Matching in a network.

Glover et al. (1976) worked with pigeons in a lab network. Birds first pecked on two keys that paid off at different rates.

After training, the team turned off all rewards. They watched whether the birds' response splits still followed the matching law.

During training, response ratios lined up with reward ratios — classic matching.

When rewards stopped, the birds did not drift toward equal pecking. They kept favoring the key that had paid more, even though it now paid nothing.

Shimp (1967) showed matching can appear without maximizing by reinforcing rare four-peck chains. Glover et al. (1976) go further: once matching forms, the bias can outlast the payoff.

Stubbs et al. (1970) found that local burst patterns drive matching. The 1976 study reveals those bursts leave a lasting footprint; birds still lean toward the once-rich side after extinction.

White (1979) seems to disagree. Adding an extinction key inside one alternative did not shift matching. The gap is procedural: M kept some rewards flowing; J removed them all. Remove every payoff and the old rich side keeps the edge.

Your client's history can lock in preference even when current rewards change. If a child always got bigger praise for math than reading, that bias may survive when you thin reinforcement. Probe unreinforced trials to see the true leftover preference, then re-balance your delivery rates to offset the historical pull.