Growth mixture modeling

Slides:



Advertisements
Similar presentations
Dummy Dependent variable Models
Advertisements

Inferential Statistics and t - tests
Latent Growth Modeling Chongming Yang Research Support Center FHSS College.
G ROWTH M IXTURE M ODELING Shaunna L. Clark & Ryne Estabrook Advanced Genetic Epidemiology Statistical Workshop October 24,
Mixture modelling of continuous variables. Mixture modelling So far we have dealt with mixture modelling for a selection of binary or ordinal variables.
TRIM Workshop Arco van Strien Wildlife statistics Statistics Netherlands (CBS)
Model Assessment, Selection and Averaging
Growth Curve Model Using SEM
Latent Growth Curve Modeling In Mplus:
Common Factor Analysis “World View” of PC vs. CF Choosing between PC and CF PAF -- most common kind of CF Communality & Communality Estimation Common Factor.
Midterm Review Goodness of Fit and Predictive Accuracy
Notes on Logistic Regression STAT 4330/8330. Introduction Previously, you learned about odds ratios (OR’s). We now transition and begin discussion of.
Latent Class Analysis Presented by Nicholas Branic UCI Stats n’ Snacks December 9, 2014.
Mixture Modeling Chongming Yang Research Support Center FHSS College.
Introduction to Multilevel Modeling Using SPSS
1 Binary Models 1 A (Longitudinal) Latent Class Analysis of Bedwetting.
Regression and Correlation Methods Judy Zhong Ph.D.
Xitao Fan, Ph.D. Chair Professor & Dean Faculty of Education University of Macau Designing Monte Carlo Simulation Studies.
Advanced Statistical Methods for Research Math 736/836
Simple Linear Regression
Trajectory 1. Physics. The path of any body moving under the action of given forces... especially the curve described by a projectile in its flight through.
Biostatistics Case Studies 2005 Peter D. Christenson Biostatistician Session 4: Taking Risks and Playing the Odds: OR vs.
Understanding Statistics
Introduction ► College-student drinking remains a significant problem on campuses across the nation. ► It is estimated that 38-44% of college students.
CJT 765: Structural Equation Modeling Class 7: fitting a model, fit indices, comparingmodels, statistical power.
Categorical and Zero Inflated Growth Models Alan C. Acock* Summer, 2009 *Alan C. Acock, Department of Human Development and Family Sciences, Oregon State.
Social patterning in bed-sharing behaviour A longitudinal latent class analysis (LLCA)
Rasch trees: A new method for detecting differential item functioning in the Rasch model Carolin Strobl Julia Kopf Achim Zeileis.
SEM: Basics Byrne Chapter 1 Tabachnick SEM
Growth Curve Models Using Multilevel Modeling with SPSS David A. Kenny January 23, 2014.
Multilevel Linear Models Field, Chapter 19. Why use multilevel models? Meeting the assumptions of the linear model – Homogeneity of regression coefficients.
Latent Growth Curve Modeling In Mplus: An Introduction and Practice Examples Part II Edward D. Barker, Ph.D. Social, Genetic, and Developmental Psychiatry.
CJT 765: Structural Equation Modeling Class 8: Confirmatory Factory Analysis.
Introduction to regression 3D. Interpretation, interpolation, and extrapolation.
1 Parallel Models. 2 Model two separate processes which run in tandem Bedwetting and daytime wetting 5 time points: 4½, 5½, 6½,7½ & 9½ yrs Binary measures.
Modeling the Course and Consequences of Parenting Self-Efficacy During Infancy and Early Childhood: Improving Estimates with an Adoption Design Chelsea.
Extending Group-Based Trajectory Modeling to Account for Subject Attrition (Sociological Methods & Research, 2011) Amelia Haviland Bobby Jones Daniel S.
Roghayeh parsaee  These approaches assume that the study sample arises from a homogeneous population  focus is on relationships among variables 
One-Way Analysis of Covariance (ANCOVA)
CJT 765: Structural Equation Modeling Class 8: Confirmatory Factory Analysis.
University Rennes 2, CRPCC, EA 1285
Exploring social mobility with latent trajectory group analysis Patrick Sturgis, University of Southampton and National Centre for Research Methods From.
ALISON BOWLING CONFIRMATORY FACTOR ANALYSIS. REVIEW OF EFA Exploratory Factor Analysis (EFA) Explores the data All measured variables are related to every.
Tutorial I: Missing Value Analysis
Two-Group Discriminant Function Analysis. Overview You wish to predict group membership. There are only two groups. Your predictor variables are continuous.
CJT 765: Structural Equation Modeling Final Lecture: Multiple-Group Models, a Word about Latent Growth Models, Pitfalls, Critique and Future Directions.
Chapter 17 STRUCTURAL EQUATION MODELING. Structural Equation Modeling (SEM)  Relatively new statistical technique used to test theoretical or causal.
Methods of multivariate analysis Ing. Jozef Palkovič, PhD.
Bootstrap and Model Validation
Latent Class Regression
Annual Meeting & Exposition of American Public Health Association
BINARY LOGISTIC REGRESSION
Dean Lauterbach, Ph.D.1, Brian Allen, PsyD2, Stefanie Poehacker, MS1,
Logistic Regression APKC – STATS AFAC (2016).
Latent Class Regression Computing examples
Notes on Logistic Regression
CJT 765: Structural Equation Modeling
CJT 765: Structural Equation Modeling
Meghan E. Martz, PhD, Robert A. Zucker, PhD, Mary M. Heitzeg, PhD
New York State Suicide Prevention Conference
Caihong R. Li, MS Latent Class Analysis in Mplus November 7, 2017
Statistical Methods For Engineers
Angela B. Bradford School of Family Life
Where did we stop? The Bayes decision rule guarantees an optimal classification… … But it requires the knowledge of P(ci|x) (or p(x|ci) and P(ci)) We.
Day 2 Applications of Growth Curve Models June 28 & 29, 2018
Special Topic: Longitudinal Mediation Analysis
Latent Variable Mixture Growth Modeling in Mplus
Faculty Fellow, University of Nebraska Public Policy Center
Rachael Bedford Mplus: Longitudinal Analysis Workshop 23/06/2015
Presentation transcript:

Growth mixture modeling Michael Moore, Ph.D. Growth mixture modeling An Introduction and Illustrative Example

Overview What is growth mixture modeling (GMM)? How do you determine how many classes are present in your data? Issues in GMM How do you do it? An example of GMM using Mplus

What is GMM? Growth mixture modeling (GMM): Goal is characterized inter-individual differences in intra- individual change over time (Nesselroade, 1991) Can we reliably classify people on the basis on how they change over time? Applications include: how people develop, naturally, over time, how people change during therapy, etc. Person-centered: classify individuals, not describe relationships between variables (unlike regression, factor analysis)

What is GMM? Extension of latent variable growth curve modeling (LGM) Goal is to characterize individuals on basis of estimate of growth parameters (intercept & slope) Intercept – where you begin (before change) Slope – describes rate & shape of change process GMM identifies subgroups on basis of intercept & slope

What is GMM? GMM similar to latent class analysis (LCGA) LCGA, unlike GMM, assumes homogeneity in growth trajectories within a class

LGM vs. LCGA vs. GMM (Shiyko, Ram, & Grimm, 2012)

How many classes? Many different indices of fit: Posterior probabilities – what is the probability of being in a given class for a given individual Entropy – aggregate of posterior probabilities (closer to 1, the better) Akaike/Bayesian Information Criteria (A/BIC) – smaller is better Lo, Mendell, Rubin (2001) Likelihood Ratio Test (LMR- LRT) – compares the estimated model with model with one fewer class, p < .05 indicates estimated model is superior Bootstrap Likelihood Ratio Test (BLRT) – LRT with bootstrapped samples, very computation intense

How many classes? Limited research (Nylund, Asparouhov, Muthén, 2007) suggests that BLRT and A/BIC perform best BUT, Nylund et al. suggest use of BIC & LMR to narrow down to a few models, then use BLRT

How many classes? Remember: GMM is exploratory technique, so a priori theory is best judge (Bauer & Curran, 2003; Muthén, 2003; Rindskopf, 2003) Like EFA Replication is important, if not essential Does this class make sense? Is Class A really different from Class B? Does each class contain a sufficient number of people to be considered reliable?

Issues in GMM Problems with convergence GMMs notorious for convergence problems May need to give the computer a lot more time to do its thing Have to worry about computer settling on solution that is not optimal (“local solution”) OK to change defaults Mplus default = 10 random starts & 2 optimizations at the final stage Correct solution only obtained in 23% of 200 data sets with default starts – OK to increase number of starts to 50 – 100 (Hipp & Bauer, 2006) Increasing starts will increase computation time Not much research as guide

Issues in GMM Sample size? Fit indices correctly identified number of classes with n = 500+ (Nylund, et al., 2007) Again, not much research to guide decision-making

How do you do it? Recommendations by Jung & Wickrama (2008) Specify a single-class LGM Look for significant (unmodeled) variability in this trajectory

LGM Mplus Syntax DATA: file is ‘C:\filename.dat’; VARIABLE: names are id sex t1 t2 t3; usevar = t1-t3; missing = all (999); ANALYSIS: type = missing H1; MODEL: i s | t1@0 t2@1 t3@2; OUTPUT: sampstat standardized;

How do you do it? Specify LCGA without covariates LCGA results will be cleaner than GMM – in case of non-convergence/ill fit, makes isolating cause(s) easier

LCGA Mplus Syntax DATA: file is ‘C:\filename.dat’; VARIABLE: names are id sex t1 t2 t3; usevar = t1-t3; missing = all (999); classes = c(3); ANALYSIS: type = mixture; starts = 10 2; stiterations = 10; MODEL: % overall % i s | t1@0 t2@1 t3@2; i-s@0; OUTPUT: sampstat standardized tech11 tech14;

How do you do it? Specify GMM without covariates Specify GMM with covariates Can examine predictor(s) of intercept/slope & class membership (not included)

GMM Mplus Syntax DATA: file is ‘C:\filename.dat’; VARIABLE: names are id sex t1 t2 t3; usevar = t1-t3; missing = all (999); classes = c(3); ANALYSIS: type = mixture; starts = 10 2; stiterations = 10; MODEL: % overall % i s | t1@0 t2@1 t3@2; OUTPUT: sampstat standardized tech11 tech14;

How do you do it? Validate classes Done via identification of correlates of class membership Can have Mplus save a data file containing which class each participant is predicted to be in Then merge with existing dataset(s)

GMM Mplus Syntax DATA: file is ‘C:\filename.dat’; VARIABLE: names are id sex t1 t2 t3; usevar = t1-t3; idvariable = id; missing = all (999); classes = c(3); SAVEDATA: file is C:\; save = filename; ANALYSIS: type = mixture; starts = 10 2; stiterations = 10; MODEL: % overall % i s | t1@0 t2@1 t3@2; OUTPUT: sampstat standardized tech11 tech14;

An Example Using Mplus Investigation of 6-week, residential ACT treatment for comorbid PTSD/SUD Very small sample size (n = 23) Participants assessed at pre-, post-treatment, & 4-6 week follow-up Treatment consisted of: ACT education group (2x weekly, 60 mins.) ACT process group (3x weekly, 90 mins.) Coping skills group (2x weekly, 90 mins.) Trauma education group (1x weekly, 60 mins.) Insomnia group (1x weekly, 60 mins.) Anger mgmt. group (1x weekly, 60 mins.) Daily living skills group (2x weekly, 75 mins.) Individual case mgmt. sessions (1x weekly, 50 mins.)

An Example Using Mplus Symptoms of PTSD: 2-Class Model 3-Class Model BIC = 178.04 Entropy = 1.00 Class 1 (n = 16); Class 2 (n = 6) 3-Class Model BIC = 131.99 Class 1 (n = 20); Class 2 (n = 1); Class 3 (n = 1)

PTSD – 2 Class Model

An Example Using Mplus Symptoms of PTSD: Class 1 – PTSD sxs decrease over acute tx, but increase over fup Class 2 – PTSD sxs increase over acute tx, but decrease over fup Predicted outcome from an ACT/exposure model Exp avoidance lower in Class 2 (vs. Class 1) – p < .001

An Example Using Mplus Symptoms of SUD: 2-Class Model 3-Class Model BIC = 167.78 Entropy = 1.00 Class 1 (n = 1); Class 2 (n = 21) 3-Class Model BIC = 166.05 Class 1 (n = 18); Class 2 (n = 3); Class 3 (n = 1)

SUD – 3 Class Model

An Example Using Mplus Symptoms of SUD: Class 1 – Relative stability of SUD sx over time Class 2 – SUD sxs increase over acute tx, but decrease over time Exp avoidance lower in Class 2 (vs. Class 1) – p < .001 Sx increase during acute tx & desistence afterwards predicted by ACT/exposure model?

Thank-You …for listening Everyone at the Baltimore VA Get slides from my ResearchGate page or e-mail mmoore@adelphi.edu