Information processing, data inferences, and scientific generalization

The philosophical view that scientific hypotheses come out of data is put to a limited test by investigating actual instances of hypothesis development in a laboratory setting. Subjects who are faculty members or graduate students are asked to find the function from which a plot of ten coordinate values affected by random error was derived. Their protocols serve as the basis for an information processing computer model of performance on the task. The model has a perceptual phase in which a pattern is found in the data, the selection of a class of hypothesized functions, a problem solving phase to find a specific function, and the ability to recycle if necessary. This seems to be compatible with reports of hypothesis development and the small amount of empirical research in other contexts. Tests of the model reveal it does a good job of getting the same answers as subjects and can explain some, not most, of the process leading up to the answers. Due to the contrived nature of the task the model is most appropriate for understanding routine scientific inference.