Levels of aggregation: Relative time allocation in concurrent-schedule performance.

Researchers watched pigeons peck two keys during 60-min sessions. Each key paid off on its own timer. The team counted pecks minute-by-minute and again for the whole hour.

They wanted to see if the birds' overall session preference matched what they did in short slices.

Session totals hid the real story. A bird could look like it favored the left key all hour, yet spend whole minutes glued to the right.

The authors warn: one big average can erase local swings you need to understand.

Garcia et al. (1973) already told us to trust response counts more than time clocks. G et al. add: even response counts can fool you if you only look at the session sum.

Rutland et al. (1996) later showed those local swings track reinforcer rate minute-by-minute, backing the warning with hard numbers.

Busch et al. (2010) stretched the idea further. over the study period the birds' quick pulses kept changing; only session-by-session graphs caught it. The 1985 paper is the seed, the 2010 paper is the tree.

When you run a concurrent-operant preference assessment, do not stop at the end-of-session score. Break the session into 1- or 2-min bins. If the learner flips between bins, you may be masking extinction bursts or emerging biases that guide your next intervention.