Course Goals

• Learn how to use modern regression methods to answer scientific questions
• Become familiar with statistical concepts including exploratory data analysis, estimation, testing in linear, logistic, and survival models
• Understand how the development of statistical methodology is motivated by biological and medical problems
• Develop data analytic skills including familiarity with several statistical programs
• Develop writing skills needed to communicate the results of a data analysis

Topics of Discussion

• Introduction to Regression Models.
• Simple Linear Regression.
• A Review of Matrix Algebra and Important Results of Random Vectors.
• Multiple Linear Regression Model. Partial F-test.
• Specification Issues in Regression Models: ANCOVA, ANOVA, Multicollinearity.
• Model Checking: diagnostics, transformations, influential observations, lack-of-fit test.
• Model Selection.
• Case Studies in Linear Regression.
• Nonlinear Regression Models
• Bayesian Linear Models
• Logistic Regression.
• Generalized Linear Models and Poisson Regression.
• Survival models

Software

• You can use any software your want other than Excel. We will primarily be demonstrating Stata and giving out Stata code. In some situations, R or WinBUGS will be used.
• See SoftwareRecs for more software recommendations
• Chun Li's Stata notes, Need-to-know commands, and R notes
• Leena Choi's Stata notes for classes, Stata Lab info, and R notes for classes

R information

• R 2.8 (optional, free from http://www.r-project.org): Powerful, versatile, and actively maintained and updated. It may require a longer learning curve than Stata and SPSS, but the effort will pay off later on. To get a feel, look at one of the following: 1 (and try the commands in "A Sample Session“) 2 3. The Department of Biostatistics has a free R Clinic every Thursday. Print the R reference card to get a list of the most commonly used commands.
• To download R: Linux Mac Windows
• Under R you can use a menu to install new packages. Install the Rcmdr package, which provides a simple SPSS-like menu system to interact with the R language. Load Rcmdr and the main menu will appear. Go here or links below for more information. The first time you load Rcmdr it will ask you if you want to download and install packages that Rcmdr depends on. Answer affirmatively, and specify CRAN as the source. This will take a few minutes but only needs to be done the first time you try to use Rcmdr.
  • Installation and usage notes from Robert Schaefer
  • Rcmdr installation instructions especially for Mac
  • To fix a bug causing the boxplot menu to not appear, install the R package aplpack.
  • To update Rcmdr to any version newer than what's on CRAN run install.packages("Rcmdr",repos="http://R-Forge.R-project.org") in the command window.

Comments on Other Software Packages

• Stata: Powerful and good graphics with an SPSS-like menu system. A good support site is at http://www.ats.ucla.edu/stat/stata. Cost is $89 for a year and $145 for life, through GradPlan. Buy Small Stata for $45 if you have to pay by yourself. Stata also is available at the College of Arts & Science Microcomputer Labs.
• SAS: The oldest survivor, with strong legacy. Hard to learn and extend, with outdated structure and the worst graphics of any major package.
• SPSS: Have “standard” methods and good graphical user interface. However, it is difficult to extend beyond the “standard” methods.
• Honorable mention: Epi Info (free from CDC), S-Plus.
• There is a long list of reasons not to use Excel. See ExcelProblems
• See StatComp for more information about statistical computing including links to several online statistical and probability calculators