Time sampling and measurement error: the effect of interval length and sampling pattern.

The accuracy of momentary time-sampling was evaluated for behaviors occurring for different proportions of time, for different momentary time-samples and for sessions of different lengths. For 1-hr sessions, momentary time-samples of 10, 20 or 30 see intervals between observations yielded 90% of estimates within 10% of the comparison standard, for behaviours of 25%, 50% and 70%. Behaviours occurring for 2% of the session were not accurately recorded even at 10-sec momentary time-samples. For a 5-min momentary time-sample, this criterion was only approached when 8 hr data was averaged to give a single score for each individual, for behaviours of 25% or more. The 5-min momentary time-sample was also compared with a 1 in 5-min cluster sample; although less accurate, cluster-sampling might be a useful strategy in some settings.