The Present, Measured: are strange crimes really becoming more common?
Blog · the-present-measured

The Present, Measured: are strange crimes really becoming more common?

Lesson from:

One feeling, three claims, and what a lab that reads 17,754 news sites can and cannot say about it

I have a feeling many people share. In the last years there seem to be more strange crimes: violent acts with no history behind them, no motive anyone can name, done by people nobody had noticed before.

I run a lab on one machine that collects news in 79 languages, keeps a library of about 76,000 works and records what the world’s press writes about, month by month. So I did not want to keep the feeling. I wanted to measure it.

This is the first post of a new kind on this blog. I call it The Present, Measured. The rule is simple: take one thing people feel strongly about, put it to everything the lab holds, and write down what the measurement shows, what it cannot show, and why.

Contents · 9 sections

One feeling, three claims #

The first thing the machine did was refuse to answer my question as I asked it. “There are more strange crimes” is three claims folded into one, and they have different answers.

  1. There are more violent crimes.
  2. A larger share of them is strange and unexplained.
  3. It feels as if there are more.

A measurement can only be made on one of them at a time.

What I measured with #

Source Reaches back Size
Articles collected by the lab 15 days 4.31 million articles from 17,754 sites
A public index of world news by theme (GDELT) 1 year 115 million records
The same project’s event counts by country and type 3 years daily totals
Published statistics, as summarised in an offline encyclopedia decades homicide by country to 2022
The library about 76,000 works

One limit decides everything else. My own articles go back fifteen days. Anything I say about years comes from thinner indexes built by others, or from published statistics.

Claim 1: more violent crime? The numbers say no #

Over centuries. Homicide in Europe fell from somewhere between 20 and 70 per 100,000 people a year in the late Middle Ages to about 1 today. A rise from the late 1950s to the early 1990s is described as a short deviation in a long decline.

Over two decades, worldwide. The published estimates run 7.6 per 100,000 in 2004, 6.9 in 2010, 6.2 in 2012, 6.1 in 2017 and 5.61 in 2022. That is a modest fall, with rises in Central America and the Caribbean.

Over three years, in news events. Assaults on persons were 1.65% of all recorded events in 2024, 1.53% in 2025 and 1.50% so far in 2026.

Over one year, in news coverage. The share of all coverage that concerns killing:

Dec 2025 Mar 2026 Jun 2026 Oct 2026
Killing 13.0% 13.5% 12.3% 12.5%
General crime 8.2% 7.3% 7.4% 7.9%
Kidnapping 1.0% 1.0% 0.9% 1.0%

Flat, or slightly down. I use shares and not counts for a reason: the index itself recorded about a tenth fewer events in 2025 than in 2024. A falling count of anything may only be the index shrinking.

One number moved the other way. Threats rose from 1.68% of events to 2.03%. There is no more violence in the record, but there is more threatening language.

Claim 2: more unexplained crime? I cannot tell, and neither can anyone with data like mine #

This is the real question, and nothing I hold can answer it.

No category in the data separates a killing in a robbery from a killing with no intelligible motive, by someone with no record, against a stranger. “Killing” is one label. A fall in total violence is fully compatible with a rise in one rare kind, because the rare kind is a tiny part of the total.

The research literature has names for this kind: sudden mass assault by a previously unremarkable person, the killing of one’s own family, grievance-fuelled violence by a lone actor, imitation after a widely covered case. My library has one strong clinical book on it and little else.

I could have filled this gap with a theory. There are several good ones: isolation, untreated illness, new drugs, imitation, online grievance, economic humiliation. Each predicts something different that could be checked. None of them is checked here, so I will not present any as a finding.

There is a seventh possibility I find the most interesting. A crime is “unexplained” on the day it is reported. Many are explained at trial, months later, in coverage a fraction the size of the original. If motives reach the public less often than they used to, more crimes would read as unexplained while nothing about the crimes had changed.

Claim 3: it feels like more. Almost certainly true #

This one needs no change in the crimes at all.

  • We judge frequency by how easily examples come to mind. A vivid, unexplained act is the most memorable kind of event there is.
  • The pool grew from a country to the world. Twenty years ago a strange killing abroad reached me days later as a paragraph, if at all. Now it arrives within the hour, with video, from every country at once. My own system reads 17,754 sites. I have built myself a machine for seeing the whole world’s worst day, every day.
  • The unexplained case is the one that gets shared. An explained crime is a closed story. An unexplained one stays open.

So the feeling is real, and it is a fact about attention before it is a fact about crime.

The rules that came out of it #

Each of these cost somebody a wrong conclusion, me included.

  1. Count shares, not totals.
  2. Ask “per how many people?” I hold no population table, so I compared no countries.
  3. Know the base rate before you interpret a case.
  4. Separate the event from its coverage. One killing can produce 500 articles.
  5. Check how far back the source goes before asking about change.
  6. Read the sample before trusting the measure. In the same week, one of my own quality measures scored 19 good out of 30. When the cases were actually read, 24 of the 30 turned out harmful or pointless.
  7. Keep only explanations that could embarrass you if they were false.

What I am missing #

A measurement that finds no gaps is an advertisement. These are mine.

  • No official statistics held locally, and nothing per head of population.
  • Fifteen days of my own articles. A trend needs years.
  • An English-speaking window. A third of my articles are in English, and half carry no country.
  • The founding books are not on my shelf: Durkheim on suicide, the classic studies of anomie and of social trust, the long historical homicide series, the work on moral panics. I have the psychiatry and the statistics, not the sociology.

What I will build #

One thing would turn the feeling into a number: a register of cases.

Each case is defined in writing before counting, counted once per event however many outlets cover it, read by hand at first, expressed per million people, and then followed to its trial to see whether it was ever explained.

After a year that gives a count. After three, a trend. And it would test the seventh possibility directly, which nothing I know of does.

Where the method comes from #

From the library, with the reference each work carries in my own catalogue. The reference is a fingerprint, not a link: it lets me find the exact copy I read, and it does not point anywhere.

  • Kahneman, Thinking, Fast and Slow, on judging frequency by ease of recall · lib:2147d579d8
  • Malmquist, Homicide: A Psychiatric Perspective · lib:ea8c535e27
  • Pinker, The Better Angels of Our Nature, for the long decline · lib:1179bbd137
  • Bregman, Humankind, as the counter-reading · lib:61880c8e10
  • Mills, Clinician’s Guide to Violence Risk Assessment, on how modestly individual violence can be predicted · lib:33c52c5356
  • Bonita, Beaglehole and Kjellström, Basic Epidemiology, for rates and denominators · lib:b54cc2dbd1
  • Kahneman, Sibony and Sunstein, Noise, on scoring forecasts · lib:8daff55106
  • Taleb, The Black Swan, on what the past does not contain · lib:6cd895c80a

The homicide figures are the published estimates of the UN Office on Drugs and Crime and earlier studies, as summarised in the encyclopedia articles listed with this post. I read the summaries, not the original studies. The seven explanations are from the research literature as I know it, not from my data.

The next post in this series takes another feeling and does the same to it.