Assessing the Accuracy of Classwide Direct Observation Methods: Two Analyses Using Simulated and Naturalistic Data

Abstract

Two studies investigated the accuracy of eight different interval-based group observation methods that are commonly used to assess the effects of classwide interventions. In Study 1, a Microsoft Visual Basic program was created to simulate a large set of observational data. Binary data were randomly generated at the student level to represent 30–min observations using a 15-s momentary time sampling procedure for 5,000 students distributed evenly across 200 classrooms. Additionally, five classroom types were generated to determine if the base rate of behavior affected the accuracy of each method. Each of the eight observation methods were applied to the simulated data and compared to a criterion to determine their accuracy in estimating rates of classroom behavior. A two-way ANOVA identified a significant main effect for observation method, F(3.40, 663.07) = 15,824.22, p < .001, and a significant interaction between coding method and classroom composition, F(13.60, 663.07) = 40.04, p < .001. In Study 2, to replicate Study 1 using authentic observation data, we compared the accuracy of the same eight methods using a small sample of observations collected in an actual classroom. The results of both studies suggest that six of the eight methods (i.e., two Individual methods and four Planned Activity Check methods) are accurate assessments of group behavior. Furthermore, although classroom composition was associated with statistically significant differences between the accurate methods, we do not consider them large enough to be practically relevant.