Of the 132 participants who had completed the PGSI, there were 26 non-problem gamblers, 63 low-risk gamblers (score of 1–4 on the PGSI), 19 moderate-risk gamblers and 24 Problem Gamblers (scoring 8 or greater on the PGSI). Seven participants withdrew from the study prior to completing all of the subjective measures leaving a sample of 125 participants for the subjective measures.
Erroneous Cognitions Specific to Slot Machine Features
Near-miss Erroneous Cognitions Frequencies pertaining to how players responded to “Near-misses reflect my skill at this slots game, and that I was close to winning” are shown in Table 1 and “Near-misses indicate that a win is imminent” are also shown in Table 1. Based on these frequencies, a small but meaningful percentage of players endorse these beliefs (16% endorsed the belief that near-misses reflected skill at slots, and just over 11% believed that near-misses were a harbinger of upcoming wins). These beliefs appeared to coincide with one another—scores on the near-miss as skill item were correlated with scores related to near-misses predicting wins (r (123) = .792, p < .001).
Table 1 (a) Frequencies of different responses to “Near-misses reflect my skill at this slots game and indicate that I was close to winning.” (b) Frequencies of different responses to “Near-misses indicate that a win is imminent”
Crucially, the near-miss as a skill item was correlated with PGSI scores (r (123) = .193, p = .031), as was the near-miss win imminence item (r (123) = .209, p = .02). Each item was also significantly related to the subscales of the GRCS as shown in Table 2.
Table 2 Correlations between the near-miss as skill item, near-miss as a harbinger of wins item, stop button as a harbinger of wins item, and stop button composite score with subscales on the Gambling Related Cognitions Scale (GRCS)
Stop Button Erroneous Cognitions To assess the prevalence of erroneous beliefs about the stop button feature being a skill-device, we tabulated how often players endorsed the statement “Using the stop button made wins more likely.” Out of the 125 responses 43 (34.4%) strongly disagreed with the statement, 46 (36.8%) disagreed with this statement, and 19 (15.2%) neither agreed or disagreed. By contrast, a concerning total of 15 participants (12%) agreed with this statement, and 2 (1.6%) participants strongly agreed with this statement.
We also calculated a composite score based on the items from Ladouceur and Sévigny (2005), designed to assess whether participants held erroneous beliefs about the stop button. This composite score was correlated with the degree to which they endorsed the statement “Using the stop button made wins more likely” (r = .667, p < .001).
This composite score was also significantly correlated with PGSI scores (r = .240, p = .007), indicating a relationship between problem gambling status and erroneous beliefs pertaining to the stop button. Composite scores were also significantly correlated with the illusion of control, predictive control and interpretive bias scales of the GRCS (See Table 2). Further, both the single item and the composite items were unrelated to gambling expectancies, and only weakly related to cognitions about the inability to stop gambling.
Thirty-four of the 125 respondents preferred the game with the stop button feature. Independent t-tests revealed that those who preferred the stop button more strongly endorsed the idea that the stop button made wins more likely (M = 1.71) than those who preferred the no stop button game (M = .87), t(123) = 4.194, p < .001. As the question format was dichotomous, the remaining 91 respondents preferred the game without the stop button.
In-Game Physiological Measures: Data Reduction and Analysis Strategy
Of the 132 participants, seven dropped out prior to completing both slot machine tasks. One participant’s force data and one participant’s SCR data was not recorded leaving samples of 124 for force, 124 for SCRs and 125 for PRPs. In analyzing our physiological measures, we partitioned our analysis in two ways. Firstly, we performed separate repeated measures analyses of variance (ANOVAs) for skin conductance (SCRs), force, and post-reinforcement pauses. The repeated factors were outcome type (losses, gramophone near-misses, LDWs, small regular wins, gramophone wins and stereo triplet wins and the stereo triplet near-misses) and button use (stop button, no stop button). We then conducted a more restricted version of the analyses above in which the outcomes measured were limited to the three types of losing outcomes that result in no credit gains. Post-hoc analyses were conducted using Fisher’s LSD procedure with alpha set at .01. Violations of sphericity were tested using Mauchly’s test, and in the event of significant violations we applied Greenhouse-Geisser corrections to the degrees of freedom which are reported in the analyses below. We made an a priori decision not to look at gambling status effects because no effects of gambling status had been shown to influence these in-game measures in our previous studies. For all repeated measures analyses effect sizes were measured using partial eta squared (reported as η
2).
Skin Conductance Responses A repeated measures ANOVA revealed a significant main effect of button use, F(1, 123) = 6.550, p = .012, (η
2 = .051). Overall, SCR magnitudes were on average greater for participants when they used the stop button (M = .163) compared to when they played without the stop button (M = .147). A significant effect of outcome was also observed, F(1.8, 221.424) = 37.863, p < .001 (η
2 = .235), but there was no stop button use by outcome interaction (F (2.48, 305.394) = 1.006, p = .380). As shown in Fig. 4, the main effect of outcome was caused by the greater magnitude of SCRs following stereo triplet wins, and following stereo triplet near-misses. Fisher’s LSD post hoc analyses indicated that, stereo triplet wins triggered greater SCRs than all the other outcomes (all p values <.001). Stereo triplet near-misses triggered larger SCRs than regular losses, gramophone near-misses, LDWs, small wins and gramophone wins (all p values <.001). The lack of interaction indicates that using the button did not preferentially inflate the SCRs on any particular outcome.
As can be seen in the left panel of Fig. 4, the five outcomes on the left side of the graph (losses, gramophone near-misses, LDWs, small wins and gramophone wins) triggered small SCRs that were relatively similar in magnitude. Among these outcomes the only statistical difference was that (inexplicably) gramophone near-misses had significantly smaller SCRs than losses—none of the other contrasts were significant.
When analyzing SCRs for outcomes that resulted in no credit gain (losses, gramophone near-misses, stereo triplet near-misses), there was a main effect of button use, F(1, 123) = 4.951, p = .028, (η
2 = .039) and a main effect of outcome, F(1.143, 140.627) = 28.695, p ≤ .001 (η
2 = .189). There was also a significant outcome by button use interaction, F(1.752, 215.528) = 4.699, p = .013 (η
2 = .037). Average SCRs following losing outcomes are displayed in the right panel of Fig. 4. The main effect of outcome is attributable to the greater magnitude of SCRs for stereo triplet near-misses, compared to gramophone near-misses or regular losses. For the no stop button condition, there was a large main effect of outcome F(1.223, 152.884) = 29.837, p < .001, (η
2 = .193) with stereo near-misses having more robust SCRs than losses (p < .001) and gramophone near-misses (p < .001). Similarly, in the stop button condition the effect of outcome was also highly significant, F(1.345, 166.797) = 14.738, p < .001 (η
2 = .106) with stereo triplet near-misses significantly larger than either gramophone near-misses (p < .001) or losses (p < .001) in the stop button condition. As can be seen in the right panel of Fig. 4 (and the effect sizes for the simple main effects reported above), the interaction was caused by the larger differences between the stereo triplet near-misses and regular losses and gramophone near-misses in the no stop button condition.
Force Applied to the Spin-Button A repeated measures ANOVA revealed a significant main effect of button use F(1, 123) = 21.39, p < .001, (η
2 = .148). Greater force was applied to the spin-button in the stop button condition (M = .416) compared to the no stop button condition (M = .333). A main effect of outcome was also observed, F(3.68, 453.040) = 16.468, p < .001 (η
2 = .118), however, there was no button by outcome interaction, F(3.217, 402.323) = 1.494, p = .212). Fisher’s LSD Post-hoc comparisons indicated that players elicited greater magnitudes of force following LDWs compared to regular losses (p = .001) as well as gramophone near-misses (p = .004). Stereo wins triggered more force than any other outcome (largest p = .008). Crucially, as shown in the left panel of Fig. 5, stereo triplet near-misses triggered more force than LDWs (p = .002), gramophone near-misses (p < .001) and regular losses (p < .001), and equivalent amounts of force as small wins, and gramophone wins.
When restricting the analysis to just the losing outcomes (losses, gramophone near-misses and stereo triplet near-misses), a main effect of button F(1, 123) = 20.388, p < .001 (η
2 = .142), and a main effect of outcome, F(1.423, 175.029) = 22.532, p < .001, (η
2 = .155) were revealed. However, there was no button by outcome interaction. The main effect of button use was attributed to players applying more force to initiate the spin button in the stop button condition (M = .410) compared to the no stop button condition (M = .325). As illustrated in the right panel of Fig. 5, the main effect of outcome was attributable to players initiating the next spin with greater force after stereo triplet near-misses than after gramophone near-misses (p < .001), or after regular losses (p < .001). Gramophone near-misses were not significantly different from regular losses (p = .942).
Post-reinforcement Pauses The repeated measure ANOVA for PRPs observed no significant main effects of stop button use F(1, 124) = .937, p = .335. However, there was a significant main effect of outcome, F(1.286, 159.458) = 455.039, p < .001 (η
2 = .786), while the button by outcome interaction fell just short of significance following a Greenhouse Geisser correction F(2.058, 255.186) = 2.82, p = .06. The main effect of outcome is shown in the left panel of Fig. 6. Post-hoc analyses showed that all outcomes were different from one another (all p values < .001) except the losses, the gramophone near-misses and the stereo triplet near misses.
The restricted analysis of PRPs for the three types of full losses (losses, gramophone near-misses, stereo near-misses), showed that the main effects of both button use and outcome were not significant. Crucially however, there was a significant button use by outcome interaction, F(1.724, 213.746) = 5.118, p = .01, (η
2 = .040). Simple effect analyses revealed that there was a main effect of outcome only in the stop button condition, F(1.379, 172.385) = 6.245, p = .007, (η
2 = .048) and not the no stop button condition. As shown in the right panel of Fig. 6, this interaction appears to be attributed to higher PRPs following stereo near-misses, but only when players were using the stop button. Fisher’s LSD post hoc comparisons revealed that while losses and gramophone near-misses did not statistically differ (p = .820), the stereo near-misses had longer PRPs than both losses (p = .008) or gramophone near-misses (p = .011) in this stop button condition.