The University of North Carolina at Chapel Hill Fall Semester, 2013

advertisement
The University of North Carolina at Chapel Hill
School of Social Work
SOWO 917 Longitudinal and Multilevel Analysis
Fall Semester, 2013
INSTRUCTOR
Shenyang Guo, Ph.D.,
Room 524j,
Tate Turner Kuralt Building
CB #3550,
School of Social Work, Chapel Hill, NC 27599-3550
Phone: (919) 843-2455
Email: sguo@email.unc.edu
CLASS MEETING TIMES & OFFICE HOURS
Class meets on Wednesdays 9:00-11:50 am
Office hours are Tuesdays 10:30 – 12:30 (Room 524j TTK)
COURSE DESCRIPTION
This course introduces statistical frameworks, analytical tools, and social behavioral
applications of three types of models: event history analysis, hierarchical linear modeling
(HLM), and growth curve analysis.
COURSE OBJECTIVES
At the completion of the course, students will have a solid understanding of the
challenges and problems in longitudinal and multilevel analysis. They will know how to
choose appropriate statistical analyses that best suit the type of data and research
questions for a given study. They are expected to be able to run, interpret, and
communicate results clearly and effectively in writing based on the following models: life
tables, Kaplan-Meier’s estimate of survivor function, discrete time model, Cox
proportional hazard model, marginal models handling multilevel event data, two-level
and three-level hierarchical linear models, growth curve analysis, and analysis of a
categorical dependent variable using HGLM.
PRE-REQUIREMENT
Students are assumed to be familiar with descriptive and inferential statistics as well as
multiple regression analysis. They should have statistical and statistical software
background at least equivalent to that provided by SOCI209, PSYC282, EDUC284
(linear regression), or SOCI211 (categorical data analysis). Students without such
prerequisites should contact the instructor to determine their eligibility to take this course.
1
STATISTICAL SOFTWARE PACKAGES
Students may choose to use Stata, SAS, or SPSS as the primary statistical software
package for the course, though the classroom lectures and materials will be based on
Stata. Specialized software package HLM will also be demonstrated.
TEXTBOOKS
Guo, S. (2010). Survival Analysis: A Practical Guide to Social Work Research. New
York, NY: Oxford University Press.
Singer, J.D., & Willett, J.B., (2003). Applied Longitudinal Data Analysis: Modeling
Change and Event Occurrence, New York, NY: Oxford University Press
RECOMMENDED TEXTBOOKS
Allison, P.D. (1995). Survival Analysis Using the SAS System. Cary, NC: SAS Institute
Inc.
Collett, D. (1994). Modeling Survival Data in Medical Research. New York, NY:
Chapman & Hall
Cleves, M.A., Gould, W.W., & Gutierrez, R.G. (2004). An introduction to survival
analysis using Stata, Rev. ed., College Station, TX: Stata Press.
Raudenbush, S.W., & Bryk, A.S. (2002). Hierarchical Linear Models: Applications and
Data Analysis Methods, Second Edition, Thousand Oaks, CA: Sage Publications
Ltd.
Rabe-Hesketh, S., & Skrondal, A. (2005). Multilevel and Longitudinal Modeling Using
Stata, College Station, TX: Stata Press.
ASSIGNMENTS
GRADE PERCENTAGE
Assignment 1
Assignment 2
Assignment 3
Assignment 4
Assignment 5
Midterm Exam (take home)
Final Exam (take home)
10%
10%
10%
10%
10%
25%
25%
GRADING SYSTEM
The standard School of Social Work interpretation of grades and numerical scores will be used.
H = 94-100
P = 80-93
L = 70-79
F = 69 and below
2
POLICY ON CLASS ATTENDANCE
Class attendance is an important element of class evaluation, and you are expected to attend all
scheduled sessions. Each class session will cover a great deal of materials, and you will fall
behind the course when you miss even one class session. It’s student’s responsibility to inform
the instructor via email in advance for missing a class session. You are expected not to miss
more than two sessions for the whole semester. Starting from the second missing, your course
grade will be reduced by 10% for each session missed.
POLICY ON INCOMPLETE AND LATE ASSIGNMENTS
Assignments are to be turned in to the professor by 5pm of the due date noted in the course
outline. Extensions may be granted by the professor given advance notice of at least 24 hours.
Late assignments (not turned in by 5pm on the due date) will be reduced 10 percent for each day
late (including weekend days). A grade of incomplete will only be given under extenuating
circumstances and in accordance with University policy.
POLICY ON ACADEMIC DISHONESTY
Students are expected to follow the UNC Honor Code. Please include the honor code statement
along with your signature on all assignments:
“I have neither given nor received unauthorized aid on this assignment.”
Please refer to the APA Style Guide, the SSW Manual, and the SSW Writing Guide for
information on attribution of quotes, plagiarism and appropriate use of assistance in preparing
assignments.
If reason exists to believe that academic dishonesty has occurred, a referral will be made to the
Office of the Student Attorney General for investigation and further action as required.
POLICY ON ACCOMMODATIONS FOR STUDENTS WITH DISABILITIES
Students with disabilities which affect their participation in the course may notify the instructor
if they wish to have special accommodations in instructional format, examination format, etc.,
considered.
COURSE OUTLINE (TOPICS, READINGS, AND ASSIGNMENTS)
8/21/13
1. Introduction and course overview
Review of longitudinal design: multi-wave panel, cohort, staggered multiple
cohorts, cohort-sequential, experimental, survey, and designs using
administrative data.
Review of conventional approaches to longitudinal analysis: transition probability
of Markov chain model, paired t test, within-subject ANOVA, repeated
measure MANOVA.
Review of statistical concepts: statistical assumptions embedded in OLS, problems
of autocorrelation, and maximum likelihood estimator.
Readings:
Singer & Willett, Chapter 1.
3
Guo, S. (2008). Quantitative research. Entry for the Encyclopedia of Social Work,
20th Edition. New York, NY: The Oxford University Press.
Guo, S. (2013). Advanced statistical analysis. Entry for the Encyclopedia of Social
Work Online. New York, NY: The Oxford University Press.
8/28/13
2. Life table and Kaplan-Meier methods
Overview of event history analysis
Censoring
Cohort life tables
Kaplan-Meier’s estimate of survivor functions
Readings:
Guo, Chapters 1 & 2.
Singer & Willett, Chapter 9.
Guang Guo (1993). “Event history analysis for left-truncated data”, Sociological
Methodology, 23, 217-243.
Schwartz, I., Ortega, R., Guo, S., & Fishman, G. (1994) "Infants in nonpermanent
placement". Social Service Review, 68(3), 405-416.
(Hand out Assignment 1)
Assignment 1 (Due: 9/11/13): (1) hand calculation of a life table; (2) use provided
data set to construct life tables by stratum, perform a Kaplan-Meier test on group
differences, and interpret findings; and (3) describe a longitudinal study that
requires event history analysis.
9/4/13
3. Discrete time models
Review of binary and multinomial logistic regression
The logit model for discrete time
Time-varying covariates
Readings:
Guo, Chapter 3.
Singer & Willett, Chapters 10-11.
Allison, P.D. (1982). “Discrete-time methods for the analysis of event histories”,
Sociological Methodology, 13, 61-98.
Harris, K.M. (1993). “Work and welfare among single mothers in poverty.”
American Journal of Sociology 99: 317-352.
9/11/13
4. Parametric models
The exponential model
The Weibull model
Overview of other parametric models
Readings:
Guo, Chapter 5.
Singer & Willett, Chapters 12-13.
Goerge, R.M. (1990). “The reunification process in substitute care”. Social Service
Review 64(3): 422-457.
4
Sandefur & Cook, (1998). “Permanent exits from public assistance: The impact of
duration, family, and work”. Social Forces, 77(2) 763-786.
(Hand out Assignment 2)
Assignment 2 (Due: 9/25/13): (1) solve problems on discrete-time and parametric
models; and (2) use provided data to estimate a discrete-time model, and interpret
findings.
9/18/13
5. Cox proportional hazards model (I)
Overview
Partial likelihood estimator
Cox regression with time-varying covariates
Readings:
Guo, Chapter 4.
Singer & Willett, Chapter 14.
9/25/13
6. Cox proportional hazards model (II)
Competing risks
Accelerated failure time models
Model-predicted survivor curves
Power analysis for survival models
Readings:
Singer & Willett, Chapter 15.
Collett, D. (1994). Modeling Survival Data in Medical Research. London,
Chapman & Hall. Chapter 9.
Guo, S., Biegel, D., Johnson, J. & Dyches, H. (2001) “Assessing the impact of
mobile crisis services on preventing hospitalization: A community-based
evaluation”. Psychiatric Services 52(2):223-228.
10/2/13
7. Cox proportional hazards model (III)
Introduction to multilevel event time data (multivariate failure time data)
Marginal approaches to multilevel event times
Unobserved heterogeneity
Readings:
Guo, Chapter 6.
Guo, S., & Wells, K. (2003). Research on timing of foster-care outcomes: one
methodological problem and approaches to its solution. Social Service
Review 77(1): 1-24.
Lin, D.Y. (1994). Cox regression analysis of multivariate failure time data: The
marginal approach. Statistics in Medicine 13: 2233-2247.
Guang Guo and German Rodriguez. (1992). “Estimating a multivariate
proportional hazards model for clustered data using the EM algorithm,
with an application to child survival in Guatemala.” Journal of American
Statistical Association 87: 969-976.
Heckman, J.J., & Singer, B. (1985), “Social science duration analysis”, in
Longitudinal Studies of Labor Market Data, New York, NY: Cambridge
University Press. Chapter 2.
5
Trussell, J., & Richards, T. (1985). “Correcting for unmeasured heterogeneity in
hazard models using the Heckman-Singer procedure.” Sociological
Methodology 15: 242-276.
(Hand out Midterm)
Midterm Exam (Due: 10/23/13): Use data sets provided by the course or data set
you choose to run a Cox regression model. Write a paper (no more than 14 pages,
double spaced) to present findings. The paper should include: (1) data and
specification of Cox regression; (2) testing interaction terms; (3) present predicted
survivor curves based on estimated model; and (4) interpret findings.
10/9/13
8. Overview of HLM and contextual analysis
Multi-level hypotheses in social sciences
Intra-class correlation
Random effects
Two-level model
Readings:
Guo, S. (2005). “Analyzing grouped data with hierarchical linear modeling”,
Children and Youth Services Review 27:637-65.
Singer & Willett, Chapter 2.
10/16/13
No Class, Fall Break
10/23/13
9. Growth curve and contextual analysis
Three-level model
Goodness-of-fit indices
Application to contextual and multilevel analysis
Readings:
Singer & Willett, Chapters 3-4
Townsend, A., Miller, B., & Guo, S. (2001). “Depression in middle-aged and older
married couples: A dyadic analysis”. Journal of Gerontology: Social
Sciences, 56B(6): S352-S364.
(Hand out Assignment 3)
Assignment 3 (Due: 11/6/13): (1) describe a multilevel or longitudinal study that
requires hierarchical linear modeling, (2) solve problems on HLM; (3) use provided
data to estimate intra-class correlation, and (4) run a two-level HLM, interpret
findings.
10/30/13
10. Computer Lab:
Using Stata to run HLM
Readings:
Singer & Willett, Chapter 5-6
11/6/13
11. Principles of estimation, hypothesis testing, and GEE method
Hypothesis testing
Multiparameter testing
6
HLM assumptions about data
Overview of estimation via ML and empirical Bayesian
Generalized-estimating-equation (GEE) method
Readings:
Singer & Willett, Chapters 7
(Hand out Assignment 4)
Assignment 4 (Due: 11/20/13): Read the Sampson, Raudenbush, & Earls’ (1997)
article. Group discussion and classroom presentation. This Science-published
study has made significant contributions to the field, both conceptually and
methodologically. On the due date, students will present findings from their group
discussions on the paper. Students will be divided into smaller groups in advance.
Sampson, R.J., Raudenbush, S.W., & Earls, F. (1997) “Neighborhoods and
violent crime: A multilevel study of collective efficacy.” Science
277(15): 918-924.
11/13/13
12. Growth curve analysis and testing mediating effects
Graphic presentation of individual trajectories
Three-level models in growth curve analysis
Predicted values of outcome variable using SAS Proc Mixed
Testing mediating effects in HLM
Readings:
Singer & Willett, Chapter 8
Guo, S. & Hussey, D. (1999). “Analyzing longitudinal rating data: A three-level
hierarchical linear model”. Social Work Research, 23(4):258-269.
Flannery, D., Vazsonyi, A., Liau, A., Guo, S., Powell, K., Atha, H., & Vesterdal,
W. (2003). “Initial behavior outcomes for PeaceBuilders universal
school-based violence prevention program”. Developmental Psychology
39 (2): 292-308.
Krull, J.L., & MacKinnon, D.P. (1999). “Multilevel mediation modeling in groupbased intervention studies”. Evaluation Review 23(4):418-444.
Krull, J.L., & MacKinnon, D.P. (2001). “Multilevel modeling of individual and
group level mediated effects”. Multivariate Behavioral Research 36(2):
249-277.
11/20/13
13. Classroom discussion & other topics in HLM
Part A (9:00-10:00):
Student presentation and classroom discussion on Sampson, Raudenbush, & Earls
(1997).
Part B (10:00-11:50):
Other applications of HLM: power analysis for HLM, HLM and SEM, multiple
outcome variables, structured covariance matrix within subjects.
Readings:
Raudenbush, SW., & Liu, X. (2000). “Statistical power and optimal design for
multisite randomized trials.” Psychological Methods 5(2): 199-213.
7
Curran, P.J. (2003). “have multilevel models been structural equation models all
along?” Multivariate Behavioral Research 38(4): 529-569.
Guang Guo & John Hipp. “The analysis of linear longitudinal data: Random-effects
models and structural equation models.” In Melissa Hardy (edited), New
Handbook on Data Analysis, London, England: Sage.
MacCallum, R.C., & Kim, C. (2000). “Modeling multivariate change”, in Little,
Schnabel, & Baumert edited, Modeling Longitudinal and Multilevel Data.
Lawrence Erlbaum Associates, Publishers.
Singer, L., Salvator, A., Guo, S., Collin, M., Lilien, L, & Baley, J. (1999).
“Maternal psychological distress and parenting stress after the birth of a
very low birthweight infant”. JAMA (Journal of American Medical
Association), 281(9): 799-805.
(Hand out Assignment 5)
Assignment 5 (Due: 11/26/13): Choose one article from two that are provided by
the course to perform a critical review. The provided articles employed (or
potentially should have employed but did not) either a survival model or an HLM.
This review (no more than two pages, single-spaced) should focus on: (1) strengths
and limitations (very briefly), (2) major statistical problems, and (3)
recommendations for revisions.
(Hand out Final)
Final Exam (Due: 12/7/13): Use data sets provided by the course or data set you
choose to run an HLM. Write a paper (no more than 14 pages, double spaced) to
present findings. The paper should include: (1) research objectives and questions;
(2) methods and model specifications; (3) findings; and (4) conclusions and
implications.
11/27/13
Thanksgiving holiday! No class.
12/4/13
14. Hierarchical generalized linear Models (HGLM)
Overview of generalized linear model (GLM)
Multilevel logistic regression
Multilevel ordered logistic regression
Multilevel Poisson regression
Readings:
Miller, B., & Guo, S. (2000) “Social support for spouse caregivers of persons with
dementia”. Journal of Gerontology: Social Sciences 55B(3): S163-S172.
Course summary
Topics for future study
12/7/13
Final Exam Due
8
Download