Search This Blog

Saturday 19 February 2011

Crime Statistics, are the Statisticians to blame for lack of confidence?


Sir Michael Scholar, the UK's National Statistician, pictured above, has been asked by the Home Secretary to carry out a review of crime statistics with the purpose of increasing public confidence in them. He has published a questionaire that can be found here that anyone can complete and submit. Below are my thoughts. I probably have not started off very well by putting some of the blame for low confidence in the crime statistics on the statisticians themselves, but I try to tell it as I see it. Having heard Sir Michael speak at the Royal Statistical Society I am sure he and his staff will be receptive to my views.

Q1: Responsibility for the publication of crime statistics is to be moved out of the Home Office. Who should now assume this responsibility to increase public trust in the crime statistics?

It should be the responsibility of a body independent of government but reports to the Home Office in a similar way that the Bank of England is independent of the Treasury but reports to the Chancellor of the Exchequer.

It would a mistake for it to be given to the UK Statistics Authority as I think that one of the major problems that needs to addressed is the fact that it is obvious that the statisticians do not understand the data fully, given its legal and procedural subtleties and have difficulty in producing reports that the policy makers or the public properly understand.

There is far more to crime statistics than just statistics, a major part of it is understanding crime and policing; which are the domains of criminology and police science. With the increasing use of maps to communicate the statistics to the public the discipline of Geographic Information Science is becoming more and more important.

Q2: Is there also a case for transferring responsibility for the management and/or compilation of data collected from the British Crime Survey and the police ? If so, where?

The National Audit Office should be seriously considered for police recorded crime. Even though the data is not financial per se, the skill sets of auditors are very compatible. One of the main challenges to the accuracy and integrity of police recorded crime statistics is that they are used as primary measures of police performance. The procedures for managing and collecting crime statistics to ensure uniformity within and between police forces needs to be overseen by people who have the forensic skills to identify deliberate under recording. As crime reduction and clear-up rates are directly related to individual police officers promotion prospects and pay the auditors should also have sanctions that they can enforce on individuals.

The British Crime Survey (BCS) should be totally independent of the police recorded crime statistics and pressure from the government. It has been a mistake to have one official publication trying to link police recorded crime with the BCS because it makes it look that the two sets of data can be compared directly. In most cases there are reasons why they cannot (due to business and youth crimes not being included for instance) the only notable exception is residential burglaries (but the fact that the BCS uses a rolling 12 month period and the police statistics use a fixed 12 month period affects this and the others crimes). As far more data is collected by the BCS than are published in the resulting dual publication it has the feel that the BCS data is being used to put a political spin on the police recorded crime data. This is at the hub of the need for independence. As important is the long term funding of the BCS so there is not a pressure to please the funding master/mistress to gain continued support.

The linking of different crime data sets should be done independently by academics and those putting them to practical use.

Another solution is to make the two data sets directly comparable which is preferable – see the 4th paragraph of the answer to Q5.

Q3: Currently, the Home Secretary determines what is recorded by the police as a crime and approves the Home Office Counting Rules for crime and statutory data requirements from the police. Should this continue or would public trust in the statistics be enhanced if this responsibility moved elsewhere? If so, where and why?

The Home Office should determine police priorities and ensure that bureaucracy is minimised; these are levers for doing this. The agency managing the collection of the police recorded data, my suggestion the National Audit Office, should be required to comment on whether the rules and requirements properly reflect the level and nature crime in the country.

Q4: The Terms of Reference for the review asks for consideration of the current definitions of crime. Do you have any comments?

Crimes in crime statistics always have a legal definition and sometimes in addition a descriptive label. For instance in law there is no offence of burglary dwelling, snatch, armed robbery, it is burglary, theft and robbery (with firearms offences). The labelling in the main allows offences against vehicles to be distinguished and with a bit of delving those offences against people; this should be made clearer. Equally important are those crimes that occurred in a private place or a public place. Under the present labelling it is often not possible to make any inferences in this regard. I think this is vital to understanding the crime problem and police performance.

There are ten Home Office (HO) crime types. The one that leads to the most misunderstanding, confusion and distrust is HO crime type 1 – violence against the person. It is not helped by the fact that statistics that are derived from it are referred as ‘violent crime’ by Home Office publications and politicians. The public order offence of violent disorder (that usually includes violence against the person but not necessarily) which is not included in HO crime type1, shows that violent crime in law applies to criminal damage as well as assaults. This aside, robberies, sexual assaults, aggravated burglaries, kidnapping and arson endangering life which are specifically violence against the person offences are not included in HO crime type 1. The inclusion of common assault (section 39 of the Criminal Justice Act 1988) in the sub-type of ’without injury’ is misleading because there almost always is an injury but not of a permanent nature; only assaults without battery should be included in this sub-type.

If the category of violent crime is important to the public and politicians, which I believe it rightly is, then all crimes that are violent should be included in that category with sub-categories as necessary. If this means double counting of certain crimes because they also meet the criteria of another category, this should not be a problem as long as it is properly explained.

Q5: It has been said that the crime statistics provide a partial picture. What, if any, are the main gaps in Home Office crime statistics that you feel should be addressed as a priority?

Official Crime Statistics have taken advantage of police computerisation of crime records in the collection of data but not in the scope of what data is collected.

Police accountability and crime prevention would be aided by counts of characteristics (age, gender, ethnicity, geodemographic group for instance) of victims of crime and suspected offenders broken down by crime types and offences. Advances have been made in providing location data about crimes through the medium of maps on the Internet, the accountability and crime prevention advantage could be further enhanced by the provision of data of when (date, day, hour) offences had been committed and/or discovered.

The dark figure of unreported and unrecorded crimes is very important to assessing police performance and confidence in police. I would argue the best proxy measure of confidence in police and police efficiency is the size of that dark figure which is known to vary from offence to offence. The BCS was originally set-up to shed light on this dark figure but as I have mentioned in my answer to Q2 the lack of direct comparability of the two data sets affects the confidence in the accuracy of the estimates produced. A reliable figure can be produced of unreported and reported BCS crimes but the figure for reported and unreported police recorded crime is less reliable. This makes the important figure of those crimes reported to the police but not recorded by the police also unreliable.

The solution to this is not to change the methodology of the BCS but to extract only those crimes that are covered by BCS from the police crime computers for comparison purposes. This is feasible given the data contained within each police crime record and the modern searching power of computers. If the characteristics of victims was also included then very useful assessments of police engagement with different communities would result and provide a good proxy indicator of the level of confidence different communities have in the police. The accurate assessment of the nature of the dark figure for different offences would provide a good measure of police effectiveness and efficiency. This would balance police recorded crime reduction targets that can be achieved through the public feeling that it is not worthwhile reporting crimes to the police.

The Reassurance Gap that led police to realign resources from response policing to neighbourhood policing arose because the public perceived that the crime problem was getting worse even though the official crime figures were showing year on year improvements. The analysis of this included an assessment that one of the reasons for this was that the official crime figures did not match the experiences of people. Drawing on Incivility Theory from the USA and the newly developed Signal Crime Perspective in the UK it is generally agreed that the observable nature of crime and disorder rather than officially collected crime figures influence people’s perception of crime and therefore their level of fear of crime. The all pervasive low level disorder (antisocial behaviour or incivilities) has as much influence, if not more, than the observation of serious crime on people’s perception. Police recorded crime, because of the way it is collected and legally categorised will never be totally in tune with people’s perceptions. My research includes analysing police incident data to see if the data can be used to produce maps that more closely match people’s observations of crime and disorder. The importance of this is that if policy makers and police have a view of the policing problem in a neighbourhood that is not matched by the communities living and frequenting there then there will be a mismatch of priorities and expectations. Research has shown that trust is based on good engagement and communication, crime statistics and maps based on them can either alienate or engage. Police incident data is being used to map anti-social behaviour but it should be used much more widely to reflect the nature and levels of violent crime. Police incident data has the added value of having good temporal attributes and shows details of the police response and priorities, and also the priorities and engagement of the public.

Q6: What are the most important considerations for trustworthy crime statistics?

The most important consideration for trustworthy crime statistics is that they are understood. The Home Office Counting rules, legal definitions, Home Office Crime Types, especially in relationship to violent crime are all barriers to proper understanding of police recorded crime. The intricacies of methodology and statistical techniques are barriers to properly understanding, evaluating and therefore trusting the BCS statistics.

The police incident data set is in everyday language, it reflects what people observe and it is easy to explain the origins and validity of the data. It cannot replace the police recorded crime statistics or the BCS but it enables the public to know and understand what the police in their local area are dealing with on a day to day basis. My evaluation of police incident data shows that police recorded crime probably accounts for fewer than 50% of their workload of crime and disorder related. This means that crime statistics would be considerably enhanced as a performance measure if police incident data were included.

Q7: What do you consider to be the main strengths of crime statistics?

The main strength of crime statistics especially when applied to a local level through mapping is that they are extremely popular. This provides an excellent communication and engagement tool that should be used positively and honestly.

Police need to exploit this popularity by including police activity data within their maps to reassure the public and demonstrate their worth to society that is far more than what is revealed in the present crime statistics.

Q8: Do you have any other views you wish to feed into this review?

The spatio-temporal characteristics of offences are extremely important to crime prevention as are the characteristics of victim and offenders for crime prevention and public engagement. This points to an integrated system of maps using police recorded crime, police crime system, police incident management system and BCS as source data. My PhD research involves designing such a system with spatio-temporal clustering and classification at its heart.

Saturday 12 February 2011

CAD incidents - validation of the k means classification method


London police violent and acqusitive crime incident in 2009 numbers each day
 This post is really quite interesting if you have following my blog.

I have carried out a simple analysis to test three important aspects of my research;
  1. How good is the SPSS 17 k means classification method
  2. Does the Metropolitan Police Service CAD data contain rhythms and cycles that reflect the lives of the people of London and  the way it is policed.
  3. What is the best CAD data to use to detect these rhythms and cycles.
From previous multiple bivariant correlations of the temporal nature of the number of different class or types CAD incidents I know that CAD incident class 1 "Violence against person" negatively correlates with the acquisitive crime incidents. I also know that there were subtle differences in the patterns of the total number of these incidents and those that had been graded as "I" for immediate - emergency response. The graph above shows lines for the number of incidents that fall into those four categories.

I carried out a k means classification using those four variables with the 365 days of the year as cases (using raw unstandardised data). I asked for seven groups to see if the k means classification could split the days into the right weekdays.

I find the result shown in the graph above quite exciting. I hope you can see why. The classification recognises a clear difference between Saturdays and Sundays and week days. The week days have been subdivided in two groups. I have been through all the dates which appear to have allocated to the wrong group and there is a very good explanation for each. For instance each Monday that was allocated to the typically Sunday group (group1) were Bank Holiday Mondays. Every other "misallocation" had a weather or holiday related explanation.

I have tried the classification with standardised data and additional and fewer variables but these four variables appear to me to produce the best results. 

So I now have even more confidence in the k means classification method, I am convinced I using a good set of data and I am sure that analysing the variations in violent and acquisitive crime incidents in the context of the police response is worthwhile.

Thursday 10 February 2011

.kmz for London and Birmingham crime and disorder


I have created .kmz files for the classification shown in the last post. They can be found here

Wednesday 9 February 2011

Comparing Crime and Disorder in London and (greater) Birmingham

What I am doing here is creating a classification by adding West Midlands police neighbourhoods (mostly Wards) to the MPS wards. The first map is of the West Midlands police area - Birmingham is in the centre, Coventry to the east and Wolverhampton to the west. I am grateful to Andy Brumwell, from West Midlands police for providing me with the shapefile.

Monday 7 February 2011

Google Map crime and disorder .kmz

I have created a .kmz file to be loaded into Google Earth to create the map above. It can be found here. Enjoy.

London Wards December 2010 crime and disorder clustered, classified and mapped

Over the weekend I have been trying to improve on my the map in my last post from a London perspective. London was clustered in a group that I classified as having robbery and burglary problems due to where the cluster centres were for that group. In fact London does have robbery problem according to the figures that were used in the K-means clustering method but the burglary figures showed that London had a burglary problem that was about average. It was the small force of Bedfordshire that was clustered with London that does have a big burglary problem and a also a robbery problem that pushed the centre of the cluster in the direction of burglary.

I have used the same methodology as in the last post to create a classification for London shown above but without dividing numbers by police officers and staff numbers. I have excluded three Wards, Heathrow Villages, St James and West End for two reasons.
  •  St James and West End Wards are not shown in the published tables, they are split into eight sub-wards with no information about their locations.
  • Heathrow Village is Heathrow Airport that clustered uniquely unsurprisingly, St James and West End always cluster together with a far higher crime rate than any other wards. Removing these gives a better scale to the other Wards.
I have been experimenting using the December 2010 crime statistics shown on the official Metropolitan Police website to split up the other crime category into, criminal damage, drugs, theft, etc. The problem with doing this is that it decreases the weighting of the burglary, robbery, etc crimes and ASB in relation to the other crime category. To me the choice of restricted crime categories and ASB reflects the importance of these crimes are to the police and the public. By keeping to those categories and weighting them equally in my clustering I do not have justify anything. Soon as I start making my own selections I have to justify my choices.

Back to the map. In broad terms it shows what I expected. There are a couple of surprises,
  • the location of the boroughs showing robbery problems, though having checked these do include personal and business robberies; my previous analysis has been exclusively on personal robbery.
  • the variation in vehicle crime numbers.
It has to be borne in mind that December 2010 was an unusual month with the heavy snowfalls. I need to analyse few more month to see how stable this classification is.

Friday 4 February 2011

Mapping and Classifying variations Crime and Disorder in forces in England and Wales in December 2010 or variations in recording practices?

In my last post I said that there are fundamental problems with the latest UK crime mapping site. I do not like to be negative about things that people have worked very hard to produce so I created the map above to illustrate  the problems and benefits of this initiative.

Firstly how did I create the map?
  1. I downloaded the neighbourhood data for all the forces (unfortunately the Sussex spreadsheet was empty) and created a force total of the five categories of crime and one category of anti-social behaviour for December 2010.
  2. I attempted to compensate for the different sizes of the forces by dividing the counts in the six categories by the latest published total police officer and staff numbers for each force. This has its problems (and benefits which I will argue elsewhere) but it is better in my opinion than dividing by resident population.
  3. I then wanted to cluster the data giving equal weight to each category (see my previous posts regarding clustering) so I standardized all the data by transforming them into Z scores. This shows the standard deviation either side of the mean creating positive and negative figures.
  4. I then decided the best probable number of clusters by using the Ward's Hierarchical method in SPSS 17 and choose 6. I then used the K means clustering method with the six clusters.
  5. The K means method allows a better understanding of what factors has influenced the clustering, allowing a classification of the clusters to be reliably undertaken. This classification is shown under the map above with a table showing the scores for the cluster centres.
  6. Now this is where I need help. I cannot find a shapefile for police forces in England and Wales. Edina has a shapefile for police force basic command units for England and a seperate one for Wales. I had to do a bit of editing to the .dbf file to show my classification at force level using ArcGIS.
Some very interesting groupings have resulted and vast differences in the recorded level of crimes and ASB has become apparant.

But what are the data I am classifying and mapping here?

To be honest you I do not know, I could find no documentation to explain. Now I am assuming that the crimes are police recorded crime in Home Office Crime Types but it would not surprise me to find out that there have been problems in ensuring that every force is submitting exactly the same types of crime. The chances of that being uniformly the same is a lot higher than the Anti-Social Behaviour figures, which I am assuming are from the police incident recording database and would not surprise me if they differ significantly in content from force to force.

So that is the number one fundamental problem - what is being counted, how is it counted and is it the same through out the country?

Wednesday 2 February 2011

New crime mapping site - impressive but with fundamental problems

The new  UK Internet Crime Mapping site was launched on Monday. It has received so many hits, reportedly 18 million an hour in the first few hours that I have had to wait until this morning before I managed to connect to the site.

I am torn between being impressed and dismayed, between thinking this is a leap in the right direction to thinking that the people who have designed this have not a clue what they are doing; between applauding the liberating of police data to castigating the fact that data without context and provenance cannot be treated as information.

First the positives.
  • Innovation and development in public services require monetary investment and risks to be taken. It requires people in power to say `yes' to ideas rather than the safe 'no'. It is the way improvements and advances are made.
  • Even though there are many problems with this site I think it should be seen leap in the right direction because it signals the acceptance of an underlying principle that the public have a right to know, in detail, what police are doing in their area through medium of police collected data.
  • Connected with that is treating the public as grown up people who are capable of dealing with the harsh realities of life. This should  lead to an understanding what the police know about the policing problems in the area thus enabling a more informed partnership between police and the public to tackle those problems.
  • The visualisation of data is new, interesting and works smoothly at various levels of resolution. Technically it is interesting; it appears to work on a grid system of squares of I estimate 50 metres by 50 metres. The point shown at the highest resolution is on a road within that grid square closest to the centre (some grid squares seem to have more than one point at the highest resolution, I think this where two distinct post codes or street name can be identified - interesting algorithm).  Everything georeferenced to a location within that grid square is shown at the point or points. The visualisation is of circles with the lowest resolution showing a circle with about a 5,000 metre diameter. As you zoom in and out the circles seamlessly change in size as the grid squares are subtracted or added. Different crimes types and antisocial behaviour counts are shown for the month of December 2010. The difficulties of counting across police borders appear to have been solved. There are links to Safer Neighbourhood Teams, with information about crime appeals, crime prevention, police Ward meetings and Neighbourhood Watch Schemes - all impressive.



I will leave the negatives which are fundamental and numerous until next time.