Stimulus disparity and punisher control of human signal-detection performance.

Twenty college students sat at a computer. They had to tell two circles apart. One circle was always a bit brighter.

If they picked the wrong circle, they sometimes lost money. The team changed two things: how different the circles looked and how often wrong picks lost money.

The goal was to see if harsher money loss made people better at spotting the brighter circle.

Making the circles more different helped people choose the right one. That was it.

Raising or lowering the money-loss rate did not change how well they saw the difference. It only made them pick the safer circle more often.

So punishment pushed response bias, not vision.

Locurto et al. (1980) showed the flip side: paying pigeons for wrong picks hurt their accuracy. Celia et al. now show that fining humans for wrong picks does not hurt accuracy—it just shifts bias. Together they map both sides of the coin.

Rider et al. (1984) found that longer delays hurt stimulus control, but payoff rates still swayed choice. Celia’s team held delay near zero and still saw payoff sway choice while leaving discriminability untouched. The two studies line up: payoff alters bias, not the stimulus itself.

Jensen et al. (1973) saw contrast effects tied to past response rates, not reinforcement rate. Celia’s data echo this: the stimulus difference mattered, punishment schedule did not. Both warn us that what the learner just did can outweigh the pay schedule we set.

When you add a penalty—time-out, response cost, a firm “no”—expect the learner to play it safe, not to see the cue more clearly. If accuracy is low, first sharpen the stimulus difference; don’t just crank up the penalty. Check bias with a signal-detection plot: hits and false alarms will show whether the learner truly can’t tell the cues apart or is simply avoiding the punished choice.