cats with stats: safety in the east

a comparison of crime rates

it’s been a while since we had a good, old fashioned “cats with stats” around here and seems like high time we remedied that, so here is some data that happened to strike me:

michael arouet, whose data and analysis i quite like, posted this yesterday:

being a data-driven sort of feline, this seemed a bit of an unfair comparison to me as it pits an urban area against a whole country and as such has the potential to embed a great deal of bias. seeking to eliminate this potential objection to the data and to make a more apples to apples comparison, i first pulled data for paris and warsaw. i also added a couple of additional categories (assault and rape) as these seemed like core quality of life and safety sort of offenses.

after running that analysis, the point still stands that paris appears far more dangerous than comparable polish cities.

the gap is no longer as large, but still remains highly significant.

warsaw apprears noticably safer than paris (this, and all other crime data, is from 2023 eurostat ICCS data)

this same relationship holds when comparing the whole of france to the whole of poland. depending upon the category, crime rates are 2-7X higher in france.

this got me wondering about how this might compare to some other locations. as can be seen, london is considerably more dangerous than paris.

and we can pull this for a large number of cities and see this result emerge in similar fashion. i added the largest city (by population) from a number of other countries as well.

the results continue to be highly favorable for warsaw which looks, at the margin, slightly safer even than zurich, a city generally regarded as “pretty fricking safe.”

one might quibble a little bit about the homicide rate in warsaw, but it’s worth realizing how low all these values are. san diego, hardly a hotbed of living dangerously, is more like 4 or 5 per 100k, so 1.15 in warsaw is nothing to lose sleep over. by american standards, there is very little murder in europe (or, at least, murder that is reported. some argue about this and i have no high conviction view on whether and to what extent this is an issue.)

this same relationship holds at national levels so it is not a purely urban phenomenon

based on anecdotal evidence, i suspect these variances may have widened since 2023, but the data are, as of yet, incomplete and fragmentary so i do not really have any ability to speak definitively.

i calculated an overall crime composite from this data (eurostat did not have one) by setting the median city in each category as 100, scaling each city relative to median and then taking the simple average of the 4 categories. this is, admittedly, a made up metric, but it’s one that seems pretty reasonable and representative.

100 would be median crime, lower is safer, higher is less safe.

let’s call this “crime composite index.” (CCI)

• warsaw (52.3)

• zurich (59.9)

• madrid (79.3)

• paris (96.7)

• berlin (103.1)

• amsterdam (108.3)

• stockholm (110.5)

• london (120.9)

• brussels (121.2)

being the sort of cat that simply cannot resist poking the pachyderm in the parlor, i then went on to address the issue that i suspect a fair few readers are beginning to wonder about.

here’s the data on immigrants as a % of the population in these cities.

i then regressed this against the CCI. the results are striking.

the correlation (R) and r2 here are quite compelling and the p value (0.0327) is already past stat sig (less than 0.05) but even this is probably not telling the full story.

the interesting outlier here is zurich: high immigration, low crime.

one might be tempted to say “see, high levels of immigration are compatible with low crime! it’s systems, not immigration!”

this is partly correct, but seemingly mostly wrong.

consider:

an obvious question thus presents itself: is all immigration equal?

eg “so, where are these immigrants from?”

and this looks to have high explanatory value as a factor.

here are the national origins of the top 5 immigrant groups in each city:

notice anyhting?

the swiss immigration is nearly all from europe.

and that looks if not fully dispositive, at least highly provocative.

to grant maximum latitude and steelman the case, i ran the regression again with OCCI against the percentage of the population of each city made up of immigrants from any non-european source.

as can be seen, this removes zurich as an outlier. r2 soars to 0.868 (an astonishingly high figure for such a potentially multifactorial cause/effect relationship) and the p value drops to an eye straining 0.0003 impliying that this outcome has an only 1 in 3,333 chance of occurring by chance.

obviously, we have some imputed and estimated values here, but i did my best to keep them solid and reasonable at each step and not to embed bias. these relationships broadly hold for each crime category as well and so it does not appear to be an artifact of OCCI calculation.

i’m sure some folks are not going to like the conclusions that arise from this, but in the end, the data is the data and don’t shoot the analyst.

if folks find fault with the methodology, please point it out. i’m just getting my paws into some of this data and so perhaps there are subtleties i’m missing or perhaps i made some sort of math or sort error. alternatively, if you have ideas about how to extend or enhance this analysis, let me know. i will make every effort to keep this tight, fair, and balanced.

as ever, thank you for shopping: