What 1980s weather models taught me about crime prediction

Weather prediction is an imperfect science - photo by Don Amaro from Madeira Islands, Portugal (upload by Herrick) CC BY 2.0 via Wikimedia Commons

by Gregory Saville

It was the summer of 1980. I was 24, a geography undergrad and junior member of a climate study team on a government contract, coding weather forecasts from old newspapers onto computer punch cards (you read that right – punch cards). Our research team was digging into two decades of Toronto forecasts. It was tedious work.

What we discovered about weather prediction wasn’t what the Canadian weather service wanted to hear. When our study was published, it caused quite the kafuffle.

Despite the rise of supercomputers and early weather models, short-term public forecasts—especially temperature—had not improved. In some seasons, they had gotten worse. The data showed it. And the implications were unsettling: more technology didn’t always mean more insight.

Years later, I saw that same blind spot in a different field: crime forecasting.

Today, predictive analytics dominate whole branches of criminology: heat maps, risk scores, and spatial data promise precision. Reviews of crime prediction programs abound.

 

Professor Robert Murdie (York University, Toronto), earned his PhD at University of Chicago, in social ecology. He led groundbreaking studies on gentrification, refugee social housing, and spatial dynamics in Toronto. In 1984 he coauthored with me one of the first Canadian studies to apply spatial analysis to motor vehicle theft and crime hotspots - photo by Phillip Kelly

PROFESSOR ROBERT MURDIE 

I learned about this work firsthand. In 1984, I conducted an early geographic crime analysis study that mapped auto theft in Peel Region. My coauthor, Professor Robert Murdie—a University of Chicago trained social ecologist—suggested using statistical mapping and something called positive regression residuals. We identified spatial patterns that pointed to where professional auto thieves or gangs might strike next.

A few years later, the Minneapolis hotspots experiment suggested some of those same things, eventually inspiring a generation of hotspot policing methods.

To me, this all traced back to that early weather prediction research. At the time, predictive modeling felt promising in crime analysis. But we had already learned something that complicated the narrative. 

Weather convection pattens - photo David Roth via Wikimedia Commons

Weather forecasting might appear chaotic and imprecise—especially compared to the confident claims in predictive policing. But it would be wrong to suggest it stood still. Since the 1980s, forecasting has made remarkable progress. Today, a 3-day forecast is as accurate as a 1-day forecast was 40 years ago. And 5- to 7-day forecasts, once nearly useless, now have real reliability. 

But in 1980, we were operating before Edward Lorenz and others brought the math of chaos and dynamical systems into mainstream science. That shift—which came in the years after our study—provided the conceptual tools to understand why nonlinear systems like weather, and crime, resist precision.

Many advances in crime forecasting came from colleagues I admire. Their commitment to evidence-based solutions isn’t in question.

But I keep returning to that weather study. Because there exists an uncomfortable truth: complex systems resist prediction. Whether forecasting tomorrow’s rainfall or next month’s burglary rate, the deeper you go, the more variables emerge: Social + contextual = unpredictable!

Hypothetical sample of crime prediction hotspot map

 

PREDICTING THE FUTURE

Like those early meteorological models, today’s crime systems often give the illusion of certainty without the annoyance of nuance. Even weather forecasters today embrace probabilities and uncertainty bands. But some criminologists still chase linear certainties in non-linear worlds.

During my master’s defense, I described this spatial analysis of auto theft. My supervisor asked about the predictive power of our model. I said, “If we had all the relevant variables… if our R² climbed to 0.95 and our prediction was more precise—that would be a dream.” 

My supervisor looked up and said:

“No… that would be a nightmare.”

He added, “Where would all that data come from? And what would we do with a prediction that accurate?”

At the time, it seemed like a preposterous question. But when I thought more carefully, I realized I didn’t have an answer. I’m not sure anyone did.

But now—amid AI, social sensing, and automated crime forecasts—we’re finally being forced to answer. And the question isn’t technical anymore. It’s ethical. It’s about the kind of society we want to build– the same paradox that triggered my fictional blog last month, The Pattern Room.

American Society of Criminology article in the journal Criminology - Each year, more studies emerge evaluating crime prediction 

HOW DOES SAFEGROWTH RESPOND?

SafeGrowth didn’t emerge as a rejection of prediction—but as a rebalancing. It asks different questions. While statistical tools estimate risk, SafeGrowth asks: Who defines risk? And who responds to it?

It treats crime not as a spatial constant to be mapped, but as a symptom—shaped by design, social ecology, and everyday experience. It doesn’t discard predictive tools. It grounds them. It recognizes that no algorithm—however calibrated—can replace local knowledge, trust, or the slow work of place stewardship.

Where predictive systems offer patterns, SafeGrowth invites relationships. Where algorithms flag where something might happen, local networks act to ensure it doesn’t.

It’s not either/or. It’s alignment: Can we design systems that serve the neighborhood, rather than just study it? It’s been four decades since that weather study. The punch cards are gone. Forecasts now come from satellites and superclusters.

Sometimes, real understanding of crime risks comes from walking and listening

And still, I think about those numbers—the ones that told us more power didn’t always mean more insight. As crime prediction tools grow more sophisticated, I wonder: have we remembered that?

Do we still hear the difference between precision and relevance—between what the model predicts and what the community needs? Because in the end, forecasting is only as valuable as what it allows us to change.

And real, lasting change doesn’t come from computation alone.

Sometimes it comes from walking the block or from listening. And sometimes… it starts with a stack of old newspapers and the realization that the model doesn’t know everything.

 

Can you avoid unintended consequences?

 

The story of unintended consequences at 
the Leaning Tower of Pisa - photo Arne Müseler 
by CC BY-SA 3.0 DE, Wikimedia Commons

by Larry Leach

When we work in large and small neighbourhoods, conversations about safety often focus on surface-level solutions. Residents may notice thefts or disturbances and jump to straightforward fixes: put up a fence, lock the gate, install cameras. While those steps can help, they rarely address the deeper roots of the problem. What if the real solution lies far beneath the surface, not just in reducing harm, but in preventing it altogether?

Jane Jacobs, in her seminal book The Death and Life of Great American Cities, wrote:

“Cities have the capability of providing something for everybody, only because, and only when, they are created by everybody.”

If the narrative is that people are stealing from our backyard to support their drug habit, does building a fence or locking your gate solve that? Probably not, unless that fence convinces that person to decide they need help to better their lives. This certainly doesn’t mean you shouldn’t build the fence, but to say it’s only a tiny part of keeping your property safe. 

Connecting with your community and establishing solid plans to work on the root causes of the problem will be the long-term solution.

Building a fence and locking a gate certainly will help for a short time, but what will change the behaviour of the would-be thief? Will they victimize your neighbour, who might be more vulnerable and unable to afford to build that fence? Will they break your fence to get in, causing you more costs? Either way, the fence is not likely to stop them from their anti-social behaviour. But what will?

The 1950s public housing "Pruitt Igoe" apartments in
St. Louis - constructed with good intentions
- Public Domain (US federal government)

By the end of the decade, crime and vacancy were 
so bad at Pruitt-Igoe, the entire project was emptied and demolished
- Public Domain (US federal government)

ENGAGEMENT IS THE MAGIC

This blog has featured many examples about how organizers use engagement tools to help trigger engagement and SafeGrowth examples from the Vancouver Strathcona Community Policing Centre.

The magic is to get involved with your community. Learn and understand your community. Who is in it? What are the local issues? Are there folks struggling? Who visits our community? These are all things that good SafeGrowth training and engagement can help to parse out.  At the end of the day, true safety includes the success of all who live, work, and visit in the community. One might call that “livability” as we do with third-generation CPTED.

In Malcolm Gladwell’s book, “Revenge of the Tipping Point”, he discusses a major unintended consequence. One community’s overstory was a monoculture of high-achievement attitudes that turned into a ripple of suicides among young people. The local society obviously didn’t intend this, but clearly didn’t factor in what kind of pressure that might potentially place on a young person to win at sports and achieve high marks academically. Dig deep and question your assumptions before deciding what your community’s story is.

A local example of this is a wonderful group in Calgary called “Brown Bagging for Calgary’s kids” that makes lunches for students every day in Calgary, who may not have one in school. The essence of this is the idea that kids can’t learn on an empty stomach. A noble and remarkable goal to achieve, but once you get to the kids that needed it (maybe only for a short time) and it’s available to all, do you demotivate parents’ responsibility to feed their own children? 

The unintended consequences of fences

I have blogged about other issues and topics for meaningful engagement, such as the Good Neighbour Agreement. 

Ultimately, without a research study, factors like parental demotivation and the effectiveness of Good Neighbourhood Agreements are difficult to assess. They may have unintended consequences that need further examination. 

DO FENCES MAKE GOOD NEIGHBOURS?

Lastly, it’s good to be aware that what looks like a good security measure can result in looking unapproachable and unwelcoming. The fence mentioned earlier is a great example of this. If a Community Hall puts up a fence, it may prevent the very thing a community hall should be welcoming. Designing a welcoming space comes with certain vulnerabilities, but if you can strike a balance between security and approachable you can hit that “sweet spot”, helping the community to feel more connected and livable. 

To hammer the point home, a local grocery store posts printed grainy photos on their front door of people that they allege stole something from their store. While this might seem like a good deterrent for criminals to the store management, what they miss is how it makes it’s customers feel as they enter the store. We have blogged before about fences and unintended consequences.

We advise in our classes that it is always wise to consider different perspectives when looking at crime prevention plans. SafeGrowth and CPTED have many examples showing how to make a safer, more livable, place, considering all the potential unintended consequences.