothesis evaluation by background knowledge about the probability of events in the environment as well as the influence of more general constraints. Experiments on information choice showed that subjects were sensitive both to explicitly presented probabilistic information and to the likelihood of evidence with regard to background beliefs. It is argued - in contrast with other views in the literature - that subjects' choice of evidence to test hypotheses is rational allowing for certain constraints on subjects' cognitive representations. The majority of experiments in this thesis, however, are focused on the issue of how the information which subjects receive when testing hypotheses affects their beliefs. A major finding is that receipt of early information creates expectations which influence the response to later information. This typically produces a recency effect in which presenting strong evidence after weak evidence affects beliefs more than if the same evidence is presented in the opposite order. These findings run contrary to the view of the belief revision process which is prevalent in the literature in which it is generally assumed that the effects of successive pieces of information are independent. The experiments reported here also provide evidence that processes of selective attention influence evidence interpretation: subjects tend to focus on the most informative part of the evidence and may switch focus from one part of the evidence to another as the task progresses. in some cases, such changes of attention can eliminate the recency effect. In summary, the present research provides new evidence about the role of background beliefs, expectations and cognitive constraints in the selection and use of information to test hypotheses. Several new findings emerge which require revision to current accounts of information integration in the belief revision literature.