There are a number of aspects about the Ruiz experiment that make it less than ideal. The interface he used perhaps structured subjects' behavior too much by forcing them to explicitly post each goal. His subjects were new to the interface and perhaps got a little faster over the problem as they became familiar with the interface. Correspondingly, the model tends to underpredict early points and overpredict late points. Pure pyramid problems he used are also quite simple and subjects quickly transition to a rote solution. This means one cannot use later trials where subjects are familiar with the interface.

To get a more demanding test of subject behavior we applied the ACT-R model to some data we collected (Anderson, Kushmerick, & Lebiere, 1993). This data comes from subjects solving the 4-disk problems in Figure 2.6. These problems are cases where one has to move from one arbitrary configuration to another so there is no possibility for subjects memorizing the sequence of moves. The interface was one in which subjects simply picked up a disk with a mouse and moved it to another peg. Unlike the Ruiz experiment, subjects were not monitored for using the subgoaling strategy. Nonetheless, they were strongly encouraged to in the instructions and the evidence is that they did. We will restrict our analysis to those move sequences where they perfectly executed the minimal 15 move sequence required to solve these problems. All problems involve setting the same numbers of subgoals before a particular numbered move in the sequence.

Model

ACT-R Model

Microsoft Excel Spreadsheet of Model