Interval sampling methods and measurement error: a computer simulation.

Wirth et al. (2014) built a computer model that pretends to be a kid.

The model acts out short bursts of behavior and long quiet spells.

They ran thousands of pretend sessions to see how close momentary time sampling, partial-interval, and whole-interval recording come to the true counts.

Each method missed the mark in its own way.

The team turned the misses into easy look-up tables.

Pick your session length, behavior rate, and interval size, and the table tells you which method adds the least error.

Virues-Ortega et al. (2022) later watched real college kids code behavior on paper or with two apps. Paper slightly beat the tech, but all three landed in the same high-accuracy range. Their live test backs up Oliver’s warning: the tool matters less than picking the right interval rule.

Chou et al. (2010) showed that feedback and small payoffs can sway what an observer writes down. Oliver’s tables help you cut the measurement noise first, so any leftover bias is easier to spot.

Alsop (2004) used the same Monte-Carlo trick to show that zero-error data sets fool standard stats. Both papers shout the same message: check your method before you trust the number.

Next time you plan 10-second partial-interval checks because “that’s what we always do,” pause. Look at Oliver’s tables first. You may see that momentary time sampling gives you cleaner data for the same effort. Less error means clearer graphs, faster decisions, and fewer program changes that weren’t needed.