Interactive Analysis

Criminogenic Entropy: Tracking the Structural Drivers of Crime

Why did U.S. crime rise for decades, peak around 1991, fall through the 2010s, spike in 2020, and fall again? This tool treats crime as a product of structural conditions — and lets you see how changing those conditions moves the national crime outlook.

Scott M. Mourtgos & Ian T. Adams · University of South Carolina

What is criminogenic entropy? It is a single summary score for how many different ways a society is, at a given moment, set up to produce crime — the size of the “opportunity space.” It is built from 15 structural conditions (jobs, housing, mobility, trust, policing, and more), combined into one index that runs from high (volatile, many pathways to crime open) to low (stable, fewer pathways open).

The headline: this purely structural index — with no crime data fed into it — closely tracks the actual rise and fall of U.S. violent and property crime from 1970 to 2023. When the structure loosened, crime rose; when it tightened, crime fell.

Criminogenic entropy tracks crime, 1970–2023
All three series shown as standardized scores so their shapes can be compared. The shaded band is the 95% credible interval for entropy.
Read it this way: entropy is highest in the 1980s, then declines for three decades to its lowest point in 2023 — the same broad arc as crime.

Adjust the structural conditions — watch entropy and crime respond

Move any slider to change a structural condition. The tool recomputes criminogenic entropy and the crime rates that level of entropy implies, relative to a baseline year you choose. Sliders are colored by their role: orange = entropy-expanding (more of it opens the opportunity space), teal = entropy-compressing (more of it closes it).

How to read this: these are model-implied associations, not proven causes. The simulator is a transparent, theory-constrained approximation of the full Bayesian model, calibrated so its scale matches the paper’s published effect sizes. It shows the direction and rough magnitude a structural shift would imply — not a guarantee that changing one dial will deliver that exact crime change. See “About & Caveats.”

Jump to a notable year
Entropy-expanding conditions
Entropy-compressing conditions
Criminogenic entropy
Lowest (2023)Highest (1980s)
Violent crime
per 100,000
Property crime
per 100,000
Where your configuration lands
If these conditions held: projection to 2033
Your configuration (held constant) compared with the paper’s three structural scenarios.

What drives entropy — and how it changes across eras

An indicator’s contribution in a given year is its loading (how strongly it belongs to entropy) times its standardized value that year (how far above or below its own historical normal it sits). Bands above zero push entropy up; bands below zero pull it down.

The same condition can switch roles over time. Residential mobility, for example, was the largest entropy-expanding force in the 1970s–80s when Americans moved often — but as mobility fell far below its historical norm, it became the largest compressing force by the 2010s. The theory of each indicator never changes; the structural context does.

Signed contribution of each condition to criminogenic entropy, 1970–2023
Stacked above/below zero. Warm = entropy-expanding conditions, cool = entropy-compressing.
A single year, ranked
Drag to inspect which conditions were pushing entropy up or down in any year.

How well does it predict — and what it can’t do

Using structural conditions alone (no past crime fed in), the entropy model reconstructs the national crime series with high accuracy for both property and violent crime.

Crime typeAvg. prediction error (MAPE)Correlation with actualDirection called correctly
Which conditions carry the signal? Factor loadings (95% credible intervals)
Larger bars = a stronger role in shaping entropy. Conditions near zero contribute little to the national signal once the others are accounted for.
Held-out test: predicting 2024 before it was known
Crime typeActual 2024Predicted (95% interval)Inside interval?

Future scenarios, 2024–2033

Because entropy moves slowly and structurally, it can be projected forward. The paper considers three illustrative paths: Continued Compression (conditions keep tightening at the recent rate), Status Quo (conditions hold), and Structural Deterioration (conditions loosen at the pace of the 1970s). These are scenario illustrations for structural monitoring — not forecasts of what will happen.

Projected crime under three structural scenarios
Lines are scenario means; the shaded band is the 95% interval for the Status Quo path.
ScenarioCrime type20242033
Want to build your own path? The Explore (What-If) tab lets you set the structural conditions yourself and projects that configuration forward against these three scenarios.

About this dashboard

The idea. Criminogenic entropy is the number of feasible crime-producing configurations in a society — the size of the opportunity space for crime. Rather than asking which single factor “explains” crime, it asks whether many structural conditions jointly expand or compress that space. It is deliberately mechanism-plural: it does not claim one causal pathway, but that these conditions together index how readily crime is realized.

It is not a relabeling of social disorganization. It places disorganization alongside strain, anomie, and routine-activity conditions in one time-varying quantity, and it formalizes the difference between durable structural change and temporary shocks.

What this tool can and cannot tell you

How the What-If simulator works

Plain-language glossary

Criminogenic entropy
The size of the opportunity space for crime — how many pathways to crime are simultaneously viable.
Entropy-expanding / -compressing
Conditions that open up (or close down) crime opportunities — e.g. high mobility expands; strong homeownership compresses.
Loading
How strongly an indicator belongs to the entropy index. Big loading = strong signal.
Standardized value (z-score)
How far a condition sits above or below its own 1970–2023 average, in standard deviations.
Contribution
Loading × standardized value — an indicator’s actual push on entropy in a given year.
Bayesian dynamic factor model
The statistical method that extracts one slow-moving index from 15 indicators measured over time.
Out-of-sample test
Checking the model against a year it never saw (here, 2024).