Summary of univariate categorical data analysis; iIntroduction to multivariate analysis of categorical epidemiologic and health sciences data using multiplicative models. Experience at interpretation; familiarity with available softwareprograms gained by analysis of bona fide data and critiques of published analyses appearing in literature. Prerequisite: BIOST 515; EPI 513 and either BIOST 513 or BIOST 518; or permission of instructor. Offered: jointly with EPI 536; A.
By the end of this course, students should be able to do the following using unconditional or conditional logistic regression, or using generalized estimating equations (GEE):
1. Perform logistic regression analyses with multiple predictors. 2. Perform tests that indicate the statistical significance of a covariate or group of covariates, with or without adjustment for other covariates included in the model. 3. Determine confidence intervals for individual regression coefficients in a model. 4. Compare different models with respect to their predictive power. 5. Determine if there is a linear trend with outcome for ordinal or quantitative covariates 6. Use graphical and other methods for assessing the adequacy of the fitted model. 7. Interpret each coefficient in the model. 8. Describe the methods and results to a non-statistical reader.
Student learning goals
General method of instruction
The lectures are prepared in advance, with hardcopies of the lecture notes distributed in class. Questions from registered students are encouraged. The questions often clarify points on which several students may share the same uncertainty. If you believe your question is not of general interest, feel free to ask your question before or after class. Brief student presentations on how the course material relates to subject matter studies of interest to them are often made at the conclusion of the formal lectures.
It is assumed that when entering BIOST/EPI 536, you have completed a course in linear regression and been exposed to logistic regression and some categorical data analysis as in BIOST 513. You should understand the basic statistical concepts of sampling variation, parameter estimation, confidence limits, and statistical hypothesis testing. You should know about simple statistical techniques for analyzing data from a binomial distribution including odds ratio estimation in 2 x 2 tables and in series of 2 x 2 tables. You should be familiar with the Mantel-Haenszel test and testing for trend in a 2 x K table.
Most students will find this course demanding. The homeworks are time-consuming and the text needs to be read several times for it to make sense. We expect you to talk to your classmates about the materials and homeworks to gain further insight
Class assignments and grading
Homework will generally be distributed one week in advance and be due in class the following Thursday. This typically involves analysis and interpretation of statistical data, and critiques of statistical designs and analyses reported in the literature. Computer output should be edited to eliminate all irrelevant material and should clearly indicate the answer to the question posed. Late assignments are not acceptable. In view of the class size, only portions of each assignment may actually be graded. Homework keys prepared by the teaching assistant will be posted to the class website.
The term project involves statistical analysis of epidemiological data and preparation of a scientific report such as would be submitted for publication to an epidemiological journal.
The following percentage distributions determine the course grade: 25% Homework; 5% Class participation; 20% Midterm (two hours in class); 25% Project term paper; 25% Final (two hour in class).