Nicholas Mackintosh's 1975 attentional model proposes that organisms selectively attend to the most reliable predictors of reinforcement. The associability (α) of a stimulus increases when that stimulus is a better predictor of the outcome than other available cues, and decreases when it is a poorer predictor. This mechanism explains phenomena where animals learn to attend to relevant dimensions and ignore irrelevant ones.
The Attentional Rule
Δαᵢ > 0 if |λ − Vᵢ| < |λ − V_other| (i is the better predictor)
Δαᵢ < 0 if |λ − Vᵢ| ≥ |λ − V_other| (i is the worse predictor)
Empirical Support
The model explains intradimensional/extradimensional shift effects: after learning to discriminate on one dimension (e.g., color), organisms are faster to learn a new discrimination on the same dimension (ID shift) than on a different dimension (ED shift), because attention to the relevant dimension has been increased. It also explains the inverse base-rate effect and blocking when understood through the lens of selective attention to the most predictive cue.
The contrast with Pearce-Hall is instructive: Mackintosh says "attend to what predicts well," while Pearce-Hall says "attend to what is surprising." Modern evidence suggests both mechanisms exist, potentially mediated by different neural systems (dopaminergic for Mackintosh, cholinergic for Pearce-Hall).