Wang, Jane Yueguang2011-07-132022-10-272011-07-132022-10-2720012001https://ir.wgtn.ac.nz/handle/123456789/25312The Rescorla - Wagner model is a well-established model that has been used for many years to study the phenomena and the underling principles of human and animal learning in conditioning experiments. However, when this model is applied to fit data from a particular experiment, namely a 2x2 contingency judgment experiment, the result is far from satisfactory. This problem has appeared in many previous studies including the recent studies by Collins which motivated this work. Our aim, in this thesis, is to develop a simple stochastic model which performs better than the Rescorla - Wagner model in fitting the data from the contingency judgment experiments. Firstly, we point out that the Rescorla - Wagner model is linear and deterministic. It does not model subject-to-subject variation or task to task variation within subjects. We show that this variation is considerable and thus point out the need for stochastic models in this area. Secondly, we note that the parameters in the Rescorla-Wagner model are non-unique. Hence the parameters do not have a real meaning and care must be taken in their interpretation. Thirdly, after considerable data analysis we have developed a simple stochastic model based on the Rescorla - Wagner approach. We show how the model can be fitted using maximum likelihood estimation and how hypothesis tests can be constructed to compare various versions of the model. This stochastic model is novel in the field of psychology. We hope that it will provide some useful insight into further study of human conditioning learning. We note that its use has already supported via hypothesis tests certain psychological theories in this area.pdfen-NZStochastic processesClassical conditioningText