Effects of reinforcement rate and delay on the acquisition of lever pressing by rats.

Researchers taught rats to press a lever for food. They compared three schedules: every press paid off (CRF), a press paid roughly every 15 s (RI 15 s), and a leaner RI 30 s. Half the rats got the food right away. Half had to wait 12 s after each press before the pellet arrived.

The team watched how fast each rat learned to press during the very first sessions.

The rats on the lean-but-steady RI 15 s schedule pressed the most. Rats that got every press rewarded (CRF) actually pressed the least. Adding a 12-s delay washed out most of the differences; all groups pressed at moderate, similar rates.

In short, during brand-new learning, a little bit of scarcity can be more powerful than constant reward.

Sutphin et al. (1998) also taught rats to press under delays, but they gave the animals a second “cancellation” lever. That study showed rats can still learn with delays up to 32 s when they can cancel a delayed reward. The current paper shows the same basic effect—delays flatten rate differences—even without the extra lever.

Coe et al. (1997) used the same unsignaled-delay setup a year earlier and found that body weight and prior “magazine training” changed whether rats learned at all. The present study keeps weight constant and instead shows that the schedule itself—how often reward arrives—matters more than sheer delay once the contingency is sensed.

Lowe et al. (1995) and Anonymous (1995) looked at force: harder levers slow rats down. The current paper flips the variable: it keeps the lever easy but changes how often and how soon reward comes. Together the two lines say “response rate drops if the work gets harder OR if reward arrives too often.”

When you shape a new skill, don’t rush to reinforce every response. A thin, steady schedule (think FI 15 s or VR 3) can produce faster early responding than a dense CRF. If you must add a delay—say the learner hands you a token and then waits for praise—keep the delay brief and consistent; long or variable delays level out the very differences you’re trying to create.