ASSESSING THE EFFECTIVENESS OF DEER WARNING SIGNS THE KANSAS DEPARTMENT OF TRANSPORTATION April 2006

Kansas Deer and Transportation News Archive 



Final Report 
Prepared by 
Eric Meyer* 
University of Kansas 
A Report on Research Sponsored By 
THE KANSAS DEPARTMENT OF TRANSPORTATION 
TOPEKA, KANSAS 
and 
UNIVERSITY OF KANSAS CENTER FOR RESEARCH, INC. 
LAWRENCE, KANSAS 
April 2006


Deer-vehicle crashes are a concern across the country, especially in states like Kansas, where
most of the highway mileage is rural. In Kansas, the concern led to passage of state statute 32-
966. One result of this legislation was the initiation of this study to consider the possible causes
of deer-vehicle crashes and the implications with respect to effective mitigation. Of particular
interest was the effectiveness of deer warning signs. A broader need lies in the development of
better means of prioritizing segments for mitigative treatments, such as warning signs or fencing.
In Kansas, the most common countermeasure is the deer warning sign, even though its
effectiveness is suspect, and accident records have traditionally been used to identify locations
for installation. This study examined the effectiveness of deer warning signs by a comparison of
crash rates before and after sign installation. Deer-vehicle crashes were then studied with respect
to an array of potential predictor variables with the intent of developing a predictive model for
deer-vehicle crash rate that could be used to prioritize segments for mitigative action. Two
separate analysis techniques were employed: Principal Component Analysis (PCA) followed by
Multiple Linear Regressssion, and Logistic Regression. Principal Component Analysis (PCA) was
used to reduce colinearities prior to applying linear regression. A total of 45 predictor variable
were considered, 20 of which required field data collection. Data was collected for 123
segments spanning 15 counties in Kansas. One hundred one data points were used for model
calibration and 22 data points were used for model validation. 

Neither analysis approach was able to generate a model with sufficient predictive
capability to justify its use in prioritizing segments, but the analysis results provided some
helpful insight into the nature of deer-vehicle crashes. The insufficiency of the database to yield a predictive model is in itself a valuable realization. Models developed with lesser data
collection efforts must be held suspect unless they are supported by a strong validation effort.

These articles have been reviewed for relevance and content.  Some links to the original article may no longer be active.