Inhibitory stimulus control in concurrent schedules.

The team ran pigeons on two levers at once. Each lever paid food on its own timer. A light or tone told the bird which lever was “hot.” The birds could hop back and forth whenever they wanted. The researchers watched where the birds spent their time and how fast they pecked.

They also slid the signal along a scale—bright to dim, loud to soft—to draw a generalization curve. The key twist: the birds themselves triggered most of the signal changes by switching levers.

When the birds controlled the signals, their time on each lever traced a smooth U-shaped curve across the test values. The curve looked like classic generalization. But the raw peck-rate numbers stayed flat and messy. Inhibitory stimulus control showed up clearly in choice and time, not in how fast they pecked.

Absolute response rates jumped around even when the birds clearly “knew” which lever was paying. The authors call this an “inconclusive” split between choice and rate.

Migler et al. (1969) saw the same split one year earlier: lever choice followed the stimulus, peck rate did not. The 1970 study adds the wrinkle that letting the bird change the stimulus sharpens the gradient even more.

Leigland (1987) went the opposite direction. When one lever paid twice as much, the generalization curve flattened. Schwarz et al. (1970) kept pay rates equal, so their curve stayed steep. The two papers together tell us: stimulus control sharpens when rewards are balanced and the learner drives the changes.

Weisman et al. (1975) later showed that as a discrimination gets harder, schedule control grows while stimulus control fades. That helps explain why raw peck rates stayed noisy here—the birds were juggling two controlling variables at once.

When you probe stimulus control, look at choice time or location, not just counts per minute. Let the client switch stimuli themselves—give them a toggle, a button, or a way to “change levers.” You will see clearer generalization gradients and spot true discrimination faster. If rewards are unequal, expect flatter curves; balance reinforcement first before you trust a flat gradient.