The accuracy of time sampling procedures for estimating behavioural frequency

Abstract

The term time sampling cover s a broad range of techniques whose common aim is to quantify observed behavior in a sample of time during which the behavior is observed. Researchers using time sampling aim to infer the true pattern of the behavior of interest from data on the observed pattern of behavior. Inaccuracy in data obtained from time sampling procedures was classified into two sources: (l) the observers and subjects, and (2) time sampling procedures themselves. This paper focused on the inaccuracy due to the procedures themselves. Inaccuracy due to time sampling procedures was classified into two sets of factors: (1) the pattern of true behavior, and (2) the design. decisions made by the researcher. Recent studies have suggested that accuracy of time sampling is affected by the interaction of these two sets of factors. Since the researcher cannot manipulate the pattern of true behavior, this paper focused on the design decisions made by the researcher in an attempt to ascertain the relationship between design decisions and accuracy. The study employed a Monte Carlo approach, using a specially written computer program to generate 1000 "true" behaviors and then to simulate time sampling studies based on these data under a wide range of values of time sampling design factors. The simulated time sampling data was analyzed by multiple linear regression techniques to ascertain factors related to accuracy of the data. For all three simulated time sampling procedures the regression analyses produced an R2 value in excess of .99. The results of the study show that accuracy of time sampling is a linear function of certain design factors. Two conclusions were drawn from the findings of this study: (1) accuracy of time sampling procedures can be perfectly predicted from mean duration of behavior, length of time interval, and length of gap between intervals, and (2) most previous studies using time sampling procedures have produced inaccurate data. Three applications of the findings of the study to time sampling studies were noted: (1) a formula was derived to enable values of design factors to be set such that the data collected is free from inaccuracy produced by time sampling procedures, (2) a formula was derived which can be used as a "correction factor" for data previously collected by time sampling, and (3) the usual method of using time sampling in settings where the true behavior pattern changes over the period of the study will result in data of differential accuracy over the period of the study, but this problem can be overcome by altering values of design variables over the period of the study.