Back to search

University of Washington

Models of human causal learning: review, synthesis, generalization. (A long argument for a short rule)

Abstract

dc:description.abstract

This dissertation is composed of three major components. The first reviews models of causal learning with special emphasis given to Bayesian approaches. The second component joins algorithmic and computational models by defining the free parameters of the former in terms of the theoretical constructs of the latter. Specifically, the weighted ΔP model can be naturally expressed as an estimator of Cheng’s (1997) causal power. This allows for a computational analysis of weighted ΔP that results in a number of insights. The analysis suggests that previous formulations of preventive weighted ΔP have been misspecified. With the correct specification, weighted ΔP is shown to be the best fitting model when entered in to Perales and Shanks (2007) model competition study. The analysis also facilitates a novel derivation of a more general Rescorla-Wagner model that attains a causal power equilibrium. Weighted ΔP is non-Bayesian, though it shares some characteristics with Bayesian estimators. Like the posterior mean, weighted ΔP can be interpreted as a compromise between a prior expectation and sample information. As such, it is also a low variance estimator of causal power. In contrast to Bayesian models, weighted ΔP predicts deterministic strengths of 0 or 1 in certain experimental conditions. Experimental results support these predictions and lead to the discovery of a “deterministic bias” in causal judgments. This phenomenon is strongly inconsistent with Bayesian models, though it also poses problems for point prediction models more broadly. The third component of the dissertation proposes capacity and response probability (CARP), a latent variable framework for models of causal inference. Under CARP, causes are associated with latent capacities. Conjoined causes are assumed to combine additively in their capacities. A response function maps capacity to the judged probability of the effect. Different response functions imply different models of causal judgment. After establishing the framework, response functions are derived for the ΔP rule and causal power, and a number of additional applications of CARP are proposed.

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Beam, Colin Stuart
Advisor dc:contributor.advisor
  • Miyamoto, John M

Subjects

dc:subject × 7

Rights

dc:rights
Statement dc:rights
  • none
Language dc:language.iso
en_US

Identifiers

dc:identifier.*
Handle dc:identifier.uri
http://hdl.handle.net/1773/38679

Chain of custody

source
Harvested from
University of Washington
Base URL
digital.lib.washington.edu/server/oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Beam, Colin Stuart. Models of human causal learning: review, synthesis, generalization. (A long argument for a short rule). 2017. http://hdl.handle.net/1773/38679