STAT/BIOSTAT/CSSS: 529 SAMPLE SURVEY TECHNIQUES
Jon Wakefield
Departments of Statistics and Biostatistics
This class concerns the analysis of data collected via surveys. Such data require specialized methods for analysis, to account for the non-random sampling that is often carried out for logistic or efficiency reasons. Ignoring the design can lead to estimators with substantial bias and inappropriate variance estimates.
Complex survey designs are now commonplace in many disciplines, including the health sciences, global health and demography. Examples from the lecturers own research will be discussed, including the estimation of health indicators (e.g., smoking, diabetes) across health reporting areas in King County using BRFSS survey data, and estimation of under 5 child mortality across regions of Tanzania using DHS and HDSS data.
In this course, modern estimation techniques will be described along with practical implementation using the R computing environment, and specifically the survey package.
A variety of complex designs will be discussed including: simple random random sampling, stratified random sampling, probability proportional to size sampling, cluster sampling, and multistage sampling.
Design-based, model-based and model-assisted methods of estimation will be described and critiqued.
Students will also learn how to unravel the mysteries of the construction and use of design weights.
So if you think you might have a weight problem, this is the course for you.
Post-stratification and raking will also be described.
The class will meet on Tuesdays and Thursdays, 10.30-11.50 in LOW 105.
References:
Complex Surveys: A Guide to Analysis Using R (2010), by Thomas Lumley.
Analysis of Health Surveys (1999), by Edward Korn and Barry Gaubard.
Sampling: Design and Analysis, Second Edition (2010), by Sharon Lohr.
Model Assisted Survey Sampling (1992), by Carl-Erik Sarndal, Bengt Swenson and Jan Wretman.
Here's the official syllabus:
CS&SS 529 Sample Survey Techniques (3)
Design and implementation of selection and estimation procedures.
Emphasis on human populations. Simple, stratified, and cluster sampling; multistage and two-phase procedures; optimal allocation of resources; estimation theory; replicated designs; variance estimation; national samples and census materials. Prerequisite: either STAT 421, STAT 423, STAT 504, QMETH 500, BIOST 511, or BIOST 517, or equivalent; or permission of instructor. Offered: jointly with BIOST 529/STAT 529.
