Risk Factors for Problematic Gambling: A Critical Literature Review (original) (raw)
Attempts to study risk factors not meeting our inclusion criteria merit mentioning in the text, without inclusion in the table, to give a better overview and understanding of the field. For condensed results, please see Table 1. For more information on results, please refer to the text.
Table 1 Risk factors for pathological gambling (dependent variable: PG) [rev 2004-04-05 final]
To give a summary of results and provide an impression about how scarce empirical studies are, Table 2 is compiled as a summary, detailing the amount of such studies that have been performed on each risk factor. Conclusions about the well-established and probable or potential risk factors are also indicated within the Table.
Table 2 Risk factors for PG—summary of results
Demographics
Demographic variables have often been mentioned as possible risk factors for PG but they have seldom been empirically tested.
Age
Ladouceur et al. (1999a) studied 3426 high school students in a correlational design using the SOGS (South Oaks Gambling Screen) (Lesieur and Blume 1987) and its relation to grades in school. A univariate ANOVA (Analysis of Variance) on SOGS scores and grade level showed a statistically significant main effect (F = 7.73, p < .001). Furthermore, a Sheffé test showed higher SOGS scores for younger students (8 grade compared to 10–11th grade; p < .05).
Bondolfi et al. (2000) used the SOGS to analyze different risk factors in a correlational designed telephone interview prevalence study (n = 2526) on gambling in Italy. The results showed that being younger than 29 years of age was a significant risk factor (X 2 = 17.1, p = .01).
Volberg et al. (2001) performed a large prevalence study using the SOGS-R and DSM-IV (APA 2000) in a Swedish sample (n = 8845). Life-time problem and pathological gamblers were compared to non-problem gamblers on different variables. Age (younger than 25) was shown to be a significant risk factor (OR = 2.51, p = .000).
These studies suggest that younger age (i.e. younger than 29 years old appears to be a significant risk factor for PG.
Gender
In a study conducted by Winters et al. (1993b), subjects aged 15–18 years (n = 702) were interviewed using the SOGS-RA (Winters et al. 1993a) on a targeted telephone list (not a random digit-dial procedure). A comparison of answers on item #1 of SOGS-RA, showed that boys had higher gambling activity than girls (t = 6.46, p < .001). The researchers compared higher degrees of problem gambling (pathological gambling and at risk gambling) to different variables. There was no comparison made between gender and problematic gambling and therefore the reference is not included in the table on gender.
In a study by Ladouceur et al. (1999a) of 3426 high school students, a univariate ANOVA on SOGS scores and male gender showed a statistically significant main effect (X 2 = 39.52, p < .001).
Feigelman et al. (1995) investigated pathological gambling in 220 methadone patients in two methadone maintenance treatment programs (MMTP), finding a high rate of gambling problems in this treatment population. Pathological and problem gambling was assessed using SOGS and, based on earlier results, other important factors were assessed. A significant relationship between problem gambling and male gender was found (r xy = 0.12, p = .04).
Male gender was also shown to be a risk factor (X 2 = 8.94, p = .01) in a telephone interview prevalence study of gambling conducted by Bondolfi et al. (2000) in Italy. This was also the case in the Volberg et al. (2001) study where male gender was again shown to be a risk factor for gambling problems (OR = 3.71, p = .000).
In four of the five gender studies where gender has been evaluated in relation to problem gambling, clear support for the notion that male gender is a significant risk factor for PG has been demonstrated. There are indications (for instance in Wardman et al. 2001) that females are at higher risk than men in aboriginals but this finding has not been replicated.
Education
In the Volberg et al. (2001) study, education level did not significantly predict risk for gambling problems. Further study on how education levels impact PG is needed.
Marital Status
In the study by Bondolfi et al. (2000), marital status (married) was shown to be a risk factor (X 2 = 7.52, p = .02). Conversely, the Volberg et al. (2001) study showed that being single was a risk factor for gambling problems (X 2 = 121.67, p = .000).
Only two studies have been directed towards marital status empirically, and with contradicting results, conclusions are not possible at this time.
Income
In the study by Bondolfi et al. (2000), higher income was shown to be a risk factor (X 2 = 10.88, p = .01) for gambling problems. The helpline study by Potenza et al. (2001) reported financial problems as a significant risk factor (X 2 = 4.21, p < .04).
Only two studies have examined income and financial problems empirically, and they have produced contradictory results.
Employment
In the study by Feigelman et al. (1995), there was a significant relationship between unemployment status within the last year and problem gambling (r xy = −0.15, p = .02).
Hall et al. (2000) studied pathological gambling among 313 cocaine-dependent outpatients. They used the DIS (Diagnostic Interview Schedule) to reach a DSM-III-R diagnosis of PG, and the ASI (Addiction Severety Index) to obtain sociodemographic and other information. Of the 313 patients, 25 (8.0%) fulfilled full DSM-III-R criteria for PG. There was a significant relationship between unemployment status and PG (t = 11.09, p < .001).
Only two studies have been directed towards employment empirically and thus, we consider employment status as a probable risk factor for PG.
Social Welfare Status
In the Volberg et al. (2001) study, being on social welfare was shown to be a significant risk factor for gambling problems (z = 2.41, p = .05).
Only one study has been directed towards social welfare status empirically and thus, we consider social welfare status as a probable risk factor for PG.
Residence
In the Volberg et al. (2001) study, living in a large city was shown to be a risk factor for gambling problems (z = 4.00, p < .01).
Only one study has been directed towards residence empirically, thus, we consider residence as a probable risk factor for PG.
Academic Achievement
In the study by Ladouceur et al. (1999a), an ANOVA performed on a five question scale for academic achievement and SOGS scores showed a significant main effect (F = 19.44, p < .001).
In the study by Winters et al. (1993b), average-to-below average school grades were related to problematic gambling (X 2 = 21.7, p < .001).
Only two studies have been directed towards academic achievement empirically. We consider academic achievement as a probable risk factor for PG.
Immigrants and Ethnic Groups
In the helpline study by Potenza et al. (2001), African-American ethnicity was identified as a significant risk factor (X 2 = 3.87, p < .05). In the Welte et al. study (2004), being African-American, Hispanic, or Asian were all risk factors for problematic gambling (IRR 1.96–4.71; p < .01). In the Volberg et al. (2001) study, being born outside the country was shown to be a risk factor for gambling problems (OR = 2.08, p = .01).
Because three studies have been directed towards immigrants and ethnic groups empirically, (but only two have focused on ethnicity), we consider immigration and ethnic groups as probable risk factors for PG.
Physiological and Biological Factors
Heart Rate and Arousal
One study used a laboratory setting with an artificial casino and compared it to a real casino situation. A group of 12 experienced gamblers showed significantly higher heart rate (HR) increases (p < .0001) in the real casino condition. The correlation between the amount of money wagered and HR increase was significant (r xy = .741, p < .0005, one-tailed). The other group consisted of students (n = 12) who did not differ from the experienced gamblers in their reactions to the artificial casino (Anderson and Brown 1984).
A second empirical study was performed by Leary and Dickerson (1985) who followed high- and low (n = 22/22) frequency players by assessing heart rate during playing. Playing was significantly associated with increases in arousal in both groups but was significantly more so by high-frequency players (p < .05).
Cocco et al. (1995) hypothesized that poker machine gamblers and horse race gamblers should differ in their state of arousal. Of the 12 problem poker machine players and 13 horse race gamblers assessed, the researchers were able to show that poker machine gamblers showed higher arousal avoidance and higher trait anxiety as compared to horse race gamblers (both p < .05). No attempt was made to predict pathological gambling.
Only two studies have been directed towards heart rate and arousal empirically. Consequently, we consider heart rate and arousal as a probable risk factor for PG.
Griffiths (1995) reviewed the literature on this field (see Griffiths 1995, Table 1.4, p. 18) and it seems the correlation of increased heart rate to increased playing among gamblers compared to non-gamblers is weak. Empirical support for such a notion seems too weak to justify further reporting.
Transmitter Activity
Bergh et al. (1997) studied monoamines and their metabolites in cerebrospinal fluid (CSF) from 10 PGs and 7 controls. There was a significant difference between the two groups (unpaired t tests): the experimental group showing a decrease in dopamine (DA) and an increase in 3,4–dihydroxyphenanylacetic acid (DOPAC) as well as in homovanillic acid (HVA). The ratio DOPAC/DA and HVA/DA was significantly different as well. Noradrenaline (NA) and its metabolite MHPG were increased, whereas 5-HT and 5-HIAA were unchanged.
Roy et al. (1988) investigated CSF levels of 3–methoxy-4-hydroxyphenolglycol (MHPG) and urinary outputs of noradrenaline (NA), in a group of PG (n = 24), and compared them to controls (n = 20). They showed that gamblers had significantly lower plasma MHPG levels than controls (t = 2.9, p < .007), and significantly greater urinary outputs of NA (F = 11.6, p < .0003). The results are discussed in a theoretical framework.
Blanco et al. (1996) studied platelet MAO activity on 27 male PGs compared to matched controls using the Jackman’s procedure. They found that MAO activity was lower in the PGs than in controls (p < .01, Wilcoxon matched-pairs signed rank test).
Three studies have been performed to examine transmitter activity in relation to PG. Several transmitters have been studied and due to the complex interactions between these neurotransmitters, the results are difficult to interpret.
Genetic Studies
The dopamine D2A1 allele has been connected to substance abuse. Comings et al. (1996) studied the presence of the dopamine (DA) D2 receptor in a PG sample without drug addiction, and a control sample. The group with PG they showed a significantly different occurrence of D2A1 allele (OR = 5.03).
Ibáñez et al. (2001) studied 69 consecutive PGs applying for treatment for their gambling. They used the SOGS to diagnosis PG as well as the full DSM-IV (APA 2000) clinical interview (both Axis I and II). They assessed allele distribution of the dopamine receptor gene (DRD2) polymorphism. The results showed that DRD2 polymorphism was different in gamblers with and without psychiatric comorbidity (X 2 = 13.9, p = .003) and the allele DRD2 C4 allele was present in 42% with comorbid psychiatric disorders compared to 5% without (X 2 = 7.0, p = .008).
Other studies assessed D2, D4 and D1 as well as serotonin and norepinephrine genes as possible risk factors for PG (Comings et al. 2001). A family study also reported that the D1 receptor gene is associated with PG (da Silva Lobo et al. 2007). We consider genetic studies as probable instruments for assessment of PG risk factors.
Cognitive Distortions
The literature reports a range of gambling-related cognitive psychopathology (Toneatto 1999). For example, magnification of gambling skill results in the gambler having an exaggerated self-confidence and ignoring the severity of losses. Superstitious beliefs are characterized by the thought that the gambler has a reliable means of manipulating outcome in his or her favour. Subcategories are talismanic superstitions that a certain object increases winning probability, behavioural superstitions that some rituals can increase winning, and cognitive superstitions that certain mental states can influence winning. The gambler’s fallacy means that a series of losses is expected to be compensated for by chasing and therefore becomes the means by which the gambler recovers losses.
Erroneous Perception or Biased Evaluations
Gilovich (1983) investigated factors leading to continued gambling behavior in spite of loosing more than winning. In one experiment he investigated how much time the subjects (29 students) used in their explanations for losses versus wins. It appeared that they used more time to discuss losses than for wins (t = 2.33, p < .05). In a second experiment, 64 students reported that their memories from the previous week correctly reported 47% of the losses but only 27% of wins (t = 2.33, p < .05).
Savoie and Ladouceur (1995) studied erroneous perceptions in two studies. The question examined was if it is possible to modify the erroneous perceptions through exact information about the probability of negative gains on gambling. The study aimed to revise the erroneous concepts among the participants and to eventually modify their playing customs.
Savoie and Ladouceur’s (1995) first study involved 100 subjects regularly participating in a lottery (53M/47F) and 100 (42M/58F) participating only occasionally. A short interview asked questions regarding their superstitious habits and preferences about lottery, choosing numbers, etc. The probability of winning was estimated as higher amongst the regular players than the occasional players (X 2 = 6.94, p < .01). The experimental group differed from the control group in that they believed that their strategies of choosing their numbers increased their chances to win (X 2 = 13.66, p < .01).
In their second study, 44 regular lottery players were asked about playing habits, concepts about lotteries, participation frequency, strategies involved in choosing numbers, degree of confidence, irregular preferences, and gains from the lottery. Participants were randomly assigned to experimental and control groups (n = 22/22). They were asked to compare the preciseness of their concepts with the actual result probabilities. A month later, they received the same questionnaire again. The results showed that the regular players had more erroneous perceptions than the control. The experimental group had become less confident to win (F = 7.38, p < .025), while the control group had reduced their playing activity during the time (t = 3.2, p < .025).
Illusion of Control
“Illusion of control” means an expectancy of a personal success probability inappropriately higher than the objective probability would warrant. It is based on the idea that factors from skill situations introduced into chance situations cause individuals to feel inappropriately confident. A series of six experiments examined the effects of competition, choice, familiarity, and active involvement on illusion of control (Langer 1975). In these experiments, a variety of factors were manipulated: bias (confidence/shyness) (t = 5.46, p < .005); choice (t = 4.33 p < .005); familiarity (t = 5.46, p < .005); involvement (F = 7.33, p < .01). Subjects did not distinguish chance from skill-determined events. The subjects acted as if they could control outcomes and they gave up the opportunity to exert real control.
In a second study, Langer and Roth (1975) performed an experiment to examine the development of illusion of control. They investigated attributions in a purely chance task (predicting coin tosses), where the task had a descending, ascending, or random sequence of outcomes. Early successes did induce a skill orientation towards the task (F = 4.20, <.05). The subjects with a descending condition rated themselves as significantly better at predicting outcomes than the other two conditions. They also selectively remembered past successes and expected more future successes.
Gilovich and Douglas (1986) also studied the development of illusion of control. They showed that evaluations of randomly determined gambling outcomes were biased (F = 10.72, p < .001). The losers appeared to use manipulated fluke events to explain away the outcome, whereas winners discounted their significance. In a second experiment, the outcomes were shown to be biased towards randomly determined gambling outcomes (t = 2.56, p < .02). Subjects were induced to perceive an “illusion of control” and the outcome affected those who had lost the first bet but had no effect on those who had won. In the no-control condition, the responses were more symmetric to the fluke manipulations by winners and losers.
There is strong evidence that illusion of control is a phenomenon that could interact with the development of PG. Nine experiments have been performed on normal subjects (students and company employers).
Varia
Availability of Plays
Countries with high level of gambling availability have among the highest prevalence rates of pathological gambling. Availability of gambling is correlated with prevalence of pathological gambling (Campell and Lester 1999; Walker 1992).
Ladouceur et al. (1999b) directly tested the effect of increased availability of gambling activities and the rate of pathological gambling in the community by conducting two prevalence studies separated by a 7-year period. The second study conducted after more gambling venues were available showed a 75% increase in the number of pathological gamblers.
Sensory Characteristics
Loba et al. (2001) studied the effect of sensory manipulations (fast speed/sound, slow speed/no sound, counter present), to examine subjective self-reported differences in reaction amongst gamblers. They used video lottery terminals (VLT), a “continuous” form of gambling, where time between wager and payout is short. Subjects included 60 (22F/38M) regular (playing at least twice a month) VLT players, recruited via advertisements. They used the SOGS to assess PG severity and used a survey of subjective reactions to VLT manipulations. The experimental condition was either a video poker game or a 20 min spinning reels game. The game versions had varied sensory characteristics (i.e. slow/no sound, fast/sound, control, counter present). The results showed that non-PG subjects were bothered by fast speed and sound, while PG subjects were bothered by slow speed and no sound. There was a significant main effect of sensory features (F = 11.29, p < .001) and a sensory feature by game interaction (F = 5.50, p < .01). The sensory features gave significant results for the following subjective variables: excitement (F = 9.85, p < .001), enjoyment (F = 7.69, p < .005), tension-reduction (F = 6.95, p < .005), easy to stop (F = 4.62, p < .05), desire to play again (F = 4.86, p < .01), notice difference (F = 19.68, p < .001), and bothered (F = 11.29, p < .001). Decreased speed and turning off the sound decreased ratings of enjoyment, excitement, and tension reduction for PG subjects compared to non-PG subjects. The study supports Griffiths (1993) notion that sensory characteristics are important in the development of PG.
This study (Loba et al. 2001) presented very robust data on seven important subjective variables. The tentative conclusion is that several sensory characteristics may play an important role in the development of PG.
Schedules of Reinforcement
In some early work, Skinner (1953, 1969) defined the schedule of reinforcement as a simple temporal order of response and consequence. The consequence, if positive, could work as a reward for the behaviour or response emitted. If negative, it could have a punishing effect. A continuous reinforcement (reward in each trial) is easy to manipulate. Withdrawing the reinforcer will cause the behaviour to eventually cease (extinguish). An intermittent reinforcement/reward, however, is more resistant against extinction. A fixed ratio (FR10, every tenth occasion), or a variable interval (VI 1 hr, every hour), has a stronger effect, and is less amenable to extinction. A random reinforcement is the strongest conditioning, which is hardest to extinguish. Therefore, the payout interval, in games of chance may be important in the development of PG.
Early wins may induce a skill orientation (Gilovitch 1983) and reinforce the need to try again, even if the contingency schedule is rather intermittent. In the same manner, a big win in gambling can also be analyzed with the operant model. The concept of the near miss has been assessed by Griffiths (1991) and is also in accordance with the operant principles.
Skinner himself did not rely on statistical probability testing, rather he designed straightforward models, where the effect was easily visible and beyond doubt. Therefore, it is unsuitable to add statistical tests to support the importance of the operant conditioning model. The results reported are clearly clinically significant.
There is a vast body of research illustrating the usefulness and importance of the operant model. This model can improve the understanding of the mechanisms of development of PG.
Age of Onset
In a study by Bondolfi et al. (2000), age of onset (before age 21) was shown to be a significant risk factor for PG (X 2 = 10.17, p = .01).
In the Volberg et al. (2001) study, age of onset was shown to be a risk factor for gambling problems (19.9 years for non-gambling problems and 15.6 years for problem gambling) (F = 52.57, p = .000).
Only two studies examined age of onset, and therefore, we consider age of onset as a probable PG risk factor.
Rapid Onset
Breen and Zimmerman (2002) studied the latency of PG-onset (from age of regular involvement, to PG criteria; in years) in 44 consecutive PG subjects. The primary form of gambling at PG-onset was the only variable retained in a stepwise multiple regression analysis (F = 8.42, p < .01). The traditional gamblers had a longer latency period than machine gamblers (3.58 vs 1.08 years; t = 2.90, p < .01).
Only one study has been performed on rapid onset, thus, we consider rapid onset as a probable PG risk factor.
Comorbidity and Concurrent Symptoms
Depression
In a study by Getty et al. (2000), a difference in depression between experimental (mean 17.73) and control (mean 8.80) groups (F = 17.43, p < .001) was noted. In the study by Ibáñez et al. (2001), higher rates of depression, as examined by the Beck Depression Inventory (Beck and Steer 1993), were found in the more serious pathological gamblers (t = 3.4, p = .0001).
A study by Potenza et al. (2005) found that in males with PG, 34% of the genetic variance for major depressive disorder (MDD) contributed to PG and vice versa.
With three studies performed on depression, we consider depression as a probable PG risk factor.
Anxiety
In the study by Ibáñez et al. (2001), trait anxiety (STAIT) was significantly higher amongst the more severe pathological gamblers (t = 2.0, =.05).
With only one study performed on anxiety, we consider anxiety as a probable PG risk factor.
OCD
Frost et al. (2001) studied the relationship between OCD symptoms and PG in 89 subjects (48F/41M). PG was diagnosed through use of the SOGS and OCD symptoms through the YBOCS (Yale-Brown Obsessive Compulsive Scale) (Goodman et al. 1989a, b) as well as a hoarding scale. Thirty-six participants met criteria for PG. The results showed higher intensity of symptoms for the PG group than for the rest, particularly for obsessions (t = 3.45, p < .001), compulsions (t = 2.77, p < .01), hoarding (t = 2.71, p < .01), urge to gamble (t = 36–16, p < .001), avoidance (t = 4.30, p < .001), and impulsivity (IES; t = 3.66, p < .005).
With six areas covered (by one paper) on the relationship of OCD to PG, we consider OCD as a probable risk factor for PG. This connection also supports the close relationship between obsessions and obsessive gambling.
Alcohol Abuse
In the study by Feigelman et al. (1995), both a lifetime alcohol problem (r xy = 0.14, p = .02) and use of alcohol within the last month (r xy = 0.14, p = .02) were significantly related to problem gambling.
In the Ladouceur et al. (1999a) study, statistical analyses showed a relationship between SOGS scores and alcohol use (F = 24.71, p < .0001).
In a twin study of genetics, between 12% and 20% of the genetic variation in the risk for PG was accounted for by the risk for alcohol dependence (Slutske et al. 2000).
We consider alcoholism as a probable PG risk factor.
Other Drugs
In the study by Feigelman et al. (1995), there was a significant relationship between problem gambling, having a major drug problem within the last year (r xy = 0.12, p = .04), and frequency of heroin use (r xy = 0.14, p = .02).
The study mentioned above (Ladouceur et al. 1999a) also showed a relationship between SOGS scores and cigarette smoking (F = 20.42, p < .0001). In the study by Winters et al. (1993b), frequent drug use was significantly connected to more problem gambling (X 2 = 46.2, p < .001). The study mentioned above (Ladouceur et al. 1999a) also showed a relation between SOGS and drug use (F = 29.09, p < .0001). The helpline study by Potenza et al. (2001) reported that drug use was identified as a significant risk factor (X 2 = 5.66, p < .02).
Because the five studies were examining different drugs, we consider drug abuse a probable PG risk factor.
Personality Disorders
In a study by Slutske et al. (2001), 7,869 men from 4,497 twin pairs (Vietnam Era Twin Registry) were diagnosed with PG using the DSM-III-R (APA 1987), with whilst using the DIS for antisocial behavior disorders. Telephone interviews were conducted to ascertain prevalence rates of PG and antisocial behavior disorders. The results showed elevated prevalence rates of antisocial personality disorder amongst individuals with a history of PG (OR = 6.4).
In a study by Ibáñez et al. (2001), there were more personality disorders in the more severe cases of PG subjects (t = 3.0, p = .004).
With only two studies performed on personality disorders, we consider personality disorder as a probable PG risk factor.
Personality Symptoms and Characteristics
Coping Styles
Getty et al. (2000) studied a group of members of Gamblers Anonymous (n = 30) compared to a matched control group (n = 30). The PG diagnosis was made using the SOGS. The study used the Problem-Focused Styles of Coping Inventory (PF-SOC). All types of coping styles, suppressive, reactive and reflective were significantly different between the experimental and control group with the experimental group being higher on suppressive (F = 13.81, p < .001) and reactive (F = 16.22, p < .001) while lower on reflective coping styles (F = 7.81, p < .007).
With only one study performed on maladaptive coping, we consider maladaptive coping as a probable PG risk factor.
Impulsivity
Vitaro et al. (1997) studied impulsivity among 754 adolescent boys using the Eysenck Impulsiveness Scale (EIS). PG severity was assessed with the SOGS. There was a clear relationship between greater PG severity and high rates of impulsivity (X 2 = 30.58, p < .01).
With only one study performed on impulsivity, we consider impulsivity as a probable PG risk factor.
Hyperactivity (ADHD)
Carlton and Manowicz (1994) found that adult pathological gamblers had a higher than average ate of childhood attention deficit hyperactivity disorder (ADHD). In a retrospective study and subsequent EEG assessments of pathological gamblers, they found that gamblers had patterns of activation similar to children with ADHD. The study does not include significance evaluations and is therefore not mentioned in the table.
Sensation Seeking
Blanco et al. (1996) investigated 27 PG and 27 matched control subjects found significant differences between the two groups (WMP = Wilcoxon Mathed Pairs) on the Sensation Seeking Scales (subscale Thrill and Adventure seeking, both p < .03; and Disinhibition, p < .01).
With only one study performed on sensation seeking, we consider sensation seeking as a probable PG risk factor.
Delinquency, Criminal and Illegal Activity
In the study by Winters et al. (1993b), delinquency status, illegal activity, or arrest were related to problem gambling (X 2 = 47.3, p < .001).
In the study by Feigelman et al. (1995), there was a significant relationship between problem gambling and criminality in general (r xy = 0.25, p = .001) as well as in number of arrests for criminal offenses (r xy = 0.16, p = .01).
The study conducted by Ladouceur et al. (1999a) also showed a relationship between SOGS scores and delinquency as measured by the Self-Reported Delinquency Scale (F = 176.18, p < .0001).
In the helpline study by Potenza et al. (2001), delinquency was identified as a significant risk factor (X 2 = 9.53, p < .002).
In the study by Hall et al. (2000), there was a clear relation between PG and duration of incarceration (t = 16.53, p < .001) and illegal activity (t = 7.83, p = .02).
With five studies performed on delinquency/illegal activity, we consider maladaptive delinquency/illegal activity as a well-established PG risk factor.