|
Activity Number:
|
253
- SPEED: Government Statistics, Health Policy, and Marketing
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, July 31, 2017 : 3:05 PM to 3:50 PM
|
|
Sponsor:
|
Health Policy Statistics Section
|
|
Abstract #325326
|
|
|
Title:
|
A Case Study in Adaptive LASSO Logistic Regression: Factors Related to Cyclist Death When Drivers Are Distracted
|
|
Author(s):
|
Lysbeth Floden* and Patrick Anthony O'Connor and Melanie Bell
|
|
Companies:
|
University of Arizona and University of Arizona and University of Arizona
|
|
Keywords:
|
Adaptive LASSO ;
Machine learning ;
Variable selection ;
Penalized regression ;
FARS data
|
|
Abstract:
|
Penalized applications of the generalized linear model (GLMs) are growing in popularity because of their ability to simultaneously perform model selection while estimating parameters. One of these methods, the Adaptive Least Absolute Shrinkage and Selection Operator (aLASSO) has been shown to outperform the popular LASSO approach when the signal-to-noise ratio is high. aLASSO imposes a differential penalty on the coefficients: small penalty on large coefficients and larger penalties on small coefficients. In this paper we demonstrate a practical application of the aLASSO for logistic regression. Specifically, we use Fatality Analysis Reporting System (FARS) data from 2010 to 2014 to examine factors related to cyclist death when drivers are distracted. We look at 2 outcomes: drivers who are districted, and drivers who are distracted by an electronic device. We illustrate a step-by-step approach to using aLASSO with logistic regression by: specifying the high-dimensional model that includes potential predictors, using an external cross-validation to choose the tuning parameter, and selecting the appropriate model. We compare the models to those obtained by using conventional LASSA.
|
Authors who are presenting talks have a * after their name.
