Validation and Parameter Estimation of a Behaviour-Modification Model
MetadataShow full item record
In this thesis, an attempt at validating and identifying parameters of a quantitative model of attitudes and behaviours is presented. The model, established in earlier work, describes the dynamics of a subject's attitude and behaviour when he or she is offered a sequence of rewards in exchange for producing some desired behaviour. The dynamics of attitude are governed by the theory of planned behaviour, cognitive dissonance theory, and overjustification theory. Validation is performed in two specific cases in which overjustification does not arise. These two cases are defined based on the sign of the attitude. In the first case, the subject's attitude is negative, while in the second case, the subject's attitude is positive and all offers are declined (this additional requirement is necessary to ensure overjustification effects are not present). The parameters to be identified are the cognitive dissonance gain parameter, K₁, sensitivity to a reward, μ₁, and the attitude measurement proportionality constant, μ₂. These parameters are unique for each person. A review of standard parameter estimation approaches concludes that the extended Kalman filter (EKF) is best suited for parameter estimation in this model. Simulations show that in this particular application, the EKF cannot accurately estimate multiple parameters simultaneously; thus, only K₁ is estimated using the EKF. An experiment is designed to produce the required conditions for each case and elicit attitude and behaviour data from human participants at 11 instances. In the initial phase of the experiment, carefully phrased questions are used to estimate μ₁ and μ₂. In the remainder of the experiment, participants are iteratively offered a reward to listen (for a specified duration) to a sound they initially rated as unpleasant or pleasant (depending on whether data is being collected for the negative or positive attitude case). Following their response to an offer, the participants are asked to rate the sound. The details of each offer, the participant's response, and the rating of the sound following each offer are used to validate the model and estimate K₁. The experimental data collected in the first case (negative attitude) shows that the model correctly captures experimental trends in 5 of 7 trials. Further, the EKF's estimates of K₁ were almost always positive and appear to be converging the majority of the time. These two observations support the validity and utility of the model when attitude is negative. The experimental data collected in the second case (positive attitude) is largely unusable. The unusable data is attributed to multiple deficiencies in the design of the experiment. The deficiencies result in the experiment not accurately producing the conditions required to excite the effects of cognitive dissonance when attitude is positive.