Sequential expectancies and decision making in a changing environment: an electrophysiological approach.

This experiment addressed two questions. First, we sought to determine how sequential expectancies, as indexed by the amplitude of P300, are formulated when subjects are presented with a series in which the event probabilities are continuously changing. Second, we wanted to determine the effect on P300 amplitude of seeking information prior to a decision when the decisions are required at random intervals. Eight subjects were presented with a Bernoulli series of high and low tones in which the event probabilities reversed at random intervals. Thus, while the event probabilities were either .33 or .67 over short segments of the series, they were .50 over each trial block. The subject's task was to count the number of low-pitched tones. In two separate conditions ("Unknown" and "Known") the subjects received different information about the series. In the Unknown condition, the subjects were not informed of the reversals in stimulus probability and were only told that the events were equally probable. In the Known condition, the subjects were informed of the probability reversals and, in addition to the counting task, were asked to detect the probability reversals. The trials were divided into two categories: "transition" and "stable." The transition trials were those that occurred between the actual probability reversal and the subject's reported detections. The stable trials were all those not included in the transition category. The data from the stable Irials revealed that, in most cases, subjects generated their expectancies, as indexed by the amplitude of P3(H), in an identical manner during both conditions. Accurate knowledge of the stimulus probabilities did, however, influence the amplitude of P300 for events that were preceded by sequences of stimuli that occur only rarely at each level of probability. These data suggest that the mechanism responsible for generating sequential expectancies is largely automatic. Certain aspects of this mechanism are, however, modulated by the subject's knowledge of the parameters of the series. The data from the transition trials revealed that, in the Unknown condition, the subjects'assignments of subjective probabilities, as indexed by the amplitude of P300, adapted rapidly to meet the new sequence-generating rules following the probability transitions. In contrast, during the Known condition, the amplitude of the P300and Slow Wave components .steadily increased, as the decision point approached, to levels well above those in the Unknown condition. By the first trial following the decision, this enhancement in F300 and Slow Wave activity was no longer evident. These results are discussed in terms ofa descriptive model of the subjects' decision making behavior. DESCRIPTORS: Event-related potentials, P300, Slow Wave, Subjective probability. Decision making. This paper is based on a portion of the research presented vai Research Contract N(XK)14-76-(HX)2 witb funds provided by Ray Johnson, Jr. to the Graduate College of the Univerby the Defense Advanced Research Projects Agency. The sity of Illinois in partial fulliltment of the requiretnents for Ihc authors wish (o acknowledge tbe helpful comments of CbrisPh,D, degree, A preliminary report on a portion of tbcsc topher D, Wickens. Connie C, Duncan-Johnson, John Podata was presented al the Eighteenth Annual Meeting of lich,andGregory Chesney onancarlierversionofthe manuthe Society for Psychopbysiological Research, Philadelphia, seript, 1978. Another part of the data reported here was presented Address requests for reprints to: Ray Jobnson, Jr., Naat the Fiftb International Symposiutn on Electrical Potentials tional Institutes of Healtb, NINCDS/Clinical Neurosciences related to Motivation, Motor, and Sensory Processes of the Braneh, Building 10, Room 4N246, Bethesda, Maryland Brain. Ulm, West Germany, 1979. 20205. Tills research has been supported by the Office of Na183 (K)48-5772/82/()2ll 183-18$ 1.80/0 © 1982 The Society for Psychophysiological Research, Inc. 184 Johnson and Donchin Vot. 19. No. 2 The amplitude of the P300 component of the event-related brain potential (ERP) is determined., at least in part, by the subjective probability that a subject assigns to each event in a random sequence: the lower the subjective probability, the larger the P300. While the objective probabilities of events may remain constant over a series, the individual's subjective probabilities, and with them the amplitude of the P300 component, vary predictably from trial to trial (Duncan-Johnson & Donchin, 1977; Johnson & Donchin, 1980; Tueting, Sutton, & Zubin, 1970). On the basis of such data. Squires, Wickens, Squires, and Donchin (1976) suggested that these variations in P3(K) amplitude are related to the subjective probability assigned to each event, according to the preceding sequence of stimuli. In this report, we extend the analysis of the relation between P300 amplitude and subjective probability to sequences in which the event probabilities do not remain constant over time. We were particularly interested in observing the assignment of subjective probability, as indexed by the amplitude of P300, during a dynamically changing series of events. We used Bernoulli sequences in which the probabilities of the two events were reversed at random intervals. Thus, whereas the event probabilities over the entire series (referred to as the "global" probability) were .50, the event probabilities over short segments of the series were either .33 or .67 (referred to as the "local" probabilities). In one condition ("Unknown"), the subjects were told that the event probabilities were .50. They were not informed about the short-term reversals in probability. In a second condition ("Known"), the subjects were presented with the identical series. They were now informed of the actual local event probabilities and were further told that these probabilities would reverse. Moreover, in the latter condition, they were instructed to identify the points in the sequence where the probabilities reversed. Thus, the first series served as a yoked control for the second. In this way, we could determine not only the extent to which knowledge about event probabilities affects the pattern of P300 amplitude, but also the degree to which the need to assess explicit changes in probability affected the ERPs. It is convenient to distinguish between the "transition" and the "stable" phases of the series. The "transition" phase includes trials that occurred between the actual changes in the series-generating rule and the subject's detections of these changes. The remaining trials are defined as the "stable" phase. The data obtained during the stable phase bear on one feature of the relation between P300 amplitude and sequential expectancies that has remained somewhat vague: namely, the extent to which the subject consciously assigns subjective probabilities to events. Some of the terms used by Donchin and his associates may be construed to imply awareness by the subject of the sequential expectancies. Terms such as "expectancy" may suggest that the processes leading to variations in P30() amplitude are conscious. Yet, this need not be the case and, in fact, subjects frequently disclaim awareness of the rules governing the generation of the series. There is, however, clear evidence that subjects do use explicit information about the prior probabilities of events in developing their estimates of sequential expectancies. Reaction time (RT) studies have shown that the presence of an additional source of information about the series (LaBerge, Van Gelder, & Yellot, 1970) or variations in instructions (Kirby, 1976) can affect the RTs obtained from identical stimulus sequences. Similar results can be found in the P300 literature. DuncanJohnson (1978), for example, has shown that when subjects are given an explicit source of information about the identity of future events, trial-to-trial variations in P3no amplitude are abolished. P300 amplitude, however, was still found to be an inverse function of the conditional probability of the events (cf. Tueting et al., 1970). The degree to which knowledge about objective event probabilities affects the amplitude of the P300 component and, by inference, the subjective probabilities for different stimuli could be assessed from the data obtained during the stable phase. If the data recorded in the Known condition display the same pattern of results as the data recorded in the Unknown condition, then we could conclude that the objective probabilities that affect the pattern of the P300 need not be explicitly known to the subject. The data from the transition phases in the same experiment allow us to address two different questions. First, we can assess the manner in which the amplitude of P300 adapts to changes in the series. Second, we can determine if there are differences in the ERP waveforms that are elicited just prior to the detection during the Known and the Unknown conditions. Since the identical sequence of events was used in both conditions, any differences between the P3OOs elicited in the two conditions can be attributed to the subject's attempt to detect the probability reversals in the Known condition. Other investigators have reported that P300 amplitude is increased during information-seeking phases of problem-solving (Wilson, Harter, & Wells, 1973) and concept-formation tasks (Stuss & Mardi. Sequential Expectancies, Decision Making, and P300 185 Picton, 1978). On this basis one could predict that P3(X) would be larger when the subjects were seeking to detect a change in the series. There are, however, a number of important differences between the studies cited and the present experiment. Because of the nature of their tasks, previous investigators could not assess trial-by-trial changes in P300 amplitude. More important, in previous studies, subjects had no way of indicating exactly when they had solved the problems. Thus, the increases in P300 amplitude that ostensibly occurred during evidence accumulation could have resulted from other factors, such as changes in the assignment of subjective probabilities for events. Therefore, the relation between the amplitude of P300 and information-seeking prior to a decision remains unclear. Our data provide an estimate of this relation.