ing in an area (Guerette and Bowers, 2009). Therefore, lower levels of crime risk in an area can be predicted by increasing guardianship. This should show the opposite effect to increasing number of suitable targets, which should in- crease crime risk.
Building on the work of Reynald (2009), that people’s willingness to act as guardians depends on their ability, their propensity, and whether individual level factors and their routine activities permit them to do so (see Chapter 2 for more detail), it can be assumed that not all individuals in that ambient population are guardians. However, it can be also assumed that some of them are. Therefore measuring the presence of active guardians in an area should show something distinct to measuring the total number of people in an area.
5.2
Measuring active guardianship using FMS
Chapter 2 detailed the great deal of innovation that has happened in recent years in the measurement of active guardianship. However measurements to date do not easily lend themselves to the mapping of micro-level dynamic fluc- tuation in active guardianship in a way to cover larger areas. To be able to map these intervening guardians specifically, it is possible to look to the wealth of information available in such crowdsourced data like FMS. It is possible to utilise FMS data to look into micro-level temporal fluctuations in active guardian- ship. Using FMS data as a proxy measure for when and where people are out and about monitoring their environments could serve as a measure for levels of active guardianship, specifically the presence of intervening guardians in neighbourhoods.
By representing active guardians with this data, micro-level temporal vari- ation in guardianship can be identified. These areas might exhibit temporary increases in crime risk as a result of a drop in guardianship. By using FMS data to identify these areas, such temporary increases could be mapped on a large spatial scope, such as across a large city. This encourages us to attempt to map active guardianship in place and time, using data that is already avail- able, and does not require the researcher to invest significant resources in data
collection.
Reporting an issue on FMS is a behavioural response to something that the person doing the reporting perceives to be an issue. The creation of the report requires that the concerned citizen be actively monitoring their environ- ment around the time of reporting, near the location of the incident, and actively taking ownership over the environment. This represents a sort of digitally en- gaged guardianship, where people use online platforms to monitor their areas, and lobby the local authority to address the issues which these guardians find problematic. Therefore these people are making behavioural responses to the environment (reporting), and as such represent intervening guardians who are in the highest tier on Reynalds’ guardianship scale; they are available, capable of supervising, and willing to actively intervene. Figure 5.1 illustrates where this new data fits in the guardianship literature.
Figure 5.1: Data for measuring active guardianship
Intervening Capable Available Invisible
Data available
Level of
guardianship
Preventative effect on crime
Population Estimates Reynald’s observational measure of guardianship in action Crowdsourced data on problem-reporting More Less
This crowdsourced data, whilst affected by external factors such as rep- resentativeness, discussed in detail in Chapter 4, can also serve as an open
5.2. Measuring active guardianship using FMS 109 source, readily available data set for estimating levels of active guardianship on a large spatial scale, complete also with fine-grain temporal data.
To test the claim that these data can be used to represent guardians within the ambient population, which is distinct from the number of people present who can be considered, on the whole, suitable targets, the following sections will test two hypotheses. First to determine whether FMS reporting is corre- lated with guardians who are willing to intervene, Section 5.3 will test whether neighbourhoods with higher levels of willingness to intervene measured with traditional survey approaches also have higher FMS reporting. After that, Sec- tion 5.4 will explore the relationship between within-day changes in levels of FMS reporting and a burglary. Burglary is chosen as it is a crime which is traditionally associated with daytime.
5.2.1
Analytical approach
Before moving on to results, a quick note about the analytical technique used in Sections 5.3 and 5.4 must be made.
The spatial unit of analysis used in this chapter is the neighbourhood level (see Chapter 3 for discussion on how neighbourhoods are defined throughout this thesis). Therefore levels of each variable are considered for each neigh- bourhood. As discussed in Section 3.6.2, there are approximately 600 house- holds in each neighbourhood and overall 4054 neighbourhoods across London. Due to the spatial nature of the data set, spatial lag regression models are used throughout. Spatial lag means that the dependent variable Y in place i is affected by the independent variables in both place i and nearby place j, which is the case in our data (Beck et al., 2006). By using a spatial lag model, the coefficient parameter reflects the spatial dependence inherent in our data, measuring the average influence of observations by their neighbouring obser- vations. This is important because spatial autocorrelation is strong with FMS data, indicated by a Moran’s I value of 0.766.
Therefore here I consider the estimation by means of maximum likelihood of a spatial regression model that includes a spatially lagged dependent vari- able. Unlike the traditional approach, which uses eigenvalues of the weights
matrix, this method is well suited to the estimation in situations with very large data sets (Anselin, 2004), and with over 4000 neighbourhoods, it was deemed appropriate for this dataset. Neighbouringness was calculated in GeoDa using a Queen contiguity weights construction (Anselin, 2003).
While it is tempting to focus on traditional measures, such as the R square, this is not appropriate in a spatial regression model (Anselin, 2004). The value listed in the spatial lag output is not a real R square, but a so-called pseudo-R square, which is not directly comparable with the measure given for Ordinary Least Square regression results. A better measure of fit is the Log-Likelihood, the result of a likelihood ratio test comparing the goodness of fit of two models, the Akaike information criterion (AIC), and the Schwarz criterion (SC) (Anselin, 2004). The higher the log-likelihood, the better the fit (high on the real line, so less negative is better). For the information criteria, the direction is opposite, and the lower the measure, the better the fit (Anselin, 2004).