Gambling Near-Misses Enhance Motivation to Gamble and Recruit Win-Related Brain Circuitry

Summary “Near-miss” events, where unsuccessful outcomes are proximal to the jackpot, increase gambling propensity and may be associated with the addictiveness of gambling, but little is known about the neurocognitive mechanisms that underlie their potency. Using a simplified slot machine task, we measured behavioral and neural responses to gambling outcomes. Compared to “full-misses,” near-misses were experienced as less pleasant, but increased desire to play. This effect was restricted to trials where the subject had personal control over arranging their gamble. Near-miss outcomes recruited striatal and insula circuitry that also responded to monetary wins; in addition, near-miss-related activity in the rostral anterior cingulate cortex varied as a function of personal control. Insula activity to near-misses correlated with self-report ratings as well as a questionnaire measure of gambling propensity. These data indicate that near-misses invigorate gambling through the anomalous recruitment of reward circuitry, despite the objective lack of monetary reinforcement on these trials.

Analysis of simple main effects looked at the effect of agency at each level of outcome. On nearmiss outcomes, ratings were significantly higher on participant-chosen trials than computerchosen trials (t 39 =4.69, p<.0001). There were no significant differences as a function of agency on wins (t 39 =.931, p=.358) or full-misses (t 39 =1.49, p=.144). On participant-chosen trials, nearmisses were rated significantly higher than full-misses (t 39 =2.66, p=.011). On computer-chosen trials, near-misses were rated significantly lower than full-misses (t 39 =-3.09, p=.004).
A further analysis looked for differential effects of near-miss position, i.e. whether the right-hand reel moved through the payline and stopped in the next position ('type I') or stopped one position short of the payline ('type II'). A 3 (outcome: near miss type I, near miss type II, full-miss) x 2 (control) ANOVA on ratings on 'continue to play' revealed a significant outcome x control interaction (F 2,78=24.5, p<.001), with a highly significant difference between participant-chosen and computer-chosen near-misses on type I trials (t 39 =7.46, p<.001) but no difference on type II trials (t 39 =1.00, p=.323). The equivalent ANOVA for ratings of 'pleased with result' did not indicate differential effects of the two near-miss types (F 2,78 =1.11, p=.336).
Finally, we conducted two sets of analyses to explore the relationship between the subjective ratings on the slot machine task, and level of gambling involvement (see Supplementary Table 1).
First, scores on the South Oaks Gambling Screen varied from 0-5, with 31 subjects scoring 0-1.
The basic analyses of the three sets of ratings were repeated in this restricted group of nongamblers (i.e. excluding those subjects with SOGS scores ≥2), and the statistical findings were unchanged. Thus, the efficacy of the personal control manipulation and the near-miss effect cannot be attributed to a minority of subjects with moderate gambling involvement. Second, we examined correlations between the GRCS (see Supplementary Table 1) and several parameters from the subjective ratings: 1) the effect of personal control ('chances of winning' on participant-chosen trials minus computer-chosen trials), 2) the effect of winning outcomes on 'pleased with result' (wins minus all non-win outcomes), 3) the effect of winning outcomes on 'continue to play' (wins minus all non-win outcomes), 4) the effect of near-misses on 'continue to play' (participant-chosen near-misses minus computer-chosen near-misses). Higher scores on the GRCS were associated with a greater effect of winning outcomes on ratings of 'continue to play' (r 40 =.328, p=.039), supporting the ecological validity of the task. There were no further significant correlations (all r<.12).

Behavioral Data
The number of response omissions during the selection phase was low overall (mean over 180 trials =2.40 sd 1.50) and the number of omissions did not differ between participant-chosen trials (mean 1.00) and computer-chosen trials (mean 1.40) (Wilcoxon signed ranks Z=.782, p=.434), indicating comparable attention across the two conditions. Subjective ratings of "How do you rate your chances of winning?" were compared across participant-chosen and computer-chosen trials using a paired t test. As in Experiment 1, subjects rated their chances of winning as significantly higher on participant-chosen trials (t 14 =5.78, p<.001). Subjective ratings of "How much do you want to continue to play the game?" were analysed using a 3 (outcome) x 2 (control) repeated-measures ANOVA, which yielded a significant main effect of outcome (F 2,28 =15.8, p<.001, Greenhouse-Geisser Epsilon=.666) and a significant outcome x control interaction (F 2,28 =8.19, p=.002). Simple main effects analysis indicated that participant-selected wins increased ratings more than computer-selected wins (t 14 =3.72, p=.002). There were no significant differences between near-misses and full-misses in the participant-chosen trials (t 14 =1.52, p=.150) or the computer-chosen trials (t 14 =1.38, p=.190).
However, inspection of mean ratings (see Supplementary Table 3) indicated the same trend as the behavioral experiment. When the participant-chosen comparison was restricted to near-misses that had passed through the payline, the near-misses were rated as significantly more motivating (mean z=+.01) than full-misses (mean z=-0.19) (t 14 =2.26, p=.040).
To summarise the ratings data, the manipulation of personal control was effective in the fMRI procedure, as evidenced by i) increased ratings of 'chances of winning' on participant-chosen trials, and ii) increased ratings of 'continue to play' after a win on participant-chosen trials. The near-miss effect was evident in a significant outcome x control interaction on ratings of 'continue to play', and in a paired t test comparing near-misses below the payline against full-misses. The attenuated effect sizes relative to the behavioral experiment are likely attributable to the smaller number of subjects and the intermittent nature of the ratings (1 in 3 trials).

FMRI Data
Contrast 1 (Win-related activity at outcome): see Supplementary Table 4.
Contrast 4 (Wins as a function of Control, at outcome): no supra-threshold voxels.
Contrast 5 (Selection phase activity): see Supplementary Table 5. Icon selection on participantchosen trials, compared to computer-chosen trials, was associated with significant activation across a distributed network comprising cerebellum, bilateral parietal cortex, bilateral premotor cortex, thalamus, striatum and cingulate gyrus.
Contrast 6 (Win-related activity during anticipation phase): no significant voxels.
Contrast 7 (Near-miss related activity during anticipation phase): no significant voxels. * The GRCS is a 23-item self-report questionnaire with each item rated on a scale from 1 (strongly disagree) to 7 (strongly agree). Example items on GRCS: