University of British Columbia
Features of information intergation in causal inference
Abstract
dc:descriptionAn important aspect of causal inference is assessing the contingency between antecedents and outcomes. Research on how people integrate contingency information has focused on identifying the "best" rule to descriptively model the information integration process. In contrast to this ruleanalytic approach, the present feature-analytic approach asks the question, "What features are important in describing the information integration process?" Five key propositions of the present account are that (a) people prefer strategies that involve contrasting data with conflicting implications to strategies that involve seeking only confirmatory or marginal-frequency data, (b) people weigh positive information more heavily than negative information, (c) people are biased toward testing sufficiency rather than necessity, (d) people are biased toward strategies that cohere with the perceived direction of time (input tests) rather than those that violate this perception (outcome tests), and (e) people are biased toward probability strategies that enable comparability across data contexts rather than frequency strategies that do not. In three experiments, subjects received contingency information on two, temporally sequenced, binary variables in numeric summary format. Subjects were asked to rate the direction and magnitude of the causal relation between the two variables based on the contingency information provided. Results of Experiments 1-3, corroborated by a reanalysis of data from two published experiments employing a discrete-trial method for presenting stimuli, strongly supported the first four propositions. To test the fifth proposition, I reanalyzed data from five published experiments in addition to an analysis of data from Experiment 3. Results indicated that within each data context preferences for either frequency, conditional-probability, or joint-probability strategies emerged, but across contexts consistent preferences for one type of combination method was lacking. Taken together, the findings indicate that invariant properties of the information integration process in causal inference can be isolated but these consist of systematic feature preferences rather than stable rankings of rules in terms of their predictive utility.
Degree
thesis:*- Name thesis:degree_name
- Doctor of Philosophy - PhD
- Level thesis:degree_level
- doctoral
- Discipline thesis:degree_discipline
- Psychology
- Grantor dc:publisher
- University of British Columbia
- Year dc:date
- 1996
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Mandel, David R.
Rights
dc:rights- Statement dc:rights
-
- For non-commercial purposes only, such as research, private study and education. Additional conditions apply, see Terms of Use https://open.library.ubc.ca/terms_of_use.
- Language dc:language
- eng
Identifiers
dc:identifier.*- Handle dc:identifier
- http://hdl.handle.net/2429/6201