Choice and the rate of punishment in concurrent schedules.

Deluty (1976) put two levers in a rat box. Pressing either lever gave food on the same random-time schedule. One lever also shocked the paw on its own random-time schedule. The other lever never shocked.

The team slowly raised the shock rate only on the first lever. They watched how this changed the rat's split of presses between the two levers.

When shocks came faster on the punished lever, rats pressed it less. At the same time, they pressed the safe lever more, even though the safe lever still gave the same food and no new shocks.

In short, punishing one choice pushed the rats toward the other choice, even though that second choice had never been made safer or richer.

DARDANO et al. (1964) saw the same bounce years earlier. They also found that punishment made rats hop between levers too often and freeze after bad presses. Deluty (1976) keeps the bounce but drops the extra hopping, showing the cleaner choice shift.

Fontes et al. (2025) later added fast-changing food rates. They found that rats still pick by relative shock danger, not by absolute shock counts. This widens Z's idea: the bounce holds even when food odds swing minute to minute.

Bland et al. (2018) swapped shock for an S- card that had meant "no food." The card alone pushed pigeons away from the key that produced it. Same story, milder tool: any signal of bad news can steer choice.

If you punish one behavior, clients may jump to another behavior that is still on extinction or still earns only thin reinforcement. Check the whole picture: the "safe" escape route might be just as problematic. Before raising the punishment level, set up a truly reinforced, prosocial option so the bounce lands in the right place.