Linear modeling of steady-state behavioral dynamics.

McIntyre et al. (2002) built a straight-line math model. The model guessed how fast rats would press a lever under brand-new reinforcement schedules.

They kept the model simple. They only added extra data points that covered the full range of past schedules.

The simple line model hit the target. It forecasted steady-state response rates under schedules the animals had never seen.

Adding wider data coverage made the guesses better, but the paper does not say how much better.

Blough (1992) saw curved scallops on fixed-interval schedules and said, "Only a wavy nonlinear model can draw these." McIntyre et al. (2002) reply, "A straight line still gets the final height right." Both can be true: one draws the shape, the other gives the level.

Chandler et al. (1992) showed that response rates rise then fall inside each session. The new linear model does not try to trace that hill; it only predicts the flat plateau at the end.

McDowell et al. (2018) used computer-evolved agents to mimic choice on concurrent ratio schedules. Their digital animals and L’s equation both hit the same endpoint without talking about mind or memory, showing that math alone can sometimes be enough.

If you track baseline response rates during assessments, a quick linear forecast can tell you where the client will land under a new schedule. No need to run days of probe sessions. Just sample rates across a few known schedules, draw the line, and read the next point. It saves time and keeps the animal—or person—from extra exposure to unstable conditions.