Robert Rescorla

Introduction

Robert Arthur Rescorla (May 9, 1940 – March 24, 2020) was an American experimental psychologist who rebuilt the theory of Pavlovian conditioning around a single claim: organisms learn about the predictive relation between events, not merely that two events happened close together in time. Working mainly with rats and pigeons, he showed that contiguity is not sufficient for learning and that a cue must carry information about what is about to happen. With Allan R. Wagner he formalised that claim in 1972 as the Rescorla-Wagner model, a compact prediction-error equation that accounted for blocking, overshadowing and conditioned inhibition in a few lines of algebra.

The model became the most influential formal theory in the history of learning research and outlived its own limitations. Its central quantity — the gap between what an organism expects and what actually occurs — turned out to be signalled by midbrain dopamine neurons and to be the same quantity that drives temporal-difference learning in modern reinforcement learning. Rescorla himself spent the last two decades of his career documenting the phenomena his own model could not handle, and remained, in the judgement of many colleagues, the most exacting experimentalist the field produced.

Early Life and Education

Rescorla was born on May 9, 1940, in Pittsburgh, Pennsylvania, to Mildred Jenkins Rescorla and Arthur R. Rescorla. His father worked in the oil business and the family moved repeatedly while he was a child, eventually settling in Westfield, New Jersey, where he completed high school. (Some reference works list Philadelphia as his birthplace; the National Academy of Sciences biographical memoir and university records give Pittsburgh.)

He entered Swarthmore College in 1958 and graduated in 1962 with a B.A. in psychology, with minors in philosophy and mathematics, and with highest honours. The Swarthmore psychology department was unusual for its time. Rescorla ran experiments on monkeys with Henry Gleitman, a student of Edward C. Tolman, worked as a research assistant to the gestalt psychologist Solomon Asch on human learning, and studied perception with Hans Wallach. Gleitman and Asch both later moved to the University of Pennsylvania.

That training mattered. Arriving at Penn in 1962 as a graduate student, Rescorla was already fluent in cognitive and purposive alternatives to the stimulus-response behaviourism that still dominated animal learning. The Penn programme was run as a research apprenticeship rather than a course sequence: a proseminar taught by Richard L. Solomon in learning, Eliot Stellar and Philip Teitelbaum in physiological psychology, and Leo Hurvich and Dorothea Jameson in vision, plus a quantitative course in which Robert Bush, David Krantz, Saul Sternberg and Duncan Luce tried to teach probability theory and calculus to a dozen students. Rescorla received his Ph.D. from Penn in 1966 under Solomon's supervision — not from Yale, as is sometimes reported; Yale was where he went to teach.

Academic and Professional Career

Rescorla joined Yale University as an assistant professor in 1966 and rose to full professor before leaving in 1981. The Yale years were the most productive of his career. There he began the collaboration with Allan R. Wagner that produced the model, and there he published the papers on conditioned inhibition and on control procedures that changed how the field designed experiments. He was elected to the Society of Experimental Psychologists in 1975.

He returned to Penn as professor of psychology in 1981 and stayed until his retirement in 2009, when he became professor emeritus. At Penn he held two successive endowed chairs — the James M. Skinner Professorship of Science and, from 2000, the Christopher H. Browne Distinguished Professorship of Psychology — and served as chair of the psychology department and, from the mid-1990s, as Dean of the College of Arts and Sciences (Penn Almanac obituary, 2020). Colleagues noted that he took the deanship seriously enough to keep teaching and running a lab through it.

His honours included a Guggenheim Fellowship (1984), election to the National Academy of Sciences (1985) and the American Academy of Arts and Sciences (2008), the American Psychological Association's Award for Distinguished Scientific Contributions (1986), the presidency of the Eastern Psychological Association, a William James Fellowship of the American Psychological Society (1989), the Howard Crosby Warren Medal of the Society of Experimental Psychologists (1991), Penn's Ira Abrams Distinguished Teaching Award (1999), the Horsley Gantt Award of the Pavlovian Society (2005) and an honorary doctorate from Ghent University (2006).

Rescorla died in Austin, Texas, on March 24, 2020, at the age of 79, following complications from a fall.

Major Contributions to Psychology

Conditioning requires information, not just contiguity

The standard account inherited from Ivan Pavlov held that a neutral stimulus becomes a conditioned stimulus by being paired with an unconditioned one. Rescorla's 1968 experiment attacked the sufficiency of pairing directly. Two groups of rats received exactly the same number of tone-shock pairings; for one group, shocks also occurred in the absence of the tone. Only the group for which the tone predicted the shock developed a strong fear response. Contiguity had been matched; contingency had not.

In the same period he argued that the field had been using the wrong control groups. A "random control" in which the unconditioned stimulus is simply omitted does not reproduce the background conditions of the paired group, because the paired animal experiences a higher overall rate of reinforcement. Rescorla's 1967 paper on proper control procedures became one of the most-cited methodological articles in the field, and a generation of learning researchers rebuilt their designs around the "truly random" control it prescribed.

Conditioned inhibition

Rescorla developed rigorous tests for inhibition — cues that signal the absence of an event — distinguishing genuine learning of a negative relation from mere inattention or performance failure. His 1969 review established the retardation and summation tests that remain standard. This work reversed the earlier assumption that inhibitory and excitatory strength are opposite ends of one scale and set up a problem the Rescorla-Wagner model handled only imperfectly.

The Rescorla-Wagner model

Published as a chapter in a 1972 edited volume, the model states that the change in associative strength of a cue on a trial is proportional to the difference between the maximum supportable strength and the total strength of all cues present:

ΔV = α · β · (λ − ΣV)

Learning therefore stops when the outcome is fully predicted. This single device explains why a redundant cue added to an already informative one acquires no strength (blocking), why two simultaneous cues each acquire less than either would alone (overshadowing), why surprising outcomes drive more learning than expected ones, and why over-training two good predictors and then presenting them together produces a drop in responding (overexpectation — a prediction the model made before anyone had looked for it).

Two-process learning theory

With his doctoral adviser Richard Solomon, Rescorla proposed in 1967 that Pavlovian and instrumental conditioning are distinct systems that interact: classical conditioning supplies the emotional and motivational states, while instrumental learning operates on the responses those states make probable. The framework guided decades of work on avoidance and on Pavlovian-instrumental transfer.

Later work

From the 1980s onwards Rescorla devoted much of his programme to phenomena his own model could not accommodate: learning about events that never occurred, the conditions under which a cue is retrospectively revalued after training has ended, hierarchical and "occasion-setting" relations in which a cue modulates the meaning of another rather than signalling the outcome directly, and the conditions under which associations are symmetrical. His 1988 American Psychologist article, "Pavlovian conditioning: It's not what you think it is," is the clearest statement of what he believed the research actually showed.

Major Publications

  • Rescorla, R. A. (1967). Pavlovian conditioning and its proper control procedures. Psychological Review, 74, 71–80.
  • Rescorla, R. A. (1968). Probability of shock in the presence and absence of CS in fear conditioning. Journal of Comparative and Physiological Psychology, 66, 1–5.
  • Rescorla, R. A. (1969). Pavlovian conditioned inhibition. Psychological Bulletin, 72, 77–94.
  • Rescorla, R. A., & Solomon, R. L. (1967). Two-process learning theory: Relationships between Pavlovian conditioning and instrumental learning. Psychological Review, 74, 151–182.
  • Rescorla, R. A., & Wagner, A. R. (1972). A theory of Pavlovian conditioning: Variations in the effectiveness of reinforcement and nonreinforcement. In A. H. Black & W. F. Prokasy (Eds.), Classical Conditioning II: Current Research and Theory (pp. 64–99). Appleton-Century-Crofts.
  • Rescorla, R. A. (1988). Pavlovian conditioning: It's not what you think it is. American Psychologist, 43, 151–160.

Impact and Legacy

Rescorla's influence runs in two directions that were not obviously compatible. Inside animal learning, he made experimental design rigorous: after his control-procedure papers, a conditioning experiment without proper controls was not publishable. Outside it, the prediction-error term of his equation supplied the language for a finding he did not make — that phasic firing of dopamine neurons encodes a reward prediction error (Schultz, Dayan and Montague, 1997) — and, through temporal-difference learning, became a foundational idea in computational reinforcement learning. His own university's obituary also drew the line from his work to cognitive-behavioural therapy, the principal modern form of psychotherapy, developed at Penn by Aaron Beck.

The model's limitations are equally well documented. Miller, Barnet and Grahame's 1995 assessment in Psychological Bulletin catalogued the phenomena it cannot handle. Spontaneous recovery and reinstatement after extinction show that extinguished responding returns, which a model in which extinction simply drives strength toward zero should not permit. Latent inhibition — retarded conditioning after pre-exposure to the cue — is unpredicted, because the model gives no role to cue exposure without an outcome. Higher-order conditioning, in which a cue paired with an already-conditioned cue acquires strength, is outside its scope. Repeated presentation of a conditioned inhibitor alone should, by the model, weaken that inhibition; in practice it often strengthens it. Reacquisition after extinction is faster than initial acquisition, contrary to the assumption that only current associative strength matters. Backward blocking, in which later training with one element of a compound retrospectively deflates a second element that was never presented again, requires the van Hamme and Wasserman revision that gives absent cues a negative learning parameter.

Two structural objections matter most. First, the model treats a compound as the sum of its elements, so it fails feature-positive and negative-patterning discriminations unless it is supplemented with configural cues — and, as critics noted, it offers no principled rule for when such cues should be invoked, so assuming them always leads to combinatorial explosion. Second, it is a model of a single associative scalar and says nothing about timing, about response form, or about the distinction between what has been learned and what is expressed. Later theorists such as Gallistel and Gibbon argued that what animals encode is closer to rates and intervals than to trial-by-trial increments.

Rescorla was unusually candid about all of this. His own later papers established several of the counterexamples, and he consistently insisted that the model was valuable because it made precise, checkable, and sometimes wrong predictions — which is a different argument from the one usually made on its behalf. Successor models (Pearce-Hall, Mackintosh, and the comparator and configural theories) were built directly on the phenomena his work exposed.

Learn More

References

  • Rescorla, R. A., & Wagner, A. R. (1972). A theory of Pavlovian conditioning: Variations in the effectiveness of reinforcement and nonreinforcement. In A. H. Black & W. F. Prokasy (Eds.), Classical Conditioning II (pp. 64-99). Appleton-Century-Crofts.

Last updated: 2026-10-20

Menu
Loading