THE INTEL ANALYST ACADEMY
Professional Intelligence Training
Printable Lesson
Crime Trend Analysis
The Intel Analyst Academy · Lesson Notes
A single snapshot of crime data tells you where trouble is today. A trend tells you where trouble is going tomorrow - and where it came from last week, last month, and last year. This lesson unpacks the difference between reacting to a spike and understanding a trend. You will learn the methods that separate the ephemeral blip from the structural shift, and you will learn why your favourite crime statistics are probably lying to you.
Imagine you arrive at a crime briefing in January. The analyst points at a map and declares that burglaries are up 40% this week compared to the weekly average. The room buzzes with concern. Resources are shuffled. A task force is formed. Then someone thinks to ask: what happened during this week last year? What about the week after Christmas, when everyone was still on holiday and houses sat empty? The question reveals the uncomfortable truth: the 40% spike was seasonal noise, not a genuine surge.
Crime does not occur uniformly across time. It clusters by hour, day, month, and season. Assaults peak on summer weekends when alcohol flows and temperatures rise. Domestic violence calls spike on Christmas Day and New Year's Eve. Burglaries follow school holidays and long weekends. Property crime often dips during extreme weather - criminals, like the rest of us, prefer not to work in a blizzard.
A snapshot that does not account for these temporal patterns is worse than useless: it is actively misleading. The analyst who reports a "surge" in domestic violence every December has not discovered a trend; they have simply discovered that the calendar has twenty-eight, twenty-nine, or thirty-one days.
Trends operate on multiple timescales, and confusing one for another is a classic analytical error. A short-term trend might last days or weeks - a gang conflict that spikes retaliatory shootings, or a holiday weekend that sees a predictable rise in DUI arrests. A medium-term trend spans months to a year or two - the displacement effect after a new police patrol strategy is implemented, or the lagged impact of an economic downturn. A long-term trend spans years or decades - the decades-long decline in U.S. violent crime since the 1990s, or the gradual shift from street-level drug markets to online narcotics distribution.
Always establish your baseline before declaring a trend. A minimum of 12 to 24 months of historical data is the floor for meaningful comparison. Anything less and you are looking at noise, not signal. Plot the data before you pontificate.
Reporting a crime trend based on two weeks of data is like declaring the start of summer because you had one warm Tuesday in March. Statistically ambitious; analytically bankrupt.
The analytical toolbox for trend detection is deep, but most analysts only reach for the simplest tools. Here we cover the essential methods that turn raw incident data into actionable trend intelligence.
A simple line chart of daily crime counts looks like the EEG of a patient having a seizure. Daily volatility obscures the underlying direction. Moving averages solve this by averaging data points over a fixed window - typically 7, 14, or 28 days - and plotting the smoothed result. A 7-day moving average eliminates day-of-week effects (because Mondays are always different from Saturdays). A 28-day moving average reveals monthly trends. When the moving average crosses a historical threshold, you have something worth briefing.
For analysts with statistical training, time series methods like ARIMA (AutoRegressive Integrated Moving Average) decompose crime data into three components: trend (the long-term direction), seasonality (the predictable cycles), and residuals (the genuinely unusual events). This decomposition is powerful because it isolates the signal you care about - the trend - from the seasonal noise that can masquerade as meaningful change. Most modern crime analysis platforms, including those integrated with COMPSTAT processes, use some form of time series decomposition under the hood.
Regression methods go beyond description to ask why a trend exists. Did the introduction of a new policing strategy cause the decline in street robberies, or was it the concurrent change in unemployment rates? Multiple regression can control for confounding variables, but it requires careful model specification and a healthy respect for the difference between correlation and causation. The analyst who mistakes one for the other will produce confident, wrong assessments.
COMPSTAT (Computer Statistics) revolutionised American policing in the 1990s by institutionalising trend analysis at the command level. Weekly COMPSTAT meetings force precinct commanders to explain crime spikes in their sectors using data, not anecdotes. The method combines statistical analysis with geographic mapping and accountability pressure. When it works, it drives rapid, targeted responses. When it fails - and it often fails - it incentivises data manipulation, under-reporting, and the classic "crime of the week" myopia that treats every blip as a crisis.
When using COMPSTAT-style analysis, apply the "three-week rule": do not treat a crime change as a trend until it has persisted for at least three consecutive reporting periods. This simple heuristic eliminates 80% of false alarms caused by random weekly variation.
COMPSTAT meetings are where data meets its match: a precinct commander with a creative definition of the word "declining." Always audit the numbers. Always.
Crime trends are not random. They follow repeatable patterns driven by human behaviour, environmental factors, and the adaptive responses of both criminals and law enforcement. Learning to recognise these patterns is what separates the trend-spotter from the trend-analyst.
The seasons write the first draft of any crime trend analysis. In temperate climates, warm weather brings people outdoors, increases social interaction, and raises the incidence of assault, robbery, and theft from vehicles. Cold weather pushes activity indoors and shifts crime toward burglary and domestic incidents. Holiday periods - Christmas, New Year, summer breaks - produce their own distinctive crime signatures. An analyst who does not seasonally adjust their data will repeatedly rediscover the same annual cycle and call it a finding.
When police crack down on crime in one area, the criminal activity does not simply disappear - it moves. Displacement is the tendency for crime to shift geographically, temporally, tactically, or target-wise in response to enforcement pressure. A successful drug bust at a street corner does not eliminate drug dealing; it moves it three blocks over, shifts it to a different time of day, or pushes it indoors. Trend analysts must account for displacement, or they will falsely attribute a decline in Sector A to good policing when the reality is simply that crime relocated to Sector B.
The flip side of displacement is diffusion of benefits - the phenomenon where crime reduction effects spread beyond the targeted area or crime type. A focused patrol initiative in a known hot spot may also reduce crime in adjacent areas, because the perception of increased enforcement deters potential offenders across a wider zone. This halo effect is real, measurable, and frequently underrepresented in trend assessments. When you see crime dropping in a comparison sector that received no additional resources, diffusion - not a coincidental parallel trend - may be the explanation.
Criminals are not consultants. When you squeeze one side of their operating environment, they do not submit a restructuring plan - they just move to the other side of town and keep working. Displacement is the oldest pattern in the book, yet analysts forget it with alarming regularity.
The hardest lesson in trend analysis is that the data itself can lie. Not maliciously - data has no intent - but because the mechanisms that produce crime data are shaped by human decisions, institutional policies, and structural conditions that have nothing to do with actual criminal behaviour.
A crime trend that appears in the data may reflect nothing more than a change in how crime is recorded. When a police department adopts a new records management system, implements mandatory reporting for certain offences, or changes its classification guidelines, reported crime rates can shift dramatically overnight - with zero change in actual victimisation. The most infamous example is the "crime decline" that followed the adoption of electronic field reporting in some departments. The decline was real in the data; in the real world, officers were simply spending less time filling out forms and incidents were being recorded inconsistently during the transition.
Policy decisions at the department, city, or state level can create apparent crime trends that have nothing to do with underlying criminality. A decision to decriminalise minor drug possession will produce a dramatic drop in drug arrests - which will appear in the data as a crime trend. A change in domestic violence reporting protocols will produce a spike, as previously uncounted incidents enter the official record. The analyst who interprets these policy-driven data shifts as changes in criminal behaviour will provide confident, policy-relevant advice that is completely wrong.
Crime rates are almost always expressed per capita, but the denominator - population - moves slowly and often invisibly. A city that experiences rapid population growth will see its crime counts rise even if the per-capita rate is stable or declining. Conversely, a shrinking city may see falling crime counts that mask a rising victimisation rate. The analyst who does not check the census data is not analysing trends; they are analysing arithmetic artefacts.
The base rate fallacy is the tendency to ignore general statistical probabilities in favour of specific, vivid information. In crime trend analysis, this manifests as the belief that a 50% increase in a rare crime type is more meaningful than a 5% increase in a common one. A neighbourhood that goes from zero homicides to one has experienced a mathematically infinite increase. A neighbourhood that goes from 200 auto thefts to 210 has experienced a 5% increase. Which trend actually matters more? The base rate fallacy tempts us to obsess over the first while ignoring the second. Good trend analysis defeats this instinct by always contextualising change against the base prevalence of the crime.
When you see a dramatic percentage change, your first question should be: what was the starting number? A 100% increase from 2 to 4 incidents is a mathematical curiosity, not a crime wave. Always report both the absolute and relative change, and let your consumer decide which matters.
The base rate fallacy is why local news hyperventilates about a single kidnapping in a town of 50,000 but says nothing about the 300 car break-ins that happened the same week. Don't let your analysis imitate local news.
You are assigned to analyse crime data for Midtown District. The dataset shows a 37% increase in reported aggravated assaults between Q1 and Q2. Before you brief your supervisor, your task is to investigate three alternative explanations:
Write a one-paragraph assessment that either confirms the trend as genuine or explains which of these distorting factors is at play. Include your level of confidence and what additional data you would need to be more certain.
Continue your training
This lesson is part of The Intel Analyst Academy — professional intelligence analysis training built for analysts. Explore the full course library, structured learning paths, and practical tools at theintelanalystacademy.com.
