Revising, refining and hopefully improving

I mentioned that the methodology included revising and refining, and hopefully improving. Today I have included a graph and map that I think is an improvement on that I posted yesterday. I have used the same downloaded data but instead of using the the average response time for each grid square on the x axis I have used the the percentage of immediate incidents that have response times of greater of 12 minutes for each grid square. I have used the same clusters and colours. The police target is to respond to 75% of immediate calls in 12 minutes. Prior to the implementation of Neighbourhood Policing the target was 85%. When I led a response team in Ealing and Acton about seven years ago we achieved the 85% without difficulty but then that was the priority and there was very little community policing................leading to the Reassurance Gap.............the rest is history.

The overall response rate is for this borough for this performance indicator was 78% for 2009 - well within target. However the blue squares show areas which did not achieve this target in 2009, that is over 25% outside 12 minutes.

At this stage I probably should mention the problem of scale, the arbitrariness of grid squares and the problem with low numbers.

• Analysis is carried out on different scales. Even though the underlying data is exactly the same the visualisation of that data will produce different results depending on the scale used. This dataset is a good example because grid squares that are on the opposite ends of the graph are geographically next to each other. This means that doubling the scale will amalgamate these squares producing an aggregate result. Halving the scale (if the dataset will allow -not in this case) will potentially give more definition but it will aggregate the problem of low numbers that is part of this dataset (see below).
• The location of the grid square frame is arbitrary. If the grid frame is moved in any direction the results will be slightly different especially if a grid gains or loses a high street.
• The problem of low numbers is a simple concept to do with probability best illustrated by an example. When you flip a properly weighted coin you know there is a 50% chance that it will land heads up and a 50% chance it will land tails up. But if you only flip the coin say three times it is not that unusual for the three results being all heads or all tails. The probability of this happening 4, 5, 6 etc in a row dramatically reduces with each flip. Therefore relying on low numbers to make inferences or hypotheses is unwise.

The problem of scale and grid are unavoidable but the problem of low numbers is simple, just get rid of them from the analysis. I have not done so up to now because of a public perception of the police reason. I am a member of the Automobile Association (AA) for which I pay about £150 a year for a premier service, anything they offer - homestart, breakdown, roadside relay, etc. I should get. I have been a member since 1988. In the early days when I had older and mechanically unreliable cars I use to call them two or three times a year and I was generally pleased with service I got because I was getting good value for money. I have called the AA only twice in the last ten years (I think) on both occasions I have been unsatisfied with the response I got. On the second occasion they refused to come out for a Homestart because they were overstretched even though I probably paid over £1000 into their coffers for that one call. (That's weird I feel an emotional release now I got that out in the open!). Coming back to the point, low numbers are very important to the police. These are the calls they really should be getting right.

This where I come back to the confidence in the police measure. What is of concern is if the blue squares are residential locations then the very people that they wish to provide a premium service to (see the end of my previous post) are the ones who are apparently getting second best.