BSTT537 Longitudinal Data Analysis - Fall 2013

Instructor: Don Hedeker

Week 1: Tuesday August 27, 2013

Reading material: Hedeker, D. and Gibbons, R.D. "Longitudinal Data Analysis"
Chapter 1: Introduction

Homework: Problem Set 1

Week 1: Thursday August 29, 2013

Reading material: Hedeker, D. and Gibbons, R.D. "Longitudinal Data Analysis"
Chapter 2: ANOVA approaches to longitudinal data

Example 2a: Analysis of vocabulary data from Bock (1975) using univariate repeated measures ANOVA (SAS code and output)

Example 2b: IML code for generating orthogonal polynomial contrast matrix (SAS code and output)

Plots: Plots of vocabulary data including matrix, spaghetti and box plots. (SAS code, and matrix plot, spaghetti plot, boxplot)

Week 2: Tuesday September 3, 2013

Example 2c: Analysis of Pothoff & Roy data using univariate repeated measures ANOVA model with two groups (boys and girls). Includes a plot of the means across time for the two groups. (SAS code and output)

Example 2d: Analysis of Pothoff & Roy data using univariate repeated measures ANOVA. This analysis is for a main effects model - no group by time interaction. (SAS code and output

Example 2e: Test of sphericity for the Pothoff & Roy data (SAS code and output)

Plots: Plots of Potthoff & Roy data including matrix, spaghetti, mean, and box plots (most by gender). (SAS code and plots)

Week 2: Thursday September 5, 2013

Reading material: Hedeker, D. and Gibbons, R.D. "Longitudinal Data Analysis"
Chapter 3: MANOVA approaches to longitudinal data

Example 3a: Analysis of vocabulary data using MANOVA (SAS code and output

Week 3: Tuesday September 10, 2013

Example 3b: Analysis of sleep data using IML to do MANOVA (SAS code and output) ***OPTIONAL***

Example 3c: Analysis of Prozac weight data using MANOVA: includes use of CONTRAST statement to obtain estimates of a-priori contrasts for the multi-group situation. (SAS code and output

Week 3: Thursday September 12, 2013

Reading material: Hedeker, D. and Gibbons, R.D. "Longitudinal Data Analysis"
Chapter 4: Mixed-effects regression models (MRM) for continuous outcomes

Homework: Problem Set 2

Week 4: Tuesday September 17, 2013

Example 4a: Analysis of Riesby dataset using MRM: this example has a few different PROC MIXED specifications, and includes a grouping variable and curvilinear effect of time. (SAS code and output)

Dataset: Riesby dataset - the variable order and names are indicated in the previous handout.

Plots: Plots of Riesby dataset including matrix, box, spaghetti, linear, quadratic, and individual plots. (SAS code and plots)

Week 4: Thursday September 19, 2013

Example 4b: Analysis of Riesby dataset. This handout shows how empirical Bayes estimates can be output to a dataset in order to calculate estimated individual scores at all timepoints. (SAS code and output)

Plots: Plots of Riesby dataset including mean and box plots by group. (SAS code and plots)

Week 5: Tuesday September 24, 2013

Example 4c: Analysis of Riesby dataset. This handout has the analysis considering the time-varying drug plasma levels, separating the within-subjects from the between-subjects effects for these time-varying covariates. (SAS code and output)

Dataset: Riesby dataset with time-varying covariates - the variable order and names are indicated in the above handouts.

Plots: Plots of time-varying drug plasma levels including matrix and box plots. (SAS code and plots)

Week 5: Thursday September 26, 2013

Reading material: Hedeker, D. and Gibbons, R.D. "Longitudinal Data Analysis"
Chapter 5: Mixed-effects polynomial regression models for continuous outcomes

Example 5a: Analysis of Riesby dataset. This handout contrasts quadratic trend models using the raw metric of week versus orthogonal polynomials. (SAS code and output)

Example 5b: Converting data from univariate to multivariate format, and getting the observed correlations and variance-covariance matrix of the repeated measures (SAS code and output)

Example 5c: IML code that illustrates the calculation of estimated means and the variance-covariance matrix (SAS code and output)

Homework: Problem Set 3

Week 6: Tuesday October 1, 2013

Reading material: Hedeker, D. and Gibbons, R.D. "Longitudinal Data Analysis"
Section 4.6: Estimation of MRMs for continuous outcomes

Week 6: Thursday October 3, 2013

Continuation of material on estimation.

Analysis of Riesby dataset: this handout includes SAS IML code for a random-intercepts model. A comparison to analysis using PROC MIXED is also included. (SAS code and output) ***OPTIONAL***

Week 7: Tuesday October 8, 2013

Reading material: Hedeker, D. and Gibbons, R.D. "Longitudinal Data Analysis"
Chapter 6: Covariance pattern models for continuous outcomes

Week 7: Thursday October 10, 2013

Continue with material from Chapter 6

Homework: Problem Set 4

Week 8: Tuesday October 15, 2013

Reading material: Hedeker, D. and Gibbons, R.D. "Longitudinal Data Analysis"
Chapter 7: Mixed regression models with autocorrelated errors

Week 8: Thursday October 17, 2013

Example 7.1: SAS code - SAS code for analysis of Bock dataset. This handout lists syntax for several PROC MIXED analyses including (a) mixed-effects models, (b) covariance pattern models, and (c) mixed-effects models with autocorrelated errors.

Example 7.2: SAS code and output - SAS IML code and output from analysis of Bock dataset. This handout uses IML to provide estimated means, variances, and correlations across time based on mixed-effects models with autocorrelated errors.

Dataset: Bock dataset - the variable order and names are indicated in the previous two handouts.

PRACTICAL ISSUES: pdf file

Week 9: Tuesday October 22, 2013

Reading material: Hedeker, D. and Gibbons, R.D. "Longitudinal Data Analysis"
Chapter 9: Mixed-effects regression models for binary outcomes.

Week 9: Thursday October 24, 2013

Example 9.1: Analysis of the NIMH Schizophrenia dataset: this handout provides SAS (PROC LOGISTIC, NLMIXED) code for running ordinary logistic regression and mixed-effects logistic regression. (SAS code)

Example 9.2: This handout shows how to use PROC IML to get the marginalized probability estimates from the above NLMIXED analysis for the random intercept model (SAS code)

Dataset: SCHIZ dataset - the variable order and names are indicated in Example 9.1.

Homework: Problem Set 5, SAS code, Dataset

Week 10: Tuesday October 29, 2013

Example 9.3: This handout shows how to use PROC IML to get the marginalized probability estimates from the NLMIXED analysis for the random trend model of the NIMH data (SAS code)

Example 9.4: GEE analysis of the NIMH Schizophrenia dataset using SAS PROC GENMOD (SAS code)

Week 10: Thursday October 31, 2013

Reading material: Hedeker, D. and Gibbons, R.D. "Longitudinal Data Analysis"
Section 9.6: Estimation of mixed-effects binary regression models

Example: SAS IML macro program for random-intercepts binary regression (SAS code)  *** OPTIONAL ***

Week 11: Tuesday November 5, 2013

Reading material: Hedeker, D. and Gibbons, R.D. "Longitudinal Data Analysis"
Chapter 8: Generalized Estimating Equations (GEE) Models.

Example 8.1: Using the NIMH Schizophrenia dataset, this handout has PROC GENMOD code and output from several GEE analyses varying the working correlation structure. (SAS code and output)

Week 11: Thursday November 7, 2013

Example 8.2: PROC GENMOD code and output from analysis of Robin Mermelstein's smoking cessation study dataset. This handout illustrates GEE modeling of a dichotomous outcome, includes CONTRAST statements to perform linear combination and multi-parameter Wald tests, and OBSTATS to yield estimated probabilities for each observation. (SAS code and output)

Dataset: Smoking data - the variable order and names are indicated in the above handouts.

Homework: Problem Set 6

Week 12: Tuesday November 12, 2013

Continue with GEE material: GEE compared with MRM

Example 8.3: PROC IML code and output showing how to get the marginalized probability estimates from GEE and NLMIXED analysis for a random-intercepts model, including using quadrature for the latter. (SAS code and output)

Week 12: Thursday November 14, 2013

Reading material: Hedeker, D. and Gibbons, R.D. "Longitudinal Data Analysis"
Chapters 10 & 11: Mixed-effects regression models for ordinal and nominal outcomes.

Examples using SAS:
schzonl.sas - SAS code for mixed-effects proportional odds regression analysis of NIMH Schizophrenia data. schizx1.dat - ASCII datafile.

schzofit.sas - SAS IML code for obtaining marginal probability estimates based on mixed-effects proportional odds regression analysis of NIMH Schizophrenia data.

sandonl.sas - SAS code for mixed-effects proportional odds and non-proportional odds analyses of San Diego homelessness data. sdhouse.dat - ASCII datafile.

sandofit.sas - SAS code for obtaining marginal probability estimates based on mixed-effects proportional odds and non-proportional odds analyses of San Diego homelessness data.

sandnnl.sas - SAS code for mixed-effects multinomial analyses of San Diego homelessness data.

sandnfit.sas - SAS code for obtaining marginal probability estimates based on mixed-effects nominal model analyses of San Diego homelessness data.

Week 13: Tuesday November 19, 2013

Continue with material from Chapters 10 & 11

Week 13: Thursday November 21, 2013

Reading material: Hedeker, D. and Gibbons, R.D. "Longitudinal Data Analysis"
Chapter 14: Missing Data in Longitudinal Studies.

Example 14.1: Analysis of NIMH Schizophrenia dataset for time to dropout using discrete-time survival analysis: shows how to create the person-period dataset. (SAS code)

Week 14: Tuesday November 26, 2013

Reading material: Hedeker, D., & Gibbons, R.D. (1997).  Application of random-effects pattern-mixture models for missing data in longitudinal studies.  Psychological Methods, 2, 64-78. (pdf

file)

Week 14: Thursday November 28, 2013 - NO CLASS - Thanksgiving Day

Week 15: Tuesday December 3, 2013

Reading material: Hedeker, D., Gibbons, R.D., & Waternaux, C. (1999).  Sample size estimation for longitudinal designs with attrition: comparing time-related contrasts between two groups.  Journal of Educational and Behavioral Statistics, 24:70-93. (pdf file)

RMASS2.EXE contains: executable program for sample size determination based on this paper.
RMASS2.PDF

Week 15: Thursday December 5, 2013

If time permits:

Reading material: Hedeker, D., Mermelstein, R.J., & Demirtas, H. (2008).  Analysis of Binary Outcomes with Missing Data: Missing=Smoking, Last Observation Carried Forward, and a Little Multiple ImputationAddiction, 102:1564-1573.

Reading material: Hedeker, D., Mermelstein, R.J., & Demirtas, H. (2008).  An application of a mixed-effects location scale model for analysis of Ecological Momentary Assessment (EMA) data.  Biometrics, 64:627-634.

-----------------------------------------------------------------------------------------------------------------------------
extras:

bockgee - SAS PROC MIXED & GENMOD code and output from analysis of Bock dataset. This handout compares results from mixed-effects modeling to GEE modeling for this dataset with no missing data across time and a continuous outcome variable.

riesgee2 - SAS PROC MIXED & GENMOD code and output from analysis of Riesby dataset. This handout compares results from mixed-effects modeling to GEE modeling for this dataset which does have missing data across time and a continuous outcome variable.

vt.dat has data on 22 subjects from a study of affective facial expressions (Vasey and Thayer, 1987).
In this study several pieces of music were played to each subject in an attempt to elicit selected affected states.  Trial 1 was a baseline, relaxing music condition.  Trial 2 was designed to produce positive effects, and trial 3 was designed to produce agitation.  Each trial lasted 90 seconds and the response variable at each trial was the mean electromyographic (EMG) amplitude from the left brow region.  The variable order in the datafile is subject number followed by the three trial EMG measurements.  For problem set 1.

box.dat has body weight data for 30 rats measured at baseline and at weekly intervals for 4 weeks. There are three groups (1=control, 2=thiouracil, 3=thyroxin) of 10 rats each.. The datafile contains, in order, GROUP, RAT, BW0, GAIN1 (BW1-BW0), GAIN2 (BW2-BW1), GAIN3 (BW3-BW2), and GAIN4 (BW4-BW3). For problem set 1.

SCHIZREP.DAT has severity of illness scores on 437 schizophrenics measured across time. Subjects were randomized to one of four treatments: placebo, chlorpromazine, fluphenazine, or thioridazine. Here the drug groups have been combined into one group. The datafile contains, in order, Patient ID, IMPS79 (7-point severity scale), week (week 0 to week 6, though most of the measurement occured on weeks 0, 1, 3, & 6), treatment group (0=placebo, 1=drug), and sex (0=female, 1=male). For problem set 2.

SCHIZX1.DAT - Same dataset as above with the following variables, in order, Patient ID, IMPS79 (7-point severity scale), IMPS79b (binary version of IMPS79), IMPS79o (ordinal version of IMPS79), intercept (a column of ones),  treatment group (0=placebo, 1=drug), week (week 0 to week 6, though most of the measurement occured on weeks 0, 1, 3, & 6), square root of week (helps to linearize the relationship of IMPS79 over time), and treatment group by square root of week (interaction term).

RIESBYT4.DAT has Hamilton Depression Rating Scale (HDRS) scores on 66 depressed subjects across time. Subjects were given imipramine for four weeks and their drug plasma levels were obtained during each week. Drug plasma levels of desimipramine (imipramine's metabolite) were also obtained. The variables, in order, are HDRS change from baseline score, a field of ones, week (coded 0 to 3), sex (0=male, 1=female), diagnosis (0=non-endogenous, 1=endogenous), imipramine plasma levels (in ln units), and desimipramine plasma levels (in ln units). For problem sets 3 and 4.

RIESORD3.RRM has Hamilton Depression Rating Scale (HDRS) scores on 66 depressed subjects across time. The variables, in order, are HDRS score, HDRS dichotomized score, HDRS trichotomized score, a field of ones, week (coded 0 to 3), DMI dichotomized value, DMI dichotomized value by week, DMI centered value (in ln units), and DMI centered value (in ln units) by week.

Any questions or comments to Don Hedeker email: hedeker@uic.edu