Richard Norrie and David Goodhart yesterday wrote an article reporting data from the new recorded hate crime statistics released by the Home Office, as well as from the Crime Survey of England and Wales (CSEW). In the article, they show that the data in the two are not in agreement: recorded hate crime has been increasingly rapidly, whilst the CSEW data shows a long-term decline in hate crime. Since the CSEW does not rely on individuals reporting hate crime, it is seen as an overall more accurate representation of the trends in England and Wales today. Their main inference is that the Brexit referendum has had no impact on actual hate crimes, and media outlets have been relying on dodgy data:
And we now have new evidence to suggest that, contrary to the widely held assumption, the Brexit blip in reported hate crime after the referendum did not lead to any more lasting increase in bigotry.
In one sense, they are right. As the authors of the Home Office report acknowledge (p.10), the police recorded statistics are not good at understanding long-term trends — the numbers are affected by improved recording/reporting — and so the long term trends reported in the CSEW towards fewer hate crimes is probably more accurate. This is good news, and is consistent with other evidence showing that racist or generally prejudiced attitudes are falling (as is attention to immigration), despite the narrative that they’re on the rise.
But as always, the devil is in the detail. There are two issues with Norrie and Goodhart’s analysis. The first is that the data are not directly comparable. I quote the report (p.9) below. Aside from general differences in sampling (the CSEW sample only those aged 16 and above, so exclude younger children), the big difference is that the CSEW does not include public order offences. And public order offences account for half the reported crime statistics. Whilst I agree with the overall trend they are highlighting, it is equally possible that whilst the crimes the CSEW does include are decreasing, public order offences (and homocides) are increasing. And it may be this increase which the reported hate crime statistics are picking up.
It also excludes homicides and crimes that are termed “victimless”, such as many public order offences, which account for over half of police recorded hate crime.
Ultimately, we can’t really know this without more work on the hate crime data. And I’m not sure that even that will help us answer the question.
Norrie and Goodhart’s analysis has a second issue. They use data averaged across years, since the sample in the CSEW is too small each year. Quoting again from the Home Office report:
The CSEW will therefore only give a very broad estimate of the level of hate crime in England and Wales across these four years and will not provide any information on whether the level of hate crime has changed in this period.
In other words, the CSEW data does not pick up fluctuations during these years, and so can only tell us about the very board and long-term trends. More specifically, then, the CSEW data does not tell us whether the referendum impacted on hate crimes, only that it does not seem to have had an impact on the overall, long-term trajectory of hate crimes.
The reported hate crime statistics — since they are recorded monthly and, during the referendum period, daily — can help us answer that question. The Home Office report answers it themselves:
While increases in hate crime over the last five years have been driven by improvements in crime recording by the police, there has been spikes in hate crime following certain events such as the EU Referendum and the terrorist attacks in 2017. (p.7)
Did the referendum cause an increase in hate crimes?
But I’d like to highlight my own analysis of the reported hate crime data to get at the crux of Norrie and Goodhart’s inference. This has two distinct parts: did the referendum cause an increase in hate crimes? And: is the increase long term? These have distinct answers.
First, let’s look at the daily data over the period of the referendum. The referendum signalled a clear increase in hate crimes in the weeks and month following the referendum result. More analysis of this — using a method called ‘time series intervention’ — shows that this is a statistically significant increase which led to approximately 31 more reported hate crimes a day. But, in line with their analysis, this decreases after this short-lived increase. The spikes at the end of the plot are from terror attacks.
We can see this again in the monthly data. There is a sudden spike, and then, two months later, we are back to normal. Once again, using time series intervention models, it shows the referendum led to a statistically significant increase in hate crimes to about 638 in a month. But this did not last.
So: did the referendum cause an increase in hate crimes? And: is the increase long term?
Yes, then no. The referendum did, based on all available evidence, cause an immediate increase in hate crimes the weeks and months following the referendum. But this did not last. It is hard to be clear due to terror attacks also increasing hate crimes, but it seems like there was no long-term impact on the time series from the referendum result. But this is shown in the reported hate crime statistics as well, something Norrie and Goodhart should also acknowledge.
Which leads me to, on the one hand, support their analysis: long-term trends are probably unaffected. But this does not mean the referendum had no impact. Just that it was short-lived and not picked up by the CSEW data, which is averaged over multiple years.
But there’s also caution in using the CSEW data. It doesn't include everybody, and it doesn’t include public order offences. Relying on this data set is potentially misleading as well. Ultimately, their main analysis is supported by the reported hate crime data they are also attacking.
 Note that this work is currently still under review at an academic journal.