Measures of response bias at minimum-detectable luminance levels in the pigeon.

The researcher tested pigeons on a dim-light detection task.

Birds pecked left or right keys when a very faint light appeared.

The team changed how often each key paid off to see if payoff odds skewed the birds’ choices.

When one key paid more often, birds favored that side even when the light was too dim to see.

Keeping payoff odds equal wiped out this bias and gave cleaner detection data.

Fahmie et al. (2013) later repeated the logic with rats and got the same quick bias shift, showing the pigeon rule works across species.

Chou et al. (2010) moved the idea to human observers; they found that feedback and bonuses swayed therapists’ data sheets just like grain swayed pigeons.

Alsop (2004) warns that zero-error datasets can fake better sensitivity than really exists; together these papers say control payoff odds first, then check for math artifacts.

If you run discrimination probes or preference assessments, keep reinforcers equal until the learner masters the task. Uneven pay will fake poor discrimination and hide true skill. After mastery, you can thin or shift rewards without the bias ghost.