The value of signals for reward: Choice in concurrent chains
Degree GrantorUniversity of Canterbury
Degree NameDoctor of Philosophy
A number of studies indicate that a signalling effect occurs when stimuli are presented during the terminal link of a concurrent-chains procedure, which signal whether or not probabilistic reinforcement is forthcoming at the end of the trial. The effect of these signals is a reduced preference for the richer alternative. This thesis includes five experiments aimed to investigate this. The overall purpose of this research is to investigate the effect on preference of differential signalling of reinforcement that differs in terms of the variable being signalled. Furthermore, these experiments are aimed at investigating the manner in which signalling affects preference when multiple dimensions of reinforcement are varied in a signalling procedure. Asking the question how does signalling affect sensitivity to various dimensions of reinforcement and how does this change when more than one dimension is varied? The first experiment involved a simple replication of previous work with percentage-reinforcement procedures. It further extends that work by signalling the size of the reinforcer rather than its probability. Pigeons were given the choice between two alternatives with identical-duration initial and terminal links. Choice of one alternative (uncertain) led to one of two possible signalled outcomes, reinforcement (3.5 s) or blackout, with a probability of .5. Choice of the other alternative (certain) always resulted in food (3.5 s) at the end of the delay. When the scheduled outcome on the uncertain alternative was differentially signalled, there was a moderate preference for the certain side. Moreover, when the scheduled outcome on the uncertain side was not differentially signalled, preference for the certain side became more extreme. This result replicates the signalling effect. With a similar design, another three conditions manipulated the magnitude of primary reinforcement, rather than its probability. When primary reinforcement was arranged like this, a weaker signalling effect was obtained. Two further experiments investigated the effect of signals on preference for alternatives that differed in either relative variability or expected magnitude of reinforcement, while the other variable was held constant. In Experiment 2, pigeons were offered the choice between two alternatives that differed in relative variability of reinforcement. Each alternative delivered one of two possible reinforcer magnitudes at the end of a terminal link: these magnitudes were either the same (fixed) or were different (mixed). In some conditions, terminal-link outcomes were signalled and in others they were not. Results showed that pigeons preferred fixed over signalled mixed magnitudes of reinforcement, and signalled mixed over unsignalled mixed magnitudes of reinforcement. Thus, these alternatives could be ordered in terms of preference: fixed, signalled mixed and unsignalled mixed. Finally, signalling does indeed moderate the preference for fixed over mixed magnitudes. In the third experiment, pigeons completed a two-component signalled concurrent-chains procedure in which the relative expected magnitude varied across alternatives as all other dimensions of reinforcement were held constant. Therefore, this experiment investigated the effects of signalling on preference between alternatives differing in relative expected magnitude of reinforcement. A slight signalling effect was obtained, with slightly reduced preference in the signalled relative to the unsignalled condition. Overall, these two experiments confirm that signalling acts on the preference for fixed magnitudes of reinforcement and to a lesser extent, preference for the larger of two expected magnitudes of reinforcement. The final two experiments investigated the effect of signals on preference in a procedure that varied parametrically either relative probability and immediacy of reinforcement, or probability and magnitude of reinforcement. The effects of relative reinforcer immediacy or magnitude and probability on choice in concurrent chains were examined, under conditions in which terminal-link outcomes (reinforcement or extinction) were either signalled or unsignalled. Pigeons responded in a three-component concurrent-chains procedure with either independent or interdependent initial links. The percentage of reinforcement was varied across conditions, while the immediacy or magnitude of reinforcement was varied across components. Both signalled and unsignalled conditions were arranged. Generalised-matching analyses revealed a strong signalling effect: sensitivity to relative reinforcer probability was greater in the unsignalled compared to signalled conditions. However, sensitivities to relative immediacy and magnitude were also greater in the unsignalled conditions. Overall, the data suggest that signalling reinforcement and extinction outcomes may attenuate sensitivity to all terminal-link variables, not just reinforcer probability. These experiments have investigated the signalling effect using a variety of methods. They have contributed several replications of the effect, and added valuable information regarding the effects on preference of signalling reinforcement outcomes. The most impressive finding of this research is that signalling has a global effect on sensitivity to all dimensions of reinforcement and that models of conditioned reinforcement are best suited to analyse these results.