Undermatching and overmatching: The fixed-ratio changeover requirement.

The team worked with pigeons in a two-key chamber. Each key paid off on its own schedule.

To switch keys the bird had to peck a small fixed number of times. The researchers tried FR 1, 5, 15, 30, 60, and 100.

They logged how many responses and how much time the birds gave each key. Then they checked if the birds matched, under-matched, or over-matched the payoff rates.

Low FR changeovers (1-5) gave almost perfect matching. Birds split responses and time just like the payoff rates.

High FR changeovers (30-100) pushed the birds toward under-matching. They stayed on the richer key longer than the payoff rate said they should.

Middle FR 15 gave a mix: some birds under-matched, a few over-matched. The size of the changeover cost predicted the pattern.

Marr (1989) later showed that when payoffs are very unequal you need a curved (power) equation, not a straight line, to predict choice. The 1981 data already hint at this: high FR changeovers flatten the slope, the same thing a power function handles.

Thrailkill et al. (2025) moved the idea from pecking to looking. Pigeons looked longer at parts of a screen that had paid off more. Same rule, new response. The 1981 paper is the root: change the cost of switching and you change how behavior spreads.

Locurto et al. (1980) looks like a clash. They found reinforcing errors hurt accuracy, yet here reinforcing key switches (changeover pecks) helped choice track payoff. The difference is what gets reinforced. Reinforcing wrong picks strengthens confusion; reinforcing a switch keeps the matching law intact.

If you want a client to split time evenly between two tasks, keep the cost of moving low. Use small response requirements or simple cues to switch. If you need them to stay on one task longer, raise the switch cost—add a few steps or a brief delay. One lever moves the whole choice pattern.