The Near-Miss in Slot Machines: Over Half a Century Later Jeffrey M

The near-miss in slot machines: Over half a century later

Jeffrey M. Pisklak1*¶, Joshua J.H. Yong1¶, Marcia L. Spetch1

1 Department of Psychology, University of Alberta, Edmonton, Alberta, Canada

* Corresponding author

¶These authors contributed equally to this work.

Abstract

In games of chance, a near miss is said to occur when feedback for what is otherwise a loss approximates a win. For instance, obtaining “cherry-cherry-lemon” on a slot machine could be considered a near miss. Sixty-six years after B. F. Skinner first proposed the idea that near- miss events might reinforce continued play in slot machines, belief in this ‘near-miss effect’ has remained strong despite some inconsistencies in the experimental literature. The present study reviewed and experimentally assessed the near-miss effect as it pertains to the frequency of the gambling response. Experiment 1 used a tightly controlled resistance-to-extinction procedure in pigeons to evaluate the putative reinforcing effect of near misses relative to a control “far-miss” reel pattern. Experiment 2 extended Experiment 1’s procedure to human participants. The results of both experiments failed to support the near-miss effect hypothesis. Experiment 3 used a further simplified procedure to assess the validity of the resistance-to-extinction paradigm when a probable conditional reinforcer was present on the reel stimuli. Although a clear discriminative function was obtained from the reel, subsequent testing in extinction revealed no reinforcing function of this stimulus.

Introduction

Near misses, also called near-hits or near wins, can generally be said to occur when the elements of a game or task “suggest” to a player that they have almost achieved a favourable result. A good example is provided by Witts, Ghezzi, and Manson [1]: Consider a novice player making repeated free throws in basketball. With each successive throw, variations in the throwing technique that get the ball closer towards the net and eventually into the hoop tend to

increase in probability (i.e., are selected for). This is an instance where the “near miss” can be said to have a clear reinforcing function on the player’s free-throw behaviour.

Near misses are not exclusive to games of skill – they also occur in games of chance.

However, the crucial difference is that in a game of chance, the outcome is practically a random event. On a standard slot machine, if a win can be signalled by “cherry, cherry, cherry” then a combination of “cherry, cherry, lemon” would be considered a near miss. Modern slot machines contain a pseudo-random number generator (RNG) that cycles through about 4.3 billion distinct values continuously at approximately one thousand values per second. When a player places a bet, the machine selects the cycle’s current position and correlated reel positions are subsequently output onto the display [2]. Unlike the basketball player practicing his free throw however, no amount of practice will ever produce the type of superhuman reflexes and discriminative capacity that a gambler would need to improve at slot machine playing.

Moreover, even if one possessed such powers, the random number generation employed by the machine means that receiving a near miss is no more informative to understanding the RNG cycle than any other type of miss (e.g. cherry, bell, lemon). In other words, the “cherry, cherry, lemon” combination is no “nearer” to a win than any other possible combination. This raises an important question: if the near miss inside a game of chance is independent of a win (i.e., cannot be used to help to determine the occurrence of a win) then why is it considered a “near” miss?

The obvious answer seems to be that the near miss is visually “nearly” a win. For example, if three cherries constitute a win, then receiving two cherries and a lemon necessarily looks more similar to that win than receiving a single cherry and two lemons. In the case of the basketball free throw, the visual aspects of the situation provide helpful feedback to the player.

Near-miss feedback on slot machines, however, offers no practical use for improving

performance. One possibility is that the visual aspect of the near miss is exploiting learning processes evolved to detect contingent (i.e., non-random) outcomes inside the environment.

Conditional reinforcement is one such learning process that has been hypothesized to fall victim to this exploitation.

Conditional Reinforcement

Conditional reinforcement has long been assumed to play a significant role in various instances of gambling behaviour. As early as 1953, B.F. Skinner utilized conditional reinforcement in what is likely the first scientific account of near misses [3]. Skinner’s account drew upon conditional reinforcement as one of a set of plausible methods casinos were using to exploit their patrons and it is still cited as an explanation of the near-miss effect – the idea that near-miss events contribute to an increased frequency of gambling by the player – a widely held belief among many gambling researchers. More formally, in this paper we define the near-miss effect as the hypothesis that near-miss events in a game of chance have a reinforcing function on the player’s total frequency of play. Skinner’s original hypothesis rested on two critical factors:

1) that the pairing (i.e., contiguity) hypothesis of conditional reinforcement is correct, and 2) that near misses do in fact cause an increase in the frequency of betting responses. At the time

Skinner proposed this, the available lines of evidence had been largely supportive of a basic pairing account of conditional reinforcement, and the reinforcing effect of near misses would have been a sensible a priori prediction. Since that time, however, research has shown that the basic pairing account is inaccurate and predicating near-miss effects based on this is questionable. Early theorizing conceptualized conditional reinforcement effects as a product of respondent (i.e., Pavlovian) conditioning, which was heavily dominated by accounts of temporal contiguity (i.e., pairing). Although it is parsimonious, referring to “pairing” as the circumstance

under which respondent conditioning will occur is now widely acknowledged as overly simplistic since it ignores the many nuances that influence and characterize this type of learning

[4-5]. Additionally – and irrespective of contemporary discussions of respondent conditioning – evidence suggests that pairing is not sufficient to produce a conditionally reinforcing effect [6-7].

Contemporary accounts of conditional reinforcement now place a stronger emphasis on the role of contingency, as opposed to strict pairing, and are usually framed in terms of two prominent mathematical models: delay reduction theory and the hyperbolic decay model [8].

Despite wide acceptance of the delay reduction theory and the hyperbolic decay model in behaviour analysis, contemporary near-miss studies still conceptualize the near-miss effect through the pairing formulation [9]. Also, although there is considerable evidence that near miss events can affect subjective measures (e.g. self reports) and physiological responses (e.g. skin conductance, heart rate or brain activity), and some studies have shown effects on betting size, there have been surprisingly few direct experimental tests of the presumed reinforcing function of near misses on gambling persistence. Effects on subjective motivation or physiological responses are consistent with a reinforcing effect on gambling frequency, but without a direct behavioral measure of persistence, they do not actually demonstrate such an effect [10-14].

Moreover, attempts at assessing near-miss effects in terms of delay reduction theory or the hyperbolic decay model are essentially nonexistent. This may be due to the manner in which conditional reinforcement studies are often conducted.

Most experimental paradigms testing conditional reinforcement employ some type of chained schedule of reinforcement. Two or more schedules of reinforcement are presented one after the other and are signalled by unique exteroceptive stimuli. Even complex paradigms such as observing response procedures and token economies are considered as such. One means of

assessing the conditionally reinforcing effect of a stimulus within the chain is to institute an extinction contingency and then compare the responding in the presence and the absence of the putative conditional reinforcer. This procedure is known as the established response technique and, as noted by Kelleher and Gollub in their exhaustive review of early conditional reinforcement literature [15], has the benefit of excluding the confounding effects of known reinforcers from testing, but the procedure may inadvertently introduce other problems. For example, extinction may alter the context such that the stimuli no longer function in the same way that they did during training (i.e., the stimuli may be “viewed” differently by the organism).

An additional problem noted by Kelleher and Gollub is that breaking the chain via extinction in the established response technique tends to produce only small effect sizes – presumably because the conditional reinforcer is being simultaneously tested and extinguished. This creates a difficulty for investigating the near-miss effect because even if near misses have a reinforcing effect, the observed effect may be small. Conventional chained procedures that successfully produce conditional reinforcement in laboratories have a logical predictability (i.e., contingency); the same cannot necessarily be said of the classic slot-machine scenario where randomness interrupts any logical predictability. This could diminish or even remove any conditionally reinforcing effect.

Near Misses as Global and Local Events

An overlooked point in near-miss articles is that near misses can be conceptualized globally or locally. For instance, the elements that occur on a basic three-reel slot machine (e.g., cherry, cherry, lemon) could be viewed globally, as a single stimulus, or they could be viewed locally, as three separate stimuli. This distinction matters according to conditional reinforcement.

Supporting the global view of near-miss events is research arguing that the conditionally reinforcing effect of win-related stimuli generalize to near misses better than to other, more dissimilar, types of misses [16-17]. These studies have argued for this explanation based on response latencies obtained in slot machine simulations. Specifically, outcomes more visually similar to wins tended to generate larger latencies than outcomes less similar to wins, which would be predicted by a stimulus generalization account. However, the reel patterns on the basic slot machine used in these studies are generally easy to discriminate which makes the generalization account less likely. If the patterns were more difficult to discern from one another, or the win and near-miss stimuli were partitioned from other reel patterns by common outcomes

[18] – as in the case of losses being disguised as wins [19] – then the generalization account would be more plausible. Furthermore, one would expect that as a person gains more experience playing a machine, they would become better at making discriminations between win and loss reel patterns. Consequently, the relevance of near misses on the gambler’s behaviour would presumably diminish. This account then would seem to predict only a temporary effect on a player’s behavior. A third concern is that other data has found that near miss latencies become shorter, not longer, the more they approximate a win [12]. A final concern is that these studies have focussed only on latency and have not reported on overall rate of responding. Reinforcers increase an organism’s rate of responding, which leads to a higher production of cumulative responses across an interval of time [20]. This occurs, even despite what is known as the post- reinforcement pause – a temporary cessation in responding that follows the delivery of reinforcement. To truly assess whether the reinforcing function of winning stimuli generalizes to near misses, it would need to be demonstrated that the presence of near misses increases the overall number or rate of bets, irrespective of response latencies.

In general, if near misses function as conditional reinforcers, they are most likely to operate locally. For instance, consider a near miss appearing on a simple hypothetical slot machine that contains only five possible reel patterns, all of which display symbols that appear from left to right and all of which are equally probable: A win (cherry, cherry, cherry), a near- miss (cherry, cherry, lemon), a flanked-miss (cherry, lemon, cherry), a far-miss (lemon, cherry, cherry), and a single-miss (a cherry and two lemons, appearing in random positions). From the point of view of a player ignorant of this machine’s odds, when a bet is initially placed there is only a 20% chance (1/5 = .2) of winning. However, when a cherry first appears on the left-most reel, a far miss outcome can be ruled out, thus the player can reasonably assume that a win is now 25% certain since only four potential reel patterns are now possible (1/4 = .25). If the middle position subsequently reveals a cherry, then both the flanked miss and the single miss can now be ruled out leaving only two possible outcomes: a near miss or a win. At this point the player might suppose that he or she has a 50% chance (1/2 = .5) of winning on this spin. Thus, from the initial moment the bet was made, the near miss has signaled a 30% increase in the probability of receiving a win. Admittedly, this is a very unrealistic example of a slot machine; nevertheless, it illustrates how near misses might have a conditionally reinforcing function on a player. Consider the 30% increase in the probability of receiving a win. One way to conceptualize this increase is temporally. Each successive cherry on the first two reels is signaling a reduction in time to a win, an idea seemingly consistent with the general notion of delay reduction theory, which states that larger signalled reductions in delay produce more conditional reinforcement. Therefore, if we compare the near miss to its conceptual opposite, the far miss, which offers no reduction and actually increases the delay to a win, we might suppose that a higher frequency of programmed near-misses, compared to far misses, would be more

reinforcing. Although this account seems plausible, the absolute amount of delay that is being reduced by the reels seems, on a practical level, trivially small and probably would not be especially reinforcing.

Animal Research

Despite calls for increased experimental analysis of gambling behaviour employing non- human animals [21], there is a scarcity of animal research to draw upon, particularly with respect to near misses. This is strange given the historical precedence animal research has set on questions of reinforcement generally. Nevertheless, a few studies warrant discussion.

Using a refined slot-machine analogue [22-23], Winstanley, Cocker, and Rogers [24] measured rates of extinction for two groups of rats’ responses on a roll lever. Alongside various neuropharmacological treatments, one group of eight rats received trials that contained near misses and another group of eight received trials that contained no near misses. No significant differences were observed between the two groups’ rates of extinction. Adjunct to this however, the rats also had to respond on a collect lever to obtain the food they had won. If they pressed this lever on trials where they received a loss, then they would incur a timeout penalty. Hence, it was to the rats’ advantage to only press the lever when the visual and auditory cues for a win were presented. The researchers found that the rats’ responses on a collect lever increased linearly as a function of win similarity. This meant that rats responded on the collect lever more after encountering a near miss than, for instance, a full loss. The authors proposed that the rats’ failure to withhold responding on losses that were similar to wins possibly reflects “a process similar to a ‘near-miss’ effect” (p. 917). As others have noted [25], this finding is probably best accounted for as an effect of stimulus generalization (i.e., a failure of discriminative control) between win and loss cues. On that basis, if the stimulus generalization account of near-misses is

correct then perhaps this finding does in fact reflect a process similar to a near-miss effect.

Though it should not be ignored that the critical extinction measure on the more relevant roll lever produced no such effect. Moreover, Peters, Hunt, and Harper [22] also tested near misses using a similar methodology to that of Winstanley et al. and they likewise found no increase in persistence on the roll lever when six rats were tested, within-subjects, on two extinction conditions.

Differing in strategy, Stagner et al. [25] assessed the reinforcing function of near misses by administering a choice task to pigeons. The task required them to select between two available alternatives. One alternative led to three possible outcomes: 1) a red key appearing for

10 s followed by food, 2) a green key appearing for 10 s followed by no food, and 3) a key that for the first 5 s appeared red and for the next 5 s appeared green with no food following it. This latter possibility was the near miss outcome. The second alternative the pigeons could choose only contained two possible outcomes: 1) a blue key appearing for 10 s followed by food or 2) a

10 s yellow key that followed with no food. Both alternatives were arranged probabilistically such that each delivered equivalent amounts of food. Across four different manipulations the birds consistently preferred the alternative containing no near misses, a finding that, given the specifics of the experimental design, was consistent with both delay reduction theory and the hyperbolic decay model interpretation of conditional reinforcement. Using a slightly different methodology a similar avoidance of the near miss by pigeons in a choice task was also obtained by Fortes, Case, and Zentall [26].

Somewhat more favourable towards the near-miss effect was a pigeon choice study by

Rice and Kyonka [27] that analyzed matching between rates of responding and reinforcer value.

The results showed a consistent bias, across three pigeons, towards a “gambling” option

containing near misses relative to a certain option that contained no losses (and therefore no near misses). Paradoxically however, this same bias was not observed when the gambling option containing near misses was tested against a similarly probabilistic option that contained no near misses. Furthermore, there were no consistent patterns of behaviour across the remaining conditions which suggests that further research should be conducted to assess the reliability of their positive findings.

Human Research

Experimental research utilizing human subjects in slot machine simulations have been slightly more favourable to the notion that near misses might have a reinforcing function, but direct evidence that near misses increase gambling persistence has also been quite limited and has not always held up to replication attempts. An early study by Strickland and Grote [28] that used a response-cost methodology had 44 subjects perform on a one-arm bandit slot machine for money. Playing the machine gradually returned an overall net loss of money. They found that

59% of the subjects who saw the winning symbols occur more frequently on earlier presented reels chose to keep playing after an initial exposure phase of playing. This is contrasted with only 23% of a second group who saw the winning symbols occur less frequently on earlier presented reels. Although this difference is supportive of a near-miss effect, other evidence they obtained was not consistent. Specifically, the average number of trials played by participants who opted to keep playing after the exposure phase did not significantly differ between the two groups. Reid [29] attempted two systematic replications of Strickland and Grote’s study. One used a card-based version (n = 56) of the slot machine task and the other used a simulated slot machine version (n = 48). Of the two, only the latter version showed the same general pattern of

results obtained by Strickland and Grote; however, neither replication obtained effects large enough to reach significance.

More recently, Kassinove and Schare [9] manipulated the frequency of near misses on a four-reel slot machine simulation that participants (n = 180) could play for money. Near misses had either a 15%, 30%, or 45% chance of occurring depending on group assignment. To assess the effect of this manipulation on gambling persistence, participants were put onto an extinction condition that removed both wins and near misses. During this condition the participants could continue to play for as long as they desired. The 30% group showed the most persistence during extinction. Interestingly, a self-report administered after the extinction phase, on willingness to return to play in the future, generated no differences between the three groups. A concomitant manipulation of big wins also revealed no main effect or interaction with either the persistence measure or the self-report. The authors argued that their findings were best accounted for in terms of conditional reinforcement. They suggested that the 45% near miss frequency likely paired the near miss too frequently with a win and thus led to respondent extinction, whereas the

15% near miss frequency paired the near miss too infrequently with a win. Across the 15%, 30%, and 45% groups, means (SD) of 5.88 (8.06), 10.26 (11.47), and 6.66 (8.22) responses in extinction were obtained respectively. Of the evidence for a reinforcing function of near misses, this is one of the most frequently cited studies. However, it has some potential problems. First,

Kassinove and Schare analyzed their extinction data using parametric statistics, which could be problematic given that their extinction data is likely to be highly skewed. This is because, in extinction, the probability of responding declines after each subsequent unreinforced response, making the data heavily bounded by zero. Additionally, if one assumes their data is symmetric then, based on their obtained variances, substantial amounts (≈ 25%) of their data points in the

30% group would be expected to fall below zero, which is not possible. Consequently, the obtained means and variances may not representative of the population’s true value, particularly in the 30% condition which displays a substantially higher variance than the other two groups, and thus a stronger relative pull of the mean under conditions of skewness. Incidentally, a similar criticism could be made for other extinction/persistence-based studies [28-30]. A second concern is that extinction was tested without the presence of near misses, i.e., without the presence of the putative conditional reinforcer. This would presumably lead to similar extinction rates across the groups since the very thing that is meant to differentiate the rates of extinction is no longer appearing, making actual differences between the groups less perceptible. Thus, the fact that there was a difference between the groups is puzzling. Finally, across three separate experiments,

Ghezzi et al. [30] attempted to replicate the basic findings of both Kassinove and Schare and

Strickland and Grote by having groups of participants play a simulated slot machine for points.

Despite having marginally larger group sample sizes than Kassinove and Schare, only the first of their three experiments revealed a main effect of near-miss density on gambling persistence, which occurred at a density of 66% − a finding that did not hold across the remaining two experiments. Overall, Ghezzi et al.’s findings were not commensurate with the two original papers. Although there were notable procedural differences between these replications and the original two studies, this failure to replicate the near miss effect is concerning for the robustness of the near miss effect on gambling persistence, leading the authors to conclude that: “The results of the three studies . . . suggest that the near-miss is either terribly elusive or overstated as a means of prolonging slot machine play” (p. 168).

More favourable evidence for a near miss effect was provide by a study by Côté, Caron,

Aubert, Desrochers, and Ladouceur [31] that had 72 participants play a three-reel video slot

machine terminal for money. The experiment comprised two phases: The first phase exposed the players to 48 trials wherein near misses were programmed to occur a total of 12 times and wins a total of 9 times. In the second phase participants were, unknowingly, placed onto extinction. For an experimental group, this meant the removal of wins only. For a control group, extinction removed both wins and near misses. Participants could choose to stop playing at any point and keep their earnings. Using non-parametric analyses, the findings revealed that participants in the experimental group chose to play significantly more during extinction than the control group.

Although this study is well cited as provided strong evidence of a near miss effect, we believe there is a noteworthy concern with this study that precludes it from providing evidence that near misses in a typical slot machine would increase gambling persistence. Specifically, in the first phase of the experiment, each win was preceded by a near miss. This contingency meant that each time a near miss was encountered a win would be 75% certain to occur on the next trial, possibly generating a conditional reinforcing effect across the reinforcement schedule. This contingency does not exist in typical slot machines. Such a contingency between scheduled winning trials and any type of miss could theoretically lead to a significant difference between the control and experimental groups, but only the near-miss was tested.

An experiment by MacLin, Dixon, Daugherty, and Small [32] gave 18 recreational slot machine gamblers three concurrently available three-reel slot machines to play for points. Each machine had a specific frequency of near-miss presentations: 15%, 30%, or 45%. Following the completion of 100 required trials, participants could keep playing until they wanted to stop. This afforded them a chance to earn a cash prize for the highest score among a pool of other participants. During this point in the experiment the machines ceased to pay out wins (i.e., extinction). Although an increase in persistence as a function of near-miss frequency was

observed during extinction − i.e., people persisted most at a 45% frequency and least at a 15% frequency on average – this difference was not found to be statistically significant.

Other studies that did not employ persistence-based methodologies have found similarly varied results ranging from support [33-34] to lack of support [1] for a near-miss effect. While not strictly experimental in nature, a notable study by Sundali, Safford, and Croson [35] examined real casino data from a set of electronic roulette terminals. They modelled the data of an estimated 36 players using regression analyses and found no evidence to support the notion that near misses had reinforced the gamblers’ playing. This was the case for both time spent playing on the machines and the number of bets placed.

An interesting commonality among all the studies mentioned here is that none of them specifically recruited problem gamblers suggesting that the variance across studies is not simply a function of samples. Rather, by using the South Oaks Gambling Screen [36] and testing populations of undergraduates, most studies excluded people that could be characterized as probable pathological gamblers [1,9,31-32]. Of these, only MacLin et al. [32] specifically sampled recreational gamblers. With one exception, the remaining studies all recruited unfiltered populations of undergraduate students [29,30,33,34]. Unusually, Strickland and Grote [28] recruited unfiltered rural high-school students.

The number of training trials employed by the studies was also similar with a range between 20-100 trials across the various studies. Of the studies that demonstrated increased persistence, Côté et al. [31] used 48 trials, Kassinove and Schare [9] used 50 trials, and

Strickland and Grote [28] used 100 trials. Interestingly, Ghezzi et al. [30] varied the number of training trials (25, 50, 75, and 100) and found a significant effect on persistence only after 25 trials, and no trends as a function of training trials were observed. Further, this was not shown to

interact with a concomitant manipulation of near-miss density. This raises an important question:

What is an appropriate amount of training in these types of studies? As a general rule, more training seems preferable to less training. This seems especially true for participants who may not have much experience with games of chance. Their expectations about gambling may conflict with the real-life gambling contingencies that experiments try to model, creating biases in the observed behaviour. At low levels of training, these biases are likely to go unchallenged.

Therefore, the results of novice gamblers may be different from more experienced gamblers particularly with low levels of training.

Perhaps because of Skinner’s influence, the suggestion that near-misses are functioning as conditional reinforcers has been largely taken for granted. Overall, the results of experimental studies testing this have not been especially compelling either for or against the hypothesis that near misses have a reinforcing function. The reasons for this murkiness are not altogether clear.

Methodological variations and other problems have made it difficult to diagnose clear areas of concern. Experimental analyses in the literature have relied primarily on human participants performing on either real lottery terminals or programs designed to mimic casino style slot machines. Although such procedures can enhance the external validity of a study, they complicate analyses of basic effects - even something as simple as having three versus four reels on a slot machine can contribute to larger variability and create difficulties in learning the contingencies. Motivation levels can differ widely between individuals and basic parameters, such as the type of population sampled and how many learning and testing trials are implemented, often vary from experiment to experiment and laboratory to laboratory. Arguably, the pursuit for external validity in these prior studies has undermined the establishment of internal validity of near-miss research. What seems required then is a more controlled analysis

reminiscent of the work Skinner himself favoured. Pigeons have had a long and storied history in the study of not only conditional reinforcement effects but learning phenomena more generally

[37]. However, their application in understanding near-miss events has been fairly minimal, despite many properties that would seem to make them desirable for a controlled analyses of near-miss effects. Pigeons have excellent visual acuity, easily regulated motivation levels, and they are well known to exhibit steep rates of temporal discounting relative to other common laboratory animals [38]. This latter fact may be especially pertinent insofar as gambling research is concerned. Studies examining the discounting functions of problem gamblers consistently show that they discount delayed rewards more than non-gamblers do [39-41]. Some have theorized that this aspect in particular, in conjunction with reinforcement schedules, is what results in much of the maladaptive behaviour they exhibit [42]. Due to enhanced experimental control, behavioural data obtained through animal models can often be more reliable than studying humans directly. However, the results from animal studies should be verified in human participants whenever feasible.

Overview of Experiments

Experiments 1 and 2 of the present study employed a novel resistance-to-extinction style paradigm to test the reinforcing function of a near miss stimulus pattern against a far-miss control pattern on both pigeons and humans respectively. Using a similar logic to that of the established response technique, and the two most notable experimental demonstrations of a near- miss effects in the gambling literature (i.e., Kassinove & Schare, [9]; Côté et al. [31]), if near- miss events function as conditional reinforcers relative to other types of misses then greater resistance to extinction should be observed on treatment conditions that contain a higher frequency of near-miss events than far-miss events. Following the results of both these

experiments, Experiment 3 more explicitly assessed the use of extinction as a means of evaluating conditional reinforcement generally. Environmental contingencies were arranged such that the discriminative (and thus conditionally reinforcing) function of a terminal-link (i.e., reel) stimulus was verified and subsequently tested to see if it could offset the effects of extinction on responding.

Experiment 1a and 1b

Methods

Subjects and apparatus

For both Experiment 1a and 1b, eight homing pigeons (Columba livia) were randomly selected from a University of Alberta Colony room for the experiment. Subjects were housed in

65×27×70 in. flight cages in a colony room maintained at 20ºC. The colony room operated on a

12-hour daylight cycle from 6:00 A.M. to 6:00 P.M. MST. All birds had free access to vitamin- enriched water and crushed oyster shell grit in the colony room. Subjects were maintained at

80% of their free-feeding weight by adjusting their post-experiment feeding of Mazuri Gamebird food pellets (PMI Nutrition International).

Six custom-built operant boxes were equipped with Carrol Touch infrared touchscreens

(Elo Touch Systems, Inc., Menlo Park, CA) to detect pecking responses. Stimuli were presented on a centrally-mounted 17” Viewsonic LCD monitor located at the back wall of each chamber.

Speakers in the operant boxes continuously played white noise to mask sounds from outside the experimental room. The sound pressure levels were equalized in each operant box at 65 dB via

A-weighting filter with a Brüel & Kjær Type 2239 Integrating Sound Level Meter. Two 2×2 in. feeding ports flanked the monitor and each port was equipped with a food hopper. Access to food was controlled by Colbourne H20-94 photocell sensors that detected entry into the ports.

Stimuli were presented and responses were logged using E-Prime® 2.0 Professional software. A

⅛ in. thick white plastic barrier was mounted in front of the screen to prevent errant behaviours

(e.g., subject’s wings, bodies, and feet contacting the screen) from interfering with the touch screen’s ability to record pecking responses. The barrier had four holes cut in it: three horizontal circles were cut to 1½ in. diameter near the top to allow for visual identification of the reel stimuli and a 1 in. diameter hole was cut beneath the middle of the three circles to allow for pecking responses.

Stimuli were presented on a black background and were aligned behind the holes cut in the barrier. A white pecking circle was presented at the smaller 1 in. hole. The win and miss patterns (i.e., reel patterns) were presented at the upper three circles. For Experiment 1a, each reel pattern consisted only of those elements comprising the winning reel pattern, which was displayed as three red circles. A win occurred when all three circles were red. The remaining miss feedback patterns were presented as such: A near miss occurred when only the left and middle circles were red. A flanked miss occurred when only the left and right circles were red. A far miss occurred when only the middle and right circles were red. Finally, a single miss occurred when only one of the three circles was red. On single misses, each of the three locations had an equal chance of turning red. Non-illuminated circles were left black. Experiment 1b was identical in this regard but with the exception that the non-illuminated circles would now be coloured blue. For example, a trial displaying a far-miss would be presented as blue-red-red, with the blue now acting as a more immediate signal for non-reinforcement. Thus we denote red circles as putative SD stimuli (i.e., stimuli that occasion reinforcement) and blue circles as putative SΔ stimuli (i.e., stimuli that occasion nonreinforcement).

Reels were always presented sequentially from left to right at a 600 ms interval in between each presentation. For instance, this was the sequence of events on a ‘winning’ trial:

Following a peck to the white circle, the white circle disappeared. After 600 ms, the left reel appeared. Then, the middle and right reel each appeared in sequence with a 600 ms interval between each presentation. In total, the sequence was 2.4 s long. Then, the right or left hopper

(randomly chosen) would rise to give the bird 1 s access to food pellets (i.e., primary reinforcement) with the reels remaining on screen during this period. Finally, the reels reset to black and the white pecking circle returned to occasion the start of the next trial. On all loss trials, each of the intervals still occurred whether or not a red circle appeared so that the total sequence was always 2.4 s long. The reels then reset back to black and the white pecking circle reappeared following the last 600 ms interval. Fig 1 diagrams this sequence for both win and far- miss outcomes.

Fig 1. Schematic illustrating the sequence of events for both a win (left column) and a far- miss (right column) during the third component of Experiment 1a’s pre-exposure phase.

The letters ‘W’ and ‘R’ indicate the colours white and red respectively. Grey circles indicate empty/unused circles. Circles always illuminate from left to right.

Ethical statement

All procedures for the care and use of animals were in accordance with the ethical standards of the University of Alberta, Canadian Council on Animal Care, and were approved by the Bioscience Animal Care and Use Committee (Protocol AUP00002018).

Procedure

The experiment was structured as a repeated-measures design with two treatments: the near-miss and far-miss treatments. The order of these treatments for each subject was randomly determined according to a Latin square design (i.e., subject × treatment). Each treatment was preceded by a pre-exposure phase comprised of three components. Only one handler conducted the experimental sessions to further reduce sources of unsystematic variation that can occur in extinction-based designs. Additionally, because the noise of hoppers in adjacent operant chambers might confound the measures taken during periods of extinction, all pigeons had their extinction sessions yoked to occur at the same time and day as other pigeons concurrently running in the experiment.

In the pre-exposure phase, the subjects began on a basic autoshaping procedure [43], with a fixed-ratio 1 (FR1) contingency built in. Here, a white circle was presented behind the lower response hole for 10 s or until the bird pecked it. After the 10 s or after one peck on the white circle, the circle disappeared and then either the left or right feeding port light illuminated and its hopper raised. When the bird’s head entered the port, it had 1 s to eat from the hopper. Then, the port light extinguished, the hopper lowered, and a 240 s interval began. This process repeated over a 90 min session. All subjects remained on this component for three consecutive sessions.

Following autoshaping, the subjects were put on a FR1 schedule. Responses were made on the lower response hole as before. Following one peck, the white pecking circle disappeared and then the subject had to wait 2400 ms until they were given access to food from either the left or right hopper. The FR1 schedule lasted for one 90 min session. On the next session, the schedule was extended to a FR3 (i.e., three pecks were required to gain access to food). The

2400 ms interval was still implemented after every response. Following one session of FR3, the schedule was extended to a FR6 for one more session. This procedure was followed for all birds with the exception of Bird 76 in Experiment 1b, who displayed unusually low responses after being moved from the autoshaping schedule to the FR1 schedule. Autoshaping was reinstated until high rates of responding on the FR1 schedule were observed to occur. Due to the necessity of yoking the extinction sessions, the FR3 and FR6 conditions were skipped in the first half of

Bird 76’s sessions to allow for maximal exposure to the RR5 contingency (described below).

In the final pre-exposure component, subjects were put on a random-ratio 5 (RR5) schedule. Similar to a variable-ratio (VR) schedule, a RR schedule typically contains an average response requirement. However, unlike a VR schedule, a RR schedule is determined by pseudo- random number generation as opposed to a predetermined number of reinforced and nonreinforced trials [44-45]. In this pre-exposure component, each peck was separated by at least a 2400 ms interval but during this interval, the reel stimuli were now presented at the upper three circles. Birds only gained access to food following wins. All reel outcomes were equiprobable.

Specifically, wins, near misses, and all other loss types each occurred 20% of the time and these probabilities were equated every 30 responses. Each bird received fifteen 90-minute-long sessions – apart from bird 76 who only received nine sessions of the RR5 schedule in the first half of Experiment 1b.

In the treatment phase, after the pre-exposure phase, all subjects were put on either the near-miss or far-miss treatment. In either treatment, responses on the white key were put on extinction. Specifically, all wins and primary reinforcement were replaced by either the near- miss or far-miss reel patterns. For instance, a subject in the near-miss treatment experienced near misses 40% of the time and no wins, and the proportions of the other loss feedback were left unchanged. The subjects were not presented with any additional cues to signal this change in the schedule. After completing the first treatment phase, all subjects were returned to the pre- exposure phase and then they completed the next treatment condition (i.e., a pigeon that began on the near-miss treatment first completed the far-miss treatment next and vice versa).

Results and Discussion

Statistical analyses were conducted using R 3.5.0 [46]. The mean difference in resistance to extinction between the Near Miss and Far Miss treatments was assessed using a paired t-test on the cumulative amount of responses made during extinction. An effect size was calculated using an unbiased estimate of Cohen’s d (see equation 11.13 in Cumming [47]). To best account for the reduction in variability offered by paired designs, the d estimate was standardized using the standard deviation of the difference scores. To obtain the relative odds in favour of the alternative hypothesis against the null, a JZS Bayes Factor (BF10) using a medium prior was calculated [48].

The results of Experiment 1a are depicted in the top row of Fig 2. Contrary to the predictions of a near-miss effect, the near miss treatment showed less overall responding (M =

613.25, 95% CI [407.26, 819.24]) in extinction than the corresponding far miss control treatment

(M = 784.12, [537.73, 1030.52]). However, the difference between the two treatments was not

found to be statistically significant, nor was the obtained Bayes Factor meaningfully large, t(7) =

−1.68, p = .136, [−410.75, 69.00], d = 0.53, BF10 = 0.92.

Fig 2. The results of experiment 1a (top) and 1b (bottom). The left-most plot shows the mean cumulative responses during extinction for the near and far miss treatments. The centre plot shows the mean difference between these treatments. All error bars indicate 95% confidence intervals. The right-most plot shows the mean cumulative responses across the five sessions of extinction testing. Circles and diamonds indicate Near Miss and Far Miss extinction treatments respectively.

The results of Experiment 1b are depicted in the bottom row of Fig 2 and, like experiment 1a, are similarly inconclusive. The mean cumulative responding for the near miss treatment (M = 677.88, [436.94, 918.81]) exhibited marginally more responses during extinction than the corresponding far miss control treatment (M = 573.12, [353.40, 792.85]), t(7) = 1.44, p

= .193, [−67.31, 276.81], d = 0.45, BF10 = 0.73.

Overall, the results of both Experiment 1a and 1b failed to provide evidence that near misses have a conditionally reinforcing function on the gambling response. However, it may be the case that the supposed near miss effect is a uniquely human phenomenon and does not apply to non-human or non-primate organisms. While the literature on operant conditioning seems inconsistent with such a hypothesis, given the strong reliability of reinforcement processes across numerous species, it is nonetheless a possibility that should be considered. The animal-based literature has, in most cases, failed to straightforwardly demonstrate the reinforcing nature of near miss stimuli whereas the human-based literature has been, at least occasionally, successful in this regard. Given this, Experiment 2 adapted and applied Experiment 1a and 1b’s procedure to human participants.

Experiment 2

Methods

Participants and apparatus

A total of 296 participants (Homo sapiens) were recruited from the research participant pool of introductory psychology courses at the University of Alberta. They were incentivized with partial course credit and a small monetary bonus of up to $9.50 CAD that depended on their performance in the task. Sessions were conducted in a space containing fifteen individual testing

rooms located around the perimeter. The task was completed on a computer with a mouse and keyboard. E-Prime® 2.0 Professional software was used to present stimuli and record responses.

This experiment was adapted from Experiments 1a and 1b. The stimuli were presented on a black background, and three 2⅛ in. circles that comprised the reel patterns were shown at the top of the screen. Their locations were outlined by a grey border. Reel patterns were presented in the same manner as Experiments 1a and 1b. A 1½ in. white circle located just below the middle reel stimulus location served as the response key. The response key was also outlined with a grey border, and text within it read “Click Me!”.

Ethical Statement

All procedures performed in studies involving human participants were approved by the

University of Alberta Research Ethics Board (Protocol Pro00058367) and were in accordance with the Tri-Council Policy Statement on the Ethical Conduct of Research Involving Humans.

Procedure

Participants first read a consent form upon entering the experimental room. Once they signed the forms to provide their consent for participation, they were assigned to the individual testing rooms and instructed to begin. After being prompted to enter their age on the computer, a brief set of instructions was presented to them on screen for a minimum of 1 min. Instructions instructed participants to try and earn as much money as possible. It also informed them that they could stop playing at any time to collect their money.

Participants were assigned randomly to one of two groups and then further randomly assigned to one of two treatments. Groups were analogous to Experiments 1a and 1b:

Participants in the SD group only saw red circles in the reel patterns, and participants in the

SD&S∆ group saw both red and blue circles but only red circles signalled wins. The treatments

were also similar to Experiments 1a and 1b: near misses or far misses replaced the wins during extinction.

In Experiment 2’s pre-exposure phase, the initial phase was similar to the third component of the pre-exposure phases from Experiment 1. The first 15 trials were structured such that the first, fifth, and eleventh trials were wins to prevent participants from encountering an early string of losses that could possibly occur due to the RR schedule and produce a negative monetary score. The remaining 285 trials proceeded according to a RR5 schedule, with the probabilities of all outcomes being equated every 15 trials. Participants began with $0.50 and their score was always shown at the bottom of the screen. It cost participants $0.05 for each play on the response key and this was immediately deducted from their score. Wins awarded $0.40 and they were signalled by the appearance of “Win! +40¢” in large bright yellow text beneath the response key.

The treatment phase occurred after 300 trials, at which point plays on the response key were put on extinction: clicking it still cost $0.05 but doing so no longer produced wins.

Participants in the near-miss treatment had the wins replaced by near misses and participants in the far-miss treatment had wins replaced by far misses. Unlike some other extinction-based studies [9,32] participants were not presented with any cues to signal this change nor were they ever told that such a change would occur. Participants could continue playing until 50 minutes had elapsed. When participants were ready to leave, they were paid according to their final score and given a debriefing form about the experiment.

Results and Discussion

Statistical analyses were conducted only on those participants for whom the task was observed to be sufficiently reinforcing. This was defined as completing more than 300 trials.

This criterion was set because 300 was the minimum number of trials a person would have to complete before entering the extinction phase. Of the total sample, 178 (60%) continued through to the extinction phase; whereas 118 (40%) did not. Of the participants who did not make it to extinction, 48% were assigned to the SD group and 52% were assigned to the SD&S∆ group.

As in Experiment 1, analyses were conducted on the cumulative number of responses made during extinction using R 3.5.0 [46]. Because visual inspection of the data revealed heavy positive skewness, a base-10 logarithmic transformation was conducted to make the data symmetrical and permit parametric analyses using a 2×2 Factorial ANOVA. Partial omega-

2 squared estimates (ωp ) of effect size were obtained for each main effect and interaction term. A corresponding JZS Bayes Factor showing the relative odds of each term against the model’s intercept is also reported [48].

Fig 3 shows the cumulative number of responses made for each group in Experiment 2 plotted in untransformed units using notched box-and-whisker plots. There was a non-significant

2 main effect of both miss-type (near miss vs. far miss), F(1, 174) = 2.12, p = 0.148, ωp = 0.01,

D D ∆ 2 BF10 = 0.42, and stimulus group (S vs S &S ), F(1, 174) = 0.01, p = 0.92, ωp = 0.01, BF10 =

0.16. A non-significant interaction between miss-type and stimulus group was also obtained, F(1,

2 174) = 1.49, p = 0.22, ωp = 0.003, BF10 = 0.03.

Fig 3. The results of experiment 2. Plot displays notched box-and-whisker plots diagramming cumulative responses during extinction for SD and SDS∆ groups across the Near and Far Miss treatments. Whiskers follow the 1.5×IQR rule. Hollow circles show data points falling outside the whisker edges.

The results of Experiment 2 corroborate the findings of Experiment 1 as well as other extinction studies that failed to find a reinforcing effect of near-miss stimuli on human participants [30]. Granted, Experiment 1 and 2 have only shown an absence of evidence for a conditional reinforcement effect. However, if it can be shown that an effect analogous to the near-miss effect in extinction could not be obtained despite the presence of a probable conditional reinforcer, then that would be indicative of evidence of absence for enhanced

persistence of conditional reinforcers in extinction. Experiment 3 simplified the reel stimuli by arranging the pre-exposure phase contingencies so that the discriminative and, by extension, conditionally reinforcing function of the feedback stimuli, could be verified independently of the extinction procedure. Following this, the effect of those stimuli were tested to see if they could be used to increase persistence during extinction.

Experiment 3

Methods

Subjects and apparatus

An additional 8 homing pigeons that were not used in Experiments 1a and 1b were randomly selected. The apparatus was the same as those used in Experiments 1a and 1b.

Ethical statement

Procedure

Like Experiment 1, a repeated-measures design with two treatments was used and the order the treatments was randomly determined by a Latin square (i.e., subject × treatment).

Subjects began on a three-component pre-exposure phase followed by a treatment phase. Only one handler conducted the experimental sessions during the treatment phase to reduce unsystematic variation.

The first component of the pre-exposure phase was the same as the autoshaping procedure from Experiment 1a. In the next component, the pigeons were put on a FR1 schedule for one session that reinforced responses to the white pecking circle at the lower hole in the

barrier. After a single response the circle turned black and was followed by a 2400 ms interval before reinforcement was delivered. The session lasted for 90 minutes or 200 reinforcements, whichever occurred first. If a bird obtained all 200 reinforcements before the 90 minutes had elapsed it remained in the box with an inoperative screen that contained no pecking circles. In the third component, the pigeons were moved to a pure RR2 schedule with the number of pecks determined randomly according to a 0.5 probability of any given response leading to reinforcement. Similar to the previous component, the session had a 90 min or 200 reinforcer limit, but a 2.4 second SD or S∆ feedback stimulus (indicating reinforcement and nonreinforcement respectively) was now presented at the center hole of the upper three circles in the plastic barrier. Following each SD, the bird was given 1 s to access food which then was followed by the disappearance of the SD and the reappearance of the white response circle. Non- reinforced trials displaying the S∆ were followed immediately by the termination of S∆ and reappearance of the white response circle. The SD and S∆ stimuli could consist of either a yellow circle with a thin black horizontal line intersecting it or a light blue circle with a thin black vertical line intersecting it. The specific stimulus-colour and function mappings were randomly assigned to each pigeon in a balanced fashion such that half the birds received a yellow SD and blue S∆ while the other half received a blue SD and yellow S∆. To allow for assessment of a conditionally reinforcing effect of the SD, pecking responses made to the SD and S∆ stimuli were recorded at the millisecond interval.

Following five sessions on the RR2 schedule, subjects were put on a treatment phase for three sessions where responses to the white pecking circle were put on extinction. Primary reinforcement was removed and the proportions of feedback were modified. In the SD 80% treatment, subjects received the SD feedback stimulus 80% of the time and the S∆ feedback

stimulus 20% of the time. Subjects in the SD 20% treatment received the SD feedback stimulus

20% of the time and the S∆ feedback stimulus 80% of the time. After the first extinction phase, subjects returned to the pre-exposure phase and then they completed the other treatment condition.

Results and Discussion

To evaluate the plausibility of a conditionally reinforcing effect occurring on the designated SD stimulus, a discrimination index (I) for each pigeon was calculated using the equation I = SD / (SD + S∆) where SD and S∆ represent the cumulative amount of responses made respectively during each stimulus across all pre-exposure phase sessions. This was tested against a prediction of no discrimination (I = 0.5). The data were collapsed across both blocks of pre- training since there was no reason to expect an effect of order. The index means across the first and second blocks of pre-exposure corroborate this supposition. The first block obtained a mean discrimination index of M = 0.985, [0.973, 0.997] and the second of M = 0.998, [0.996, 1.001] respectively. Across all pre-exposure sessions, an extreme effect on responding to the SD relative

∆ to the S was observed, t(7) = 187.73, p < 0.000, [0.986, .999], d = 118.90, BF10 > 150). This indicates that the presence of the SD relative to the S∆ had a strong discriminative function on pecking, a robust and critical feature of conditionally reinforcing stimuli [49-51]. This occurred despite the fact that pigeons were never trained to respond to this location, had no prior experience with these stimuli, and that responses made to this stimulus had no impact on the sequence or timing of events.

Statistical analyses on cumulative number of responses made during extinction was calculated as per Experiment 1. The results (Fig 4) showed a non-significant effect and no support of a difference between the tested extinction treatments, t(7) = 0.02, p = 0.985, [−162.38,

165.13], d = 0.01, BF10 = 0.34, despite the strong evidence that the feedback stimulus functioned as a conditional reinforcer during the pre-exposure phase. This strongly implies that whatever reinforcing function the reels on a slot machine may have, it is likely independent of and thus does not reliably reinforce prolonged play during a period of extinction (i.e., a period of consistent losses).

Fig 4. The results of experiment 3. Left-most plot shows the mean cumulative responses during extinction for the SD 80% and SD 20% treatment. The centre plot shows the mean difference between these treatments. All error bars indicate 95% confidence intervals. The right-most plot shows the mean cumulative responses across the three sessions of extinction testing. Circles and diamonds indicate the SD 80% and SD 20% extinction treatments respectively.

General Discussion

At present, some of the most prominent lines of experimental evidence in favour of the near-miss effect come from a select set of studies employing extinction-based persistence tasks

[9,31]. The present set of studies failed to replicate positive evidence of a near-miss effect and instead corroborated other, less cited, experimental studies that have not been able to

demonstrate a reliable reinforcing effect of near-miss stimuli on extinction-based, choice-based, or observing-response-based tasks [1, 22, 25-26, 30]. More notably, Experiment 3 of the present study demonstrates that the use of an extinction-based treatment in evaluating the function of a putative conditional reinforcer is, at best, unreliable and, at worst, ineffective. This raises a salient point about the nature of reinforcing effects: Even if it is assumed that slot machine contingencies produce some type of conditionally reinforcing function on the reels, there is no guarantee that the generated function will act as reinforcement for the frequency of the gambling response itself. Specifically, there is no guarantee that it will reinforce gambling persistence.

Instead, the reinforcement generated on the reels may only serve to reinforce other types of responses such as machine selection or amount bet [25-26,30,32]; or perhaps may only serve a respondent function whereby various conditional emotional responses are elicited [12, 52].

A possible criticism of the present work is that the contingencies in effect across all three experiments’ pre-exposure phases only ever produced a net gain – a fact uncharacteristic of slot machines and casino games in general. However, this was done to maximize the possibility of obtaining a conditionally reinforcing effect; incorporating a net loss would only serve to hinder this outcome. The experimental parameters were chosen to be as favourable as possible to the generation of a conditionally reinforcing effect of near misses. A second criticism concerns the simplistic nature of the stimuli used. Slot machines generally incorporate all manner of varied stimuli; some of which are quite complex whereas the present set of experiments employed only very basic shapes, patterns, and outcomes. However, this too was by design: If near misses do produce a meaningful effect on gambling persistence, then it should be an effect independent of conventional gambling stimuli, which are highly varied and largely arbitrary. In an effort to reduce unsystematic variation, the present studies used basic neutral stimuli and patterns that

would limit pre-existing biases and make learning of the contingencies more straightforward. A final concern is that certain boundary conditions might be necessary to generate the near miss effect. That is, our studies may not have included the specific set of variables needed to produce the near miss effect. While we do not discount this possibility, the strongest positive evidence for near-miss effects on gambling persistence [9, 31] does not make explicit any such boundary conditions. Consequently, there was no a priori reason to believe anything but near misses should need to be manipulated to see a resulting effect on behaviour. Moreover, from our examination of the literature, we were unable to determine any conditions that consistently predicted whether a near-miss effect would emerge.

Near-miss research is a topic that has gained much attention in recent years; however, most of that attention has been directed towards correlational studies or studies examining effects that are presumed to be consistent with prolonging slot machine play, rather than demonstrating this reinforcing effect explicitly. If near misses are truly prolonging gambling behaviour, then why are there not numerous experimental studies that have conclusively and consistently demonstrated this effect? It is profoundly telling that decades after B.F. Skinner first acknowledged the possibility that near misses might be reinforcing gambling responses, there is still no experimental paradigm that seems able to consistently demonstrate this effect. On the other hand, if the near-miss effect on gambling persistence is an elusive phenomenon, or one that is subject to many boundary conditions, then what might account for its staying power in the literature as a major contributor to prolonged gambling?

One possibility may be publication bias resulting from the so-called file-drawer problem which is the tendency for journals and researchers to only publish and cite results that obtain statistical significance – a trend that has been increasing over time across multiple scientific

fields [53-54]. A second may be confirmatory bias, which refers to the tendency to seek and attend to only that information that conforms to an individual’s existing beliefs [55-56]. A third and lesser known problem may involve what has become colloquially known in the social sciences as the “Woozle effect” [57]. Broadly speaking, this is when erroneous or, in the case of the near-miss literature, incomplete information is perpetuated through citations, resulting in a misleading or exaggerated portrayal of a purported phenomenon. In the case of near misses, a good example is illustrated by studies citing Clark et al. [10] which found evidence that near misses enhanced subjective levels of motivation on a two-reel simulated slot machine when participants had personal control of one of the slot machines reel’s outcomes. Citations of this study often omit the fact that the opposite result was obtained when the reels were purely computer selected – a finding that has been replicated [52]. Since the computer-selected trials could be regarded as being more representative of typical slot machine play, the results suggest that, under some conditions, people express less motivation to play following near misses compared to full misses. Nevertheless, the study has been regularly and recently cited as evidence that near misses enhance motivation without qualifying statements [58-60]. As a result, one might infer from a quick reading of the literature that near misses make people gamble more.

However, key facets of the study, such as the subjective nature of the measure or its contrasting condition, show a more nuanced picture of what is reasonable to conclude. The Woozle effect has also occurred in other notable studies of the near-miss effect. For instance, studies citing

Côté et al. [31] have failed to acknowledge the confound of the contingency between near misses and wins which is not present in normal slot machines and could have accounted for their results

[1, 24, 26]. Thus, the positive findings from near-miss studies are often portrayed as being more robust or general than they actually are. We are not implying that the resulting distortions in the

literature are intentional; rather, we see this problem as an emergent property of the imperfect citation system academia relies on. Information often needs to be presented as succinctly as possible, but sometimes this comes at the expense of important detail. Like the degradation of an image that has been compressed too many times, this loss can accumulate as the information is continuously transmitted.

Given that the near-miss effect on gambling persistence was founded on an early and imprecise account of conditional reinforcement [3,61], near-miss research may have been misguided from the very start. Furthermore, some studies − including the present work − seem to do more to challenge the belief that near misses prolong gambling. If near misses lead to prolonged gambling at all, the effect is limited or idiosyncratic [1]. Nevertheless, 66 years after

B.F. Skinner first proposed the idea, adherence to the belief that near-miss outcomes reinforce gambling persistence has remained strong. Our research questions the underlying premise that conditional reinforcement by near-miss stimuli should increase persistence of gambling behavior during extinction.

Acknowledgements

The authors wish to acknowledge the helpful support of lab members Shima Afhami,

Nuha Mahdi, Dr. Neil McMillan, and Vivian Tran as well as the technical support of Isaac Lank and Al Denington. Additionally, the authors would like to thank Dr. Luke Clark for his helpful input and critique of an earlier version of the manuscript.

This research was completed as part of the second author’s Master's thesis in Psychology at the University of Alberta.

References

1. Witts BN, Ghezzi PM, Manson M. Simultaneously observing concurrently-available

schedules as a means to study the near miss event in simulated slot machine gambling.

Psychol Rec. 2015 Mar 1;65(1):115-29. doi:10.1007/s40732-014-0095-y

2. Schüll ND. Addiction by design: Machine gambling in Las Vegas. Princeton, NJ:

Princeton University Press; 2012.

3. Skinner BF. Science and human behavior. New York: The Free Press; 1953. p. 397.

4. Rescorla RA. Pavlovian conditioning: It's not what you think it is. Am Psychol. 1988

Mar;43(3):151. doi:10.1037/0003-066X.43.3.151

5. Lattal KA. Pavlovian conditioning. In: Madden GJ, editor. APA handbook of behavior

analysis, Vol. 1: Translating principles into practice. Washington, DC: American

Psychological Association; 2013. p. 283-306. doi:10.1037/13937-013

6. Schoenfeld WN, Antonitis JJ, Bersh PJ. A preliminary study of training conditions

necessary for secondary reinforcement. J Exp Psychol. 1950 Feb;40(1):40.

doi:10.1037/h0062153

7. Stubbs DA. Second-order schedules and the problem of conditioned reinforcement. J Exp

Anal Behav. 1971 Nov;16(3):289-313. doi:10.1901/jeab.1971.16-289

8. Mazur JE, Fantino E. Choice. In: McSweeney FK, Murphy ES, editors. The Wiley

Blackwell handbook of operant and classical conditioning. Malden, MA: John Wiley &

Sons; 2014 Jun 23. doi:10.1002/9781118468135

9. Kassinove JI, Schare ML. Effects of the" near miss" and the" big win" on persistence at

slot machine gambling. Psychol Addict Behav. 2001 Jun;15(2):155. doi:10.1037/0893-

164x.15.2.155

10. Clark L, Lawrence AJ, Astley-Jones F, Gray N. Gambling near-misses enhance

motivation to gamble and recruit win-related brain circuitry. Neuron. 2009 Feb

12;61(3):481-90. doi:10.1016/j.neuron.2008.12.031

11. Clark L, Liu R, McKavanagh R, Garrett A, Dunn BD, Aitken MR. Learning and affect

following near-miss outcomes in simulated gambling. J Behav Decis Mak. 2013

Dec;26(5):442-50. doi:10.1002/bdm.1774

12. Dixon MJ, MacLaren V, Jarick M, Fugelsang JA, Harrigan KA. The frustrating effects of

just missing the jackpot: Slot machine near-misses trigger large skin conductance

responses, but no post-reinforcement pauses. J Gambl Stud. 2013 Dec 1;29(4):661-74.

doi:10.1007/s10899-012-9333-x

13. Habib R, Dixon MR. Neurobehavioral evidence for the “near-miss” effect in pathological

gamblers. J Exp Anal Behav. 2010 May;93(3):313-28. doi:10.1901/jeab.2010.93-313

14. Scarf D, Miles K, Sloan A, Goulter N, Hegan M, Seid-Fatemi A, et al. Brain cells in the

avian ‘prefrontal cortex’code for features of slot-machine-like gambling. PLoS One. 2011

Jan 25;6(1):e14589. doi:10.1371/journal.pone.0014589

15. Kelleher RT, Gollub LR. A review of positive conditioned reinforcement. J Exp Anal

Behav. 1962 Oct;5(4):543-97. doi. 10.1901/jeab.1962.5-s543

16. Belisle J, Dixon MR. Near misses in slot machine gambling developed through

generalization of total wins. J Gambl Stud. 2016 Jun 1;32(2):689-706.

doi:10.1007/s10899-015-9554-x

17. Daly TE, Tan G, Hely LS, Macaskill AC, Harper DN, Hunt MJ. Slot Machine Near

Wins: Effects on Pause and Sensitivity to Win Ratios. Anal Gambl Behav. 2014 Dec

1;8(2).

18. Urcuioli PJ. Stimulus Control and Stimulus Class Formation. In: Madden GJ, editor.

APA handbook of behavior analysis, Vol. 1: Translating principles into practice.

Washington, DC: American Psychological Association; 2013. p. 361-86.

doi:10.1037/13937-013

19. Sharman S, Aitken MR, Clark L. Dual effects of ‘losses disguised as wins’ and near-

misses in a slot machine game. Int Gambl Stud. 2015 May 4;15(2):212-23.

doi:10.1080/14459795.2015.1020959

20. Ferster CR, Skinner BF. Schedules of reinforcement. New York: Appleton-Century-

Crofts; 1957.

21. Weatherly JN, Phelps BJ. The pitfalls of studying gambling behavior in a laboratory

situation. In: Ghezzi PM, Lyons CA, Dixon MR, Wilson GR, editors. Gambling:

Behavior theory, research, and application. Reno, NV: Context Press; 2006 May 1:105-

26.

22. Peters H, Hunt M, Harper D. An animal model of slot machine gambling: The effect of

structural characteristics on response latency and persistence. J Gambl Stud. 2010 Dec

1;26(4):521-31. doi:10.1007/s10899-010-9183-3

23. Weatherly JN, Derenne A. Rats playing a slot machine: a preliminary attempt at an

animal gambling model. Anal Gambl Behav. 2007 Dec 1;1(2). 79-89.

24. Winstanley CA, Cocker PJ, Rogers RD. Dopamine modulates reward expectancy during

performance of a slot machine task in rats: evidence for a ‘near-miss’ effect.

Neuropsychopharmacology. 2011 Apr;36(5):913. doi:10.1038/npp.2010.230

25. Stagner JP, Case JP, Sticklen MF, Duncan AK, Zentall TR. Do pigeons prefer

alternatives that include near-hit outcomes?. J Exp Psychol Anim Learn Cogn. 2015

Jul;41(3):247. doi:10.1037/xan0000069

26. Fortes I, Case JP, Zentall TR. Pigeons, unlike humans, do not prefer near hits in a slot-

machine-like task. Behav Processes. 2017 May 1;138:67-72.

doi:10.1016/j.beproc.2017.02.012

27. Rice NC, Kyonka EG. Pigeons Choose to Gamble in a Categorical Discrimination Task.

Anal Gambl Behav. 2017;11(1):2.

28. Strickland LH, Grote FW. Temporal presentation of winning symbols and slot-machine

playing. J Exp Psychol. 1967 May;74(1):10. doi:10.1037/h0024511

29. Reid RL. The psychology of the near miss. J Gambl Stud. 1986 Mar 1;2(1):32-9.

doi:10.1007/BF01019932

30. Ghezzi PM, Wilson GR, Porter JC. The near-miss effect in simulated slot machine play.

In: Ghezzi PM, Lyons CA, Dixon MR, Wilson GR, editors. Gambling: Behavior theory,

research, and application. Reno, NV: Context Press; 2006 May 1:155-70.

31. Côté D, Caron A, Aubert J, Desrochers V, Ladouceur R. Near wins prolong gambling on

a video lottery terminal. J Gambl Stud. 2003 Dec 1;19(4):433-8. doi:

10.1023/A:1026384011003

32. MacLin OH, Dixon MR, Daugherty D, Small SL. Using a computer simulation of three

slot machines to investigate a gambler’s preference among varying densities of near-miss

alternatives. Behav Res Methods. 2007 May 1;39(2):237-41. doi:10.3758/BF03193153

33. Kurucz G, Körmendi A. Can we perceive near miss? An empirical study. J Gambl Stud.

2012 Mar 1;28(1):105-11. doi:10.1007/s10899-011-9240-6

34. Tan G, Macaskill AC, Harper DN, Hunt MJ. Derived relations and the slot machine near-

win effect. Eur J Behav Anal. 2015 Jan 2;16(1):21-30. doi:

10.1080/15021149.2015.1065638

35. Sundali JA, Safford AH, Croson RT. The impact of near-miss events on betting behavior:

An examination of casino rapid roulette play. Judgm Decis Mak. 2012;7(6).

36. Lesieur HR, Blume SB. The South Oaks Gambling Screen (SOGS): A new instrument

for the identification of pathological gamblers. Am J Psychiatry. 1987 Sep;144(9).

doi:10.1176/ajp.144.9.1184

37. Logue AW. The living legacy of the Harvard Pigeon Lab: Quantitative analysis in the

wide world. J Exp Anal Behav. 2002 May;77(3):357-66. doi:10.1901/jeab.2002.77-357

38. Stevens JR ,Stephens DW. The adaptive nature of impulsivity. In: Madden GJ, Bickel

WK, editors. The Behavioral and Neurological Science of Discounting. Washington, DC:

American Psychological Association; 2009. p. 361-88.

39. Dixon MR, Marley J, Jacobs EA. Delay discounting by pathological gamblers. J Appl

Behav Anal. 2003 Dec;36(4):449-58. doi:10.1901/jaba.2003.36-449

40. Petry NM, Madden GJ. In: Madden GJ, Bickel WK, editors. Impulsivity: The behavioral

and neurological science of discounting. Washington, DC: American Psychological

Association; 2010. p. 273-294.

41. Reynolds B. A review of delay-discounting research with humans: relations to drug use

and gambling. Behav Pharmacol. 2006 Dec 1;17(8):651-67.

doi:10.1097/FBP.0b013e3280115f99

42. Rachlin H, Safin V, Arfer KB, Yen M. The attraction of gambling. J Exp Anal Behav.

2015 Jan;103(1):260-6. doi:10.1002/jeab.113

43. Schwartz B, Gamzu E. Pavlovian control of operant behavior: An analysis of autoshaping

and its implications for operant conditioning. In: Honig WK, Staddon JER, editors.

Handbook of operant behavior. Englewood Cliffs, NJ: Prentice-Hall; 1977. p. 53-60.

44. Zeiler M. Schedules of reinforcement. In: Honig WK, Staddon JER, editors. Handbook of

operant behavior. Englewood Cliffs, NJ: Prentice-Hall; 1977. p. 202-32.

45. Hurlburt RT, Knapp TJ, Knowles SH. Simulated slot-machine play with concurrent

variable ratio and random ratio schedules of reinforcement. Psychol Rep. 1980

Oct;47(2):635-9. doi:10.2466/pr0.1980.47.2.635

46. Team RC. R: A language and environment for statistical computing. R Foundation for

Statistical Computing, Vienna, Austria. 2018.

47. Cumming G. Understanding the new statistics: Effect sizes, confidence intervals, and

meta-analysis. Routledge; 2013 Jun 19.

48. Morey RD, Rouder JN. BayesFactor: Computation of Bayes Factors for Common

Designs; 2015. R package version 0.9. 2015:12-2.

49. Gollub L. Conditioned reinforcement: Schedule effects. In: Honig WK, Staddon JER,

editors. Handbook of operant behavior. Englewood Cliffs, NJ: Prentice-Hall; 1977. p.

288–312.

50. Davison M, Baum WM. Do conditional reinforcers count?. J Exp Anal Behav. 2006

Nov;86(3):269-83. doi:10.1901/jeab.2006.56-05

51. Shahan TA. Conditioned reinforcement and response strength. J Exp Anal Behav. 2010

Mar;93(2):269-89. doi:10.1901/jeab.2010.93-269

52. Clark L, Crooks B, Clarke R, Aitken MR, Dunn BD. Physiological responses to near-

miss outcomes and personal control during simulated gambling. J Gambl Stud. 2012 Mar

1;28(1):123-37. doi:10.1007/s10899-011-9247-z

53. Rosenthal R. The file drawer problem and tolerance for null results. Psychological

bulletin. 1979 May;86(3):638. doi:10.1037/0033-2909.86.3.638

54. Pautasso M. Worsening file-drawer problem in the abstracts of natural, medical and

social science databases. Scientometrics. 2010 May 5;85(1):193-202.

doi:10.1007/s11192-010-0233-5

55. Wason PC. On the failure to eliminate hypotheses in a conceptual task. Q J Exp Psychol.

1960 Jul;12(3):129-40. doi:10.1080/17470216008416717

56. Mahoney MJ. Publication prejudices: An experimental study of confirmatory bias in the

peer review system. Cognit Ther Res. 1977 Jun 1;1(2):161-75. doi:10.1007/BF01173636

57. Gelles RJ. Violence in the family: A review of research in the seventies. J Marriage Fam.

1980 Nov 1:873-85. doi:10.2307/351830

58. van Timmeren T, Zhutovsky P, van Holst RJ, Goudriaan AE. Connectivity networks in

gambling disorder: a resting-state fMRI study. Int Gambl Stud. 2018 Apr 5:1-7.

doi:10.1080/14459795.2018.1449884

59. Wu Y, Sescousse G, Yu H, Clark L, Li H. Cognitive distortions and gambling near-

misses in Internet Gaming Disorder: A preliminary study. PloS one. 2018 Jan

18;13(1):e0191110. doi:10.1371/journal.pone.0191110

60. Xia J, Wu D, Xu X. Susceptibility to the gambling near-win effect in optimists versus

pessimists: An event-related potential study. Pers Individ Dif. 2018 Jul 15;129:159-65.

doi:10.1016/j.paid.2018.03.032

61. Fantino E. Conditioned reinforcement: Choice and information. In: Honig WK, Staddon

JER, editors. Handbook of operant behavior. Englewood Cliffs, NJ: Prentice-Hall; 1977.

p. 313–39.