Near repeat analysis

The near-repeat phenomenon was first discussed and published in 2003 by Townsley and colleagues (Townsley, Homel, & Chaseling, 2003), and soon after by Bowers and Johnson (Bowers & Johnson, 2004; Shane D. Johnson & Bowers, 2004). It is the realization that when a location in the target of a crime (such as a burglary for example), there are two interesting features that are common. First, the same location is at an elevated risk of further burglary for a certain amount of time. This is repeat victimization. Second, nearby locations to the first burglary are also at a heightened chance of being a crime target for a limited number of weeks. This second feature is the near repeat phenomenon. It is the risk to nearby locations after an initial crime event. 

In the nearly two decades since the near repeat hypothesis was suggested, the phenomenon has be identified in a variety of crime types. Not just burglary (S.D. Johnson et al., 2007; Moreto, Piza, & Caplan, 2014), but also street robberies (Haberman & Ratcliffe, 2012), vehicle theft (Piza & Carter, 2018), shootings (Ratcliffe & Rengert, 2008), and even insurgent attacks on coalition forces in Iraq (Townsley, Johnson, & Ratcliffe, 2008). It seems to be a fairly ubiquitous feature of spatial crime patterns. 

Detecting the near repeat phenomenon

There are a number of free software options for detecting the near repeat phenomenon. All of them use the Knox test to compare the spatio-temporal relationship in space and time between all events. The pattern expected under the null hypothesis (where no near repeat pattern exists) is estimated by simulating the absence of a pattern. This usually done by randomizing the date between events, using a process called Monte Carlo simulation. If the last couple of sentences had your head spinning, don’t worry – there are free software programs that can do all of the hard work for you. There are a variety of programs that do it, but the following three options show that you can explore your own near repeat pattern as a Python script, an R package, and a standalone executable program that can be downloaded to your computer, called the Near Repeat Calculator.

 
The NearRepeat R package

In 2019 Dr Wouter Steenbeek published the R package ‘NearRepeat’. The program, as well as support materials can be accessed here. The website also include information about installing the program into R and details regarding the parameter choices. It requires some understanding of the R statistical language. 

The NearRepeat Python script

In collaboration with Dr Steenbeek, Dr. Toby Davies published a similar Python script to undertake the same function. The code as well as useful documentation demonstrating the implementation can be found here. 

The Near Repeat Calculator

The Near Repeat Calculator was developed over a decade ago as a standalone executable program to estimate the near repeat hypothesis. In the development of their programs, Drs. Steenbeek and Davies identified discrepancies in the Near Repeat Calculator that existed in version 1.3 and earlier. These discrepancies are detailed here. The current version has corrected the discrepancies and adjusted some other aspects of the software so that the Near Repeat Calculator and its output is more in line with the R and Python scripts, enabling analysts to validate findings across platforms. It also provide a simple language interpretation of the analysis if you are new to near repeat analysis.