Toward an empirically based parameter selection for time-sampling observation systems

The measurement error of time-sampling observation systems, used to estimate the frequencyof behavioral events, was analyzed by means of a five-factor design, computer simulation experiment. The first three factors represented response parameters: the relative frequency and duration of the behavior and the pattern of response distribution. For each combination of frequency, pattern, and duration, five simulated behavior events were generated within a simulated observation period of 900 sec. A total of 21 different time-sampling systems was employed. As a fourth factor three different observe and record lengths were included (6, 12, and 60 sec). In addition, there were seven different ratios of “observe” to “record” interval length (5∶1, 3∶1, 2∶1, 1∶1, 1∶2, 1∶3, and 1∶5) representing the fifth factor. Thus, 21 time-sampling systems scanned a total of 120 different behavior simulations from 24 types of behavior parameter combinations. The data were analyzed by means of a five-factor (2 × 3 × 4 × 3 × 7) analysis of variance with repeated measures on two factors. The study demonstrated that time sampling leads to high average measurement errors, which are determined by complex interrelationships among a variety of variables. Choosing a time-sampling system arbitrarily may lead to highly erroneous data. It was also shown, however, that time-sampling systems have the potential to yield very accurate results. An empirical selection procedure for time-sampling intervals is proposed to minimize measurement error. Exemplary tables are presented from which time-sampling parameters can be chosen given that the rate, duration, and pattern of occurrence of a behavior to be observed are known.