Robert A. Rescorla, working at Yale and later the University of Pennsylvania, fundamentally changed how psychologists understand associative learning. His theoretical and empirical work demonstrated that conditioning depends not on mere contiguity (pairing of CS and US) but on contingency -- the informational relationship between stimuli. The Rescorla-Wagner model, developed with Allan Wagner, formalized this insight into one of the most influential mathematical models in all of psychology.
The Rescorla-Wagner Model
V_i = associative strength of CS_i
alpha_i = salience of CS_i
beta = learning rate for the US
lambda = maximum conditioning supportable by the US
Sum V = total associative strength of all present CSs
The model's key insight is that learning is driven by prediction error -- the discrepancy between the expected and actual outcome. When the total associative strength of all present cues matches the outcome (Sum V = lambda), no further learning occurs. This simple principle explains blocking (a pre-trained cue prevents learning about an added cue), conditioned inhibition, and the nonlinear acquisition curves observed in conditioning experiments.
Rescorla's landmark 1968 paper demonstrated that conditioning depends on the contingency between CS and US, not mere contiguity. When the US occurs equally often with and without the CS (zero contingency), no conditioning develops despite numerous CS-US pairings. This finding required a model that computes prediction error relative to a baseline expectation -- exactly what the Rescorla-Wagner model provides.
Beyond the Rescorla-Wagner Model
Rescorla continued to refine understanding of associative learning throughout his career, demonstrating that extinction is not simply unlearning but new learning, that the content of associations is richer than simple stimulus-response links, and that within-compound associations play a critical role in many conditioning phenomena. His experimental rigor and theoretical precision set standards that learning researchers continue to follow.
Legacy and Impact
The Rescorla-Wagner model's prediction-error learning rule anticipated the temporal difference learning algorithm and the discovery that dopamine neurons encode reward prediction errors. The model remains the standard starting point for teaching associative learning and has been extended in numerous directions, including attentional models (Mackintosh, Pearce-Hall) and configural theories (Pearce).