Graph of the number of calls to the police graded
as Immediate Response in 2009 for grid square
plotted against the average Response time
for that grid square
As I mentioned in a previous post I have started carrying out analysis on another London Borough. I am doing this with the full knowledge and co-operation of the senior police managers on that borough. They have given me permission to publish my analysis on my blog and I have permission to identify the borough. I decided not to identify the borough for a number of reasons. The main reason is I do not want my analysis to be quoted out of context and used to criticize the police. Nothing within this analysis is intended to be critical of the police, though the main reason why the police are facilitating this research is to learn ways of becoming even more efficient and effective. Another reason is that this is experimental research, in a sense that it is new and previously untested. I am publishing it as I do it. This means that there is a high risk that I may make mistakes and that I may come to the wrong conclusions, which I will subsequently have to revise. I do not want this to cause any difficulties to the police. Lastly just a quick note to say that the data I download from the police computer systems do not contain information which can identify individuals, therefore the data published in map or graph form cannot either.
The graph that I have included at the top of this post is from the Metropolitan Police Service (MPS) incident logging computer system, Computer Aided Dispatch CAD. When a call comes in from the public, usually by telephone, it is graded depending on its assessed urgency. The most urgent calls are graded as to be dealt with immediately and are known as "I" calls. There is a target time of police arriving at the scene of the incident the call relates to within 12 minutes. The types of calls which are graded "I" are those where there a possibility the police can provide assistance to a person or people in danger and/or prevent or detect a crime by rapid attendance.
The CAD system is now almost 30 years old, meaning that it was designed at the same time as the Sinclair ZX81 (my first computer with a memory of 16kb!) and in my opinion one of the best designed MPS computer systems. There have obviously been numerous upgrades but there are certain design features that remain. One of these is how locations are recorded. Without going into too much detail, locations are only accurate to a grid square that is 250 meters by 250 metres. The incidents that occur within the grid are recorded at the central point of the grid. As it is imperative for calls that occur within a borough to be assigned to the right borough police to deal with there is a detailed gazetteer within CAD that can assign all locations to the right borough that operates separately to the grid system. This means that even if a grid square straddles a borough boundary, as many do, the calls will be accurately split between the two or more boroughs which share that grid square. This level of sophistication does not occur at sub-borough level meaning that it is difficult to totally accurately plot calls at ward and sub-ward level.
My analysis therefore uses the same grid squares as CAD. The graph shows each grid square as a point. The location of the point on the x and y axis is determined by the number of I calls that occurred with in that grid square in 2009 on y axis and the average response time for those calls on the x axis. There are over 400 grid squares that are contained within the borough or straddle the boundary. 317 had 1 or more "I" calls in 2009.
This graph provides a simple analysis of how these two variables relate to each other. In this dataset there is almost perfect non correlation between the two datasets. If the higher the number of calls a grid square had the higher the average response time the correlation would be positive and the equation that work these things out would have produced a result of 1. If the higher the number of calls the lower the average response rate the correlation would have been negative, that is -1. In this case the result is very close to zero.
Surprisingly, even though the dataset looks complex and disorganised it can be split quite easily into four clusters that have police operational relevance. That will be the subject of my next post.