An empirical method for determining an appropriate interval length for recording behavior.
Test two interval lengths before you start — let the data, not the manual, pick the slice.
01Research in Context
What this study did
Capehart et al. (1980) built a simple test grid. They tried many interval lengths on the same behavior.
They asked: does the length change the picture you get? They watched mixed-clinical clients and counted both duration and events.
What they found
Duration scores stayed almost the same no matter how long the interval was.
Event counts jumped around. Shorter slices gave very different totals than longer ones.
The team said: pick your length with data, not with a guess.
How this fits with other research
Killeen (1978) came first. It told us when behavior is stable enough to trust. Capehart et al. (1980) adds the next step: once it is stable, choose the right slice size.
Klapes et al. (2021) updates the same line of work. They tuned lab tasks to get steady data faster. Their 0.5-s COD and blackout tricks replace the old long watches.
Vos et al. (2013) looks close but answers a different question. They pick the best math index for contingencies. Capehart et al. (1980) picks the best time window before you even run the math.
Why it matters
Next time you open a 10-s, 15-s, or 30-s grid, stop. Run a quick probe with two lengths. If duration is the target and both give the same number, use the longer one to save staff effort. If event counts differ, stay with the shorter slice. One extra pre-check saves weeks of shaky data.
Want CEUs on This Topic?
The ABA Clubhouse has 60+ free CEUs — live every Wednesday. Ethics, supervision & clinical topics.
Join Free →Open your next observation sheet, run one session with both 10-s and 30-s intervals, and keep the length that gives the same duration score.
02At a glance
03Original abstract
The study sought to examine the effects of varying interval length on the representation of data obtained using modified frequency time sampling. A 7-category scale was used to observe reliably the behavior of eight psychiatric inpatients. Using electronic real time recording equipment, it was possible to computer analyze the obtained data at varying interval lengths, the shortest interval being 1.0 seconds. It was found that increasing the interval length had little effect on the percentage of total duration recorded within each behavioral category, suggesting that this is a relatively stable measure of behavior. Percentage total events for each category was less stable with increasing interval lengths. The number of recorded events within each category tended to decrease, while their average durations tended to increase, as a function of increasing the interval length. The data suggest that the current practice of determining interval length in an arbitrary fashion, or on the basis of convention, should be abandoned. Rather, such a decision should be empirically determined for each particular observation scale and subject group. One method by which this might be achieved is presented.
Journal of applied behavior analysis, 1980 · doi:10.1901/jaba.1980.13-493