Trends in research methodology A workshop for early stage investigators Contact info Bethany C Bray PhD bcbraypsuedu Research Associate The Methodology Center Penn State httpmethodologypsuedu ID: 177904
Download Presentation The PPT/PDF document "Innovative methods for gambling data" is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
Innovative methods for gambling data
Trends in research methodology:
A workshop for early stage investigatorsSlide2
Contact info
Bethany C. Bray, Ph.D.
bcbray@psu.edu
Research Associate, The Methodology Center, Penn State
http://methodology.psu.edu
Slide3
Goals
Brief overview of three innovative methods
Research questions
Modeling approach
Tools needed (i.e., software)
Gambling
applications
Resources for more informationSlide4
Questions to ask …
What is my
research question
?
What are the
data
I have to address my research question?Slide5
For example …
Question: What are the risk factors for developing a gambling disorder?
Data: total number of DSM-5 diagnostic criteria endorsed
Question: Are there types of gamblers at higher risk for developing gambling disorder compared to others?
Data: multiple indicators of gambling activity engagementSlide6
For example …
Question: Does the relation between gender and gambling vary across time?
Data: amount wagered online every day for two yearsSlide7
What do I do when ……
I want to model a
count outcome
?
e.g., total number of DSM-5 diagnostic criteria endorsed
I want to identify
types of individuals
?
e.g., using multiple indicators of gambling behavior
I want to model
intensively-collected data
?
e.g., daily online wagering over timeSlide8
Count outcomesSlide9
Research questions
What are the predictors of
number
of days
gambled?
e.g., measured by total number of days in a month during which an individual gambled
What are the predictors of
severity
of gambling behavior?
e.g., measured by total number of endorsed DSM-5 diagnostic criteriaSlide10
Research questions
What are the risk factors (i.e., gambling behaviors, substance abuse, other problem behaviors, sociodemographic characteristics) for
disordered
gambling
?
(Welte et al., 2004)
e.g., frequency of 15 types of gambling, count of diagnostic criteria
Does the
“gambler's fallacy”
predict fluctuations in lottery play?
e.g., number of winning bets conditional on the history of draws (Papachristou, 2004
)Slide11
Research questions
Do college student athletes have higher levels of disordered gambling than non-athletes? What are the risk factors for athletes and non-athletes?
(Weinstock et al., 2007)
e.g., gambling frequencySlide12
Research questions
Do gender, age, race/ethnicity, family socioeconomic status, sensation seeking, and participation in risky behaviors predict problematic financial behavior among college students?
(Worthy et al., 2010)
e.g., number of problematic financial behaviorsSlide13
At-risk sampleSlide14
What would this distribution look like for a general population sample?Slide15
Modeling approach
Poisson regression
Negative binomial regression
Zero-inflated Poisson regression
Zero-inflated negative binomial regressionSlide16
Modeling approach
Behaviors like gambling can generate data that are
characterized by excess
zeros
e.g., much
of the population
may
not engage in the
behavior or not have problems
These behaviors can also generate data with over-dispersion
Variance
substantially exceeds the mean
e.g., because
a small number of individuals engage in extreme levels of the
behaviorSlide17
Modeling approach
Several
closely related models are available for predicting a count
variable…
Poisson regression (mean
and variance
assumed
to be
equal)
Negative binomial regression (adds over-dispersion)Slide18
Modeling approach
Models can
be
extended to accommodate excess zeros
Zero-inflated
Poisson (
ZIP) regression
Zero-inflated
NB (ZINB)
regression
These
models posit two types of individuals who report zero
gambling…
Those
who are
non-gamblers
And
those who have the potential to engage in
gambling
but report zero acts during that
timeSlide19
Modeling approach
Models estimate
population-average associations between
predictors and gambling
Can do model comparisons between options to determine which is optimal for examining the association between predictors and outcome
e.g., fit
indices
like AIC
and
BICSlide20
Estimated Coefficient
(Exponentiated Coefficient)
Intercept
(Mean
Gambling Count
)
0.309**
(1.36)
Dispersion Parameter
4.418
(n/a)
Community Risk
Community Cohesion
0.028*
(1.03)
Community Drug/Guns
0.143**
(1.15)
School Risk
School Prosocial
0.141**
(1.15)
Family Risk
Family Cohesion
0.112**
(1.12)
Family Risk
0.305**
(1.36)
Peer Risk
Antisocial Peers
0.823**
(2.28)
Individual Risk
Antisocial Attitudes
0.408**
(1.50)
Risky Behaviors
0.436**
(1.55)Slide21
Modeling approach
Total
number of
gambling
acts
expressed
as a function of an intercept, a dispersion parameter, and eight risk
indices
The
intercept, 0.309, corresponds to the log of the mean
gambling
count among all adolescents when all predictors in the model are set to
zero
e.g., for
adolescents with average levels on all risk indices, the expected number of
gambling
acts in the past year was
e
0.309
=1.3
A
positive dispersion parameter (4.418) indicates that the outcome was
over-dispersedSlide22
Modeling approach
All
risk indices were significantly and positively associated with the number of
gambling
acts in the overall
population
Coefficients
reflect
association
between each risk index and
gambling
count, after adjusting for other predictors in the
model
For
example, a one-standard-deviation increase in family risk corresponds to e
0.305
=1.36 times more
gambling
acts, holding all other predictors
constant
For example, associating
with antisocial peers had the largest coefficient, corresponding to e
0.823
=2.28 times more delinquent acts for every one-unit
increaseSlide23
Software
SPSS
http
://
www.ats.ucla.edu/stat/spss/dae/poissonreg.htm
http://www.ats.ucla.edu/stat/spss/dae/neg_binom.htm
SAS
http://
www.ats.ucla.edu/stat/sas/dae/poissonreg.htm
http://www.ats.ucla.edu/stat/sas/dae/negbinreg.htmSlide24
Gambling applications
Papachristou, G. (2004). The British gambler's fallacy.
Applied Economics
,
36
,
2073-2077.
Weinstock, J., Whelan, J. P., Meyers, A. W., & Watson, J. M. (2007). Gambling behavior of student-athletes and a student cohort: What are the odds
?
Journal of Gambling Studies
,
23
,
13-24
.Slide25
Gambling applications
Welte
, J. W., Barnes, G. M., Wieczorek, W. F., Tidwell, M. C. O., & Parker, J. C. (2004). Risk factors for pathological gambling.
Addictive behaviors
,
29
,
323-335.
Worthy
, S. L., Jonkman, J., & Blinn-Pike, L. (2010). Sensation-seeking, risk-taking, and problematic financial behaviors of college students.
Journal of Family and Economic Issues
,
31
,
161-170.Slide26
Resources
Comprehensive texts…
Agresti, A. (
2012).
Categorical data
analysis
. Hoboken, NJ: Wiley.
Cameron, A. C., & Trivedi, P. K. (2013).
Regression analysis of count data
. New York, NY: Cambridge University Press.Slide27
Resources
Recommended journal articles…
Atkins, D. C., & Gallop, R. J. (2007). Rethinking how family researchers model infrequent outcomes:
A
tutorial on count regression and zero-inflated models
.
Journal
of Family Psychology
,
21
, 726-735.
Coxe
, S., West, S. G., & Aiken, L. S. (2009). The analysis of count data: A gentle introduction to Poisson regression and its alternatives.
Journal of
personality assessment
,
91
,
121-136
.Slide28
Resources
Recommended journal articles continued…
Lanza
, S. T., Cooper, B. R., & Bray, B. C. (in press). Population heterogeneity in the salience of multiple risk factors for adolescent delinquency.
Journal of Adolescent Health
.Slide29
Identifying subgroupsSlide30
Research questions
Are there identifiable patterns of gambling behaviors? If so, what are the related individual characteristics and health consequences?
(Boldero et al
., 2010;
Cunningham-Williams & Hong, 2007; Lloyd
et al.,
2010)
Are there identifiable types of gamblers based on the DSM diagnostic criteria?
(Carragher & McWilliams, 2011;
McBride et al.,
2010; Xian et al., 2008)Slide31
Modeling approach
Latent class analysis (LCA)
Individuals can be divided into subgroups, or latent classes, based on unobservable construct
True class membership is unknown
Classes are mutually exclusive and exhaustiveSlide32
Modeling approach
Measurement of construct typically based on several categorical indicators
There is error associated with the measurement of the latent classes
Like factor analysis in that you have to identify the number and structure of the classes, but the latent variable is categoricalSlide33
Modeling approach
Interested in two sets of parameters…
Latent class prevalences
e.g., probability of membership in the ‘table and sports gambling’ latent class
Item-response probabilities
e.g., probability of responding ‘yes’ to a question about betting on sports in the past month given membership in the ‘table and sports gambling’ latent classSlide34
Gambling
Classes
lotto
poker
sports
… Slide35
Modeling approach
Conduct model selection procedure to determine optimal number of latent classes
Use model fit criteria like the AIC and BIC
Somewhat similar to process for exploratory factor analysis
After model selection, use item-response probabilities to interpret the latent classesSlide36
Non-Gamblers
(30%)
Lotto
Only
(10%)
Lotto
& Cards
(25%)
Table & Sports
(25%)
Multi-Game
(10%)
Lotto
.10
.95
.90
.45
.98
Poker
.02
.01
.95
.20
.95
Other
card games
.02
.02
.92
.10
.90
Dice
games
.01
.05
.30
.85
.90
Other table games
.00
.10
.20
.80
.95
Games
of personal skill
.10
.05
.10
.15
.85
Horse racing
.00
.01
.05
.05
.80
Other parimutual
betting
.00
.02
.03
.05
.70
Sports
.05
.20
.25
.95
.98
Modeling approachSlide37
Modeling approach
Might also want to examine group differences in…
Latent class structure
e.g., test measurement invariance
Latent class prevalences
e.g., test distribution across groups
e.g., gender differences in gambling behavior patterns…Slide38
Non-Gamblers (30%)
Lotto
Only
(10%)
Lotto
& Cards
(25%)
Table & Sports
(25%)
Multi-Game
(10%)
Males
25%
5%
20%
35%
15%
Females
35%
15%
30%
15%
5%
Modeling approachSlide39
Modeling approach
Lots of extensions to these models…
e.g., add covariates to predict subgroup membership
e.g., use subgroup membership to predict a distal outcome
e.g., examine changes in subgroup membership over timeSlide40
Gambling
Classes
casino games
sports
lotto
…
Gender
Drug UseSlide41
Gambling
Classes
casino games
sports
lotto
…
Mental Heath DisordersSlide42
Time 1
Gambling
Classes
Time 2
Gambling
ClassesSlide43
Software options
SAS (PROC LCA)
http://methodology.psu.edu/downloads
Mplus
http://www.statmodel.com/
Latent Gold
http://statisticalinnovations.com/products/latentgold.htmlSlide44
Gambling applications
Boldero, J. M., Bell, R. C., & Moore, S. M. (2010). Do gambling activity patterns predict gambling problems? A latent class analysis of gambling forms among Australian youth.
International
Gambling
Studies
,
10
,
151-163
.
Carragher
, N., & McWilliams, L. A. (2011). A latent class analysis of DSM-IV criteria for pathological gambling: Results from the National Epidemiologic Survey on Alcohol and Related Conditions.
Psychiatry
Research
,
187
,
185-192
.Slide45
Gambling applications
Cunningham-Williams
, R. M., & Hong, S. I. (2007). A latent class analysis (LCA) of problem gambling among a sample of community-recruited gamblers
.
The
Journal of
Nervous
and
Mental
D
isease
,
195
,
939-947
.
Lloyd, J., Doll, H., Hawton, K., Dutton, W. H., Geddes, J. R., Goodwin, G. M., & Rogers, R. D. (2010). Internet gamblers: A latent class analysis of their behaviours and health experiences.
Journal of Gambling Studies
,
26
,
387-399
.Slide46
Gambling applications
McBride
, O., Adamson, G., & Shevlin, M. (2010). A latent class analysis of DSM-IV pathological gambling criteria in a nationally representative British sample.
Psychiatry
Research
,
178
,
401-407
.
Xian, H., Shah, K. R., Potenza, M. N., Volberg, R., Chantarujikapong, S., True, W. R., ... & Eisen, S. A. (2008). A latent class analysis of DSM-III-R pathological gambling criteria in middle-aged men:
Association
with psychiatric disorders.
Journal of addiction medicine
,
2
,
85-95.Slide47
Resources
Comprehensive text…
Collins
,
L.
M.,
& Lanza, L. M. (2010).
Latent class and latent transition analysis: With applications in the social, behavioral, and health sciences
.
New York, NY: Wiley.Slide48
Resources
Recommended journal articles…
Lanza, S. T., Bray, B. C., & Collins, L. M. (2013). An introduction to latent class and latent transition analysis. In J. A
. Schinka
, W. F. Velicer, & I. B. Weiner (Eds.),
Handbook of psychology
(2nd ed., Vol. 2, pp. 691-716). Hoboken, NJ
: Wiley.
Lanza, S. T., Collins, L. M., Lemmon, D. R., & Schafer, J. L. (2007). PROC LCA: A SAS procedure for latent class analysis
.
Structural
Equation Modeling,
14
,
671-694
.Slide49
Resources
Recommended journal articles continued…
Lanza
, S. T., Collins, L. M., Lemmon, D. R., & Schafer, J. L. (2007). PROC LCA: A SAS procedure for latent class
analysis.
Structural Equation Modeling, 14
(4), 671-694.
Lanza
, S. T. & Rhoades, B. L. (2013). Latent class analysis: An alternative perspective on subgroup analysis in
prevention and
treatment.
Prevention Science, 14,
157-168
.Slide50
Resources
Recommended journal articles continued…
Lanza
, S. T., Rhoades, B. L., Nix, R. L., Greenberg, M. T., & the Conduct Problems Prevention Research Group (2010
). Modeling
the Interplay of Multilevel Risk Factors for Future Academic and Behavior Problems: A
Person-Centered
Approach.
Development and Psychopathology, 22
, 313-335.Slide51
Resources
Recommended journal article on power in latent class models that is helpful for grant applications…
Dziak, J. J., Lanza, S. T., & Tan, X. (in press). Effect size, statistical power and sample size requirements for the
bootstrap likelihood
ratio test in latent class analysis.
Structural Equation Modeling.Slide52
Intensive longitudinal dataSlide53
Research questions
Do baseline characteristics like gender exert differential effects at various points in the gambling onset process?
How do we deal with excessive wins and losses in understanding the onset process
?Slide54
Research questions
Does pharmacological treatment for gambling disorder continue to suppress craving over the long-term?
Which psychological characteristics like depression present greatest risk for relapse?
Do these differed based on duration of cessation?Slide55
What are EMA data?
E
cological
Real-world environments and experiences
Provides ecological validity
M
omentary
Real-time assessments
Avoids recall bias
A
ssessment
Self-report or automatic
Repeated, intensive, longitudinal
(Stone & Shiffman, 1994)Slide56
What are EMA data?
Key idea is that…
…these data allow analysis of psychological, behavioral, and/or physiological
processes over time
(Stone & Shiffman, 1994)Slide57
What are EMA data?
Collected using…
Tablets
Smartphones
Biological/physiological devices
(Stone & Shiffman, 1994)Slide58
What are EMA data?
Also called intensive longitudinal data (ILD)
e.g., repeated internet-based assessments
e.g., daily or weekly assessments over a long period of time
e.g., longitudinal burst designs
bwin gambling data available from The
Transparency Project
(
www.thetransparencyproject.org
) Slide59
What are EMA data?
Traditional longitudinal dataSlide60
Irregular longitudinal data
What are EMA data?Slide61
Irregular longitudinal data
What are EMA data?Slide62
Modeling approach
Time-varying effect model (TVEM)
(
methodology.psu.edu/ra/inten-longit-data
)
Examines the time-varying relation between an outcome and predictor(s)
Let’s consider two examples…Slide63
Modeling approach
Does
gender
exert differential effects
across time on amount of money wagered after the opening of an online gambling account?
e.g., time-invariant predictor
What varies across time?
Mean amount of money wagered (i.e., intercept function)
Effect of genderSlide64
Modeling approach
The complex function is approximated using non-parametric smoothing techniques
Coefficients are not single-number summaries, but are expressed as functions over time
Interpretation must take time into account
Look at time-varying effects using graphsSlide65
Treatment
Placebo
Days since account opening
poker
c
asino
games
Time-varying intercept function by activity.Slide66
Days since account opening
poker
c
asino
games
Time-varying effect of gender by activity.Slide67
Modeling approach
When does
depression
present the greatest risk for gambling disorder relapse after cessation?
e.g., time-varying predictor
What varies over time?
Engagement in gambling behavior (i.e., intercept function)
Effect of depressionSlide68
Days since cessation attempt
males
females
Time-varying effect of depression by gender.Slide69
Software
SAS suite of %TVEM macros
Binary outcomes
Normally-distributed continuous outcomes
Count outcomes (Poisson)
Zero-inflated count outcomes (ZIP)
Yang, J., Tan, X., Li, R., & Wagner, A. (2012).
TVEM (time-varying effect model) SAS macro suite users'
guide
(
Version 2.1.0). University Park: The Methodology Center, Penn State. Retrieved from
http://methodology.psu.edu
methodology.psu.edu/downloadsSlide70
Modeling approach
TVEM enables us to think differently about effects
Effects of predictors can strength or weaken with time
In particular, note that the effect of “baseline” characteristics can change over time
Could lead not only to tailoring interventions and treatment, but also to adaptive treatment designsSlide71
Gambling applications
None to date
But, there are applications of TVEM to other behaviors like smoking and smoking cessation that may provide inspiration!Slide72
Resources
No comprehensive text, but…
Walls, T. A., & Schafer. J. L. (2006).
Models for intensive longitudinal data
. New York, NY: Oxford University Press.Slide73
Resources
Recommended journal articles…
Lanza, S. T., Vasilenko, S., Liu, X., Li, R., & Piper, M. (2013). Advancing the
understanding
of
craving during smoking cessation attempts
: A
demonstration
of the
time-
v
arying effect model
.
Nicotine and Tobacco Research
. doi:
10.1093/ntr/ntt128
Liu, X., Li, R., Lanza, S. T., Vasilenko, S., & Piper, M. (in press). Understanding the
role
of
cessation fatigue
in the s
moking cessation process
.
Drug and Alcohol Dependence
.Slide74
Resources
Recommended journal articles continued…
Shiyko, M. P., Lanza, S. T., Tan, X., Li, R., & Shiffman, S. (2012). Using the time-varying effect model (TVEM) to examine dynamic associations between negative affect and self-confidence on smoking urges: Differences between successful quitters and relapsers.
Prevention Science, 13
, 288-299
.
doi:
10.1007/s11121-011-0264-z
Tan
, X., Shiyko, M. P., Li, R., Li, Y., & Dierker, L. (2012). A time-varying effect model for intensive longitudinal data.
Psychological Methods,
17
,
61-77.
doi:10.1037/a0025814Slide75
Resources
Recommended journal articles continued…
Vasilenko
, S., & Piper, M., Liu, X., Lanza, S. T., & Li, R. (in press). Time-varying processes involved in smoking lapse in a randomized trial of smoking cessation therapies.
Nicotine and Tobacco Research
.Slide76
Thank you!!