Observational Procedures and Markov Chain Models for Estimating the Prevalence and Incidence of a Behavior

• In Event: 16.025 - Division D Exemplary Work From Promising Researchers

Abstract

Data based on direct observation of behavior are used extensively in certain areas of educational and psychological research. A number of different procedures are used to record data during direct observation, including continuous recording, momentary time sampling, and interval recording (partial interval recording and whole interval recording). Among these, interval recording has long been recognized as problematic because the mean of such data measures neither the prevalence nor the incidence of a behavior. In this project, I do two things. First, I propose a model for interval recording data based on a latent alternating Poisson process for the behavior being observed. The model can be used to recover estimates of the behavior's prevalence and incidence, using maximum likelihood methods. Second, I propose some new recording procedures that involve combining momentary time sampling and interval recording. By comparing the asymptotic efficiency of the new methods versus existing methods, I show that for certain behaviors (i.e., certain parts of the prevalence-incidence parameter space), the new methods yield more precise estimates of prevalence and/or incidence. In further work, I am planning to field-test the new procedures I've proposed. I'm also planning to study the robustness of the parameter estimates to violations of the alternating Poisson process modeling assumptions.

Author

• James Eric Pustejovsky, The University of Texas - Austin