Associative change to cues conditioned in compound

Models of associative learning that rely on a common error term to determine associative change, such as the Rescorla-Wagner model, assert that all cues on a conditioning trial undergo equivalent amounts of change. Recent evidence from Rescorla (2000, 2001) himself, as well as Spicer and colleagues (2020) have been interpreted to be incompatible with this fundamental assertion. This thesis evaluates both sources of evidence and, thereby, the challenge they present to the Rescorla-Wagner model. Chapter 2 provides an assessment of Rescorla's (2000, 2001) compound test data, his interpretation of unequal change to cues conditioned in compound, and an alternative explanation for the results that preserves the key insight of the Rescorla-Wagner model by appealing to non-linearity in the function that maps (or translates) learning into observable behaviour. Chapter 3 evaluates Spicer et al.'s (2020) findings which have been taken to indicate that prediction certainty rather than prediction error is the greater determinant of associative change in people. These results too can be accounted for under the mapping function introduced in Chapter 2. Finally, Chapter 4 presents a series of experiments that examine the effect of certainty/uncertainty on associative change in human causal learning using designs that address the interpretive issues raised in Chapter 3. These experiments use a compound test procedure in a continuous outcome allergist task and provide evidence for greater change to more uncertain cues in the case of associative gains, but not associative losses. The results are discussed with respect to the Rescorla-Wagner model, Rescorla’s proposed modification to the model, and the Spicer et al. notion of “theory protection” which is beyond the explanatory powers of the model