|
| International Paper Session - Conceptual Issues in the Experimental Analysis of Behaviour |
| Tuesday, August 14, 2007 |
| 1:00 PM–1:50 PM |
| L2 Room 2 |
| Area: EAB |
| Chair: M. Jackson Marr (Georgia Institute of Technology) |
| |
| A Mathematical Theory of Behavior Chains. |
| Domain: Experimental Analysis |
| JACK J. MCDOWELL (Emory University) |
| |
| Abstract: Two mathematical and two computational theories can be combined to produce a more comprehensive theory of adaptive behavior that describes reinforced responding in the presence of a discriminative stimulus. The cornerstone of this theory is an evolutionary algorithm that instantiates the idea that behavior evolves in response to selection pressure from the environment in the form of reinforcement. The evolutionary algorithm has been shown to generate steady-state behavior that is consistent with matching theory. This evolutionary dynamics and matching-theory statics can be combined with a theory of stimulus control to produce a more general theory. Rescorla and Wagner’s computational theory of associative learning may be used to describe the development of conditioned reinforcing value in a discriminative stimulus. An extension of Mazur’s hyperbolic delay theory of conditioned reinforcement constitutes the corresponding equilibrium theory. The resulting general theory of reinforced responding in the presence of a discriminative stimulus provides a quantitative account of the development, maintenance, and dissolution of behavior chains, as is observed on chained schedules of reinforcement. It also provides a biologically plausible method of generating adaptive state action sequences in a virtual agent or physical robot, which is a canonical problem in artificial intelligence and machine learning. |
| |
| Residual Meta-Analysis: A New Frontier in the Quantitative Analysis of Behavior? |
| Domain: Experimental Analysis |
| RANDOLPH C. GRACE (University of Canterbury) |
| |
| Abstract: Meta-analysis has become increasingly popular among researchers in the social sciences as a method of aggregating effect sizes across studies that overcomes some of the problems with statistical power. However, it has generally not been used in the quantitative analysis of behavior. The reason is simple. Our models are so good that effect size is not the issue. We face essentially the opposite problem: How to distinguish between models that typically account for over 90% of the variance. The critical question is then whether the remaining 10% variance is systematic, allowing for models to be falsified. But typically this cannot be answered by a single study, because of low statistical power. Here I describe a technique for aggregating residuals from fits of competing models to archival data, and using polynomial regression to test whether they are systematic. Residual meta-analysis solves the problem of low statistical power, and provides a stringent test of quantitative models. Several examples are presented, involving models for choice in concurrent chains and concurrent schedules, and temporal discounting of delayed rewards. |
| |
| Tweedledum and Tweedledee: Symmetry in Behavior Analysis. |
| Domain: Experimental Analysis |
| M. JACKSON MARR (Georgia Institute of Technology) |
| |
| Abstract: Symmetry is revealed when some non-trivial transformation of a system leaves the system unchanged or invariant. Symmetry is a pervasive feature in many sciences from physics to embryology. In physics, for example, symmetry is reflected in fundamental laws. Can this be said of behavioral principles? I explore this question by discussing several examples from behavior analysis including the operational and functional aspects reinforcement, stimulus control, the three-term contingency, and putative scale invariance of behavioral principles. |
| |
|
| |