Skip to main content
PMC home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2020 May 1.
Published in final edited form as: Clin Psychol Sci. 2019 Jan 29;7(3):582–596. doi: 10.1177/2167702618812700

Potential Causal Influence of Neighborhood Disadvantage on Disordered Gambling: Evidence From a Multilevel Discordant Twin Design

Wendy S Slutske 1, Thomas M Piasecki 2, Arielle R Deutsch 3, Dixie J Statham 4, Nicholas G Martin 5
PMCID: PMC6521693  NIHMSID: NIHMS1510372  PMID: 31106041

Abstract

The quality of the neighborhood in which one lives has been linked to disordered gambling (DG), but whether this reflects a causal relation has not yet been empirically examined. Participants were 3,450 Australian twins who completed assessments of past-year DG and personality and for whom census-derived indicators of disadvantage were used to characterize their neighborhood. Multilevel models were employed to estimate within-twin-pair and betweentwin-pair effects of neighborhood disadvantage on DG, with the within-twin-pair effect representing a potentially causal association and the between-twin-pair effect representing a noncausal association. There was robust evidence for a potentially causal (as well as a non-causal) effect of neighborhood disadvantage on DG (in contrast, parallel analyses of past-year alcohol use disorder failed to find evidence of a potentially causal effect). These results support efforts focused on identifying the active ingredients contributing to the effect of neighborhood disadvantage on DG and developing interventions to limit their impact.

The quality of the neighborhood in which one lives has been linked to disordered gambling (DG) in the United States, Great Britain, New Zealand, and Australia (Welte et al., 2004; Wardle et al, 2011; Barnes et al., 2013; Pearce et al., 2008; Slutske et al., 2015). In Australia, the prevalence of past-year DG was nearly eight times higher among those living in the highest decile of neighborhood disadvantage (3.3%) compared to those living in the lowest decile (0.4%; Slutske et al, 2015). In the United States, those living in the highest quintile of neighborhood disadvantage reported 2.5 times as many DG symptoms as those in the lowest quintile, and the association between neighborhood disadvantage and DG remained even after a host of demographic, personality, family, and gambling accessibility measures were entered as predictors in a multivariate analysis (Welte et al., 2017). The authors of the study concluded by noting that “…there is something about poor neighborhoods that in itself promotes problem gambling…we have speculated that individuals living in disadvantaged neighborhoods have few examples of financial success by conventional means and are therefore more vulnerable to the attraction of gambling” (p. 339, Welte et al, 2017).

Despite the robustness of these findings, it may be premature to conclude that the neighborhood in which one lives is truly a measure of the environment because, like many other putative environments (Kendler & Baker, 2007; Plomin et al, 2016), living in a deprived neighborhood is partially heritable (Marioni et al, 2014; Sariaslan et al, 2016; Slutske et al, 2015). In observational studies (as in life), environments are not randomly assigned to people. Rather, our environments arise in part from genetically-influenced choices based on our abilities, interests, talents, and proclivities and are inherited along with our genes from our parents (Scarr & McCartney, 1983). Establishing that a risk factor is truly environmental requires more sophisticated research designs than have typically been employed in the study of DG.

This is a question that is of critical importance to public health (Kendler, 2017). Establishing that a risk factor is truly environmental is synonymous with establishing that the association between the risk factor and the outcome is causal. The discordant twin design is a “natural experiment” or “quasi- experimental” method (McGue, et al, 2010; Rutter, 2007a; Rutter, 2007b), in which an unexposed twin serves as the control for an exposed cotwin in examining potentially causal associations. It is based on the knowledge that monozygotic (MZ) twins are perfectly correlated for genetic and family environmental background factors, and dizygotic (DZ) twins are perfectly correlated for family environmental background factors and correlated 0.5 (on average) for genetic factors. A comparison of DG among MZ twins who are discordant for an exposure such as neighborhood disadvantage allows one to control completely for genetic and family environmental background factors (that is, between-family risk factors), and a comparison among discordant DZ twins allows one to partially control for genetic and completely control for family environmental background factors. This type of control is more powerful than the standard statistical controls that are commonly used in the behavioral sciences because it controls for a whole host of possible confounds – including those that we know about and even those that we do not. A potentially causal effec1 of neighborhood disadvantage on DG would be implicated when the twin living in a more disadvantaged neighborhood has a greater likelihood of DG than the cotwin living in a less disadvantaged neighborhood. On the other hand, if there are no differences between the twin living in a more disadvantaged neighborhood and the cotwin living in a less disadvantaged neighborhood in the likelihood of DG, then the association between neighborhood disadvantage and DG is more likely to be due to between-family differences that are related to both living in a disadvantaged neighborhood and DG. These two scenarios are not mutually exclusive, and neighborhood disadvantage could be both a causal factor as well as non-causally associated via between-family differences.

Although increasingly deployed in the behavioral sciences (D’Onofrio et al, 2013; Lahey & D’Onofrio, 2010), this research design has only been used in one previous investigation of an important putative risk factor for DG – early age of gambling initiation (Lynch et al, 2004; Kessler et al, 2008). The discordant twin design was employed to answer the question of whether an earlier age of initiation of gambling was causally related to later regular gambling and DG (Slutske et al, 2014). There was little evidence that early gambling was causally related to later gambling involvement and disorder after shared familial risk factors were controlled. That is, the earlier-gambling twin was no more likely to proceed to gambling regularly or to develop symptoms of DG than was the later-gambling cotwin. This suggests that early gambling initiation is better thought of as an early marker of genetic and family environmental risk factors for gambling disorder. This finding has potentially important public health significance in that it suggests that efforts to delay gambling among youth may not have the intended consequence of reducing the numbers of individuals that develop a gambling disorder.

The present study represents an investigation of another putative risk factor for DG -- residing in a disadvantaged neighborhood. A previous investigation found that genetic, shared environmental, and unique environmental factors explained 25%, 20%, and 55% of the variation in residing in a disadvantaged neighborhood (Slutske et al, 2015). One of the main questions of interest was the extent to which the genetic and environmental contributions to moving to or remaining in a disadvantaged neighborhood overlapped with genetic and environmental contributions to gambling involvement and disorder. The association between neighborhood disadvantage and the frequency of gambling was completely explained by overlapping familial (genetic and shared environmental) risk factors, which is inconsistent with a potentially causal association. However, the association between neighborhood disadvantage and DG among men was explained in part by non-familial (unique environmental) factors, which is consistent with a potentially causal association (Slutske et al, 2015).

Because comparing discordant MZ twins only controls for genetic and shared environmental background factors, there still exists the possibility that there may be individual-specific factors related to twin discordance, that is, the differential exposure of the two twins, that are also related to the outcome. Thus, when conducting a discordant-twin study, it is also important to explore the factors associated with twin discordance in exposure (McGue, et al, 2010). For example, socioeconomic and personality differences may lead to one twin living in a more disadvantaged neighborhood than the other twin, and this difference may also be related to the one twin developing a gambling disorder. Therefore, we also compared the results of models that included covariates, such as educational attainment, household income, and the Big 3 personality traits of positive emotionality, negative emotionality and behavioral undercontrol, to unadjusted models in order to establish the robustness of the effects.

Methods

Participants

Participants were selected from a sample of 4,764 members of the Australian Twin Registry Cohort II (57% female, age range = 32–43 years [mean = 37.7]). Nearly all of the participants were White (~97%) with the next most common racial category being Asian (~2%). The participants were overwhelmingly of European origin, particularly northwest Europe, and mainly the United Kingdom and Ireland (see Slutske et al, 2009 for more participant details). The participants represented a broad cross-section of the general population of Australia. Similar to most community-based twin cohorts, the sample was relatively socio-economically advantaged (median household income of approximately AU $62,500 [equivalent to US $49,375]). Data from participants who were living overseas or whose state of residence was unknown were excluded. Because complete pairs were necessary for the multilevel models the sample was limited to the 3,596 individuals whose cotwin’s data was also available. Informed consent was obtained from all participants and the study was approved by the Institutional Review Boards at the University of Missouri and the QIMR Berghofer Medical Research Institute.

Procedure

Respondents completed a structured psychiatric telephone interview conducted in 20042007 (response rate=80.4%) during which gambling behaviors were assessed. Interviews were conducted by trained lay-interviewers who were blind to the status of the co-twin. The participants who completed the telephone interview were mailed a paper-and-pencil personality questionnaire that was usually returned within one month of the interview. Personality questionnaires were available for 3,106 (86%) of the participants whose data were included in this study.

Environmental Exposure Measure

Neighborhood disadvantage.

Neighborhood socioeconomic disadvantage was obtained by matching the participants’ postal codes to information provided by the Australian Bureau of Statistics using data from the 2006 Australian census (Pink, 2008a)2. Census data were matched successfully for 3,450 (96%) of the complete twin pairs (970 MZ female, 632 MZ male, 692 DZ female, 404 DZ male, 752 unlike-sex DZ); this was the final analytic sample. Out of 2,515 postal codes in the 2006 Australian census, there were 1,245 different postal codes represented in the sample, with up to 23 participants residing in each (M = 2.95, SD = 2.76). Seventy-nine percent of the participants resided in a city and 21% in a rural area. The median population size of the postal code areas of the participants in this sample was 16,305 residents. Forty percent of the postal code areas included a single participant, and 81% included fewer than five participants. The majority of twins and co-twins (80%) resided in a different postal code area.

An index of relative socioeconomic disadvantage (IRSD) created by the Australian Bureau of Statistics was used (Pink, 2008a, Pink, 2008b). This index has been used in its present form since 1986, and is updated every five years subsequent to each new Australian census. The construct of relative socio-economic disadvantage was defined in terms of “people’s access to material and social resources, and their ability to participate in society” (Pink, 2008a, p. 17). The IRSD is based on the results of a principal components analysis of 17 census-derived indicators and takes into account the proportion of households in the specified geographic area characterized by (for example) low income, low educational attainment, unskilled employment or unemployment, single-parenthood, subsidized living or low rent housing (see Slutske et al, 2015 for more details about the IRSD). The IRSD index was reversed so that higher scores reflected greater disadvantage. For these analyses we used the IRSD decile, which ranged from 1 (lowest disadvantage) to 10 (highest disadvantage). The postal codes represented in this sample were slightly less disadvantaged on average than the Australian population. Although the entire range of disadvantage was represented, postal codes in the top three deciles were underrepresented, and in the bottom three deciles of disadvantage were over-represented.

Outcome measures

The main outcome measure of interest was symptoms of past-year disordered gambling. Analyses were also conducted for symptoms of past-year alcohol use disorder for comparison.

Disordered Gambling.

Past-year symptoms of DG were assessed using the National Opinion Research Center DSM Screen for Gambling Problems (Gerstein et al., 1999) and the South Oaks Gambling Screen (SOGS; Lesieur & Blume, 1987). These two measures of DG were correlated r = .70 (p < .0001). The nine DSM-53 disordered gambling symptoms and 20 items from the SOGS were summed to form the DG outcome measure (coefficient alpha = 0.85). A one-factor model provided an excellent fit to these 29 indicators (Slutske et al., 2015). Biometric modeling revealed that the familial sources of variation in liability to DSM disordered gambling symptoms completely overlapped with the familial sources of variation in liability to SOGS disordered gambling symptoms (Slutske et al, 2011). Using the SOGS in conjunction with the DSM symptom set more thoroughly describes the full DG continuum (Slutske et al, 2011). A previous study reported that genetic and unique environmental factors explained 45% and 55% of the variation in past-year DG among men, and 42% and 58% of the variation among women (Slutske et al, 2015).

Alcohol use disorder.

Past-year symptoms of DSM-IV alcohol use disorder were assessed using the World Health Organization Composite International Diagnostic Interview (Kessler et al, 1998; World Health Organization, 1992). The seven alcohol dependence symptoms and the four alcohol abuse symptoms were summed to form the alcohol use disorder outcome measure (coefficient alpha = 0.73).

Covariate Measures

Personality dimensions.

The personality questionnaire contained a modified 177-item version of the Multidimensional Personality Questionnaire (MPQ; Tellegen & Waller, 2008; Slutske, Cho, Piasecki, & Martin, 2013). The wording of some of the MPQ items was modified for use in Australia. The MPQ contains items that measure ten lower-order factors which in turn comprise three distinct superfactors (the “Big Three”): Positive Emotionality includes Wellbeing, Social Potency, Achievement, and Social Closeness; Negative Emotionality includes Stress Reaction, Alienation, and Aggression; and Constraint includes Control, Harm Avoidance, and Traditionalism. Scale scores were created by dividing the sum of the items by the number of non-missing items; scales with more than 3 missing items were considered missing. These scale scores were converted to T-scores (i.e., mean = 50, SD = 10) prior to inclusion as predictors in multi-level model analyses to produce more interpretable scaling of model coefficients. The internal consistency reliabilities in the present study for the three superfactors (α = 0.85 to 0.88, mean α = 0.87) were good. The percentage of variation in the three superfactors attributable to genetic, shared, and nonshared environment were 48%, 0, and 52% for Positive Emotionality, 43%, 1%, and 57% for Negative Emotionality, and 40%, 7%, and 53% for Constraint, which is consistent with the worldwide twin literature on personality (Polderman et al, 2015).

Socioeconomic status.

Participants were asked to report their current combined household gross income before taxes using a respondent booklet with a list of 12 income ranges to use as response options. These 12 ordered categories were used as an ordinal income variable in analyses. Participants also reported their highest educational level with the options of 8–10 years schooling (coded ‘1’), matriculation/year 12 (‘2’), technical, teachers' college, TAFE, business or secretarial college (‘3’), university under-graduate training (‘4’) and university post-graduate training (‘5’). These ordered categories were treated as an ordinal educational attainment variable in analyses.

Analytic Plan

Descriptive analyses.

Associations between study variables were estimated using a survey data analysis procedure in SAS (SAS Institute Inc, 2009) that took into account the non-independence of twin pair observations.

Multi-level modeling analyses.

Multilevel modeling (also known as hierarchical linear, mixed, or random effects modeling) is an extension of conventional regression analysis that has become increasingly popular in the behavioral sciences (Snijders & Bosker, 1999). It is used to take into account and to properly model levels of nesting, or clustered observations, within a dataset. In the case of data from pairs of twins, there are two levels of nesting – the individual twin level is nested within the twin pair level. Multilevel modeling is a flexible and powerful data analytic approach that can accommodate a continuous index of twin discordance and include covariates for the identification of potential mediators of effects.

Estimation of the multilevel models was conducted using PROC GLIMMIX for generalized linear mixed models (GLMMs) in SAS software version 9.4 (SAS Institute Inc, 2009). GLMM is a form of multilevel modeling for the analysis of clustered data with nonnormal dependent measures (Hedeker, 2005). A negative binomial distribution and log link function were used because the DG symptom count was skewed with a large proportion of respondents reporting zero symptoms (81.3% of analyzed sample). Both level 1 and 2 variances were estimated, along with a random intercept of twin pair. Coefficients from these models were exponentiated and expressed as incidence rate ratios (IRRs).

An initial set of co-twin control models (McGue, et al. 2010) was performed to (a) test whether there was an overall association between neighborhood disadvantage and DG symptoms, and (b) to identify potential sources of confounding that might contribute to the overall effect. First, the overall association, or “individual level” effect of exposure was tested in the full sample of twins. Next, separate models were performed that were limited to either DZ or MZ twins. In these zygosity-limited models, the neighborhood disadvantage indicator was decomposed into two variables indexing level 1 (i.e., within twin pair) and level 2 (i.e., between twin pair) sources of variance (Begg & Parides, 2003). The between-pair effect was represented using the mean score across both twins. The within-pair effect was indexed by subtracting the pair mean from each twin’s own score. In this parameterization, the within-pair effect (i.e., comparison against the cotwin) represents the quasi-causal unique effect of an environmental exposure, and the between pair effect measures between-family genetic and shared environmental confounds that contribute to the phenotypic association between the environmental measure and outcome (McGue, et al., 2010; Turkheimer & Harden, 2014).

The estimates of the within-pair effect among DZ twins accounts for shared environmental and partially accounts for genetic confounds. Thus, if this effect is markedly smaller than the individual-level estimate, this indicates that shared environmental or genetic factors may be inflating the overall association. The within-pair effect among MZ twins represents the most stringent test, as it fully controls for both genetic and shared environmental confounds. If this estimate is significant, this provides strong evidence for a quasi-causal effect of neighborhood disadvantage exposure on DG. Comparison of the magnitude of the within-pair MZ effect to the individual-level and within-pair DZ estimates can indicate the presence and likely sources of confounding, if any, affecting the overall association (McGue, et al., 2010).

Next, a series of models was tested in the pooled sample of twins, incorporating additional explanatory variables that might help account for any disadvantage effects observed in the co-twin control analyses. A base model predicted DG from the within- and between-pair neighborhood disadvantage variables, adjusting for sex and age4. Because neither the main effect for zygosity (IRR = 1.05, 95% CI = 0.83 to 1.34, p = .675) nor the zygosity x within-pair disadvantage effect (IRR = 0.95, 95% CI = 0.85 to 1.07, p = .3924) was significant, these predictors were not included in the models. Subsequent models added within- and between-pair effects of (a) Big 3 personality traits, (b) educational attainment, and (c) household income. A final, fully adjusted model incorporated all these explanatory covariates simultaneously. Of particular interest were the significance of the within- and between-pair effects included in each of the models, as well as the extent to which the inclusion of the predictors reduced or completely eliminated the within- and between-pair effects of neighborhood disadvantage.

Results

Descriptive Findings

Most twin pairs (69%) were discordant for their level of disadvantage; 19% of pairs differed by one decile, 15% by two, 11% by three, 9% by four, and 15% by five or more deciles. The average discordance was 2.09 (SD = 2.14) deciles.

The correlation between DG and neighborhood disadvantage was significant but modest (r = 0.10, p < .0001; converted to standardized mean difference [effect size]: d = 0.20). DG was significantly associated with nearly all of the covariates (see Table 1). Negative emotionality, educational attainment, and household income were significantly associated with both DG and neighborhood disadvantage, implicating these covariates as potentially accounting for the within- twin-pair (i.e. potentially causal) or between-twin-pair (i.e. family-level confounding) effects of disadvantage on DG in the multilevel models.

Table 1.

Correlations between study variables

DG AUD Sex Age Disad PE NE Constraint Educational attainment Income
Disordered gambling
Alcohol use disorder .27
Sex .07 .10
Age .02 −.03 −.02
Disadvantage .10 .01 .01 −.01
Positive emotionality −.04 −.03 .10 −.02 −.15
Negative emotionality .21 .17 .08 −.03 .12 −.17
Constraint −.08 −.15 −.32 .01 .03 −.15 −.14
Educational attainment −.10 −.05 −.02 −.04 −.23 .19 −.18 −.10
Household income −.06 −.05 .10 .01 −.28 .23 −.20 −.03 .29
Mean 0.42 0.35 0.41 37.60 4.38 0.58 0.26 0.68 2.92 8.20
Standard deviation 1.46 1.00 0.49 2.30 2.66 0.15 0.15 0.14 1.32 2.22
Open in a new tab

Note: Significant correlations (p < .05) are in bold font. Sex is coded 0 = female, 1 = male.

Multi-level Models

Findings from the co-twin control analyses are depicted in Figure 1. The individual-level effect of neighborhood disadvantage was significant (IRR = 1.11, p < .001), indicating that a 1decile difference in disadvantage was associated with an 11% increase in the expected DG symptom count. The within-pair disadvantage effect among DZ twins was similar in magnitude, though not statistically significant (IRR = 1.08, p = .075). The within-pair effect in MZ twins was significant and comparable in magnitude to individual-level and DZ within-twin estimates (IRR = 1.14, p < .001). The presence of a within-pair effect among MZ twins provides strong evidence for a quasi-causal effect of neighborhood disadvantage on gambling symptoms. A supplemental model in the pooled sample of MZ and DZ twins indicated that zygosity did not moderate the within-pair effect of neighborhood disadvantage, interaction IRR = 0.95, 95% CI = 0.85 to 1.07, p = .392, meaning that the within-pair effect did not significantly differ among MZ and DZ twins. The fact that the magnitude of this within-pair effect was comparable to the individual-level effect despite co-twin controls is potentially consistent with a fully causal exposure effect (McGue, et al., 2010).

Figure 1.

Figure 1.

Open in a new tab

Effect sizes (expressed as incidence rate ratios) and 95% confidence intervals for the association between neighborhood disadvantage and DG symptom count at the individual-level and at the within-pair level in DZ and MZ twins. Estimates of each effect are depicted when covarying sex and age (initial models) and when fully adjusting (fully adjusted models) for all covariates listed in Table 2. The within-pair effect of neighborhood disadvantage did not significantly differ among DZ and MZ twins in the initial models (p = .392) or in the fully-adjusted models (p = .3881). Note: IRR = incidence rate ratio, DZ = dizygotic, MZ = monozygotic, DG = disordered gambling.

Findings from models investigating effects of measured explanatory variables are summarized in Table 2. In the base model, both within-pair (IRR = 1.11, p < .001) and between-pair (IRR = 1.13, p < .001) differences in neighborhood disadvantage were significantly associated with gambling symptom count. When the base model was expanded to include personality traits as additional predictors, the quasi-causal within-pair disadvantage effect remained (IRR = 1.09, p = .006). Additionally, significant within-pair effects for two traits indicated quasi-causal influences, such that the twin reporting higher Negative Emotionality (IRR = 1.04, p < .001) and lower Constraint (IRR = 0.96, p < .001) experienced more gambling symptoms in the past year compared to the co-twin. The between-pair effect for disadvantage also remained significant but was somewhat diminished (IRR = 1.08, p = .015). Between-pair Negative Emotionality (IRR = 1.08, p < .001) was significant, and thus potentially accounts for some of the familial association between living in disadvantaged neighborhoods and gambling symptoms. Neither the within-pair nor between-pair effects for Positive Emotionality were significant (ps > .597)

Table 2.

Results from multilevel models predicting disordered gambling symptom counts

Predictor Base Model
 Personality
 Education
 Income
 Fully Adjusted
IRR 95% CI IRR 95% CI IRR 95% CI IRR 95% CI IRR 95% CI
Sex 1.80*** (1.45, 2.24) 1.46** (1.14, 1.87) 1 78*** (1.43, 2.21) 1.88*** (1.50, 2.35) 1 47** (1.14, 1.90)
Age 1.02 (0.97, 1.08) 1.01 (0.95, 1.07) 1.02 (0.97, 1.07) 1.03 (0.98, 1.08) 1.02 (0.96,1.08)
WP Disadvantage 1.11*** (1.05,1.18) 1.09*** (1.02, 1.16) 1.11*** (1.05, 1.17) 1.11*** (1.05, 1.18) 1.08* (1.02, 1.15)
BP Disadvantage 1 13*** (1.07, 1.20) 1.08* (1.01, 1.14) 1.09** (1.03, 1.15) 1.10*** (1.04, 1.17) 1.04 (0.98, 1.11)
WP Positive Emotionality 1.00 (0.98, 1.01) 1.00 (0.98, 1.02)
BP Positive Emotionality 1.00 (0.98, 1.01) 1.00 (0.98, 1.02)
WP Negative Emotionality 1.04*** (1.03, 1.06) 1.04*** (1.02, 1.06)
BP Negative Emotionality 1.07*** (1.06, 1.09) 1.07*** (1.05, 1.09)
WP Constraint 0.96*** (0.95, 0.98) 0.96*** (0.95, 0.98)
BP Constraint 0.99 (0.97, 1.00) 0.99 (0.97, 1.00)
WP Educational Attainment 0.86* (0.75, 0.97) 0.90 (0.77, 1.04)
BP Educational Attainment 0.77*** (0.69, 0.86) 0.85* (0.75, 0.96)
WP Household Income 0.94 (0.88, 1.00) 0.98 (0.91, 1.05)
BP Household Income 0.90** (0.84, 0.97) 0.97 (0.89, 1.05)
Open in a new tab

Note: Because neither the main effect for zygosity (IRR = 1.05, 95% CI = 0.83 to 1.34, p = .675) nor the zygosity x within-pair disadvantage effect (IRR = 0.95, 95% CI = 0.85 to 1.07, p = .3924) was significant, these predictors were not included in the models.

WP = within-pair, BP = between-pair

*

= p < .05

**

= p < .01

***

= p < .001

The within-pair (IRR = 1.11, p < .001) and between-pair (IRR = 1.09, p = .002) effects of neighborhood disadvantage both remained significant in the model accounting for educational attainment. Lower education potentially exerted a causal effect on gambling symptoms (within pair IRR = 0.86, p = .019). Twins from families with lower average levels of educational attainment experienced more past-year gambling symptoms (IRR = 0.77, p < .001).

In the model accounting for household income, disadvantage effects remained significant both at the within-pair (IRR = 1.11, p < .001) and between-pair (IRR = 1.11, p < .001) levels. The within-pair effect for household income was nominally significant (IRR = 0.94, p = .052). Twins from families with lower incomes tended to report more past-year DG symptoms (between-pair IRR = 0.90, p = .005).

In the fully adjusted model, the within-pair disadvantage effect remained significant but was reduced in magnitude (IRR = 1.08, p = .016). Significant within-pair effects were found for Negative Emotionality (IRR = 1.04, p < .001) and Constraint (IRR = 0.96, p < .001), suggesting that these traits may be important mechanisms accounting for some of the apparent quasi-causal effect of neighborhood disadvantage on disordered gambling symptoms. The between-pair disadvantage effect was not significant in the fully adjusted model (IRR = 1.04, p = .198). At the between-pair level, Negative Emotionality (IRR = 1.07, p < .001) and educational attainment (IRR = 0.85, p = .011) were the only explanatory variables associated with gambling symptoms.

Given evidence that some explanatory variables were related to DG symptoms at the within-pair level, we conducted the sequence of co-twin control models after fully adjusting for these variables. Results were similar to the initial models (Figure 1). Disadvantage remained significantly associated with DG symptoms at the individual-level (IRR = 1.07, p = .003). The within-pair disadvantage effect was not significant among DZ twins (IRR = 1.06, p = .232) but was significant among MZ twins (IRR= 1.13, p =.016), and the within-pair effect of neighborhood disadvantage among DZ and MZ twins did not significantly differ (p = .3881).

Probing the Effects of Negative Emotionality and Constraint

Additional analyses sought to further characterize the effects of Negative Emotionality and Constraint in the personality-adjusted models by omitting these traits and substituting their corresponding facet scale scores. All three facets of Negative Emotionality (Stress Reaction, Aggression, and Alienation) were significantly associated with both DG and neighborhood disadvantage (see Supplemental Table 1), suggesting that they may account for the within- and between-twin-pair effects of disadvantage on DG in the multilevel models. Facets of Constraint were associated with DG (Control and Harm Avoidance) or neighborhood disadvantage (Traditionalism), but not both.

In a multilevel analysis, within-pair differences in Stress Reaction (IRR = 1.03, p = .006), Aggression (IRR = 102, p = .009) and Control (IRR = 0.97, p < .001) were significant predictors of DG (Table 3). At the between-pair level, Negative Emotionality facets of Stress Reaction (IRR = 1.02, p = .047), Alienation (IRR = 1.04, p < .001) and Aggression (IRR = 1.03, p = .001) and the Constraint facets of low Control (IRR = 0.91, p < .001) and high Traditionalism (IRR = 1.02, p = .027) were significant predictors of DG.

Table 3.

Results from multilevel models testing facets of MPQ Negative Emotionality and Constraint predicting disordered gambling symptom counts

Predictor IRR 95% CI p
Sex 1.40 (1.06, 1.85) 0.016
Age 1.00 (0.95, 1.06) 0.945
WP Disadvantage 1.09 (1.03, 1.16) 0.006
BP Disadvantage 1.05 (0.99, 1.12) 0.081
WP Positive Emotionality 0.99 (0.98, 1.01) 0.492
BP Positive Emotionality 0.99 (0.97, 1.01) 0.238
Negative Emotionality
    WP Stress Reaction 1.03 (1.01, 1.05) 0.001
    BP Stress Reaction 1.02 (1.00, 1.04) 0.047
    WP Alienation 1.00 (0.99, 1.02) 0.698
    BP Alienation 1.04 (1.02, 1.06) < .0001
    WP Aggression 1.02 (1.01, 1.04) 0.009
    BP Aggression 1.03 (1.01, 1.05) 0.001
Constraint
    WP Control 0.97 (0.95, .99) 0.0004
    BP Control 0.96 (0.95, .98) < .0001
    WP Harm Avoidance 0.99 (0.97, 1.01) 0.168
    BP Harm Avoidance 1.00 (0.99, 1.02) 0.669
    WP Traditionalism 0.99 (0.98, 1.01) 0.452
    BP Traditionalism 1.02 (1.00, 1.04) 0.027
Open in a new tab

Note: Because neither the main effect for zygosity (p = .675) nor the zygosity x within-pair disadvantage effect (p = .3924) was significant, these predictors were not included in the models.

MPQ = Multidimensional Personality Questionnaire;

WP = Within-pair; BP = Between-pair

Probing the Effects of Using Alternate Measures of DG

All of the models were re-run substituting as the dependent variable either a DSM-5 DG symptom count (Supplemental Table 2) or SOGS symptom count (Supplemental Table 3). Using the DSM-5 measure, there was evidence for within-pair and between-pair effects in the baseline analyses that were no longer significant after inclusion of the full set of covariates. Using the SOGS measure, the within-pair effect and between-pair effects were significant in the baseline model. The within-pair effect persisted after including covariates but the between-pair effect did not.

Probing the Similarity of the Results Obtained for Another Addictive Disorder

Analyses were conducted for alcohol use disorder to discern whether the results were similar to those obtained for DG. Inspection of Table 1 reveals that alcohol use disorder and DG had similar correlates except for neighborhood disadvantage, which was not significantly associated with past-year alcohol use disorder symptoms (r = .01). Co-twin control analyses with the DSM-IV alcohol use disorder symptom count as the dependent variable indicated there was no effect of neighborhood disadvantage on AUD symptoms at the individual level (IRR = 1.00, 95% CI = 0.97 to 1.03, p = .995). The within-pair effects for disadvantage were not significant in either DZ (IRR = 1.01, 95% CI = 0.93 to 1.10, p = .749) or MZ twins (IRR = 0.99, 95% CI = 0.99 to 1.07, p = .740). In multilevel modeling analyses incorporating explanatory variables (Supplemental Table 4), there was no evidence for a within-pair effect of neighborhood disadvantage on AUD (ps > .233). After including covariates, however, there was a significant inverse relation between the between-pair effect of neighborhood disadvantage and AUD wherein twins from families who were living in less disadvantaged neighborhoods reported more past-year symptoms of alcohol use disorder (IRR = 0.92, 95% CI = 0.87 to 0.97, p = .003).

Discussion

The purpose of this study was to examine the relation between neighborhood disadvantage and DG using a research design that can determine whether the association is truly environmental and potentially causal. Using a discordant twin design, levels of DG were compared within twin pairs who were discordant for neighborhood disadvantage. A potentially causal effect of neighborhood disadvantage on DG would be implicated if the twin residing in a more disadvantaged neighborhood exhibited more symptoms of DG than the co-twin living in a less disadvantaged neighborhood.

Potentially Causal Effect of Neighborhood Disadvantage.

There was a significant association between the MZ within-pair difference in neighborhood disadvantage and DG that was as large as the overall association, suggesting that the relation is not confounded. However, because comparing discordant MZ twins only controls for genetic and shared environmental background factors, there still exists the possibility that there may be unique (non-shared) environmental factors related to twin discordance, that is, reasons why one twin is living in a more disadvantaged neighborhood than the other, that is also related to DG. Therefore, we examined whether the effect remained after controlling for educational attainment, household income, and the Big 3 personality traits of Positive Emotionality, Negative Emotionality and Constraint. The evidence for a potentially causal effect of neighborhood disadvantage on DG persisted even after controlling for these covariates. These analyses also pointed to two facets of Negative Emotionality, Stress Reaction (similar to Big 5 neuroticism, or emotional instability), and Aggression (similar to Big 5 low agreeableness) and a fact of Constraint, low Control (impulsivity) in accounting for a portion of this effect. That is, high stress reactivity, disagreeableness, and impulsivity were associated both with living in a disadvantaged neighborhood and with symptoms of DG even after accounting for overlapping genetic and shared environmental background factors.

The potentially causal effect of neighborhood disadvantage on DG contrasts with the results of parallel analyses of past-year alcohol use disorder that failed to find evidence of a potentially causal effect. A national US study also found a significant association between neighborhood disadvantage and DG, but not AUD (Barnes et al, 2013). This distinction provides a clue to the potential mechanisms underlying the neighborhood effect. Perhaps feelings of hopelessness that may be engendered by living in a neighborhood in which there are higher rates of unemployment (Morselli, 2017) are relieved by the hope of a potential life-transformative gambling win (Clotfelter & Cook, 1991; Downs, 2008). This false hope of a future big win is also one of the diagnostic criteria for DG, “chasing losses,” wherein the solution to one’s gambling problem lies in more gambling. Alcohol use disorder and other addictions are not characterized by such motivations and beliefs.

Non-Causal Effect of Neighborhood Disadvantage.

Multilevel models are also useful in probing the non-causal, or between-family association, between neighborhood disadvantage and DG. These between-family effects are the genetic and environmental factors that twins from the same family have in common. In contrast to the within-twin effects, the significant between-twin effect of neighborhood disadvantage on DG could be completely explained by individual differences in personality. In particular, all three facets of Negative Emotionality and two facets of Big 3 Constraint, low Control (impulsivity) and Traditionalism, appeared to explain the between-twin effect. This builds upon a previous paper (Slutske et al. 2015) which presented evidence for a significant overlap in the genetic factors associated with moving to or remaining in a disadvantaged neighborhood and DG, but could only speculate about what was driving this genetic association. These new analyses suggest that the genetic factors that are associated with both neighborhood disadvantage and DG are genetically-influenced personality traits, with the strongest contributors being Alienation (distrustfulness), Control (impulsivity), and Aggression (disagreeableness). The gene-environment correlation observed between neighborhood disadvantage and DG could be completely explained by individual differences in personality that were related to both the quality of the neighborhood in which a twin lived and the development of gambling pathology. In other words, in addition to a potentially causal effect of living in a disadvantaged neighborhood, individuals at genetic risk for DG were actually more likely to be exposed to this causal risk factor, primarily via genetic influences on distrustfulness, disagreeableness, and impulsivity.

The Importance of the Measure of Disordered Gambling

It is noteworthy that the evidence for a quasi-causal effect of neighborhood disadvantage on gambling disorder was not firmly established when restricting the analyses to the DSM-5 symptom set, but was robustly demonstrated with a composite measure that incorporated SOGS items. We previously advocated using this approach in etiologic research (Slutske et al, 2011), and our suggestion was put to good use in a genome-wide association study of a quantitative disordered gambling trait (Lind et al, 2010), and in a recent paper in which the neighborhood effect on gambling disorder in the United States was demonstrated (Welte et al, 2017).

A concern for the future of the DSM diagnosis of gambling disorder is how the DSM can move forward in offering a dimensional approach to describing DG (Regier et al, 2013). The nine DSM-5 symptoms do not do a very good job of this because the symptoms tap only a very narrow high end of the latent DG continuum. In a combined IRT analysis of the DSM and SOGS symptoms, the severities (difficulties) for the DSM-5 symptoms were uniformly high, whereas the severities for the SOGS items were more variable (Slutske et al, 2015). There were a number of items unique to the SOGS that were relatively less psychometrically severe than the DSM symptom set (for example, “Did you ever gamble more than you intended to?” and “Have you ever felt guilty about the way you gamble, or what happens when you gamble?”), meaning that they were endorsed by individuals lower on a latent DG continuum. These are just the sort of items that would be eliminated were the focus on the identification of items with high specificity for a categorical disorder. In contrast, for the development of a dimensional approach to describing DG, it is desirable to have items that represent a broad range of severities because this will provide better measurement of DG across the full continuum. This was not accomplished in this study, as the average item difficulty was 2.5, the lowest item difficulty was 1.0 (Slutske et al, 2015), and 95% and 83% of the participants had scores of zero on the DSM and SOGS measures, respectively5. Although borrowing items from the SOGS (or other measures) is a good place to start in characterizing variation in DG above and below the DSM threshold, future efforts might focus on developing DG assessments that are informative across all levels of gambling pathology found in the general population. If we truly think that DG lies on a continuum this is just what is needed.

Possible Explanations for the Potentially Causal Effect of Neighborhood Disadvantage

Exactly how neighborhood disadvantage causes DG cannot be deduced from these analyses, although there are clues from established risk factors. Neighborhood disadvantage may be causally related to DG via increased access to gambling opportunities. In the United States (Welte et al., 2004), New Zealand (Pearce et al., 2008), Canada (Wilson et al., 2006), England (Wardle et al., 2014), and Australia (Marshall & Baker, 2002), there is a greater density of gambling outlets in relatively disadvantaged neighborhoods. There is consistent evidence that access to electronic gambling machines (variously known as slot machines, pokies, fruit machines, or VLTs in different countries), in particular, is greater in disadvantaged than in more advantaged neighborhoods (Marshall & Baker, 2002; Pearce et al., 2008; Wardle et al., 2014; Wilson et al., 2006).

Other aspects of neighborhood disadvantage not related to gambling access may also be causally related to the development of DG. Neighborhood disadvantage may function as a chronic stressor or persistent strain (e.g., Boardman, 2004; Steptoe & Feldman, 2001) that may encourage gambling as a means to cope or escape from problems (Sinha, 2008; Stewart & Zack, 2008), or by rendering one less able to regulate impulses (e.g. Duckworth et al., 2013; Muraven & Baumeister, 2000; Sinha, 2008). There is a robust association between adverse life events and gambling problems among adolescents (Bergevin, Gupta, Derevensky, & Kaufman, 2006; Cheung, 2013) and adults (Eitle & Taylor, 2011; Ronzitti et al, 2018), and the added strain associated with living in a disadvantaged neighborhood may strengthen the association between experiencing negative life events and DG. Such an interaction has been observed in which neighborhood disadvantage strengthened the association between negative life events and the subsequent onset of an episode of depression (Cutrona et al, 2005). Further support for a chronic stress interpretation of neighborhood disadvantage comes from emerging evidence demonstrating how neighborhood deprivation “gets under the skin” (Prior et al, 2018). Using data from a large United Kingdom survey, investigators found that allostatic load (13 biomarkers from the cardiovascular, inflammatory, lipid and glucose metabolism systems, and the hypothalamic-pituitary axis) was significantly associated with neighborhood disadvantage and partially mediated the associations between neighborhood disadvantage and mental health. Individuals residing in neighborhoods characterized as more deprived had worse allostatic load scores, that is, they showed more physiological ‘wear-and-tear’ across multiple organ systems as a consequence of chronic exposure to stressful experiences (Prior et al, 2018).

In addition to increased exposure to chronic stress and strain, individuals in disadvantaged neighborhoods have less exposure to protective factors, such as access to institutional resources, mental health facilities, alternative recreational activities, or high-quality support systems (Cutrona et al, 2006; Sampson, Morenoff, & Gannon-Rowley, 2002). In sum, disadvantaged neighborhoods provide environments that are high in risk and low in protective influences, and when combined with the increased access for gambling opportunities, could potentially create a “perfect storm” environment that fosters the development of gambling problems. Evidence in support of this comes from a previous paper demonstrating a significant interaction between neighborhood disadvantage and the density of local gambling venues in predicting DG; the association between neighborhood disadvantage and DG was stronger in states or territories in which there was a greater density of gambling venues (Slutske et al, 2015).

Limitations

This study has several limitations. Because the study was conducted in Australia, the extent to which these findings are generalizable to other settings can be questioned. Although there are important differences in the gambling milieu in Australia compared to the United States, it is reassuring that the association between gambling disorder and neighborhood disadvantage observed in Australia (r = 0.10) was of similar magnitude to that observed in the United States (r = 0.08; Welte et al, 2017), and there was a similar lack of association of neighborhood disadvantage with AUD (Barnes et al, 2013). More generally, the ubiquity of the neighborhood effect across locales suggest that the effect observed in Australia is likely to be found elsewhere in the world (Sampson, 2011).

The measures of DG and the covariates were collected between 2004 and 2007, and the measure of neighborhood disadvantage was based on 2006 census indicators; for 71% of the participants the measure of neighborhood disadvantage was actually collected before the self-report measures. This might be a concern if the neighborhood disadvantage index evidenced substantial changes from year to year, but this was not the case. The census-derived index of neighborhood disadvantage was very stable (Pink, 2008a) and there were minimal changes in the level of disadvantage in the neighborhoods between 2004 when the earliest interviews were conducted and 2006 when the census indicators of disadvantage were collected. It was a limitation, however, that we were not able to characterize the duration of exposure of the participants to neighborhood disadvantage.

The cross-sectional design of the study precluded establishing the temporal relation between neighborhood disadvantage and the personality covariates. The personality covariates were included to provide the most stringent test of a quasi-causal association between neighborhood disadvantage and DG by accounting for potential confounding variables that were associated with DG and also may have played a role in twin discordance for neighborhood disadvantage. The rationale behind this was that personality differences between twins could have been a cause of twin discordance in neighborhood disadvantage. For example, the more emotionally unstable twin may have been more likely to move to or remain in a more disadvantaged neighborhood than the more emotionally stable cotwin. On the other hand, it is also plausible that higher levels of emotional instability may be a consequence of living in a more disadvantaged neighborhood (Kim, 2008). The latter scenario would not alter the conclusions drawn because the quasi-casual effect remained even with this potentially overly stringent control.

“The idea of ‘neighborhood effects’ has emerged as a sharp point of contention in the social sciences…. Indeed, disputes have erupted across multiple disciplines over the proper level of analysis for assessing neighborhood effects” (Sampson, 2011, p. 227). There may be concerns about the use of census-defined geographic regions to define neighborhoods in the present study, but this was the only feasible approach with these national data. Also, it is not always clear what the correct level of aggregation is when using census-derived geographic entities (e.g. census blocks, block groups, census tracts, or postal code areas) because some effects may be more diffuse and others may be more localized and may vary by the specific neighborhood characteristic and outcome studied (Hipp, 2007). We have followed the lead of previous national studies conducted in Australia (Azar et al., 2015; Bayer et al., 2011) to use the postal code area. Fortunately, a review of the literature on neighborhood effects suggests that study results tend to be robust across varying definitions of neighborhood (Sampson et al, 2002), “suggesting a general form to neighborhood effects” (Sampson, 2011, p. 229).

Conclusions and Implications

Based on evidence from multivariate analyses of epidemiologic survey data, Welte et al (2017) concluded that “there is something about poor neighborhoods that in itself promotes problem gambling.” This paper provides new evidence obtained from a quasi-causal research design that supports this conclusion. Establishing that neighborhood disadvantage is a potentially causal and malleable environmental risk factor for DG is of great public health importance. It supports efforts focused on identifying the active ingredients contributing to the neighborhood effect and in the development of preventions and interventions. Some of these efforts are currently underway: harm minimization practices are being developed to help mitigate the negative impacts of the worldwide proliferation of gambling opportunities (Blaszczynski, 2001; Ginley et al, 2017; Parke et al, 2015; Planzer & Wardle, 2012; Tanner et al, 2017), and there are major initiatives directed at overall health equity by improving the quality of neighborhoods and communities in which people live (Arcaya et al, 2016; Marmot et al, 2008).

However, in addition to identifying a potentially causal effect of living in a disadvantaged neighborhood, the results of this study suggests that individuals at genetic risk for DG were actually more likely to be exposed to this causal risk factor, primarily via genetic influences on individual differences in personality. This highlights the complex nature of the relation between neighborhood disadvantage and DG and the importance of taking into account the multiple contexts in which problematic gambling behavior can emerge, from genetic risk factors, to personality differences, to the neighborhood in which one lives, as well as the ways in which such contexts may interact with each other.

Supplementary Material

1

Acknowledgements

This work was supported by National Institute of Mental Health grant MH66206. Preparation of this manuscript was also supported in part by a Center of Excellence award from the National Center for Responsible Gaming. We thank Bronwyn Morris and Megan Fergusson for coordinating the data collection for the twins, David Smyth, Olivia Zheng, Anthony Conciatore, and Harry Beeby for data management, Alex Blazczynski and Rachel Volberg for lending their expertise to the overall data collection project. We are grateful for the participation

Footnotes

1

The phrase “potentially causal” is used throughout this paper because causality cannot be definitively established using observational data. For example, in the discordant twin design the underlying reason for twin discordance can never be known with certainty thereby making it impossible to rule out alternative non-causal explanations.

2

Because the census is only conducted every five years in Australia, it was not possible to perfectly align the year of the derivation of the neighborhood disadvantage index with the self-report data collection. Of the participants in this study, 18% and 53% were interviewed in 2004 and 2005 (prior to the 2006 census), 18% were interviewed in 2006 (concurrently with the census year from which the indicators of neighborhood disadvantage were obtained), and 11% were interviewed in 2007 (after the 2006 census). Fortunately, the census-derived index of neighborhood disadvantage was very stable – the neighborhood disadvantage index based on the 2006 census was correlated r = .94 with the disadvantage index based on the 2001 census data.

3

The data collection pre-dated the release of the DSM-5. Because the DSM-5 symptom set was identical to that of the DSM-IV with the exception of one eliminated symptom (committed illegal acts to finance gambling), we were able to create a DSM-5 symptom set.

4

Because sex did not significantly interact with within-pair (interaction IRR = 1.09, 95% CI = 0.96 to 1.23, p = .176) or between-pair (interaction IRR = 0.94, 95% CI = 0.86 to 1.04, p = .260) disadvantage variables in expanded base models, sex moderation of disadvantage effects was not explored.

5

The metric of the item difficulties are in z-score units from the z (standard normal) distribution. Higher difficulties indicate that the item is less likely to be endorsed and lower difficulties indicate that the item is more likely to be endorsed at a given level of a latent DG trait. The interpretation of an item difficulty of 1.0 is that the item has a 50% probability of being endorsed by an individual who is about one standard deviation above the mean on the latent DG trait; the interpretation of an item difficulty of 2.5 is that the item has a 50% probability of being endorsed by an individual who is about two and one half standard deviations above the mean on the latent DG trait.

Contributor Information

Wendy S. Slutske, Department of Psychological Sciences and Center of Excellence in Gambling Research, University of Missouri

Thomas M. Piasecki, Department of Psychological Sciences and Center of Excellence in Gambling Research, University of Missouri

Arielle R. Deutsch, Sanford School of Medicine, University of South Dakota

Dixie J. Statham, University of the Sunshine Coast, Sunshine Coast Mind and Neuroscience - Thompson Institute, Birtinya, Australia

Nicholas G. Martin, QIMR Berghofer Medical Research Institute, Brisbane, Australia

References

  1. Arcaya MC, Tucker-Seeley RD, Kim R, Schnake-Mahl A, So M, & Subramanian SV (2016). Research on neighborhood effects on health in the United States: a systematic review of study characteristics. Social Science & Medicine, 168, 16–29. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Azar D, White V, Coomber K, Faulkner A, Livingston M, Chikritzhs T, Room R, Wakefield M. (2015). The association between alcohol outlet density and alcohol use among urban and regional Australian adolescents. Addiction, 111, 65–72. [DOI] [PubMed] [Google Scholar]
  3. Barnes GM, Welte JW, Tidwell MO, & Hoffman JH (2013). Effects of neighborhood disadvantage on problem gambling and alcohol abuse. Journal of Behavioral Addictions, 2, 82–89. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Bayer JK, Ukoumunne OC, Lucas N, Wake M, Scalzo K, & Nicholson JM (2011). Risk factors for childhood mental health symptoms: national longitudinal study of Australian children. Pediatrics, 128, e865–879. [DOI] [PubMed] [Google Scholar]
  5. Begg MD, & Parides MK (2003). Separation of individual-level and cluster-level covariate effects in regression analysis of correlated data. Statistics in Medicine, 22, 2591–2602. [DOI] [PubMed] [Google Scholar]
  6. Bergavin T, Gupta R, Deverensky J, & Kaufman F (2006) Adolescent gambling: Understanding the role of stress and coping. Journal of Gambling Studies, 22, 195–208. [DOI] [PubMed] [Google Scholar]
  7. Blaszczynski A (2001). Harm minimization strategies in gambling: An overview of international initiatives and interventions. Melbourne: Australian Gaming Council. [Google Scholar]
  8. Boardman JD (2004). Stress and physical health: The role of neighborhoods as mediating and moderating mechanisms. Social Science and Medicine, 58, 2473–2483. [DOI] [PubMed] [Google Scholar]
  9. Cheung NW (2016). Social strain, self control, and juvenile gambling pathology: Evidence from Chinese adolescents. Youth and Society, 48, 77–100. [Google Scholar]
  10. Clotfelter CT, & Cook PJ (1991). Selling hope: State lotteries in America. Harvard University Press. [Google Scholar]
  11. Cutrona CE, Russell DW, Brown PA, Clark LA, Hessling RM, & Gardner KA (2005). Neighborhood context, personality, and stressful life events as predictors of depression among African American women. Journal of Abnormal Psychology, 114, 3–15. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Cutrona CE, Wallace G, & Wesner KA (2006). Neighborhood characteristics and depression: An examination of stress processes. Current Directions in Psychological Science, 15, 188–192. [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. D’Onofrio BM, Lahey BB, Turkheimer E, & Lichtenstein P (2013). Critical need for family-based, quasi-experimental designs in integrating genetic and social science research. American Journal of Public Health, 103, S46–S55. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Downs C (2008). The changing face of gambling: The Gambling Act (2005) and working class gambling cultures Relocating the Leisure Society: Media, Consumption and Spaces. Leisure Studies Association, Eastbourne, 97–119. [Google Scholar]
  15. Duckworth AL, Kim B, & Tsukayama E (2013). Life stress impairs self-control in adolescence. Frontiers in Psychology (Developmental Psychology), 3, 1–12. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Eitle D, & Taylor J (2011). General strain theory, BIS/BAS levels, and gambling behavior. Deviant Behavior, 32, 1–37. [Google Scholar]
  17. Elton-Marshall T, Wijesingha R, Veselka L, Williams C, Carton M, Pradeep B, Hudson R, Chiodo C, & Turner NE (2017). A public health approach to gambling: A report prepared for Gambling Research Exchange Ontario (GREO). London, Ontario: Centre for Addiction and Mental Health. [Google Scholar]
  18. Gerstein D, Hoffmann JP, Larison C, Engelman L, Murphy S, Palmer A, et al. (1999). Gambling Impact and Behavior Study: Report to the National Gambling Impact Study Commission. Chicago: National Opinion Research Center. [Google Scholar]
  19. Ginley MK, Whelan JP, Pfund RA, Peter SC, & Meyers AW (2017). Warning messages for electronic gambling machines: evidence for regulatory policies. Addiction Research & Theory, 25, 495–504. [Google Scholar]
  20. Hedeker D (2005). Generalized linear mixed models In Everitt BS & Howell D (eds.), Encyclopedia of Statistics in Behavioral Science. Chichester, UK: Wiley. [Google Scholar]
  21. Hipp JR (2007). Block, tract, and levels of aggregation: Neighborhood structure and crime and disorder as a case in point. American Sociological Review, 72, 659–680. [Google Scholar]
  22. Kendler KS (2017). Causal inference in psychiatric epidemiology. JAMA Psychiatry, 74, 561562. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Kendler KS, & Baker JH (2007). Genetic influences on measures of the environment: A systematic review. Psychological Medicine, 37, 615–626. [DOI] [PubMed] [Google Scholar]
  24. Kessler RC, Andrews G, Mroczek D, Ustun B, Wittchen HU The World Health Organization Composite International Diagnostic Interview Short‐Form (CIDI‐SF). Int J Methods Psychiatr Res 1998; 7: 171–185. [Google Scholar]
  25. Kessler RC, Hwang I, LaBrie R, Petukhova M, Sampson NA Winters KC et al. (2008). The prevalence and correlates of DSM-IV pathological gambling in the National Comorbidity Survey Replication. Psychological Medicine, 38, 1351–1360. [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Kim D (2008). Blues from the neighborhood? Neighborhood characteristics and depression. Epidemiologic Reviews, 30, 101–117. [DOI] [PubMed] [Google Scholar]
  27. Lahey BB, & D’Onofrio BM (2010). All in the family: Comparing siblings to test causal hypotheses regarding environmental influences on behavior. Current Directions in Psychological Science, 19, 319–323. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Lesieur HR, & Blume SB (1987). The South Oaks Gambling Screen (SOGS): A new instrument for the identification of pathological gamblers. American Journal of Psychiatry, 144, 1184–1188. [DOI] [PubMed] [Google Scholar]
  29. Lind PA, Zhu G, Montgomery GW, Madden PA, Heath AC, Martin NG, & Slutske WS (2013). Genome‐wide association study of a quantitative disordered gambling trait. Addiction Biology, 18, 511–522. [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Lynch WJ, Maciejewski PK, & Potenza MN (2004). Psychiatric correlates of gambling in adolescents and young adults grouped by age at gambling onset. Archives of General Psychiatry, 61, 1116–1122. [DOI] [PubMed] [Google Scholar]
  31. Marioni RE, Davies G, Hayward C, Liewald D, Kerr SM, Campbell A, ... & Hastie ND (2014). Molecular genetic contributions to socioeconomic status and intelligence. Intelligence, 44, 26–32. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Marmot M, Friel S, Bell R, Houweling TA, Taylor S, & Commission on Social Determinants of Health. (2008). Closing the gap in a generation: health equity through action on the social determinants of health. Lancet, 372, 1661–1669. [DOI] [PubMed] [Google Scholar]
  33. Marshall DC & Baker RGV (2002). The evolving market structures of gambling: Case studies modelling the socioeconomic assignment of gaming machines in Melbourne and Sydney, Australia. Journal of Gambling Studies, 18, 273–291. [DOI] [PubMed] [Google Scholar]
  34. McGue M, Osler M, & Christensen K (2010). Causal inference and observational research: The utility of twins. Perspectives on Psychological Science, 5, 546–556. [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Morselli D (2017). Contextual determinants of hopelessness: Investigating socioeconomic factors and emotional climates. Social Indicators Research, 133, 373–393. [Google Scholar]
  36. Muraven M, & Baumeister RF (2000). Self-regulation and depletion of limited resources: Does self-control resemble a muscle? Psychological Bulletin, 126, 247–259. [DOI] [PubMed] [Google Scholar]
  37. Parke A, Harris A, Parke J, Rigbye J, & Blaszczynski A (2015). Responsible marketing and advertising in gambling: A critical review. The Journal of Gambling Business and Economics, 8, 21–35. [Google Scholar]
  38. Pearce J, Mason K, Hiscock R, & Day P (2008). A national study of neighborhood access to gambling opportunities and individual gambling behavior. Journal of Epidemiology and Community Health, 62, 862–868. [DOI] [PubMed] [Google Scholar]
  39. Pink B (2008a). An Introduction to Socio-economic Indexes for Areas (SEIFA) 2006. Australian Bureau of Statistics. [Google Scholar]
  40. Pink B (2008b). Socioeconomic Indexes for Areas – Technical Paper 2006. Australian Bureau of Statistics. [Google Scholar]
  41. Planzer S, & Wardle H (2012). What we know about the comparative effectiveness of gambling regulation. European Journal of Risk Regulation, 3, 410–416. [Google Scholar]
  42. Plomin R, DeFries JC, Knopik VS, & Neiderhiser JM (2016). Top 10 replicated findings from behavioral genetics. Perspectives on Psychological Science, 11, 3–23. [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Polderman TJ, Benyamin B, De Leeuw CA, Sullivan PF, Van Bochoven A, Visscher PM, & Posthuma D (2015). Meta-analysis of the heritability of human traits based on fifty years of twin studies. Nature Genetics, 47, 702. [DOI] [PubMed] [Google Scholar]
  44. Prior L, Manley D, Jones K (2018). Stressed out? An investigation of whether allostatic load mediates associations between neighbourhood deprivation and health. Health and Place, 52, 2533. [DOI] [PubMed] [Google Scholar]
  45. Regier DA, Kuhl EA, & Kupfer DJ (2013). The DSM‐5: Classification and criteria changes. World Psychiatry, 12, 92–98. [DOI] [PMC free article] [PubMed] [Google Scholar]
  46. Ronzitti S, Kraus SW, Hoff RA, & Potenza MN (2018). Stress moderates the relationships between problem-gambling severity and specific psychopathologies. Psychiatry Research, 259, 254–261. [DOI] [PMC free article] [PubMed] [Google Scholar]
  47. Rutter M (2007a). Proceeding from observed correlation to causal inference: The use of natural experiments. Perspectives on Psychological Science, 2, 377–395. [DOI] [PubMed] [Google Scholar]
  48. Rutter M (2007b). Identifying the environmental causes of disease: how should we decide what to believe and when to take action? London: Academy of Medical Sciences; Available at (http://acmedsci.ac.uk/policy/policy/identifying-the-environmental-causes-of-disease/) [Google Scholar]
  49. Sampson RJ (2011). Neighborhood effects, causal mechanisms and the social structure of the city In: Analytical Sociology and Social Mechanisms, pp 227–249, Demeulenaere P, ed. New York: Cambridge University Press. [Google Scholar]
  50. Sampson RJ, Morenoff JD, & Gannon-Rowley T (2002). Assessing “neighborhood effects”: Social processes and new directions in research. Annual Review of Sociology, 28, 443–478. [Google Scholar]
  51. Sariaslan A, Fazel S, D'onofrio BM, Långström N, Larsson H, Bergen SE, ... & Lichtenstein P (2016). Schizophrenia and subsequent neighborhood deprivation: revisiting the social drift hypothesis using population, twin and molecular genetic data. Translational Psychiatry, 6, e796. [DOI] [PMC free article] [PubMed] [Google Scholar]
  52. SAS Institute Inc. (2009) Base SAS ® 9.2 Procedures Guide. Cary, NC: SAS Institute Inc. [Google Scholar]
  53. Scarr S, & McCartney K (1983). How people make their own Environments: A theory of genotype → environment effects. Child Development, 54, 424–435. [DOI] [PubMed] [Google Scholar]
  54. Sinha R (2008). Chronic stress, drug use, and vulnerability to addiction. Annals of the New York Academy of Sciences, 1141, 105–130. [DOI] [PMC free article] [PubMed] [Google Scholar]
  55. Slutske WS, Cho SB, Piasecki TM, & Martin NG (2013). Genetic overlap between personality and risk for disordered gambling: Evidence from a national community-based Australian twin study. Journal of Abnormal Psychology, 122, 250–255. [DOI] [PMC free article] [PubMed] [Google Scholar]
  56. Slutske WS, Deutsch AR, Richmond-Rakerd LS, Chernyavskiy P, Statham DJ, & Martin NG (2014). Test of a potential causal influence of earlier age of gambling initiation on gambling involvement and disorder: A multilevel discordant twin design. Psychology of Addictive Behaviors, 28, 1177–1189. [DOI] [PMC free article] [PubMed] [Google Scholar]
  57. Slutske WS, Deutsch AR, Statham DJ, & Martin NG (2015). Local area disadvantage and gambling involvement and disorder: Evidence for gene-environment correlation and interaction. Journal of Abnormal Psychology, 124, 606–622. [DOI] [PMC free article] [PubMed] [Google Scholar]
  58. Slutske WS, Meier MH, Zhu G, Statham DJ, Blaszczynski A, & Martin NG (2009). The Australian twin study of gambling (OZ-GAM): Rationale, sample description, predictors of participation, and a first look at sources of individual differences in gambling involvement. Twin Research and Human Genetics, 12, 63–78. [DOI] [PubMed] [Google Scholar]
  59. Slutske WS, Zhu G, Meier MH, Martin NG. (2011) Disordered gambling as defined by the DSM-IV and the South Oaks Gambling Screen: Evidence for a common etiologic structure. Journal of Abnormal Psychology, 120, 743–751. [DOI] [PMC free article] [PubMed] [Google Scholar]
  60. Snijders T, & Bosker R (1999). Multilevel analysis: An introduction to basic and advanced multilevel modeling. London, UK: Sage. [Google Scholar]
  61. Steptoe A, & Feldman PJ (2001). Neighborhood problems as sources of chronic stress: Development of a measure of neighborhood problems and associations with socioeconomic status and health. Annals of Behavioral Medicine, 23, 177–185. [DOI] [PubMed] [Google Scholar]
  62. Stewart SH, & Zack M (2008). Development and psychometric evaluation of a threedimensional Gambling Motives Questionnaire. Addiction, 103, 1110–1117. [DOI] [PubMed] [Google Scholar]
  63. Tanner J, Drawson AS, Mushquash CJ, Mushquash AR, & Mazmanian D (2017). Harm reduction in gambling: a systematic review of industry strategies. Addiction Research & Theory, 25, 485–494. [Google Scholar]
  64. Tellegen A, & Waller NG (2008). Exploring personality through test construction: Development of the Multidimensional Personality Questionnaire In: The SAGE Handbook of Personality Theory and Assessment, Vol 2: Personality Measurement and Testing, (pp. 261–292). Boyle GJ, Matthews G, Saklofske DH (Eds). Thousand Oaks, CA, US: Sage Publications. [Google Scholar]
  65. Turkheimer E, & Harden KP (2014). Behavior genetic research methods: Testing quasi-causal hypotheses using multivariate twin data In Reis HT and Judd CM, eds., Handbook of Research Methods in Social and Personality Psychology, pp. 159–187. New York: Cambridge University Press. [Google Scholar]
  66. Wardle H, Moody A, Spence S, Orford J, Volberg R, Griffiths M, et al. (2011). British gambling prevalence survey 2010. London: The Stationery Office. [Google Scholar]
  67. Wardle H, Keily R, Astbury G, & Reith G (2014). ‘Risky places?’: Mapping gambling machine density and socio-economic deprivation. Journal of Gambling Studies, 30, 201–212. [DOI] [PubMed] [Google Scholar]
  68. Welte JW, Barnes GM, Tidwell MCO, & Wieczorek WF (2017). Predictors of problem gambling in the US. Journal of Gambling Studies, 33, 327–342. [DOI] [PubMed] [Google Scholar]
  69. Welte JW, Wieczorek WF, Barnes GM, Tidwell MC, & Hoffman JH (2004). The relationship of ecological and geographic factors to gambling behavior and pathology. Journal of Gambling Studies, 20, 405–423. [DOI] [PubMed] [Google Scholar]
  70. Wilson DH, Derevensky J, Gilliland J, Gupta R, & Ross NA (2006). Video lottery terminal access and gambling among high school students in Montreal. Canadian Journal of Public Health, 97, 202–206. [DOI] [PMC free article] [PubMed] [Google Scholar]
  71. World Health Organization. The ICD-10 Classification of Mental and Behavioural Disorders: Clinical Descriptions and Diagnostic Guidelines, World Health Organization; 1992. [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

1
NIHMS1510372-supplement-1.pdf (133.6KB, pdf)

RESOURCES