Methods for improving regression analysis for skewed continuous or counted responses

Standard inference procedures for regression analysis make assumptions that are rarely satisfied in practice. Adjustments must be made to insure the validity of statistical inference. These adjustments, known for many years, are used routinely by some health researchers but not by others. We review...

Full description

Saved in:
Bibliographic Details
Published in:Annual review of public health 2007-01, Vol.28 (1), p.95-111
Main Authors: AFIFI, Abdelmonem A, KOTLERMAN, Jenny B, ETTNER, Susan L, COWAN, Marie
Format: Article
Language:English
Subjects:
Biological and medical sciences
Data Interpretation, Statistical
Epidemiology
General aspects
Health Care Costs
Hospitalization - economics
Humans
Linear Models
Medical sciences
Models, Statistical
Poisson Distribution
Prospective Studies
Public health. Hygiene
Public health. Hygiene-occupational medicine
Regression Analysis
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Standard inference procedures for regression analysis make assumptions that are rarely satisfied in practice. Adjustments must be made to insure the validity of statistical inference. These adjustments, known for many years, are used routinely by some health researchers but not by others. We review some of these methods and give an example of their use in a health services study for a continuous and a count outcome. For the continuous outcome, we describe re-transformation using the smear factor, accounting for missing cases via multiple imputation and attrition weights and improving results with bootstrap methods. For the count outcome, we describe zero inflated Poisson and negative binomial models and the two-part model to account for overabundance of zero values. Recent advances in computing and software development have produced user-friendly computer programs that enable the data analyst to improve prediction and inference based on regression analysis.
ISSN:0163-7525
1545-2093
DOI:10.1146/annurev.publhealth.28.082206.094100