Effects of point-loss punishers on human signal-detection performance.

Adults played a computer game. They had to decide if a faint light flashed on the left or right. Correct picks earned points. Wrong picks lost points.

The team added two conditions. In one, only correct answers gave points. In the other, correct answers gave points and wrong answers took points away. They tracked how the extra point-loss changed the players' choices.

Point-loss pushed players away from the punished side. It acted like a mirror of reinforcement, not an extra boost.

When both rewards and losses ran together, response bias shrank below the level seen with rewards alone. The data fit a subtractive model: punishment takes away bias rather than adding new strength.

Lie et al. (2010) ran the same game a year later. They changed how often punishers appeared instead of how big they were. Bias still moved away from the punished side, but discriminability stayed flat. Together, the two studies show the effect is robust across different punisher setups.

Reberg et al. (1979) first showed the flip side in pigeons. More food for one choice biased pigeons toward that side while leaving discrimination untouched. The 2009 human study mirrors that pattern, but with losses instead of gains. The pair confirms that consequences steer bias, not perceptual skill.

Locurto et al. (1980) reinforced pigeons for errors and watched accuracy drop. The 2009 paper adds the missing piece: taking points away for errors steers bias in the opposite direction. The studies line up to show that both adding and removing consequences shift choice, but in opposite directions.

If you use both rewards and penalties in a task, know they do not simply stack. Losses cancel out some of the bias that rewards create. When you want neutral responding—say, during preference assessments or conditional-discrimination probes—pairing small penalties with rewards can keep skewed choices in check. Start by adding a mild penalty for incorrect responses while keeping reinforcement equal across options, then watch for more balanced responding.