Richard Shiffrin
Memory and Perception Lab
Indiana University

Cursor movements analyzed with a Hidden Markov Model provide rapid reliable inference about cognition

Individuals’ use of attention is studied by use of a mouse to move a cursor to rapidly appearing and disappearing targets, and not toward foils. Four aspects of the cursor movement are measured every ten ms. The data are analyzed with a Hidden Markov Model (HMM) to infer the screen position at which attention resides every ten ms. Doing so requires estimation of 13 parameters that map the cursor measurements onto location of attention. The first five minutes of data are used to find the parameters that maximize the probability that the subject is carrying out the task as instructed. The parameters for each subject are then fixed and used to predict the movements of attention for the following fifty minutes of data collection. The inferred positions of attention are then used as data for a cognitive model positing how attention moves in accord with the sequence of presented stimuli. This method can be used for many other cognitive and perceptual tasks. Diagnostic data are obtained in large amounts very rapidly and coupled with the analysis methods provide rapid and reliable inference about cognitive processes for each individual tested.