The role of affect-driven impulsivity in gambling cognitions: A convenience-sample study with a Spanish version of the Gambling-Related Cognitions Scale.

Background and aims Abnormal cognitions are among the most salient domain-specific features of gambling disorder. The aims of this study were: (a) to examine and validate a Spanish version of the Gambling-Related Cognitions Scale (GRCS; Raylu & Oei, 2004) and (b) to examine associations between cognitive distortion levels, impulsivity, and gambling behavior. Methods This study first recruited a convenience sample of 500 adults who had gambled during the previous year. Participants were assessed using the Spanish version of GRCS (GRCS-S) questionnaire, the UPPS-P impulsivity questionnaire, measures of gambling behavior, and potentially relevant confounders. Robust confirmatory factor analysis methods on half the sample were used to select the best models from a hypothesis-driven set. The best solutions were validated on the other half, and the resulting factors were later correlated with impulsivity dimensions (in the whole n = 500 factor analysis sample) and clinically relevant gambling indices (in a separate convenience sample of 137 disordered and non-disordered gamblers; validity sample). Results This study supports the original five-factor model, suggests an alternative four-factor solution, and confirms the psychometric soundness of the GRCS-S. Importantly, cognitive distortions consistently correlated with affect- or motivation-driven aspects of impulsivity (urgency and sensation seeking), but not with cognitive impulsivity (lack of premeditation and lack of perseverance). Discussion and conclusions Our findings suggest that the GRCS-S is a valid and reliable instrument to identify gambling cognitions in Spanish samples. Our results expand upon previous research signaling specific associations between gambling-related distortions and affect-driven impulsivity in line with models of motivated reasoning.

Introduction

Gambling disorder (GD) is defined as persistent and recurrent problematic gambling behavior leading to clinically significant impairment or distress (American Psychiatric Association, 2013). In the European Union, for example, problem gambling rates have been found to range from 0.3% to 3.1% (Planzer, Gray, & Shaffer, 2014). In Spain specifically, epidemiological research points to high rates of gambling behavior and specific culturally bound types of gambling (e.g., state lotteries and pervasive slot machines) are thought to encourage such behaviors (Jiménez-Murcia, Fernández-Aranda, Granero, & Menchón, 2014). Demographic variables (gender, age, and education levels), personality traits, schedules of reinforcement, comorbid states (drug and alcohol abuse, obsessive-compulsive disorder, and personality disorders), and delinquency/illegal acts have been identified as risk factors for problematic gambling (del Pino-Gutiérrez et al., 2016; Johansson, Grant, Kim, Odlaug, & Götestam, 2009; Petry, Stinson, & Grant, 2005). Cognitive behavioral therapy (CBT) is regarded as being effective at treating GD (Oei, Raylu, & Casey 2010; Raylu & Oei, 2010, 2016; Yau & Potenza, 2015), though relapse and dropout rates in such interventions remain high (Aragay et al., 2015). To enhance currently available treatment options, a better understanding of the mechanisms underpinning GD is crucial.

Recently, the role of altered gambling cognitions in the etiology and maintenance of GD has received increased interest from researchers and clinicians alike (Fortune & Goodie, 2012; Goodie & Fortune, 2013; Raylu & Oei 2002); although it has also been suggested that the causal link between GD and gambling cognitions could be bidirectional (so that GD and its accompanying cognitive distortions can remit spontaneously, or as a consequence of treatments not explicitly targeting such distortions, e.g., Echeburúa, Báez, & Fernández-Montalvo, 1996). Numerous studies have identified common patterns of distorted thinking in individuals with GD, and such patterns have been linked to the frequency of gambling behavior and GD severity (Emond & Marmurek, 2010). For example, the term the gambler’s fallacy refers to the cognitive distortion that a win will follow a sequence of losses even though outcomes occur independently of each other and are therefore unpredictable (Delfabbro, 2004). Similarly, gamblers may be of the belief that they themselves are able to influence gambling outcomes through the use of strategies, rituals, or lucky charms [Illusion of control (IC); Teed, Finlay, Marmurek, Colwell, & Newby-Clark, 2012]. Other cognitive distortions involve the over-interpretation of signals of gambling skills, attributional errors, selective memory, and probabilistic bias (Cantinotti, Ladouceur, & Jacques, 2004; Goodie, 2005; Ladouceur & Sévigny, 2005).

The Gambling-Related Cognitions Scale (GRCS) is an instrument that assesses five domains of gambling-related cognition in clinical and non-clinical gamblers (Raylu & Oei, 2004). These domains consist of cognitions related to: interpretative bias (IB), IC, predictive control (PC), gambling expectancies (GE), and perceived inability to stop gambling (ISG; see the Measures section for details on the meaning of each domain). The GRCS has been found to be highly reliable (scale Cronbach’s α = 0.93, domain Cronbach’s α = 0.77–0.91) and criterion valid when tested against the South Oaks Gambling Screen (SOGS) (Whelan, Meyers, & Steenbergh, 2007), a widely used instrument for assessing gambling-related behavior (Lesieur & Blume, 1987; Stinchfield, 2002). Validated versions of the GRCS have been developed for Chinese (Oei, Lin, & Raylu, 2007), Japanese (Yokomitsu & Takahashi, 2015), Turkish (Arcan & Karanci, 2013), French (Grall-Bronnec et al., 2012), and Italian samples (Donati, Ancona, Chiesi, & Primi, 2015; Iliceto et al., 2015).

Gambling cognitions have clinical implications: gamblers who identify and correct their gambling cognitions have greater treatment adherence than gamblers who do not (Ladouceur et al., 2001). Thus, a Spanish validated, self-report questionnaire for the assessment of gambling cognitions is an essential preliminary step for conducting research on gambling cognitions and to determine their role in the efficacy of treatment interventions for GD patients in the Spanish population. [Preliminary tests of a provisional version of the scale suggest that it is also usable, with minor alterations, in other Spanish-speaking communities (Jara-Rizzo, Navas, & Perales, 2016).]

Interestingly, impulsive personality traits in problem gamblers have been found to correlate with gambling cognitions, and it has been hypothesized that an impulsive decision-making style could increase the acceptance of erroneous beliefs during the execution of gambling behavior (Michalczuk, Bowden-Jones, Verdejo-García, & Clark, 2011). Somewhat counterintuitively, gambling cognitions and particularly the IC bias were observed to be more closely linked to affect-driven impulsivity (positive and negative urgency and the tendency to act rashly under the influence of emotional states) than to cognitive impulsivity (lack of premeditation and lack of perseverance). In line with this evidence, gambling-related cognitions also correlate with anomalies in emotion-regulation abilities (Navas, Verdejo-García, López-Gómez, Maldonado, & Perales, 2016), as commonly observed in other putative behavioral addictions (Wolz et al., 2015). This supports the idea that gambling cognitions are self-serving, that is, they help gamblers curb negative affective states generated by aversive events (e.g., losses), and encourage them to keep on gambling, as suggested by motivated reasoning models (Kunda, 1990).

The aims of this study were: (a) to develop a Spanish version of the GRCS (GRCS-S) and examine its validity and reliability in Spanish samples and (b) to assess associations between cognitive distortion levels, measures of impulsivity, and gambling behavior. In line with previous evidence, we hypothesize cognitive distortions in gamblers to be linked, on the one hand, to problematic aspects of gambling behavior and gambling severity and, on the other hand, to affect-driven impulsivity traits.

Methods

Participants and procedure

Data were collected between October 2012 and March 2015. A first convenience sample of 500 individuals was selected for factor analyses (henceforth, factor analysis sample). The only criterion for inclusion in this sample was having gambled at least once during the year before the assessment. Subsequently, using random number generation, this sample was split into two groups of 250 subjects (henceforth, Subsets A and B; see Table 1, upper panel).

Table 1.

Participant characteristics

	Factor analysis sample
	Subsample A	Subsample B
n	250	250
Age, mean (SD)	21.56 (7.04)	23.22 (9.63)
% of females	40.5	42.08
	Validity sample
n	137
Age, mean (SD)	34.96 (0.99)a
% of females	16.78b
% of n diagnosed as GD	37.23
% of n with SOGS ≥ 5	40.88

Age information was lost for four participants.

Gender information was lost for three participants.

For validity analyses, a different convenience sample of 137 participants (henceforth, validity sample) was used (Table 1, lower panel). This sample was composed of (a) treatment-seeking gamblers from three outpatient clinics (Asociación Granadina de Jugadores de Azar en Rehabilitación, Asociación Provincial Linarense de Jugadores de Azar en Rehabilitación, and Asociación de Ludópatas Jiennenses en Rehabilitación), in the cities of Granada, Linares, and Jaén (Andalusia, Spain), and (b) non-treatment-seeking gamblers meeting the same inclusion criterion described above. Exclusion criteria for this sample were: (a) having a history of neurological disease or brain trauma causing unconsciousness for 10 min or longer and (b) having been in treatment for a psychiatric disorder other than GD, in the case of GD patients. Time in treatment for GD patients ranged between 1 and 21 months (mean = 5.206 months and SD = 4.963). Fifty-two percent of these patients had been in treatment for three complete months or less.

Most of the factor analysis sample (n = 500) was extracted from the same pool of participants assessed for a study by Navas, Torres, Cándido, and Perales (2014). These participants were contacted in the University of Granada faculties, by means of Internet posting, or through people who had already participated in the study (so that most were college students) and took part in-group assessment sessions carried out in several lecture rooms at the University of Granada. Only those who had gambled at least once in the last year were later selected as members of the factor analysis sample. Extra participants were recruited and assessed in the same way until an n = 500 sample size was reached. In the group sessions, after being debriefed about aims and procedures (via written and read-aloud instructions), and providing informed consent, they filled out all the questionnaires described below except the SOGS (Spanish version; Echeburúa, Báez, Fernández-Montalvo, & Páez, 1994) in random order. Participants in this sample were not paid, although college students taking courses from the Department of Experimental Psychology, University of Granada, at that time obtained course credits for their participation. Each session lasted approximately 30 min.

The validity sample (n = 137) partially overlaps with the one in Navas et al. (in press) and consisted of both disordered and non-disordered gamblers. Treatment-seeking gamblers in this sample (51 participants) were contacted through their treatment centers. For these gamblers, all assessments were conducted individually and face-to-face. After welcoming the participant and obtaining his/her consent, the assessment began. In these patients, assessments were part of a larger protocol aimed at carrying out a detailed evaluation of behavioral and neurobiological correlates of gambling (see, e.g., Navas et al., 2016; Perales, Navas, Ruiz de Lara, Maldonado, & Catena, in press). The whole protocol was divided into two sessions, and all the instruments mentioned here were administered during the first session, lasting for approximately 3 hr. Participants were not compensated in this session (although they were paid approximately €10/hr in the second session, with payment being channeled through the treatment center or a responsible relative). The clinical assessment, including GD and potential comorbidities diagnosis, was carried out by a professional therapist in the treatment center. All other assessments were carried out by a trained psychologist with extensive experience in clinical evaluations (third author). Beyond clinical diagnosis, fulfillment of inclusion/exclusion conditions was reported by the participant.

Non-treatment-seeking gamblers in the validity sample (86 participants) were contacted through people who had already participated in the study, by Internet posting or by personal contacts. They were individually assessed or, alternatively, received the materials via e-mail, and were asked to fill the questionnaires at home, and returned them again via e-mail. These off-site participants were informed about the participation conditions (exclusion/inclusion criteria) and debriefed about their task by phone. They were then sent an e-mail with a fact sheet about the study aims and the conditions of consent. After formal consent was obtained, they received a second e-mail with full instructions and the questionnaires. On-site assessments were carried out by the third author on an individual basis. In this case, the presence of psychiatric comorbidities was assessed by means of a structured interview carried out by the same evaluator. Completion time for the set of questionnaires was approximately 45 min.

Measures

GRCS-S

The original GRCS questionnaire (Raylu & Oei, 2004) assesses five gambling-related cognitive domains through twenty-three 7-point Likert style items. ISG (e.g., “I’m not strong enough to stop gambling”) and GE (e.g., “Gambling makes things seem better”) refer to personal beliefs of lacking the ability or capacity to control gambling impulses and overvaluing the joy, reward, or relief that can be obtained from gambling, respectively. IC (e.g., “Praying helps me win”), PC (e.g., “Once I have a won, I will definitely win again”), and IB (e.g., “Relating my losses to bad luck and bad circumstances makes me continue gambling”) are cognitive distortions involving causal attribution processes.

The English version of GRCS questionnaire was translated into Spanish, and then back-translated into English by a native, English-speaking bilingual translator. Potential discrepancies between the original and back-translated versions of the questionnaire were discussed and eventually polished from the Spanish version by the translator and one of the authors (see Appendix A for the final version).

UPPS-P impulsive behavior scale (Whiteside & Lynam, 2001)

The brief Spanish version of the UPPS-P scale used here (Cándido, Orduña, Perales, Verdejo-García, & Billieux, 2012) contains 20 items (four items per dimension), and allows for a quick multidimensional assessment of impulsivity: positive urgency (e.g., “I tend to lose control when I am in a great mood”), negative urgency (e.g., “When I am upset I often act without thinking”), (lack of) premeditation (e.g., “My thinking is usually careful and purposeful”), (lack of) perseverance (e.g., “Once I get going on something I hate to stop”), and sensation seeking (e.g., “I quite enjoy taking risks”).

MultiCAGE CAD-4 (Pedrero Pérez et al., 2007)

This instrument is a quick screening tool to detect alcohol misuse and illegal drug misuse among other problematic behaviors beyond the scope of this study. Each subscale consists of four yes/no items, checking for current feelings of craving, others’ complaints about the potential problematic behavior, guilt or shame feelings and/or lack of self-acknowledgment, and self-reported compensatory behaviors. The scales of alcohol and illegal drug misuse have shown appropriate psychometric properties and predictive validity of alcohol and drug abuse. In this study, the illegal drug and alcohol subscales were used as control variables to check for GRCS-S domain specificity.

SOGS (Lesieur & Blume, 1987)

This questionnaire is aimed to assess gambling severity, dependence, and debt accrual. It is the most commonly used tool in gambling research, a fact that allows comparisons across studies. The Spanish version used in this study has shown good psychometric properties (Echeburúa et al., 1994). It comprises 16 items, 12 of which are worded as yes/no questions, and count for calculating the SOGS dependence, debt, and total severity indices.

Statistical analysis

Preliminary analyses and selection of estimation method

Preliminary analyses showed that all variables of interest had non-normal distributions in the factor analysis sample, with strong violations of skewness and kurtosis. This fact, along with the use of ordinal measure scales (Muthén & Kaplan, 1985), led us to consider all the variables as discrete, and to use analyses and estimation methods appropriate for them (Babakus, Ferguson, & Joreskog, 1987). Hence, we used a robust method for the estimation of models [diagonal weighted least square (DWLS); Babakus et al., 1987; Schumacker & Beyerlein, 2009]. The inputs for DWLS implementation are a correlation matrix and a covariance matrix, suitable for ordinal data, and an asymptotic covariance matrix, that the software uses to adjust the estimation in the presence of extreme values.

Factor analyses

For testing and comparison purposes, we took four different nested factorial solutions into consideration: (a) the original five-factor model composed of GE, IC, PC, ISG, and IB; (b) a four-factor solution formed by GE, ISG, and PC, plus a fourth factor obtained by merging IC and IB items; (c) a three-factor solution composed of GE, ISG, and a third factor obtained by merging IC, PC, and IB items; and (d) finally, a one-factor solution resulting from loading all the items on a single factor. Model b was based on the assumption that IC and IB refer to illusory perceiving a connection between one’s behavior and gambling outcomes (either a priori or retrospectively). Model c, on the other hand, reflects the fact that IC, PC, and IB are distortions in the realm of causal and contingency-based cognition (see Perales et al., in press).

The traditional goodness-of-fit measure (χ2) assesses the magnitude of discrepancy between the sample and fitted covariance matrices (Hu & Bentler, 1999). However, χ2 is sensitive to sample size and tends to reject the model as the sample grows larger (Bentler & Bonnet, 1980; Jöreskog & Sörbom, 1993). In addition, it does not provide enough information to discriminate between different models applied upon the same data set. In view of these features, we used several alternative indices (Schermelleh-Engel, Moosbrugger, & Müller, 2003), namely Δχ2, relative χ2, root mean square error of approximation (RMSEA), normed fit index (NFI), comparative fit index (CFI), and expected cross-validation index (ECVI).

The statistic Δχ2, suitable for nested models, is the difference between the χ2 values of two hierarchical models estimated with the same data. This difference is tested on the χ2 table with degrees of freedom equal to the difference between the two respective values of degrees of freedom. A non-significant result leads to choosing the most parsimonious model (i.e., the model with fewer factors). Contrarily, when Δχ2 is significant, the best fitted solution is the least parsimonious model (i.e., the model with more factors) (Cheung & Rensvold, 2002; Gallucci & Leone, 2012). Relative χ2 is the odds between χ2 and its degrees of freedom, and its acceptance criterion varies among researchers, ranging from a value of less than 2 to a value of less than 5 (Schumacker & Lomax, 2004; Tabachnick & Fidell, 2007; Ullman & Bentler, 2003). RMSEA is a measure of the distance between the perfect model and the estimated model. It is not affected by sample size and values lower than 0.05 are assumed to imply a good fit (Hu & Bentler, 1995). NFI (Tucker & Lewis, 1973) and CFI (Bentler & Bonnet, 1980), compare the model of interest with alternative ones, such as the null or independence model. Values higher than 0.95 are acceptable, and those higher than 0.98 are optimal (Bentler, 1990). ECVI is the discrepancy between the covariance matrix in the analyzed sample and the covariance matrix that would be expected in a different sample with the same size (Jöreskog & Sörbom, 1993). The model with the smallest ECVI indicates the best fit.

As noted above, for factor analyses, the sample was divided into two subsamples (n = 250). The first (Data set A) was used for model estimation, whereas the second (Data set B) was used for model replication. Factor intercorrelations were computed using a non-parametric method (Spearman’s rho), separately for the two subsamples. Reliability of the factors was computed using the intraclass correlation coefficient (ICC; Gallucci & Leone, 2012), again separately for the two subsamples.

Concurrent and criterion validity

Non-parametric correlational analysis (Spearman’s rho) was used to test GRCS-S convergent validity with SOGS and specificity with MultiCAGE alcohol and drug scores [using the validity sample (n = 137), and the whole factor analysis sample (n = 500), respectively]. For criterion validity, a classification tree (Breiman, Friedman, Olshen, & Stone, 1984) was performed, using the variables from the best-fitting GRCS-S model as discriminative factors and the recommended threshold for clinical significance (SOGS severity ≥ 5) as criterion of group membership. Non-parametric correlation analyses were used to test GRCS-S/UPPS-P links in the factor analysis sample (n = 500). Parametric partial correlation analyses (controlling for clinically significant gambling) were used to further test and replicate GRCS-S/UPPS links in the validity sample (n = 137).

Preliminary and correlation analyses were conducted using Statistical Package for the Social Sciences (SPSS 20) software. Factor analyses were performed using LISREL 8.80 software (Jöreskog & Sörbom, 2006). PRELIS software was used for goodness-of-fit estimation (Jöreskog & Sörbom, 2006).

Ethics

The study procedures were carried out in accordance with the Declaration of Helsinki and its later amendments. The Ethics Committee of Clinical Research at the University of Granada approved the study as part of the funded PSI2013-45055-P project. All subjects were informed about the study and all provided signed consent.

Results

Factor analysis

Fit indices for Subset A from the factor analysis sample are shown in Table 2 (left panel). Although all the solutions have good or very good fit indices, the four- and the five-factor solutions stand out as the best models. The comparison indices between these two solutions showed a lower ECVI for the four-factor solution (1.65 vs. 1.68, for four- and five-factor solution, respectively) and a non-significant Δχ2. In other words, although models a (the original one) and b (four factors, merging the two biases on own behavior’s impact on gambling outcomes) are viable, a parsimony criterion would lead to selecting the latter.

Table 2.

Indices of goodness-of-fit for each model for both data sets from the factor analysis sample (see text for the meaning of the different indices)

	Subset A (n = 250)				Subset B (n = 250)
	One-factor solution	Three-factor solution	Four-factor solution	Five-factor solution	One-factor solution	Three-factor solution	Four-factor solution	Five-factor solution
χ2 (df)	417.14 (230)	326.07 (227)	307.49 (224)	305.62 (220)	503.94 (230)	421.23 (227)	360.37 (224)	349.64 (220)
Δχ2	/	91.07 (<0.01)	18.58 (<0.01)	1.87 (ns)	/	82.71 (<0.01)	60.86 (<0.01)	10.73 (ns)
Relative χ2	1.83	1.44	1.37	1.39	2.19	1.86	1.61	1.59
RMSEA (CI)	0.06 (0.05; 0.07)	0.04 (0.03; 0.05)	0.04 (0.03; 0.05)	0.04 (0.03; 0.05)	0.07 (0.06; 0.08)	0.06 (0.05; 0.07)	0.05 (0.04; 0.06)	0.05 (0.04; 0.06)
NFI	0.98	0.98	0.98	0.98	0.97	0.98	0.98	0.98
CFI	0.99	0.99	1.00	0.99	0.99	0.99	0.99	0.99
ECVI (CI)	2.04 (1.83; 2.29)	1.70 (1.53; 0.91)	1.65 (1.48; 1.85)	1.68 (1.51; 1.88)	2.39 (2.15; 2.67)	2.09 (1.87; 2.33)	1.86 (1.67; 2.09)	1.85 (1.67; 2.07)

Note. RMSEA = root mean square error of approximation, NFI = normed fit index, CFI = comparative fit index, ECVI = expected cross-validation index, ns = not significant.

In general, all the results for Subset B (Table 2, right panel) confirmed the ones obtained in Subset A, except for a slightly weaker strength. Again, the four- and five-factor solutions clearly beat the other models. ECVI for the four- and the five-factor solutions were virtually identical (1.86 vs. 1.85) and the Δχ2 value was not significant. The four- and five-factor models thus provide equivalent fits, with the former being preferable in terms of parsimony.

Factor intercorrelation

Intercorrelation analyses were run on eight factors (the five original ones, plus IC + IB, IC + IB + PC, and the one merging all the items together). All the correlations were significant and above 0.40 for Subsets A and B (Table 3). Intercorrelations between individual factors are in a range slightly higher than the one reported for the original scale (0.49–0.62 in Raylu & Oei, 2004). As in the original scale, the highest correlation was observed between PC and IB.

Table 3.

Factor intercorrelation for Subsets A and B from the factor analysis sample

	GE	IC	PC	ISG	IB	IC + IB	IC + PC + IB	TOT
Subset A
GE	1
IC	0.57	1
PC	0.69	0.64	1
ISG	0.70	0.65	0.64	1
IB	0.75	0.60	0.79	0.67	1
IC + IB	0.75	0.87	0.80	0.74	0.92	1
IC + PC + IB	0.76	0.82	0.92	0.73	0.91	0.97	1
TOT	0.88	0.79	0.88	0.84	0.90	0.95	0.97	1
Subset B
GE	1
IC	0.54	1
PC	0.60	0.60	1
ISG	0.64	0.56	0.48	1
IB	0.67	0.55	0.76	0.58	1
IC + IB	0.70	0.84	0.78	0.64	0.92	1
IC + PC + IB	0.70	0.79	0.91	0.61	0.91	0.97	1
TOT	0.85	0.77	0.85	0.76	0.88	0.94	0.96	1

Note. GE = gambling expectancies, IC = illusion of control, PC = predictive control, ISG = inability to stop gambling, IB = interpretative bias, TOT = total.

Reliability

Reliability outcomes are presented in Table 4. The overall scale showed an excellent value. For Subset A of the factor analysis sample, values were good for IC + PC + IB, IC + IB, ISG, and PC. Values were acceptable for GE and IB, and just sufficient for IC (for threshold values, see George & Mallery, 2007). Reliabilities were very similar for Subset B, although with slightly lower values. These results provide further proof of the goodness of the four-factor solution. Indeed, this solution does not isolate the factor IC, which is the only one with low internal consistency in the validation subsample.

Table 4.

Reliability of all the factors in both data subsets from the factor analysis sample

	ICC Subset A (n = 250)	ICC Subset B (n = 250)	Number of items on the factor	Items that load on the factor
GE	0.80	0.72	4	1–6–11–16
IC	0.72	0.64	4	3–8–13–18
PC	0.80	0.78	6	4–9–14–19–22–23
ISG	0.85	0.79	5	2–7–12–17–21
IB	0.78	0.75	4	5–10–15–20
IC + IB	0.83	0.80	8	3–8–13–18 + 5–10–15–20
IC + PC + IB	0.90	0.88	14	3–8–13–18 + 4–9–14–19–22–23 + 5–10–15–20
One-factor solution	0.94	0.92	23	All

Note. ICC = intraclass correlation coefficient, GE = gambling expectancies, IC = illusion of control, PC = predictive control, ISG = inability to stop gambling, IB = interpretative bias.

Although the size of the validity sample (n = 137) was not large enough for factor analysis purposes (see results below for validity analyses), we did compute reliability indices for all factors in this sample. ICC values were 0.77 (GE), 0.68 (IC), 0.84 (PC), 0.91 (ISG), 0.89 (IB), 0.88 (IC + IB), 1.00 (IC + IB + PC), and 0.95 (full scale). That is, despite the smaller sample size, increasing the range of gambling severity scores (this sample was composed of both GD patients and non-disordered gamblers) slightly improved reliability of the subscales. IC remained, however, on the limit of acceptability.

In previous reports, reliability values (Cronbach’s α) were 0.87 (GE), 0.87 (IC), 0.77 (PC), 0.89 (ISG), 0.91 (IB), and 0.93 (full scale), for the original scale, and 0.73 (GE), 0.75 (IC), 0.76 (PC), 0.84 (ISG), and 0.79 (IB), averaged across translations (French, Turkish, Japanese, Chinese, and Italian). In terms of both order and magnitude, these reliabilities are well matched by the ones found in our validity sample.

Validity

Concurrent validity

Correlations of GRCS-S factors with SOGS scores (in the validity sample) and MultiCAGE CAD-4 subscores (in the factor analysis sample) are shown in Table 5. As expected, GRCS-S scores strongly correlated with SOGS measures (dependence, debt, and total severity score). ISG was the factor showing the strongest correlation with SOGS scores, followed by the factors including IB (either isolated or in combination with IC and PC), and by PC, GE, and IC. Interestingly, GRCS-S factors did not significantly correlate either with alcohol- or illegal drug-related problems, which unveils a very strong specificity of the measured cognitions with regard to potentially problematic aspects of gambling.

Table 5.

Correlations of GRCS-S factors with gambling severity (SOGS) and alcohol and substance use (MultiCAGE CAD-4)

	GE	ISG	IC	PC	IB	PC + IC + IB	PC + IB
SOGS total	0.55**	0.74**	0.53**	0.61**	0.66**	0.66**	0.67**
SOGS dependence	0.54**	0.74**	0.51**	0.60**	0.66**	0.65**	0.66**
SOGS debt	0.41**	0.51**	0.34**	0.42**	0.44**	0.46**	0.46**
Alcohol use (MC)	0.15	0.16	0.08	0.05	0.12	0.09	0.08
Substance use (MC)	0.03	−0.07	0.02	0.01	0.04	0.03	0.03

Note. Spearman’s rho correlations. MC = MultiCAGE CAD-4, GE = gambling expectancies, IC = illusion of control, PC = predictive control, ISG = inability to stop gambling, IB = interpretative bias.

p < .05.

p < .01.

Criterion validity

Using the recommended threshold (Lesieur & Blume, 1987), the validity sample (137 participants) was divided into two groups of problematic (SOGS ≥ 5, n = 51) and non-problematic (SOGS < 5, n = 86) gamblers. [All treatment-seeking gamblers had very high SOGS scores, which means 5.8% of the total number of non-treatment-seeking gamblers presented a SOGS score equal to or above 5 (the problem gambling threshold). This percentage is approximately what would be expected, based on problem gambling prevalence data in the population of regular gamblers in Spain. So, it is unlikely that a desirability bias in SOGS responses could have significantly distorted our results. Still, a large majority of participants were assessed face-to-face. The presence of some desirability biases in these individuals can be somewhat reduced by ensuring confidentiality and introducing the evaluator as an experienced psychologist subject to strict ethical standards, but are not completely eliminated.] A classification tree (Breiman et al., 1984) was performed using the variables from the four-factor model to discriminate between groups. Results yielded a three-node tree wherein three of the four factors (i.e., ISG, IC + IB, and GE) correctly categorized 95.3% of non-problematic and 84.3% of problematic gamblers (total = 91.2%). After the third node, PC did not improve the percentage of correct categorization. The I effect size index (the improvement in classification capacity beyond chance; Henson, Natesan, & Axelson, 2014) yielded a value of 0.82 (customarily interpreted as very good; Hess, Olejnik, & Huberty, 2001).

Correlations between impulsivity traits and gambling cognitions

As shown in Table 6, all GRCS-S scores significantly correlated with positive and negative urgency, and GE, PC, and all factors including IB correlated with sensation seeking. Most interestingly, none of the GRCS-S scores correlated with cognitive impulsivity traits (lack of premeditation and lack of perseverance).

Table 6.

Correlations of GRCS-S factors with impulsivity (UPPS-P) measures

	GE	ISG	IC	PC	IB	PC + IC + IB	PC + IB
Negative urgency	0.25**	0.35**	0.20*	0.23**	0.28**	0.27**	0.26**
Positive urgency	0.33**	0.30**	0.28**	0.41**	0.37**	0.40**	0.41**
Sensation seeking	0.18*	0.14	0.14	0.31**	0.31**	0.31**	0.32**
Lack of premeditation	−0.03	0.05	−0.11	0.02	0.01	0.01	0.03
Lack of perseverance	0.06	0.07	−0.05	−0.01	0.04	0.01	0.02

Note. Spearman’s rho correlations. GE = gambling expectancies, IC = illusion of control, PC = predictive control, ISG = inability to stop gambling, IB = interpretative bias.

p < .05.

p < .01.

To ensure that this set of correlations does not merely reflect the potential confounding between GRCS-S scores and GD severity, we carried out partial correlation analysis between GRCS-S and UPPS-P measures, controlling for SOGS total score, in the validity sample (see Appendix B). The majority of the correlations lose strength, but the general pattern was very similar to the one in the first sample. GE lost significance with positive urgency and sensation seeking, and PC lost significance with negative urgency. The correlations between SOGS measures and the other factors did not show any qualitative changes in significance values.

Discussion

The first aim of this study was to develop a Spanish version of the GRCS (GRCS-S) to assess gambling cognitions in Spanish communities and to determine the reliability and validity of this scale. Second, and more importantly, we sought to confirm and extend previously identified associations between gambling-related cognitive distortion levels, measures of impulsivity, and gambling behavior (Michalczuk et al., 2011).

Previous studies have identified a five-factor model in the GRCS: perceived ISG, GE, IB, IC, and PC. This study confirms the five-factor model to provide good fit for a Spanish sample and supports the soundness of the psychometric properties of the GRCS-S. However, a more parsimonious, four-factor model was also found to be viable. The superiority of the four-factor model is attributable to parsimony, and, quite likely, also to the relatively lower internal reliability of the IC factor. Tentatively, the lower reliability of the IC subscale, in turn, is attributable, first, to the fact that it is computed from only four items, and second, to the wording of one of such items in religious terms (“Praying helps me win”). Potentially, religiosity differences can have a large impact on responses to this item, and thus on global internal consistency. Conversely, the virtually constant IC internal consistency values observed across samples (from the factor analysis to the validity data sets) virtually discards the possibility that sample composition, different severity ranges, or treatment stage (in treatment-seeking gamblers) accounted for its relatively low value.

This study also verifies the concurrent and criterion-related validity of the GRCS-S, and replicates previous reports that the GRCS model provides specific correlates of gambling clinical features. However, in Michalczuk et al.’s (2011) study, the strongest effect of group (GD patients vs. controls) was found for IC, and the weakest effect for PC and IB. In spite of the good general predictive value of the scale for clinically relevant features in both studies (group in one case, gambling severity in the other), this ordering is partially at odds with our finding that SOGS is strongly related to IB, but much more weakly to IC. This discrepancy can be attributed to differences across the study samples, to the fact that group categorization and SOGS severity are not equivalent variables, or to the differences in subscales reliabilities in the two versions of the scale. At this point, we have no evidence to favor any of these interpretations. Nonetheless, this study replicated the main finding from Michalczuk et al.’s (2011) study, in that impulsivity is highly correlated with increased cognitive distortion levels.

Most interestingly, correlations were only significant for affect- and motivation-driven components of impulsivity. The privileged link between gambling-related biases and emotion-driven impulsivity might be seen, in principle, as counterintuitive, considering that the UPPS-P contains specific dimensions straightforwardly assessing purely cognitive impulsivity (lack of premeditation and lack of perseverance). Recent, highly powered factorial analyses confirm this distinction and separate a conscientiousness/planning impulsivity component from inadequate coping of negative emotionality (Knezevic-Budisin, Pedden, White, Miller, & Hoaken, 2015; Sharma, Markon, & Clark, 2014). This view is also consistent with the proposal that poor executive and decision-making functioning is further qualified in accordance with the stronger or weaker involvement of abnormal emotion processing (Billieux, Gay, Rochat, & Van der Linden, 2010; Chester et al., 2016; Gunn & Finn, 2015).

Independent of the soundness of that factorization of impulsivity, evidence in the framework of the Theory of Planned Behavior has shown that heavy gambling is not necessarily accompanied by lack of planning (Martin et al., 2010), and in a previous study, we have observed a less future-oriented decision-making style only in a subgroup, but not the general population of disordered gamblers (Perales et al., in press). Our results corroborate that gambling-related cognitive distortions bear no strong connections with the conscientiousness/planning impulsivity dimension, as measured by UPPS-P dimensions lack of premeditation and lack of perseverance. Instead, strong cognitive distortions belong – along with affect-driven impulsivity – to a general executive and decision-making profile, in which emotions play a key role.

Several lines of evidence converge in stressing the importance of the link between emotion-driven processes and gambling-related cognitive distortions. For example, Clark, Studer, Bruss, Tranel, and Bechara (2014) have recently shown that lesions of the insula, an area strongly involved in emotional appraisal, abolish the near-win effect (known to be strongly linked to the illusory control bias in gamblers). On the other hand, positive and negative urgency empirically and theoretically overlap with emotion dysregulation in GD and other risky behavior patterns (Weiss et al., 2015; Wolz et al., 2016), and a connection between abnormal emotion regulation and the cognitive symptoms of GD has been recently reported by Navas et al. (2016).

These recent developments point out to the importance of dealing with cognitive distortions in therapy, while making sure their emotional and motivational underpinnings are not neglected, in line with models in which cognitions are shaped by emotions and motivations (Kunda, 1990). According to these models, the emotional stakes in the conclusions influence the neurocognitive mechanisms by means of which such conclusions are reached (Westen, Baglov, Harenski, Kilts, & Hamann, 2006).

In the realm of GD, cognitive distortions are likely to play a subjectively protective role against the distress generated by losses, mounting debt, or the very fact of regarding oneself as a disordered gambler (which would explain why most relevant cognitive biases have to do with reinterpreting or underestimating losses, overestimating future gains, or fueling one’s sense of mastery).

Our findings provide some useful indications to be considered in GD treatment and prevention. Although the cognitive differences between GD patients and non-problematic gamblers, and the association of such differences to gambling severity, have been convincingly replicated, treatments specifically addressing cognitive distortions are less powerful than expected (Goodie & Fortune, 2013). In other words, the neglect of emotion regulation in attempts to restructure cognitive distortions in GD could account for the fact that patients with GD often experience problems generalizing cognitive change from therapeutic to daily-life settings (Ladouceur & Sevigny, 2003). Moreover, the abovementioned study by Navas et al. (2016) suggests that some emotion-regulation strategies usually regarded as adaptive, and frequently included in CBT packages, could be counterproductive in GD, as they can blunt the emotional impact of losses and fuel cognitive distortions. Taken together, these results support the use of metacognitive techniques, aimed at increasing awareness and reconfiguring responses to inner states, as for example, detachment mindfulness, attention retraining, virtual reality exposure, and neurofeedback (Chu & Clark, 2015; Fernández-Aranda et al., 2012).

Limitations

In spite of the importance of these conclusions, this study has a number of non-trivial limitations. First, only convenience samples were used. Therefore, further research with systematic sampling methods is needed to replicate the present findings in representative samples of sporadic, regular, problematic, and disordered gamblers. In other words, equivalence of the scale’s factor structure across different populations should not be taken for granted. Second, and in relation to the previous point, our factor analysis sample was mostly composed of college students, and very few of them showed potentially risky gambling behavior. The sample was also more homogeneous than the one used for validation of the original scale and contained more females than most studies with regular (both disordered and non-disordered) gamblers. This could compromise generalizability and could also account for the low reliability of the IC dimension in subsample B. This problem is compensated in part by the fact that, in contrast, the validity sample was mostly composed by disordered and non-disordered gamblers recruited from gambling treatment centers and the general (non-college) population. In spite of its smaller size, this sample yielded good levels of reliability and usability for the GRCS-S scale.

Third, our clinical subsample was largely made up of male gamblers. Although the prevalence of problem gambling has repeatedly been found to be higher in male samples (Planzer et al., 2014), future research using the GRCS-S should aim to include more diverse samples.

Finally, this study was cross-sectional and did not assess the sensitivity of the GRCS-S across time. Other authors have recommended that follow-up assessments take place at four time points (short term, medium term, long term, and post-treatment) (Walker et al., 2006). Future studies are needed to examine the test–retest reliability of the GRCS-S.

Final remarks

The GRCS has been proven to be a valuable tool in the study of gambling behavior correlates and has helped to better understand GD symptomatology and temporal dynamics. In this paper, we have shown the Spanish version of this questionnaire to have adequate psychometric properties in two samples of young, mostly non-problematic, sporadic gamblers, and older, heavier gamblers (some of whom had been previously diagnosed as disordered gamblers). Most importantly, however, we have replicated and extended previous results on the tight bond between gambling cognitions and emotional and motivational aspects of impulsivity, which points out to the importance of including emotion-regulation training in psychoeducative interventions for GD prevention, treatment, and relapse control.

Authors’ contribution

FDP and TS have equally contributed to this study and should be considered at the same level of authorship. FDP, JFN, and JCP contributed to the development of the study concept and design. FDP performed the statistical analysis. All authors aided with interpretation of data. TS wrote the first version of the manuscript. All authors were involved in further discussion and rewriting. JFN, FF-A, SJ-M, and JCP supervised the study. JCP is the principal investigator for the research project.

Conflict of interest

The authors declare no conflict of interest.

Acknowledgement

We thank Asociación Granadina de Jugadores de Azar en Rehabilitación (AGRAJER), Asociación Provincial Linarense de Jugadores de Azar en Rehabilitación (APLIJER), and Asociación de Ludópatas de Jaén en Rehabilitación (ALUJER) for their fruitful collaboration and assistance in the recruitment of participants.

Table B1.

Partial correlation analysis between GRCS and UPPS-P measures controlling for SOGS total score

	GE	PC	ISG	IC + IB
Negative urgency	−0.09*	0.08	0.23**	0.16*
Positive urgency	−0.06	0.22**	0.18**	0.20**
Sensation seeking	−0.02	0.17**	0.03	0.11*
Lack of premeditation	−0.14**	0.02	0.05	0.00
Lack of perseverance	−0.07	0.03	0.07	0.02

Note. Pearson’s r correlations. GE = gambling expectancies, IC = illusion of control, PC = predictive control, ISG = inability to stop gambling, IB = interpretative bias.

*p < .05. **p < .01.