An analysis of the interaction of incentive magnitude and relative event frequency in probability learning.

Abstract

Each of 160 Ss made 400 two-choice predictions which consisted of a random sequence of high-and low-risk trials. Eight groups were formed by varying event frequencies under the two risk levels. Choice proportions revealed a strong over-all incentive effect within Ss, whereas between Ss incentive studies have generally failed to show differences. Choice proportions under each incentive level were related to event frequencies under the second incentive level. Subjective estimates of event proportions also showed an incentive effect. Observed conditional statistics were found to be at variance with basic reinforcement assumptions contained in several mathematical models of choice behaviour. RECENT EXPERIMENTS ON INCENTIVE EFFECTS in probability learning haveshown a greater effect of incentive magnitude on choice proportions under a within-subjects design (i.e., each subject has more than one incentive level) than under a between-subjects design (Le., each subject has only one incentive)

Human incentive learning as a function of reinforcement schedule and experimental paradigm.

“…The results contrast with those reported by Lipkin (1968), who employed a similar RA paradigm with concurrent ir and incentive variation. The present data suggest that Lipkin's within-,?…”

“…TT differences. Thus, it seems quite likely that CEs similar to those obtained with differential incentive will occur with differential v. Lipkin (1968) combined v and incentive in a within-,? design where each 5 experienced two levels of w along with two different monetary incentives and obtained the usual within-.?…”

Positive contrast effects as a function of method of incentive presentation.

“…Considering the fact that the number of outcomes favorable to the reinforcement prediction is 2, while the total number of possible rankings is 24, then finding 17 such favorable outcomes of 24 four-tuples would be expected to occur by chance with a probability considerably less than .01. This finding is especially noteworthy with respect to the rankings reported by Schnorr and Myers (1967) and Lipkin (1968), who used very similar methodology and found the reinforcement rank ordering to prevail in surprisingly few instances.…”

Incentive effects in the discrimination of two simultaneously occurring probability distributions.

“…Such observations of negative but not positive contrast are consistent with the findings of a majority of learning experiments involving successive or simultaneous shifts in reward magnitude. Specifically, in-vestigators who varied incentive levels within 5s in one-dimensional PL have typically demonstrated only negative contrast (e.g., Halpern et al, 1968;Lipkin, 1968;Schnorr et al, 1966). Additionally, incentive contrast studies with animals have consistently found negative contrast, and have only rarely demonstrated positive contrast (see review by Black, 1968).…”

“…Negative but not positive contrast was also demonstrated when the payoff 5 received for a correct prediction was varied randomly on each trial (Halpern et al, 1968;Lipkin, 1968;Schnorr et al, 1966). For this within-5 manipulation of incentive, one of two incentive magnitudes was stamped on the back of each card in a shuffled deck (e.g., l£ or Sji), and when predicting which of two stimulus alternatives was on the face of a given card, 5s "risked" the incentive magnitude indicated on the back.…”