Run length, visit duration, and reinforcers per visit in concurrent performance.

Researchers set up two side-by-side keys for pigeons. Each key paid off on its own VI schedule. The birds could stay on one key or hop to the other at any time.

The team recorded every peck and switch. They measured how long each bird stayed on a key (visit duration) and how many pecks it made before leaving (run length).

Visit durations and run lengths tracked the programmed payoff odds. When the left key paid twice as often, birds stayed there about twice as long and pecked about twice as much before switching.

The numbers fit the generalized matching law. Stay/switch contingencies alone produced orderly, predictable choice patterns.

Mechner (1958) first showed that run lengths depend on what happened before. Macdonall (1998) adds the matching law to that story: the longer runs aren't random; they scale with reinforcement rate.

Gabriels et al. (2001) extend the idea. They used log-survivor plots to prove that reinforcement boosts two things: starting a bout and keeping it going. Macdonall (1998) shows those same bouts in plain time and count units.

Hachiga et al. (2010) look like they clash. They broke up long runs on purpose with a special algorithm. Macdonall (1998) lets long runs grow naturally. Both studies agree: run length is under environmental control; the difference is whether you let it grow or step in to shrink it.

You can now see matching in micro-behavior, not just overall rates. If a client hops between tasks, measure seconds or responses per bout. When the richer task pays more, longer bouts there tell you matching is working. Use that data to fine-tune reinforcement before problem switching appears.