By Shane Henry, The University of Melbourne
Arson is a major issue in Australian summers, particularly in Victoria, where there have been approximately 5,000 maliciously lit fires per year since 2000. Our project was aimed at statistically examining ignition patterns of arsonists, to determine if there are any links between locations of fire ignition points and variables such as temperature, rainfall, distance to road, etc.
The 6 week project consisted of two parts. In the first part, we looked at the distributions of distance from ignition point, and the nearest road, as well as the type of road, which could be ‘primary’, ‘secondary’ or ‘non-road’ (paths, tracks etc). There is a popular theory that says that arsonists tend to head just off secondary roads to light a fire, to reduce their chance of being caught. We found that there was some validity to this, however the secondary roads account for around 60% of the road network by kilometre. The more interesting finding was that there was many more ignitions per kilometre of road closer to the non-road network, and to a lesser extent, the primary network. This suggests that arsonists may tend to head of tracks or paths, as there is even less chance of being caught.
In the second part of the project, we used logistic regression models on 20km x 20km grid squares across Victoria, for the summers of 2008-09 to 2010-11. Each of the 4 models we looked at contained roughly 10-15 variables. We wanted to determine which variables had a significant effect on the probability of a fire being lit in a certain grid square on a certain day. In the main model we looked at, with ‘nearest distance to any road’ as a variable, without ‘density of roads’, we found that of the 11 variables we started with, only 4 were definitely statistically significant; daily maximum temperature, daily precipitation, population density and proportion of urban area in the square. Unfortunately due to time constraints, we could only compelte 2 of the 7 steps of the purposeful selection method described by Hosmer & Lemeshow, but future research would allow this to be completed to find other significant variables.
The AMSI VRS program provided me with a valuable insight into life as a researcher, and I enjoyed my time investigating this project. I would like to thank my supervisors Professor Peter Taylor, and special thanks to Nicholas Read for all his time and effort helping me during the project. I would also like to thank AMSI for providing me with this great research opportunity, and funding to complete the project.
References:
Hosmer Jnr., DW, Lemeshow, S & Sturdivant, RX 2013, Applied Logistic Regression 3rd Ed., Wiley, New Jersey, USA.
Shane Henry was one of the recipients of a 2015/16 AMSI Vacation Research Scholarship.