StatLinks

From Martin Bland

Childhood Cancer Epidemiology
Childhood Leukemia Epidemiology
MHC and Leukemia (PPT)
Gender Effect in Childhood Cancer
Genetic Epidemiology & Glossary (PPT)
Statistical Analysis of Genetic Association Studies (PPT)
Pitfalls in Genetic Association Studies {PPT}
Epidemiologic Study Designs {PPT}
Bias and Confounding in Epidemiology (PPT)
Principles of Infectious Disease Epidemiology {PPT}
Glossaries: Biostatistics Genetics Genetic Epidemiology

Internet Links
Extensive Epidemiology and Biostatistics Links
Online Epidemiology Textbook (BMJ) Principles of Epidemiology (CDC)
Epidemiology & BioStatistics Super Lectures
Human Genome Epidemiology Online Book (CDC, 2004)

From Department of Biostatistics, Vanderbilt University

By Ewout Steyerberg and Frank Harrell

There are two general applications for multiple regression (MR): prediction and explanation. These roughly correspond to two differing goals in research: being able to make valid projections concerning an outcome for a particular individual (prediction), or attempting to understand a phenomenon by examining a variable's correlates on a group level (explanation). There has been debate as to whether these two applications of MR are grossly different, as Scriven (1959) and Anderson and Shanteau, (1977) asserts, or necessarily part and parcel of the same process (e.g., DeGroot, 1969, Kaplan, 1964; for an overview of this discussion see Pedhazur, 1997, pp. 195-198). Regardless of the philosophical issues, there are different analytic procedures involved with the two types of analyses. The goal of this paper is to present: (a) the concept of prediction via MR, (b) the assumptions underlying multiple regression analysis, (c) shrinkage, cross-validation, and double cross-validation of prediction equations, and (d) how to calculate confidence intervals around individual predictions.

StatTools is a collection of statistical programs that have accumulated over the years in response to needs, requests, and enquiries, from colleagues and students. They should therefore be viewed more as a handy set of tools, all of which were found to be needed or useful by some clinicians or researchers. Users should therefore search through StatTools to see if a program that is useful to them exists, but should not see StatTools as a structured set of instructions or a system of teaching.

StatTools contains no original algorithms or statistical theories. All the programs in StatTools have been described in statistical packages, text books, published papers, and on web sites in the public domain. What the authors had done was to study the materials, write the programs that convert the maths into a usable program, and place these in a user friendly format on the Internet. Although StatTools provides a quick solution and easy calculation that may immediately solve a problem, users are nevertheless encouraged to follow the reference trail provided, accessed the original materials, and learn in details what a particular program is all about, how the results should be interpreted, and what the pitfalls are.

StatTools, in one form or another, has existed on the Internet for over 10 years. Although some of the more esoteric pages have received relatively few hits, on the whole StatTools has a few thousand hits per month, and its pages are linked to many other sites. StatTools has received much feedback from users, and this helped us to correct errors in the early stages, and improve our explanations and examples more recently. The authors hope that users will continue to offer feedbacks to these pages.

By Neil W. Henry
March, 2001
Virginia Commonwealth University