The generalization‐across‐dimensions model applied to conditional temporal discrimination

Davison et al. (2024) tested pigeons in a matching-to-sample task where the sample stayed on for either 2 s or 8 s. The birds had to pick the comparison key that matched the sample duration to earn food.

The team tweaked the generalization-across-dimensions model so it could handle this time-based choice. They ran two lab experiments and checked how well the new equations predicted which key the birds would peck.

The updated model fit the data from each experiment almost perfectly. It tracked how often birds chose the short or long key across different food-payoff ratios.

But when they took the numbers from Experiment 1 and tried to forecast Experiment 2, the fit fell apart. The reason: small, unseen differences in where food was delivered changed the birds’ timing behavior.

The result lines up with Campbell (2003), who showed that extreme payoff ratios make pigeons ignore the sample and lock onto position. Both papers warn that reinforcer location can quietly warp matching performance.

It also echoes Collier et al. (1986): when stimulus location drifts, some discriminations break down while others survive. Davison’s team now adds ‘timing’ to the list of features that can collapse if reinforcer placement shifts.

Finally, the failure to predict across setups looks like the context effects seen in Ferrari et al. (1991). Just as pigeons treat the same odor cue differently in different chambers, they treat the same time cue differently when food moves.

If you teach conditional discriminations—whether colors, shapes, or durations—lock down reinforcer location early and keep it fixed across sessions. A tiny move of the treat bag or the token dispenser can undo the stimulus control you just built. Before you brag that a kid has ‘mastered’ a temporal discrimination, run a quick probe with the reward in a new spot; if accuracy drops, you’ve found hidden control by place, not time.