Why you can't identify changes in crime by comparing this month to last month – a new post by me on the @TheSRAOrg blog: https://the-sra.org.uk/SRA/Blog/whyyoucantidentifychangesincrimebycomparingthismonthtolastmonth.aspx

And a thread …

Lots of people in the police, academia and the media need to know if crime has gone up or down, but there are at least five reasons why it's just not possible to reliably tell using binary (month-to-month or year-on-year) comparisons

Using Atlanta PD data for some examples …

1. Binary comparisons throw away too much information: a 25% drop in homicides in 2017 looks impressive on its own ('great, whatever police are doing works!'), but not when you include all the available information

Next time someone mentions a month-to-month or year-on-year change in crime, remember that a 25% year-on-year drop in crime might be part of any of these longer-term patterns:

2. Binary comparisons ignore any long-term trend in crime. An analyst using year-on-year comparison to work out if a burglary initiative was effective would get unreliable results unless they factored in that burglary has been decreasing by 13% per year for the past decade

3. Months are a terrible unit of analysis, not only because they're different lengths (March is 11% longer than Feb) but because they don't have the same number of each weekday, meaning one month can have four weekends (when many crimes are most frequent) and the next have five