Continued from page 1.

The experimental apparatus consisted of a 1978 Bally's Quarter Slot Machine with a single payoff line that was modified to accept tokens. Figure 1 provides a picture of the machine. This specific slot machine had three reels that would spin when the participant pulled a handle on the right side of the machine. Each reel consisted of 22 symbols, three positions of which were visible to the player via the payout window. Of these visible symbols, only ones aligning on the middle location, or the "payout line" were considered possible winning payouts. Symbol variations that would result in a payout were displayed on the upper section of the slot machine. Reels would stop sequentially from left to right approximately 3 s following a handle pull. If a winning combination of symbols were aligned on the middle payout line, the corresponding number of tokens was dispensed into a payout tray located at the bottom of the machine. All sessions were conducted and videotaped in a small quiet room approximately 10' by 11' containing other gambling devices, microcomputers, and a one-way observation mirror.

Procedure

All participants were told the following instructions before beginning the experiment.

Please play this slot machine one token at a time. Each time you do not win, please estimate the degree to which your losing spin was close to a win using a scale from 1 to 10 with 1 being far from a win, and 10 being as close to winning as possible. Do not play any tokens that might be dispensed from the slot machine on your winning spins. Continue playing until all the tokens I gave you are gone.

An 8'' × 11'' piece of paper was affixed to the wall approximately 6'' above the slot machine with the 10-point scale printed on it in 24-point font. If the participant had no other questions, the experimenter gave the participant 100 tokens (which would correspond to 100 total trials), pressed record on the video camera, and left the room.

Session Analysis

Two dependent measures were later obtained from each participant's videotaped session. First, we recorded the response latency between trials or spins. Latency was defined as the time from the stopping of the reels of the slot machine until the participant inserted the next token to initiate the next trial. Second, we recorded the trial type as a win, a near miss, or a loss. A win was defined as a trial that resulted in three identical symbols (or one "Cherry" symbol) appearing on the payout line, which subsequently resulted in the payout of tokens from the slot machine. A near miss was defined as a trial that resulted in two identical and one different symbols appearing on the payout line, which did not result in the payout of any tokens (the picture of the slot machine in Figure 1 illustrates such a combination). A loss was defined as any trial not meeting either of the above two definitions.

All postsession analysis of participant behavior was conducted by one of the two authors reviewing each participant's videotape. During this review, the experimenter considered latency and trial type information using a computer program requiring a simple mouse click to indicate trial type and start-stop times of each trial. Mark Dixon created this computer program using Microsoft Visual Basic 6.0.

A second observer who was not aware of the purpose of the study also coded the performance of 6 participants, or 50% of the sample size. An agreement was noted for response latency if both observers scored the same trial's latency within .25 s of each other. An agreement was noted for trial type if both observers scored the same trial's type identically. Interobserver agreement was calculated by dividing the number of agreements by the number of agreements plus disagreements and then multiplying by 100. The resulting agreement was 91% for response latency and 100% for trial type.

Statistical analyses were conducted on resulting data solely to support visual inspection of that data. These analyses are somewhat controversial to include in a behavioral analysis (Baron, 1999) and may distract from understanding the behavior of the single-subject (Perone, 1999). We have chosen to include them as supplemental information in an attempt to strengthen the acceptability of our findings to the broader psychological community which may not be as easily persuaded by a descriptive single-subject analysis as suggested by Ator (1999) and Crosbie (1999).

Results

All participants played the slot machine for 100 trials. Distribution of wins, near misses, and losses varied somewhat in density and magnitude across participants because the apparatus was an actual slot machine from a casino. Table 1 displays a detailed description of the contingencies contacted by each participant. In summary, participants contacted losing spins between 45-60% of the time, winning spins 8-21%, and near-miss spins 27-42%. Total payouts ranged from 19 to 93 tokens, while the average payout ranged from 2.32 to 7.50 tokens. While these contingencies did vary across participants, no reinforcement variation produced observable response variations at the level of the single subject.

Regardless of the specifics of the individual contingencies contacted, participants' responses to slot machine trial outcomes were notably similar. Figure 2 displays each participant's average subjective estimation of how close the losing trial was to a win. Again, the 10-point Likert-type scale indicates that lower estimations were deemed far from a win, and high estimations were deemed closer to a win. These data show that all 12 participants estimated that near-miss type trials as being somehow closer to a winning spin than non-near miss, or total loss trials. Five paired t tests were conducted on the present data set to support visual inspection of the single-subject data. Because multiple f tests were conducted, we set the critical alpha level at .008 (.05 alpha / 6 tests) to control for an over inflation of the probability of a Type I error occurring. A significant difference was observed between the two trial types, t(11) = 6.916, p

To further determine if specific topographies of near misses influenced estimation ratings, all near-miss trials were separated into three subcategories: (a) trials with two winning symbols located in the left and middle payout line position (left), (b) trials with two winning symbols located in the left and right payout line position (split), and (c) trials with two winning symbols located in the middle and right payout line position (right). Figure 3 displays the resulting mean subjective estimations of win approximation for each participant. Of 12 participants, 11 emitted higher estimations for near-miss types of left or right than they did for a split. Only participant 210 had an equal value for right and splits. Near misses with winning symbols on the left and middle or right and middle of the payout line were estimated as closer to a win than near misses with winning symbols split on the left and right. Four paired t tests were conducted to support visual inspection of the data. Significant differences were observed between the left versus split, t(11) = 3.895, p

Figure 4 displays each participant's mean response latency following winning and all types of losing trials (i.e., included all types of near misses). Of the 12 participants, 8 displayed longer response latencies following a winning trial than following a losing trial. Only participants 201, 203, 205, and 209 displayed the opposite pattern of losing trial latencies being longer than winning trial latencies. As expected by visual inspection, a paired t test yielded a nonsignificant difference between trial types, t(11) = 1.767, p = .105.

Because variations were observed across participants, further analyses were conducted in an attempt to determine a potential variable responsible for these single-subject differences. Figure 5 again displays each participant's mean response latency following winning trials, yet now also displays a separation of the two types of losing trials: total loss trials and near-miss trials. When reexamining the latencies from participants 201, 203, 205, and 209, it can be seen that these 4 participants had much greater response latencies following near-miss trials than total loss trials. In fact, these 4 participants, and only these four participants, had response latencies greater following near-miss trials than following winning trials. Visual inspection of the data without these 4 participants suggests that there are noticeable differences between winning and all types of losing trials for the remainder of the participants.

Discussion

Continued from page 2.

Our current data lend an initial empirical foundation to Skinner's theoretical comment that "almost hitting the jack pot increases the probability that the individual will play the machine" (1953, p. 397). The near miss, or in Skinner's words "almost a win," appears to be evaluated by participants as approximating a winning trial more so than a total loss trial type. This belief by our participants was entirely erroneous. No losing trial was any closer to a win than any other. All losses were in fact equal. The pure randomness of the variable ratio schedule of the slot machine results in each upcoming trial having the exact probability of a win as the one before it, the one after it, and so on. There was no internal mechanism within the slot machine that somehow came close to wins when two winning symbols appeared on the payoff line. The observed fluctuations in subjective estimations of trial approximations to wins also may suggest that a slot machine with near misses may be favored by participants over one without near misses. Although our study did not directly examine preference for machines containing more near-miss trial types over machines containing few or no near-miss trial types, or the length/probability of playing near-miss machines versus non-near-miss machines, future research could easily do so. The implications regarding participant preference among alternatives and their resistance to extinction are substantial both experimentally and clinically.

For example, if a person with a gambling problem is exposed to a machine with a high density of near misses, they may have to continue to play for longer periods of time than they would on a non-near-miss machine. Initial behavioral treatment attempts might include teaching appropriate discrimination of losses-of which any type of loss is just as far from a win as any other loss. This could be done by having a preparation similar to the current study's method, videotaping the participant, having them review their tape, and providing them feedback when they identified incorrect estimations of losing trials. Future sessions could be conducted and examination of the frequency of incorrect estimations could be assessed. This approach is similar to what has been described from a cognitive behavioral standpoint by Bujold, Ladouceur, Sylvain, and Boisvert (1994) regarding gambling fallacies more generally, and it may merit exploration by behavior analysts attempting to treat problem gambling.

In the present study, there also appeared to be various topographies of near misses which had different levels of control over participants' verbal estimations. Two winning symbols on the left or right and middle positions of the payoff line resulted in higher estimations than two winning symbols split between the right and left position of the payoff line. These effects may have been caused by the physical proximity of the "winning" stimuli or symbols, which resulted in a discriminative stimulus that more physically resembled a win resulting in a higher estimation, perhaps as a product of stimulus generalization. Although our 25-year-old slot machine randomly generated a good number of near-miss trials to discover some order in participant responding, there are other topographies of near misses found in newer casino slot machines that may warrant investigation which ours simply did not have. For example, it has been suggested (by Reid, 1986) that many newer machines have a higher than random density of two winning symbols on the payoff line with the third winning symbol directly above or below the payoff line. If this is true, together with our results, it may be conceivable that this type of near miss will be evaluated as even closer to a win, preferred by participants, and results in extended play. A next logical step for future research would be to manipulate near-miss density and topography to assess a participant's resistance to extinction. Other preparations might include varying the size of the near-missed jackpot, or in other words the types of winning symbols on the payoff line to determine if there is a near-miss magnitude effect, as it appears to exist for actual jackpots (Delfabbro & Winefield, 1999) resulting in longer response latencies and greater physiological arousal following larger jackpots. Here a participant might have a given percentage of near misses that would have approximated a win of small magnitude of payout, and another percentage of near misses that would have approximated a win of a quite large magnitude of payout. If a near-miss magnitude effect exists, behavioral clinicians might attempt interventions that initially expose the participant to near misses of small jackpots, and gradually fade in near misses of larger jackpots in a similar fashion as what has been done with phobies and their aversive stimuli via gradual systematic desensatation (McGlynn & Rose, 1998). Because game manipulations such as these are impossible with actual casino slot machines, we suggest utilizing the slot machine simulation described by MacLin, Dixon, and Hayes (1999).

The differences observed in response latencies following winning versus total losing trials in the current experiment support previous observations by Dixon and Schreiber (2002) and Schreiber and Dixon (2001). Together these results may be conceptualized by a negative reinforcement and avoidance paradigm (Hineline, 1977) whereby a losing trial is an aversive stimulus that results in the participant initiating the onset of the next trial at a faster rate (shorter latency) to escape the continued presentation of that aversive stimulus. Furthermore, because 9 of our 12 participants displayed longer response latencies for near-miss trials compared to losing trials, it may suggest that near misses were not as aversive as a total loss, and in fact may actually have some reinforcing property, as illustrated by participants 201, 203, 205, and 209. Intervention strategies might be targeted at these pausing points (i.e., after a win or near miss) for participants to count their losses/earnings, assess the true probability of the game, or recall how much they planned on spending before pulling the lever again and initiating the next trial.

Where the current response latency findings differ from previous research is that higher response latencies following wins than following losses were not observed for all participants. Because this was the first study to include near-miss trial types, it appears reasonable to assume that this may have in fact been responsible. It is possible that to these 4 participants near-miss trials were a somewhat different environmental event than they were to other participants. Unfortunately, our current experimental arrangement could not detect what such a difference might have been. All participants were screened for potential gambling pathology using the South Oaks Gambling Screen (Lesieur & Blume, 1987) and no differences between these 4 and the other 8 participants were observed, which may have suggested some type of setting event (Kantor, 1924) or establishing operation (Michael, 1993) in effect. Likewise no differences could be determined between exposures to a specific type of contingency arrangement (i.e., the density or magnitude of winning payouts, strings of losses, etc.), which may have suggested some type of slight variation in the current context was responsible for changes in behavior.

The current findings extend previous research findings that greater response latencies tend to follow wins to actual slot machine players on an actual slot machine instead of novice undergraduates playing computer simulations. Although the difference in the apparatus may seem trivial, the current findings closer approximate the natural stimuli and contingencies contacted in a real casino. Because experimentation including game manipulations within casinos in the United States are not possible (Ghezzi et al., 2000), the current experimental arrangement may be as legally close as behavior analysts can come to an observation of gambling behavior. Furthermore, wagering for course extra-credit and probabilistic cash prizes may be as ethically close to real gambling that we can expose our experimental participants to without obtaining a casino license, or running the risk of them developing a gambling problem.

Continued from page 3.

We propose that the construction of a "near miss" is a verbal event. Almost catching that big fish, almost becoming famous, or almost hitting the jackpot are verbal descriptions of almost reinforcing events, which in turn are reinforced by the culture via responses such as "wow, so close" or "you'll get it next time." These in turn result in the almost events being somewhat reinforcing themselves. The near-miss type of trial was never paired with reinforcement, nor did it signal an upcoming reinforcer delivery. Yet, all our participants appeared to believe that it did. We did not tell participants that a near miss was somewhat closer to a win than a total loss; they told us. Furthermore, the varying response latencies across participants suggest that certain participants respond to near misses differently from others. These variations might be explained in subsequent studies using think-aloud procedures to examine rule following where subjects are required to speak aloud everything they are thinking about during the experiment ( see Hayes, 1986, for a review of this procedure). This procedure might reveal certain rules participants were generating during the experiment, erroneous beliefs they had about the game, or incorrect conceptualizations of randomness.

There are many avenues to explore towards a behavioral analysis of gambling. The importance of verbal behavior's inclusion in the conceptualization and treatment should not be overlooked. We believe that the current data, along with previous behavior analytic work by Lyons and Ghezzi (1995) and Dixon and colleagues (Dixon, 2000; Dixon, Hayes, & Aban, 2000; Dixon & Schreiber, 2003; Schreiber & Dixon, 2001), suggest that gambling is not a pure product of programmed contingencies. If gambling were completely under contingency control, people would not gamble and continue to gamble at the lottery when they have never won. This behavior has never contacted reinforcing consequences, and therefore cannot be explained simply by obtained intermittent programmed reinforcers. Furthermore, types of rules given to players have been shown to change betting strategies (i.e. Dixon et al., 2000) and produce risk-level alterations (Dixon, 2000), while programmed contingencies remained essentially unchanged. The current study now also demonstrates that participants under similar programmed contingencies will respond differently to the various types of gambling trials, both in terms of response latency and in terms of subjective win estimations. Whereas Skinner (1953) stated that the near miss is a "reinforcer (that) costs the owner of the device nothing," we suggest that if it alters a gambler's behavior, its costs to society may be quite substantial.

References

AMERICAN PSYCHIATRIC ASSOCIATION. (1994). Diagnostic and statistical manual of mental disorders (4th ed.). Washington, DC: Author.

ATOR, N. A. (1999). Statistical inference in behavior analysis: Environmental determinants? The Behavior Analyst, 22, 93-98.

BARON, A. (1999). Statistical inference in behavior analysis: Friend or foe? The Behavior Analyst, 22, 83-86.

BLASZCZYNSKI, A. P., & MCCONAGHY, N. (1994). Criminal offenses in Gamblers Anonymous and hospital treated pathological gamblers. Journal of Gambling Studies, 10, 99-127.

BROWN, R. I. F. (1986). Arousal and sensation-seeking components in the general explanation of gambling and gambling addictions. International Journal of the Addictions, 21, 1001-1016.

BUJOLD, A., LADOUCEUR, R., SYLVAIN, C., & BOISVERT, J. M. (1994). Treatment of pathological gamblers: An experimental study. Journal of Behavior Therapy and Experimental Psychiatry, 25, 275-282.

CROSBIE, J. (1999). Statistical inference in behavior analysis: Useful friend. The Behavior Analyst, 22, 105-108.

DELFABBRO, P. H., & WINEFIELD, A. H. (1999). Poker-machine gambling: Analysis of within-session characteristics. British Journal of Psychology, 90, 425-439.

DIXON, M. R. (2000). Manipulating the "Illusion of Control": Variations in risk-taking as a function of perceived control over chance outcomes. The Psychological Record, 50, 705-720.

DIXON, M. R., HAYES, L. J., & ABAN, I. (2000). Examining the roles of rule following, reinforcement, and preexperimental histories on risk-taking behavior. The Psychological Record, 50, 687-704.

DIXON, M. R., & SCHREIBER, J. (2002). Utilizing a computerized video poker simulation for the collection of experimental data on gambling behavior. The Psychological Record, 52, 417-428.

GHEZZI, R, LYONS, C., & DIXON, M. R. (2000). Gambling from a socioeconomic perspective. In W. K. Bickel & R. E. Vuchinich (Eds.), Reframing health behavior change with behavioral economics. New York: Erlbaum.

GRIFFITHS, M. R. (1999). The psychology of the near-miss (revisited): A comment on Delfabbro & Winefield (1999). British Journal of Psychology, 90, 441-445.

HAYES, S. C. (1986). The case of the silent dog - verbal reports and the analysis of rules. A review of Ericsson and Simon's Protocol Analysis: Verbal Reports as Data. Journal of the Experimental Analysis of Behavior, 45, 351-363.

HINELINE, P. N. (1977). Negative reinforcement and avoidance. In W. K Honig & J. E. R. Staddon (Eds.), Handbook of operant behavior. New York: Prentice Hall.

KANTOR, J. R. (1924). Principles of psychology. New York: Alfred A. Knopf.

KNAPP, T. J. (1998). Behaviorism and public policy: B. F. Skinner's views on gambling. Behavior and Social Issues, 7, 129-139.

LADOUCEUR, R., BOISVERT, J., PEPIN, M., LORANGER, M., & SYLVAIN, C. (1994). Social costs of pathological gambling. Journal of Gambling Studies, 10, 399-409.

LESIEUR, H. R., & BLUME, S. B. (1987). The South Oaks Gambling Screen (SOGS): A new instrument for the identification of pathological gamblers. American Journal of Psychiatry, 144, 1184-1188.

LYONS, C. A., & GHEZZI, P. M. (1995). Wagering on a large scale: Relationships between public gambling and game manipulations in two state lotteries. Journal of Applied Behavior Analysis, 28, 127-137.

MACE, F. C., & WACKER, D. P. (1994). Toward greater integration of basic and applied behavioral research: An introduction. Journal of Applied Behavior Analysis, 27, 569-574.

MACLIN, O. H., DIXON, M. R., & HAYES, L. J. (1999). A computerized slot machine simulation to investigate the variables involved in gambling behavior. Behavior Research Methods, Instruments, and Computers, 31(4), 731-735.

MCGLYNN, R. D, & ROSE, M. P. (1998). Assessment of anxiety and fear. In A. S. Bellack & M. Hersen (Eds.), Behavioral assessment. New York: Allyn and Bacon.

MICHAEL, J. (1982). Distinguishing between discriminate and motivational functions of stimuli. Journal of the Experimental Analysis of Behavior, 37, 149-155.

NATIONAL GAMBLING IMPACT STUDY COMMISSION (1999). NGISC Final Report. Retrieved January 30, 2004, from http://www.ncfpc.org/specialngisc.html

NATIONAL RESEARCH COUNCIL. (1999, April 1). Pathological gambling: A critical review. p. Exec.-2

NEVADA STATE GAMING CONTROL BOARD. (1997). Gaming revenue report (Vol. 4). Carson City, NV: Author.

PERONE, M. (1999). Statistical inference in behavior analysis: Experimental control is better. The Behavior Analyst, 22, 109-116.

PHILLIPS, D. P., WELTY, W. R., & SMITH, M. M. (1997). Elevated suicide levels associated with legalized gambling. Suicide and Life-Threatening Behavior, 27, 373-378.

REID, R. L. (1986). The psychology of the near miss. Journal of Gambling Behaviour, 2, 32-39.

SCHREIBER, J., & DIXON, M. R. (2001). Temporal characteristics of behavior on random-ratio schedules observed during slot machine play. Psychological Reports, 89, 67-72.

SKINNER, B. F. (1953). Science and human behavior. New York: Appleton-Century-Crofts.

STROMER, R. (2000). Integrating basic and applied research and the utility of Lattal and Perone's Handbook of research methods in human operant behavior. Journal of Applied Behavior Analysis, 33, 119-136.

WACKER, D. P. (1996). Behavior analysis research in JABA: A need for studies that bridge basic and applied research. Experimental Analysis of Human Behavior Bulletin, 14, 11-14.

MARK R. DIXON

Southern Illinois University

JAMES E. SCHREIBER

Duquesne University

A Faculty Research/Creative Grant awarded to Mark Dixon by Southern Illinois University in part supported this research. We thank Daniel Fienup for assistance with participant recruitment and data collection. We also thank Dr. Thomas Critchfield for his constructive comments on an earlier version of this manuscript. Address all correspondence regarding this paper to Mark R. Dixon, Behavior Analysis and Therapy Program, Rehabilitation Institute, Southern Illinois University, Carbondale, IL 62901.

Copyright Psychological Record Summer 2004
Provided by ProQuest Information and Learning Company. All rights Reserved