In the Diagnostic and Statistical Manual of Mental Disorders (DSM), gambling disorder refers to a persistent and recurrent problematic gambling behavior that disrupts several spheres of an individual’s life [1]. Whereas the dichotomous approach of the DSM may be useful in a clinical setting, it can easily lead to problematic subclinical behaviors not being considered. From a prevention standpoint, a continuous conceptualization of problem gambling is more useful. While cutoffs and nomenclature vary between existing screening tools, they mostly overlap with “Recreational” or “Non-problem” on the lower bound and “Pathological” reserved for severe problem gambling meeting the clinical threshold [2-8]. In between are the designations “At risk” and “Problem gambling”. The former regroups people experiencing some adverse consequences from their gambling habits and exhibiting some signs of impaired control. The latter designates people experiencing significant harm from or impaired control over their gambling habits, but at a subclinical level [8]. Finally, recent developments in the literature have begun to shift the focus from counting people afflicted by them to counting harms (see the Conceptual Framework of Harmful Gambling for more information [9]). This framework allows for a more granular study of the impacts of gambling, notably the recognition that relatively small harms can be significant when experienced by a large population.

Data from a recent meta-analysis indicates that the worldwide prevalence in the adult population for any risk gambling (ie, anything higher than recreative or no problem gambling) is 8.7% and is 1.41% for problem gambling [10,11]. Problem gambling is associated with a range of negative effects encompassing multiple domains. For example, Cowlishaw and Kessler [12] found associations between problem gambling and poorer outcomes on mental health, psychosocial, and financial indicators. These effects are largely confirmed in a recent systematic review by Allami et al [13]. More generally, recent frameworks of gambling harms have been developed to classify and itemize different gambling harms across dimensions and temporal spans [14], among which the taxonomy of harms by Langham et al [15] is considered especially thorough. These frameworks of gambling harms are especially useful to show that harms are not limited to the individual gambler but can also affect their immediate surroundings and society at large. Furthermore, harms are not limited to pathological gamblers and can persist even after complete abstinence from gambling. To limit and prevent these harms, a wide range of measures grouped under the term “responsible gambling” (RG) have been put in place (see the study by Blaszczynski et al [16] for a detailed definition of the RG concept and see the study by Williams et al [17] for a review of different RG measures).

Among RG measures are pop-up messages. In the context of RG, pop-ups, dynamic warnings, or on-screen messages are presented during gambling activities to prevent or reduce gambling-related harm [18-20]. A common strategy used when designing such messages is providing factually accurate information used to correct cognitive distortions and irrational beliefs about gambling (eg, “The way you play will not increase your chances of winning” [21]). Other pop-up messages prompt gamblers to set limits on time or money allocated to gambling and then reappear when those limits are reached (eg, “You have reached your preset limit of 20 credits. You still have 60 credits in the slot machine…” [22]). Pop-up messages have also been used to remind gamblers about their behavior in-play (“You have now played 1000 slot games. Do you want to continue?” [23]) with the rationale that this combats compulsive play and dissociative-like state [24-27]. This strategy is sometimes coupled with normative feedback (eg, “…Only a few people play more than 1000 slot games” [28]). Finally, pop-up messages have used self-appraisal wording to foster self-awareness in gamblers by inducing a deeper and more personal processing of the information (eg, “Have you spent more than you can afford?” [29]).

The effectiveness of pop-up messages is still debated in the literature. On one hand, some results suggest that pop-up messages can be an effective and minimally invasive strategy for introducing responsible gambling information and reinforcing responsible gambling behaviors [18-20]. They also have the potential to reach a wide variety of gamblers at minimal cost and can be tailored to better fit their receiver’s characteristics (eg, age, sex, etc) and the context in which they are presented (eg, if the message is presented during a winning or a losing streak). Furthermore, the interest in pop-up messages is heightened by the fact that they serve as a prevention strategy for types of gambling regarded as the most damaging from a public health perspective, namely electronic gambling machines (EGMs) and online gambling [10,30-33]. On the other hand, numerous methodological limits affect the RG pop-up messages literature, including the paucity of replication studies for a same message parametrization [18-20], use of self-reported data instead of objective observations (eg, [29,34]), limited sample sizes (eg, [35]) and ecological validity flaws (eg, [36-38]). Thus, it is evident that more data is necessary to evaluate the effectiveness of RG pop-up messages and their potential moderators.

While the literature is not settled on the best design for effective RG pop-up messages, some elements are more supported than others. To synthesize available data in a useful manner, it is pertinent to start with an overarching model. For this paper, we based our strategy on Comprehensive Messaging Strategy For Sustained Behavior Change (CCSSBC) [39], a multitheoretical persuasive communication strategy combining both message tailoring strategy of the Transtheoretical Model (TTM) [40] and message framing strategy of the Self-Determination Theory (SDT) [41,42]. “Tailoring” is the adaptation of messages’ content to better fit the characteristics (eg, sociodemographics, psychological, etc) of its target individuals or populations [39]. On the other hand, “framing” refers to the way a message is conveyed, not its content per se [39].

The TTM describes the process of change in human behavior and proposes tailoring interventions to the stage of change of the receiver [40]. In their review, Pope et al [39] build on this model and propose 4 stages: detection (characterized by unawareness of a problem or failure, willfully or not, to acknowledge said problem); decision (characterized by acknowledging a problem exists, which leads to a state of uneasiness, and therefore higher receptiveness to information on potential solutions or reliefs); implementation (where the intentions about the problem are formed but have still not been enacted); and maintenance (where behavior is produced and the challenge is now about sustaining in the long term). Tailoring RG messages to gamblers’ stage of change would require some form of evaluation that would also need to be periodically updated. This could be done by a questionnaire via a gambling website or tied to an identity card required to play EGMs (eg, [43,44] for the use of a loyalty card to track behavior on EGMs). While this kind of evaluation would be desirable, it is not actually done by most, if any, gambling website, even less for offline play. Therefore, the next best option would be to tailor messages to the stage of change of gamblers that would benefit the most from it: gamblers at the detection stage. As said earlier, RG pop-up messages are a low-level prevention tool pushed to gamblers while in play, without any action needed on their part to receive them. This makes RG pop-up messages ideal for those who are the most oblivious to their own gambling behavior and are the least inclined to search for protective strategies by themselves. Gamblers at other stages of change could still benefit from these messages even though they are less tailored to them. This (presumably) reduced effectiveness is thought to be offset by the increased awareness of gamblers at higher stages of change, which in turn should make them more autonomous in protecting themselves and less reliant on the “imposed” protection of RG pop-up messages.

SDT describes how people regulate and act according to their social environment, goals, and motives [41]. Broadly speaking, both goals (what behavior is done) and motives (why said behavior is done) can be intrinsic (comes from the self, for their own sake) or extrinsic (determined by outside influences, like shame from peers). In their review, Pope et al [39] show that messages fostering both intrinsic goals and motives lead to better outcomes by almost every metric (ie, increased well-being and quality of life, less stress and anxiety, deeper processing of information, more commitment to positive change, and sustained positive behavior, etc). Whatever stage of change, messages should always strive to foster intrinsic-oriented goals and motives. This is readily compatible with the inclusion of self-appraisal statements in RG messages, which are the most promising feature in the literature for this specific prevention tool [28,29,45-47]. At the detection phase, Pope et al [39] suggest that messages should “(a) identify intrinsic risks or negative consequences associated with avoiding the behavior; (b) introduce small feasible options that serve as solutions to the problem; (c) provide a self-determined rationale; and finally, (d) use images/titles that reflect intrinsic goals, such as fitness and well-being, as opposed to appearance, social comparisons, or appealing to others.”

Apart from the “content” portion of effective RG pop-up messages are the “container” features. These features refer to strategies aimed at better capturing the attention and maximizing readability and easiness of understanding. First is the necessity to periodically disrupt the game by forcing a pause. An explication for why EGMs are so much linked to problem gambling is their capacity to induce dissociative-like state, which makes gamblers lose track of time and money spent [24-27]. Therefore, a good pop-up message should be disruptive enough to break this mental state and give enough time to gamblers to think more rationally about their behavior. This is exemplified by Williams et al [17], who argue that smoking bans in public spaces could have been one of the more effective RG measures, because it forces gamblers, who are predominantly smokers, to take pauses outside gambling venues. Having a minimum duration between the appearance of the RG pop-up message and being allowed to continue is a common feature in the literature. Sometimes, the forced pause was as effective as the pause plus RG message condition to reduce gambling behavior [21,48]. The duration of the pop-up message should be sufficient, so gamblers have enough time to read and reflect on them, but still not be so long that they become excessively obstructing. Similarly, presentation frequency should be sufficient so that gamblers are exposed to them but not so much as they become harassing. There is much variation on these two parameters in the literature. On the higher end are studies like Ladouceur and Sevigny [48] with a 7s message each 15 games and on the lower end are studies, like Auer et al [23] or Schellink and Schrans [49], with a single message after a continuous hour of play and no minimum duration. A common observation by researchers was that few participants were exposed to their RG pop-up messages (eg, [23,28,49]). Schrans et al [50] tried a 30-minute presentation frequency that had the benefit of elevating the proportion of gamblers exposed to the RG pop-up messages but failed at lowering gambling behavior more than the 1-hour presentation frequency.

RG pop-up placement should be such that they are readily noticeable (other than the fact that they block the gambling activity). Gainsbury et al [34] compared the performance of RG pop-up messages presented in the center versus on the periphery of EGMs’ screens. Pop-up messages at the center of the screen were more noticeable and remembered by participants and were perceived as more useful. Participants exposed to the centered pop-up messages also reported gambling less than those exposed to the peripheral ones.

The combination of a signal word [51] and a pictogram [51-53] can be useful to capture attention. Data on the subject mainly comes from the products safety literature where physical injuries are the main concerns. Because gambling deals with psychological and social harms which can nevertheless be devastating, the lower-level signal word “Notice” (prescribed by ANSI for practices not related to physical injury [54]) could be too tame to convey the proper level of danger. We argue that higher-level signal word “Caution” (prescribed by ANSI for hazardous situations which, if not avoided, could result in minor or moderate injury [54]) is more adequate for this purpose.

To improve readability, messages should be designed with an appropriate text-background contrast, font, font size, letter case, and text disposition [51,55]. Because people’s perceptual abilities can vary, it is preferable to design messages to maximize the number of people that can read them easily. Messages accessible for disadvantaged people can still be easily read by the regular population, while the reverse is not necessarily true [56]. Good text-background makes the message stand out [51,55] and improves information processing [57]. Black characters on a white background are considered easier to read than white characters on a black background [57].

There is no consensus in the literature regarding which of serif or sans-serif fonts are optimal for readability [58-61]. However, a more recent study by Rello and Baeza-Yates [62] found that sans-serif fonts were easier to read for people with dyslexia. Current ANSI norms also recommend sans-serif fonts [54].

Font size should be optimized so that under normal reading conditions the message is entirely contained in one’s field of vision [60,61]. Because readability tends to increase with font size [60,61,63], selected font size should err on the higher side of the recommended size [54].

Messages should be written with a mix of capital and lower-case letters. The general rule is that lower-case letters are more distinguishable from one another than capital letters [55]. Capital letters are more useful to mark the beginning of phrases or statements in the messages, or to emphasize the signal word in the message’s header [54].

Wogalter et al [51] recommend segmenting messages by units of meaning written in a simpler telegraphic style and disposed of in a point-form manner. Text should be left-aligned. This disposition allows for an easier identification of each “argument” of the message and improves memorization. The ANSI proposes using bullet points to highlight even more each message segments [54].

Finally, designing effective messages necessitates accounting for habituation. Indeed, repeated exposure to prevention messages has the potential to lower their effectiveness because people pay less attention to them [64]. Hitchman et al. [64] suggest that an optimal strategy is to circulate different messages at the same time and periodically change the whole set of messages for a new one.

Ecological validity poses a more complex problem than parametrization of the “ideal” prevention pop-up message. Testing the effectiveness of RG pop-up messages in a natural environment poses major ethical, organizational, and financial challenges, mainly related to the need to reach an agreement with a gambling operator. Testing in a natural environment makes inevitable some sort of collaboration with a gambling operator to access their infrastructure and player base. This places the researchers at the mercy of the gambling operator who can interfere with the prevention measure tested or the research design itself. For example, in their study testing the effect of RG messages on EGMs in real gambling venues, Gainsbury et al [29] were asked by one of the venues’ owners to reduce the presentation from 15 seconds four times per hour to 10 seconds once per hour. Even without such interference, these sorts of arrangements can incentivize researchers to self-censor to maintain access to their partnered gambling operator’ infrastructure for future studies [65,66]. Finally, there are technical difficulties regarding gambling operators’ proprietary software that can limit the possibilities of alterations to insert RG pop-up messages. For the aforementioned reasons, it is generally more practical to test RG measures in a laboratory setting, even if this approach comes with its own set of problems related to ecological validity. Indeed, recent systematic reviews evaluating the effect of RG pop-up messages show that most studies have been conducted in a laboratory environment, whereas only a few were conducted in a natural setting [18,19].

In laboratory settings, volunteers are typically invited to gamble by wagering virtual credits (eg, [47,67]) or their monetary compensation received for their participation in the study (eg, [21,35,36,48,68]). Their behavior in that setting is then deemed to resemble that of gamblers in a real gambling environment. Surprisingly, this proposition has received very little scientific verification. Data suggests that gamblers would be more emotionally engaged in play in a real gambling environment [69,70] but would take more risks in a laboratory setting [71]. These results could be explained in part by the fact that laboratory research in the field of gambling neglects a fundamental aspect of its definition: the wagering of money or an object of value, a wager which, once placed, cannot be taken back [72]. Indeed, a constant of laboratory studies is both the absence of monetary risk and low possible prizes. In the best of cases, wagering a certain amount of money provided upfront only risks participants losing something that was not in their possession before participating in the study. This is, by definition, enforcing some kind of preset limits as it is not possible for participants to bet money they cannot afford to lose. In cases where only virtual credits are gambled, the disconnect with risk is even greater. Prize money, for its part, is never really addressed in laboratory studies. Maximum winnable prizes are often quite low (eg, AU $40 [47], some chances to win a 19-inch TV [36], etc) compared to what is theoretically possible in a real gambling environment (eg, in Quebec, Canada, maximum jackpot per game on video lottery terminals, a subtype of EGMs, is CA $1000 [73]).

The absence or reduction of participants’ sense of risk while gambling in a laboratory setting could reduce people’s gambling inhibitions. This raises doubts about the seriousness with which they would take any warning message about the consequences of their actions. Indeed, without any serious motivation to gamble (eg, desire to win big, desire to make up for a major loss, etc), it is quite possible that participants would be more intensely subject to what Orne [74] calls “demand characteristics of the experimental situation,” which are “…the totality of cues which convey an experimental hypothesis to the [participant and] become significant determinants of [participants’] behavior...” Participants exposed to RG measures in a laboratory setting could understand that they are being encouraged to moderate their gambling by the researchers. Participants could then modify their behavior mainly to either confirm or contradict the researchers’ hypotheses. This bias could artificially inflate the effectiveness of certain RG measures when tested in the laboratory. These RG measures may subsequently fail to fulfill their protective role when deployed in the field.

Apart from the sense of risk, most laboratory studies on RG prevention messages were conducted in a significantly lesser immersive or restrictive environment. For example, in the study by McGivern et al [38], participants were proposed a computerized gambling task. In the study by Cloutier et al [21] and Ladouceur and Sevigny [48], participants gambled on real EGMs in a properly decorated gambling environment, but the gambling session was limited to how long it took to respectively spend CA $20 and CA $10. At this point, it is worth noting that Bjørseth et al [18] did not find a significant moderation effect of setting (real gambling setting vs laboratory). However, this absence of effect could as easily be explained by the extreme unexplained heterogeneity between studies. More studies are needed to determine the source of this heterogeneity.

This study addresses the aforementioned ecological validity concerns in multiple ways (see “Duplicity” and “Gambling session in the laboratory” subsections of the “Methods” section for details). First, the study’ environment replicates as closely as possible a real gambling setting. This is done by using real EGMs and giving attention to decor and ambiance (eg, lighting, music, etc). Second, participants are allowed to behave in a more natural and unrestricted way. For example, they can take pauses from gambling without ending the study, they are not informed about the maximum duration that the protocol allows for gambling, and they are told they can gamble as much as they want. Finally, and most importantly, efforts are done to both elevate the perception of personal risk and provide a comparable incentive to gamble as a real gambling setting. The elevation of risk is done by use of duplicity to make participants believe they are betting their own money. Participants are also provided with a way to withdraw more money from their bank account to prevent situations where they would end their gambling session, not because they chose to, but because they did not bring enough money with them. An incentive to gamble is done by offering a comparable jackpot to the one offered in local gambling venues.

This study aims to test (1) the potential of pop-up messages as a prevention measure in a gambling setting and (2) whether this potential is moderated by characteristics of people exposed to the messages. Secondary aims also tackle some fundamental assumptions of gambling studies conducted in a laboratory setting.

Specifically, the main objectives are to test to what extent prevention pop-up messages affect participants:

Secondary objectives are:

This study was approved by the Comités d'éthique de la recherche avec des êtres humains de l'Université Laval (CÉRUL; reference 2020-076 A-1 / 27-05-2024). Preliminary consent is obtained verbally during the intake interview and confirmed with a signature just before the gambling session in the laboratory. Some aspects of the study are withheld from the participants and are only revealed during debriefing after the gambling session has ended (see "Duplicity" and "Gambling session in the laboratory" subsections of the "Methods" section for details). After debriefing, participants are asked to confirm their consent one last time. Participants can remove consent at any point of the study without prejudice. After debriefing, participants who wish to quit without completing the study can ask that all their data be destroyed. Compensation for participation can vary between CA $30 and CA $500 depending on the results of the gambling session, which is completely random (see "Presentation of the Study and Verbal Consent" and "Duplicity" subsections of the "Methods" section for details). Some transportation expenses were also reimbursed on top of the compensation, namely transit fare, parking validation at the university, and taxi for the elderlies and people with mobility impairments. All data is kept in secured servers (files) and locked drawers (physical forms). Participants' data is anonymized with a nonsequential ID number.

This study exposes participants to potentially stressful situations. Notably, by gambling with their own money, participants risk gambling more than what they can afford and find themselves in a situation where they think they have lost a lot of money. This stress is a byproduct of the duplicity required to ensure maximal ecological validity of the study. It is mitigated by the fact that any potential stress is limited to a maximum of 2 h (the time limit for the study), after which debriefing ensues and they learn their “lost money” will be returned to them.

The use of duplicity poses some concerns over the importance of free and informed consent. While participants initially enroll in the study without knowing all the parameters, they have the option to retroactively remove their consent and ask for the destruction of their data during the debriefing. Some steps are also taken to minimize aspects of the study that are masked by duplicity. For example, even though losses are reimbursed to participants, gains are still paid as advertised. This prevents a potential situation where a participant who wins a lot of money and expects to leave with it is instead given a standardized (and lesser) amount.

It is also worth noting that this study does not expose participants to higher levels of risk and stress than what they already frequently experience in their own lives. Inclusion criteria target participants who regularly gamble on EGMs in their everyday lives. In the context of the study, participants are provided with the same incentives to gamble, but nothing is done to exacerbate these behaviors (eg, altering the EGMs’ payout rate, offering loans to gamble, etc). On the contrary, half the participants are provided with prevention messages, and any money lost while gambling is reimbursed at the end of the session. Also, because this study is conducted in a psychology laboratory, people on site can provide crisis support should a participant feel unwell after a significant loss. This kind of support is not usually available where EGMs are situated. Finally, we chose not to recruit the most vulnerable gamblers (ie, people trying to reach or maintain abstinence). Despite those protections, the ethics committee felt that probable pathological gamblers should also be excluded from the study.

This is a 2-arm stratified block randomized controlled study. Participants are randomized to one of the two arms in a 1:1 ratio:

There are 3 assessment phases in this study (see Figure 1). Recruitment began in February 2024 and concluded in December 2024. Study completion is expected in February 2026. Study outcomes will be reported in line with the Consolidated Standards of Reporting Trials guidelines and the extension for social and psychological interventions trials (CONSORT-SPI) [75].

Using G*Power, we estimated the required sample size for an α level of 5% and ß level of 80% for a 2-tailed t test for a range of effect sizes (see Table 1). This was done both for testing differences in outcomes between 2 independent groups and for repeated measures.

A recent meta-analysis by Bjørseth et al [18] on the effects of RG pop-up messages on gambling behaviors and cognitions reports effect sizes respectively of g=0.62 (95% CI 0.36-0.87 and g 0.58 (95% CI 0.33‐0.83) when adjusting for publication bias. Based on those values, the total sample size would be between 86 and 96. Note that sample size requirements for 1-way analysis of covariances (ANCOVAs) and repeated measures analyses would be equal to or smaller. Considering the potential difficulty in recruiting participants accepting to play with their own money, we aimed at a sample size of 80. Therefore, we can expect a power range from 72.34% to 77.66% with effect sizes comparable to what is reported by Bjørseth et al [18].

We aim to recruit 80 participants in total. While many may refuse to participate when presented with the study’s specifics, we do not anticipate a significant number of dropouts after consent is obtained. Textbox 1 shows the inclusion and exclusion criteria.

The rationale for the age cut-off is that it is the legal age to gamble where the study is conducted (Quebec, Canada). The literacy requirement is justified by the fact that this study is conducted in French and requires being able to read and understand simple texts in that language. Questions are also asked and answered in French.

The age cut-off is the legal age to gamble where the study is conducted (Quebec, Canada).

The rationale for a floor limit on play frequency is that prevention pop-up messages are expected to produce a small effect that becomes more impactful when cumulated over multiple sessions. Therefore, these messages are thought to be more suited to people who gamble on a regular basis. Furthermore, regular EGM gamblers are more likely to feel compelled by a message warning about the dangers of excessive gambling than people who gamble very sporadically.

The rationale for a floor limit on play frequency is that prevention pop-up messages are expected to produce a small effect that becomes more impactful when cumulated over multiple sessions. Therefore, these messages are thought to be more suited to people who gamble on a regular basis. Furthermore, regular EGM gamblers are more likely to feel compelled by a message warning about the dangers of excessive gambling than people who gamble very sporadically.

The rationale for the exclusion criteria is ethical considerations. The first criterion is mandated by the ethics committee to prevent possible adverse effects on the most vulnerable gamblers. The other 2 criteria aim not to interfere with someone’s recovery process or wishes to stop gambling.

From February 2024 and on, the study was advertised online via social media, our research team’s website, and our university’s mailing list. An email was also sent to participants from a previous study who agreed to be contacted again. The study was advertised offline via ads in a local newspaper 2 times in February 2024. Recruitment posters were placed in gambling venues in the same city of our university.

To boost recruitment, we started a second recruitment wave during the summer of 2024. This wave mainly used online ads targeted at people geolocated at gambling venues. Inclusion criteria were also relaxed on minimal gambling frequency (see the “Participants and Eligibility Criteria” section).

This condition is the same as the active control group with the addition of prevention pop-up messages delivered during the gambling session on a fixed presentation schedule.

The participant is brought in a separate office after the end of the gambling session for a series of questionnaires (outcomes listed in “Secondary Outcome” and “Other Outcomes” sections) which are, in order:

These questionnaires are comprised of Likert scales, multiple-choice questions, and open-ended questions. Contrary to more standard qualitative designs, open-ended question responses are not recorded in an audio format, nor are they transcribed verbatim [82]. Instead, the interviewer takes notes, which the participant can see on a computer screen. Participants are then asked to confirm that the interviewer’s notes accurately reflect their opinion. It has been suggested that note-taking is superior to audio recording for obtaining qualitative data [83].

Debriefing about the true goals of the study and final validation of consent to participate in the study are done between the evaluation of participants’ perception of the realism and the evaluation of participants’ ability to recall prevention pop-up messages.

Table 3 shows the primary outcome measures.

Sociodemographic variables, past 12 months of gambling-related variables, and PGSI category will be presented in a table with appropriate central tendency estimates. This table will be divided by assignation group.

A total of 3 statistical models will be used to analyze the outcomes presented in Tables 3 and 4. All these models will control for the time-varying covariate “gains-losses differential” (ie, remaining balance on the EGM at measured time point minus money inserted in the EGM from the start of the study). All models will also test the moderation effect of the following variables: gender (men or women), age (continuous), education (5 categories), and PGSI category (nonproblem, low risk, and moderate risk). Holm-Bonferroni correction [85] will be used to counteract family-wise error rate due to multiple comparisons in each model for each statistically significant categorical variable (ie, time for repeated measures ANCOVAs; moderators with ≥3 categories).

The first model uses ANCOVAs to examine the differences between the two groups at the end of the gambling session on outcome variables 1, 3, 4, 6‐9, 14, 16, 17, and 19, separately.

The second model uses repeated measures ANCOVAs to examine the differences between the two groups at specific time points during the experiment corresponding to the presentation of the prevention pop-up messages in the experimental group. This model is computed on outcome variables 1 and 10‐13, separately.

The third model uses repeated measures ANCOVAs to examine the differences between self-reported measures (past 12 months) and the observed behaviors during the experiment. This model is computed on outcome variables 2, 5, 15, and 18, separately.

Secondary and other outcomes will be documented but will not be statistically compared. Descriptive estimates will be provided for each quantitative variable. Open-ended questions will be analyzed in a manner inspired by the thematic analysis method (see Textbox 3) [86].

The main objective of this study is to test the potential of pop-up messages as a prevention measure in a gambling setting. Effects of messages are measured on behaviors, cognitions, and emotions. The optimal benefit from the messages would be observing a reduction in money bet during the gambling session. However, other behavioral effects, like reduction in gambling intensity or session lengths, are considered indirectly beneficial because they foster conditions for further preventive actions. For example, all things being equal, a reduction in gambling intensity reduces the amount of money lost per unit of time. This, in turn, gives more time to the person gambling to think about their behavior and potentially stop their session with some money left. The improved realism of our research design heightens the confidence that the effects observed in our study can successfully be transposed to a real gambling setting.

Similarly, cognitive and emotional responses to the pop-up messages are of interest for the ease of adoption and indirect protection. Modifying cognitions is among the first steps in initiating behavioral change. Even if exposure to pop-up messages does not cause immediate reduction in gambling behavior, it is possible that they initiate a reflection about one’s gambling habits. This may lead to behavioral change at some point in the future or make the person more amenable to other gambling prevention strategies. Another aspect to consider regarding cognitive and emotional response to pop-up messages is psychological reactance, which is “the motivational state that is hypothesized to occur when a freedom is eliminated or threatened with elimination” [87]. In essence, the theory of reactance posits that, when a person’s freedom is threatened or removed (eg, being exposed to an RG pop-up message they feel is telling them how to feel or to act), they are motivated to restore it, whether directly by rebelling against the instructions (eg, gambling more) or indirectly, where the rebellion is done cognitively (eg, derogating the source of the message, denying there is even a threat, etc) [84,87-89]. Measuring how people cognitively and emotionally react to the pop-up messages is useful for understanding their effect. Hypothetical negative results could be due to psychological reactance. Data collected in this study, namely open-ended questions, could help understand how to better design pop-up messages to reduce psychological reactance. Furthermore, observing that pop-up messages do not negatively affect the participants’ level of pleasure could be a good argument to reduce potential reticence of the industry to broaden their usage [90].

Finally, this study will generate valuable data regarding the feasibility of studies conducted in a laboratory setting with the use of real money. Results will help determine if concerns are founded regarding participants’ potential uneasiness with gambling their own money in the laboratory and choosing not to participate in the study. Furthermore, results will show if the use of duplicity was successful in making participants believe they were really at risk of losing money. Feedback from participants could help improve this aspect if necessary.

While this study diverges from Auer and Griffiths [28] on certain theoretical underpinnings of the pop-up messages, their conceptualizations are mostly similar. This study has the advantages of measuring more thoroughly the effect of pop-up messages. This is reflected in the fact that this study records individual performance instead of a collection of gambling sessions (with the possibility that some of them have been done by the same individual). Furthermore, this study incorporates a variety of indicators, behavioral and psychological, whereas Auer and Griffiths [28] only evaluate the effect on ending of the gambling session immediately after the apparition of the pop-up messages.

This study also improves on other studies conducted in a real gambling setting using objective observed behavioral data instead of self-reported ones [29,34,49,50]. This limits the potential imprecision associated with the recollection of past events.

Finally, this study innovates by the level of ecological realism of the main task. To our knowledge, no other study conducted in the laboratory has so closely replicated the conditions of a real gambling setting.

This study’s main strengths are the use of a highly realistic setting for the gambling session and the use of duplicity to make participants believe that they are risking their own money. Furthermore, the maximal possible gain (CA $500) is sufficiently high to elicit an incentive to gamble seriously even though the (perceived) level of risk. This level of realism has seldom been achieved in a laboratory setting. Another strength is the diversity of outcomes observed. More specifically, the use of open-ended questions on multiple aspects heightens the chance to be able to interpret unexpected results.

On the other hand, limitations include potential difficulties in recruiting enough participants. First, regular EGM gamblers do not constitute a relatively small population in which this study is conducted. Furthermore, being a regular EGM player significantly overlaps with probable pathological gambling [91,92], which the ethics committee prohibits us from recruiting. The exclusion of probable pathological gamblers could affect the generalizability of the results. While pathological gamblers would be better served by more intensive prevention measures (or even therapy), they could still theoretically benefit from RG pop-up messages. Therefore, if this study goes well, it would be pertinent to repeat it while including this vulnerable population. In the meantime, it is possible that the pop-up messages observed in this study are higher than if deployed in a real setting due to the exclusion of pathological gamblers.

A second limitation is the impossibility of programming the same sequence of gains and losses for all participants. This is due to hardware limitations on the EGM. While randomness should provide balance between both groups, there is always the possibility that one group is significantly “luckier” than the other, altering how they respond to RG pop-up messages [45,93]. This could limit our ability to observe an effect. The gains-losses differential (see the “Statistical methods” section) is also used as a covariate in the analysis to mitigate this potential source of bias.

Third and finally, it is possible that the protection effect from pop-up messages is best observed over repeated exposure over multiple sessions. Therefore, a single gambling session as used in this study might not be enough to observe the full effect. Nevertheless, short of committing to operate a real gambling venue, there are no real feasible solutions to the “single session” in a laboratory setting with the use of real money. Indeed, duplicity works so long as participants do not know they will be reimbursed for their losses. In a study setting, it is not feasible to keep their money longer than a single session because (1) a participant might not come back for another session and be unreachable, making it impossible for the researchers to give them their money; and (2) even assuming all participants stay engaged in the study to the end, some of them might incur financial hardship that is not compensated by recuperating their money at a later date (eg, a participant might need to pay their rent now, a problem not fixed by getting their money back after months).

This study will provide new insights on the efficacy of pop-up messages as a prevention measure for gambling as well as the appropriateness of laboratory studies as a substitute to a real-life setting.