Crime epidemic or never had it so good? Drilling into statistics is murder

Britain is in the grip of a crime epidemic, the likes of which we have never seen before. Knife crime. Stabbings. And if you're out after dark, make sure your will is written and posted before you close the front door.

Or how about the alternative reality of... we've never had it so good. Violent crime was down last year, down the year before. And this follows a decade of similar reductions.

Which is "true"? Well, both. And neither. The issue lies in the statistics: the way we measure "crime" and, more fundamentally, what we mean by crime. Though if you are looking to the popular press for explanation, you can be forgiven for not getting that.

As with crime, and pretty much everything else in statistics, measurement is rarely simple, interpretation never is. Crime stats though are a useful illustration of the issues.

Where are they pulling this data from?

Let's start with that rising epidemic of violence. When reporting on crime, as the Office for National Statistics (ONS) does, two statistical series are regularly referenced. The first, the Crime Survey for England and Wales (CSEW) published late last year, is a survey of households in England and Wales in respect of their experience of crime.

It is broad brush. The ONS admits it is not good at drilling down to detailed figures on rarer crimes such as murder or stabbings whose incidence is too low for accurate analysis. That is not the same, as some then imply, as the CSEW being less "accurate" than the other major set of crime figures out there. The Police Recorded Crime and Outcomes (PRC), published in April, captures millions of individual incidents. The CSEW captures trends.

These are two different things, providing two different and equally valuable insights into the overall picture. Unfortunately, the public response is not unlike that of the marketing director of one high street retailer, now defunct, who when presented with two different pictures of their customers – one based on attitudinal and behavioural data, the other economic and demographic – demanded to know "which was right". The idea that both could be correct was unthinkable.

As marketing, so crime. For many years both CSEW and PRC have been running in the same direction: downwards. But in the last two years they have diverged sharply. PRC shows violent crime up for the last two years, with the last 12 months alone showing a rise of 21 per cent. CSEW, by contrast, showed violent crime continuing to fall, down 5 per cent in 2017 compared to the previous year.

While accepting that the PRC stats provide useful insight, the ONS has said that they do not meet the required standard for designation as National Statistics. Of course they would, the cynics declare. After all, as a government body, they have an interest in spreading complacency. That, though, is to ignore the ONS track record across a wide range of issues, including crime, where they have a long history of publishing unpopular figures.

It ignores, too, a fundamental issue with PRC, which is that the PRC reports precisely what it claims to report: recorded crime. That is, reports of crime at the point at which someone deems a crime to have been committed and which the police accept is a crime. This is not the same thing as a crime pursued to a (successful) outcome through the criminal law. Matters rejected by the police – "no crime" – are not included. So PRC reflects changes in police policy, as well as public willingness to come forward and report crime.

The incidence for recorded sexual offences has increased greatly in recent years, but a constant debate, both in academic and law enforcement circles, is how far that represents a real increase, how far an increase in willingness to report offences, as well as increased police willingness to listen to victims.

Change criteria for recording crime, recorded crime figures change

One standout fact mostly ignored by press and public is that there have been major changes to the way in which crimes are recorded. Because in 2013/14, Her Majesty's Inspectorate of Constabulary took a look at how police forces were doing, and declared a massive under-recording of violent and sexual crimes. The police needed to do better.

Changes in protocols were put in place in 2015, designed to increase reporting of violent crime. And hey presto! In the two years since those changes were instituted, crime rates have "soared".

It would not be unreasonable to suggest that sensationalism plays some part. It's the old story that "good news is rarely news". PRC figures are quarterly, while CSEW figures are annual, which means four times as many opportunities for "bad news" on the crime front. Add, too, that players – from politicians to police chiefs opposed to cuts – have an interest in demonstrating a larger issue than is actually the case, and it is not surprising that the focus is on the negative picture.

What this illustrates is an issue that anyone who has ever been called upon to measure some critical quantity needs to bear in mind. Few of us, unless we work for the police, relevant government departments or interest groups, are likely to be called upon to generate statistics about crime. Many working in the tech sector will be asked, at some point, to create reports on some critical aspect of business performance.

Some questions – how many widgets were sold last month? – are straightforward and the way we answer is often contained in the wording of the question. Though even there is wiggle room. Do we mean how many we sold and weren't returned? Or just how many were sold, period? How do we account for bulk orders? Is a box of nails one sale or dozens? Which is more important, purchases? Or purchasers?

Similar questions turn up in the crime stats. Are we, for instance, talking crimes, criminals or victims? Which is a more meaningful representation of crime?

Then there are issues with how crime, or sales, have evolved over time. Because if data is heterogeneous – that is, populations, samples or results are based on different data sets or collection methods – comparisons are at best compromised, at worst useless. Is PRC data from 2017 really comparable to PRC data from 2013. Yes. Probably. But with serious caveats, rarely expressed in the press.

Are we getting more murderous?

Beyond that are questions more philosophical than factual. What do we mean when we talk about crime "getting worse". That is particularly the case when it comes to murder, another favourite topic for press and politicians – but less so for those concerned with analysing crime trends. In fact, even though discussion may be about "murder", the official UK figures record a combined tally for "homicide", which includes infanticide and manslaughter. Teasing out a pure figure for murder is not straightforward.

Then there's the issue of how we count mass homicide events, like Dr Shipman (218 victims). In recent years, the ONS has provided two sets of figures: including and excluding those events it considers anomalous, as well as a health warning that the year in which a homicide event gets counted is itself subject to reporting convention and is not necessarily the year it happened. What, though, is the basis for deciding an event, such as a terror attack, anomalous? And just because the ONS does not count it as such does not make it any the less significant for the victim.

If, for instance, it is eventually determined that the more than 400 patients who died after inappropriate prescribing of painkillers at the Gosport War Memorial Hospital in Hampshire count as homicide, then this will face the ONS with further challenges. Technically, they should be added to the tally in the year on which such a determination is made. But it does not require a degree in statistics to understand that adding 450 to an annual tally that normally fluctuates between 600 and 700 would be massively distorting.

Because murder is both anomaly (it is highly exceptional: around 600 to 700 instances a year compared to around 1.3 million reported instances of violence against the person) and it is quasi-random – accidental, almost – in when and whether it gets committed. That is, as Crown Prosecution Service guidelines make clear, murder happens when a person ends up dead and the perpetrator intended to kill or to inflict grievous bodily harm.

Do most murderers "intend to kill"? A good question, since most would claim they did not. What we do know is that the most common method of killing someone in the UK tends to be using a knife or sharp instrument. The recent rise in murder levels has taken place almost exclusively among men and, separately, the police have recorded a significant increase in knife crime.

In other words, we have grown marginally more murderous as a country over the last couple of years (but still less so than we were a decade or so ago). But the outcome (more murders) may simply reflect a growing tendency to carry – and use – a knife than any genuine rise in murderous intent among the wider population.

So what have we learnt? Crime may or may not be getting worse in the UK. But what remains constant is that in answering such a question we must always dig below the surface to understand not only the data that is being used to provide an answer – but also just what we mean when we ask the question. ®