Whole document restricted, see Access Instructions file below for details of how to access the print copy. / The two experiments conducted were designed to investigate behavioural models of signal-detection performance. The detection environment was made difficult, or confusing, for the pigeon subjects by either varying the number of stimuli presented or by varying the disparity of the stimuli. These experiments provided thorough data sets from which the adequacies of current behavioural-detection models were able to be evaluated. The models evaluated were developed from two different areas of research: One was based on the generalised matching law and the others were based on the confusion between stimuli. One of the most fundamental assumptions of these models is that their parameters are assumed to be independent of each other, i.e., parameter invariant. When a variable known to affect measures of stimulus discriminability is manipulated, estimates of stimulus discriminability only are predicted to vary. Similarly, when a variable known to affect sensitivity to reinforcer frequency (or response-reinforcer discriminability) is manipulated only this parameter is predicted to change. The experiments conducted aimed to determine whether this assumption was true not just theoretically, but also empirically. The models were assessed in terms of whether their assumptions were met, whether the parameters estimated had sensible relations to the experimentally arranged stimulus disparities, and whether they provided a good description of the data. The first experiment investigated the effects on matching-to-sample performance of increasing the sample stimulus size pool from two to four. Of interest was whether stimulus discriminability between any pair of stimuli would remain constant when a third or fourth sample was added. Six pigeons were trained on a symbolic matching-to-sample task involving the discrimination of various intensities of yellow light. The difference in the arranged physical "intensity" between each successive sample stimulus in the set was 1 Volt. The numbers of sample stimuli presented were varied between two (Part l), three (Parts 2 and 4), and four (Part 3). Across conditions, the ratios of reinforcers obtainable for correct matching responses were varied over at least five levels, allowing behavioural models of signal-detection to be fitted to the data. In Part 5, the numbers of sample stimuli were varied between two, three, and four and the reinforcer ratio between consecutive remaining samples was 2:1. All response alternatives remained available in the choice phase and as a result once a sample had been presented, subjects continued to emit error responses to stimuli that were no longer presented. Current behavioural models of detection were used to investigate the effects of varying the number of sample stimuli on matching performance. The analyses from the generalised-matching-law based model indicated that this model was unable to deal with the data effectively. In order for the analyses to be conducted, pairwise combinations of the sample stimuli had to be made. Analyses using only the two errors relevant to the pairwise combination of the stimuli indicated that adding or removing a sample stimulus did not affect the ability of the subjects to discriminate between the remaining samples, following the principle of indifference from irrelevant alternatives. Estimates of stimulus discriminability followed the order of the arranged physical disparity between the stimuli. An interaction between stimulus discriminability and sensitivity to changes in the frequency of reinforcement was found when there should theoretically have been independence between these two measures. The rate of reinforcement on another alternative caused fluctuations in the behaviour ratios between a discriminable pair of stimuli, violating the principle of indifference. Stimulus discriminability, however, remained unchanged. However, by using only the two errors relevant to a pair of stimuli, a large amount of behaviour for other error types was omitted. The alternative, using all errors, provided even poorer data fits to the model and the number of errors overwhelmed the ratio of correct to incorrect behaviour resulting in insensible parameter values being estimated. The results from more recent, confusion-based, detection models indicated that these models were able to effectively deal with and describe these data. No attempt was made to degrade the disparity of the choice-reinforcer contingency and estimates of response-reinforcer discriminability were high and constant throughout the experiment. Estimates of stimulus-response discriminability from these later models also decreased as the arranged physical disparity between the sample stimuli decreased. In accordance with the theory, stimulus-response discriminability, was independent of response-reinforcer discriminabiliry, preserving parameter invariance. The extended confusion-based model provided an easy and effective numerical analysis of detection tasks in which n-stimuli and m-choices are arranged. The discussion raised the issue of why subjects should continue to emit error responses to stimuli that were no longer or had never been presented as samples. It also commented on how parameter estimates for the 4 × 4 matrix tended to be smaller due to the finite bound of confusion probabilities. Overall, a version of the principle of indifference from irrelevant alternatives was not supported by this study. In the second experiment, six pigeons were trained on a standard 2-stimulus,2- response symbolic matching-to-sample task involving the discrimination of sample and choice stimuli varying in density (numerosity). The disparities of the sample and choice stimuli were varied systematically across experimental parts, allowing sample and choice confusion to be investigated independently. Degradation of the stimuli was carried out by manipulations of the probabilities of pixels illuminated on the stimulus areas of a computer monitor. Across conditions, the ratios of reinforcers obtainable for matching responses were varied over five levels. Parameter invariance was the main topic of investigation. Current behavioural models of signal-detection were investigated. Again the model based on the generalised matching law provided poor descriptions of the data both empirically and theoretically. Interactions between stimulus discriminability and sensitivity to reinforcement were found, making these parameters unreliable estimates of sample discriminability and response bias. Furthermore, the effects on detection performance of reducing the disparity of the sample or choice stimuli were difficult to evaluate independently as either sort of manipulation had the same effect on stimulus discriminability due to the structure of the model. The more recent behavioural models of detection based on the confusion between the stimulus and response relations provided sound descriptions of the data. Reductions in the disparity of the sample stimuli led to decreases in stimulus-response discriminability and left the response-reinforcer discriminability unchanged. Similarly, reductions in choice-stimulus disparity caused a decrease in estimates of response-reinforcer discriminability and not in stimulus-response discriminability. Therefore, parameter invariance was again obtained and the variables known to affect these parameters (sample and choice stimulus disparity) were correctly identified. The confusion model, therefore, provided a befter description of performance for this experiment. Discussion centred on how previous criticisms of the model in regard to a failure to demonstrate parameter invariance were unsupported by both the original data and the present study. In summary, confusion-based behavioural-detection models were quite adequate in describing the data from more complicated symbolic matching-to-sample tasks than have traditionally been employed. Their theoretical assumptions were supported empirically and high percentages of the data variance were accounted for by the model. As a word of caution, the general discussion concentrated on the mathematics of fitting non-linear models. Importantly, in order to attain systematic replication of results, details of the exact fitting process need to be specified. Differences in fitting procedures caused differences in parameter estimates. Extension of the confusion-based models into other areas of research (e.g., free-operant studies and delayed matching-to-sample tasks) will increase the usefulness of these models. Future directions for detection research were suggested. Keywords: Detection, behavioural models, parameter invariance, matching-to-sample task, disparity, principle of indifference, stimulus discriminability, response-reinforcer discriminability, numerosity, confusion, pigeons.
| Identifer | oai:union.ndltd.org:ADTP/278025 |
| Date | January 1997 |
| Creators | Godfrey, Rebecca |
| Publisher | ResearchSpace@Auckland |
| Source Sets | Australiasian Digital Theses Program |
| Language | English |
| Detected Language | English |
| Rights | Whole document restricted. Items in ResearchSpace are protected by copyright, with all rights reserved, unless otherwise indicated., http://researchspace.auckland.ac.nz/docs/uoa-docs/rights.htm, Copyright: The author |
Page generated in 0.0072 seconds