Abstract
The R package blm provides functions for fitting a family of additive regression models to binary data. The included models are the binomial linear model, in which all covariates have additive effects, and the linear-expit (lexpit) model, which allows some covariates to have additive effects and other covariates to have logisitc effects. Additive binomial regression is a model of event probability, and the coefficients of linear terms estimate covariate-adjusted risk differences. Thus, in contrast to logistic regression, additive bino- mial regression puts focus on absolute risk and risk differences. In this paper, we give an overview of the methodology we have developed to fit the binomial linear and lexpit models to binary outcomes from cohort and population-based case-control studies. We illustrate the blm package's methods for additive model estimation, diagnostics, and in- ference with risk association analyses of a bladder cancer nested case-control study in the NIH-AARP Diet and Health Study.
Original language | English (US) |
---|---|
Journal | Journal of Statistical Software |
Volume | 54 |
Issue number | 1 |
DOIs | |
State | Published - 2013 |
Keywords
- Absolute risk
- Binary outcome
- Constrained optimization
- Logistic regression
- Risk difference
ASJC Scopus subject areas
- Software
- Statistics and Probability
- Statistics, Probability and Uncertainty