Discovering the Causes of Problem Gambling: Overcoming Methodological Challenges
Donald Schopflocher, PhDAssociate Professor,
School of Public Health, University of Alberta
Preliminaries
• Datasets– LLLP 3 (of 4) waves, Alberta, initial N= 1808– QERI 4 (of 5) waves, SE Ontario, initial N= 4121 – ‘07-’08 CCHS cross section, Ont., Quebec, Sask. N=81427
• Analyses are exploratory – Sketch out some research questions– Examine some methods especially for
• Causal analysis• Longitudinal (panel) data
• Analyses (so far) focus upon gambling as measured by the CPGI
Focus Questions
• Are there enough Problem Gamblers to study by survey methods?
• Is Problem Gambling a category, or part of a continuum?
• What causes changes in gambling?
Focus Methods
• Latent Variable and Structural Equation Models • Fixed Effects Regression Methods• Multivariate Visualization Techniques
CCHS 2007-2008Quebec, Ontario, SaskatchewanN=81427
CPGI Problem Score
0 5 10 15 20 25
Num
ber
0.1
1
10
100
1000
10000
100000
Non Problem Gambler 40.5%Low Risk Gambler 1.7.%Moderate Risk Gambler 0.9%Problem Gambler 0.3%Not a Current Gambler 53.4%
Question 1: Are there enough Problem Gamblers to study?
CCHS 2007-2008Quebec, Ontario, SaskatchewanN=81427
CPGI Problem Score
0 5 10 15 20 25
Num
ber
0
5000
10000
15000
20000
25000
30000
35000
Non Problem Gambler 40.5%Low Risk Gambler 1.7.%Moderate Risk Gambler 0.9%Problem Gambler 0.3%Not a Current Gambler 53.4%
LLLP
QERI
Wave 1 Recruitment: CPGI categories
Loss to Followup LLLP Wave 1 to Wave 2 QERI Wave 1 to Wave 4 by CPGI Problem Score Wave 1
CPGI Problem Score
NON-PROBLEM GAM
BLER
LOW RISK GAM
BLER
MODERATE RISK GAM
BLER
PROBLEM GAM
BLER
Odd
s of
Los
s to
Fol
low
up
0.5
1.0
1.5
2.0
2.5
3.0
LLLPQERI
QERI CPGI Problem Score Trajectories
Wave
1 2 3 4
CP
GI
Sco
re
0
5
10
15
20
8
35
7
5
1312
35
2152
Question 2: Is Problem Gambling a category, or part of a continuum?
QERI Waves 1-4
QERI Waves 1-4
Further indications that gambling and problem gambling may be stable characteristics on a continuum:
• QERI Intraclass Correlations (proportion of o/a variance between individuals)– Gambling Activities 0.78– Gambling Frequency 0.76– Ln (Gambling Expenditures) 0.71– CPGI Problem Score 0.77
• QERI Autocorrelations in change scores – Lag 1 Change in Gambling Activities -0.176– Lag 1 Change in Gambling Frequency -0.254– Lag 1 Change in Ln (Gambling Expenditures) -0.279– Lag 1 Change in CPGI Problem Score -0.212
Question 3a: What causes gambling ?
• Traits
Question 3a: What causes gambling ? Mental disorders implicated?
CCHS Discriminant Function Analysis of CPGI Risk Groups
Discriminant Function 1
-0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8
Dis
crim
inan
t F
unct
ion
2
-0.6
-0.4
-0.2
0.0
0.2
0.4
0.6
NGNEVER
NRG
LRGMRGPG
+
PHlthMHlth
AlC
IncEd
PHlthMHlth
SMK
ALC
GVar
DEPR
ANX
happy
stress
CCHS Discriminant Function Analysis of CPGI Risk Groups
Discriminant Function 1
-0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8
Dis
crim
inan
t F
unct
ion
2
-0.6
-0.4
-0.2
0.0
0.2
0.4
0.6
NGNEVER
.
..PG
+
PHlthMHlth
AlC
IncEd
PHlthMHlth
SMK
ALC
GVar
DEPR
ANX
happy
stress
Biplot: Variables composing Canonical Variates, Axes representing High Frequency Gamblers with and without Problem Gambling status
Loadings on Canonical Variate1
-0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6
Loa
ding
s on
Ca
noni
cal V
aria
te 2
-0.8
-0.6
-0.4
-0.2
0.0
0.2
0.4
0.6
GambFreq
DSMProb
Smoker
single
Religiosity
Excitement
Impulsive
Antisocial
Depression
Anxiety
1st age gamb
Child Trauma
IQ
Gamb Fallacies
older
Drugs
Alcohol
OCD
male
Gamb-Friends
Gamb-ParentsGamb-Sibs
Gamb-Young
+
High Frequency Problem Gamblers
High Frequency Non Problem Gamblers
Infrequent/Non Problem Gamblers
LLLP Wave 1
Aside: Maybe Gambling behaviour and Gambling Problems are dissociated.
QERI Waves 1-4
• Note that if we accept the dissociation of gambling behaviour from gambling problems there can now be path models of this type:
Question 3b: What causes gambling ?
• What causes CHANGES to gambling behaviour &/or gambling problems ?
OR• What’s time got to do with it?
• Total model
Yit = B1TXit + B2TZi + eit
where i indexes persons
t indexes occasions
Problem:– Persons will generally be more similar to themselves than to
random others
ReviewRegression Analysis of Longitudinal Panel Data
• Between Model
Problem-Ignores change
Yi = B0B + B1BXi + B2BZi + ei
Traits related to Trait Gambling and Trait Gambling Problems
Weights
-0.25 -0.20 -0.15 -0.10 -0.05 0.00 0.05 0.10 0.15
age
male
Neuroticism
Extraversion
Agreeableness
Conscienciousness
Openness
Gambling ProblemsGambling Behaviour
+-QERI Waves 1-4
• Within Model (Fixed Effects)
yit = B0W + B1Wxit + AiDi + eit
uses Ai as a single person specific coefficient to
stand in for the effects of all variables constant over time, measured or unmeasured.
Weights for Fixed Effect Regression of CPGI Total Problems
weights
-0.10 -0.05 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35
Marital
Social Support
Family
Community
Stressful Events
Stress
PTSD
Depression
Mania
Anxiety
Panic Attacks
OCD
Schizophrenia
Substance Abuse
Gambling
Trait (Between)State (Within)
+-QERI Waves 1-4
Weights for Fixed Effect Regression of Gambling Behaviour
Weights
-0.10 -0.05 0.00 0.05 0.10 0.15 0.20
Marital
Social Support
Family
Community
Stressful Events
Stress
PTSD
Depression
Mania
Anxiety
Panic Attacks
OCD
Schizophrenia
Substance Abuse
- +
Trait (Between)State (Within)
QERI Waves 1-4
Coefficient
-0.04 -0.02 0.00 0.02 0.04 0.06
dropped out of school had serious conflicts or difficulties with spouse
borrowed a significant amount of money (e.g., morhad serious conflicts with neighbor(s)
suffered a significant financial losshad a significant change in work hours
had serious conflict(s) with coworkerdeveloped a serious mental illness
received serious threats or harassmenthad serious conflicts with ex-spouse
suffered a serious injury as a result of an accidhad serious conflicts with close friend(s)
had serious conflicts with family member(s)
declared bankruptcy
had a new addition to the family through birth orwas a victim of some other crime
started schoolreceived an important promotion
death of important family pet moved to new location/house
was laid offwas disciplined at work
suffered a significant loss or damage of property had a significant financial improvement
suffered a significant business loss or failuredeveloped a serious physical illness
QERI Waves 1-4
Some tentative conclusions
• ‘Gambling behaviour’ is largely a stable characteristic• ‘Having gambling problems’ is largely a stable characteristic • The two are dissociated, but related (here as elsewhere
behaviour ‘causes’ problems).• Personality traits are differentially related to gambling
behaviour and having gambling problems.• Changes in mental health and in stressful life events are
related to changes in gambling behaviour and gambling problems
• Relationships in general are quite small
• Random Effects Model (Multilevel Models)
Yit = B0R + B1RXit + B2RZi + Vi + eit
Assumes a specific distribution for V. A critical assumption is that V and E are
independent of X and Z
Appendix: An Alternative to Fixed Effect Regression