Research · 2 min read
Last Year’s Hotspot May Still Matter—But It Is Not a Set-and-Forget Security Plan
A useful plan combines a persistent baseline with a refreshed recent-change layer instead of freezing one hotspot map.
A block lands in the top 10% for property crime, so it goes on the map and into the budget. Reasonable. But what are the odds it's still elevated a year later? Across 25 rolling windows of Seattle data: about two in three. History means something here — just a lot less than a permanent label.
Our analysis of 1,918 eligible blocks across 25 rolling 12-month windows, july 1, 2023–june 30, 2026.
How much should a current plan rely on a block that was elevated last year?
Across 1,918 eligible Seattle blocks in 25 rolling windows from July 2023 through June 2026, 63.4% of eligible top-decile memberships remained elevated 12 months later, while only 32.2% of ever-elevated blocks met the study's stability rule. Use history, but refresh it.
The numbers behind the answer
Selected measures only. Denominators and interpretation stay attached so the headline cannot stand alone.
63.4%
12-month membership retained
1,656 of 2,613 eligible top-decile membership comparisons remained elevated 12 months later.
32.2%
Ever-elevated blocks classed stable
125 of 388 ever-elevated blocks were elevated in at least 75% of their eligible windows.
46%
Median persistence
The median ever-elevated block appeared in the elevated set in 46% of eligible windows.
Hotspots can persist even while membership turns over
The tension sits in one pair of numbers. When a block held a top-decile spot in one twelve-month window, it kept that spot twelve months later 63.4% of the time — 1,656 of 2,613 comparisons. Recent history is clearly worth something.
Now flip the lens. Of the 388 blocks that were ever elevated during the study, only 125 — 32.2% — stayed elevated in at least three quarters of their eligible windows. Both numbers are true. One asks whether a specific membership survives a year; the other asks how reliably a block stays hot across the whole study.
Planning rule
Keep a persistent baseline, but refresh the recent-change layer.
63.4% retention and 32.2% stability answer different questions
The difference is the denominator. Retention counts 2,613 membership comparisons — a block can appear in several, and 957 of those memberships ended within the year. Stability counts 388 distinct blocks and grades each one on consistency.
The median ever-elevated block was elevated in 46% of its eligible windows: meaningfully present, far from a fixture. Neither percentage is a prediction. They're two honest ways to describe the same churn.
Evidence visual
Persistence among 388 ever-elevated blocks
Bars show the share of eligible rolling windows in which each block remained in the elevated group.
The label changes when the threshold changes
How much does the answer depend on the rule you pick? Plenty. Define "stable" as elevated in at least half of windows instead of three quarters and the stable share rises from 32.2% to 46.7%. Define "elevated" as top 5% or top 20% instead of top 10% and you get different sets again.
There's a subtler trap, too. To qualify for the study, a block needed at least 12 records across the full three years — 1,918 blocks made it. But only 524 met that bar in the first rolling window. Eligibility judged with future data is fine for description. Used for prediction, it leaks the future into the past.
Build a two-layer review instead of freezing one map
The working answer is two layers, not one frozen map. A baseline layer: has this area recurred over the long run? A change layer, refreshed on a stated cadence: did it recently enter, stay in, or exit the elevated set? The first layer stops recency overreaction. The second stops stale allocation.
Through it all, an "elevated" label is a prioritization tool — not an SPD classification, not a risk score for your property, not a promise about next year. It's a reason to look again, on a schedule, with fresh data.
- Keep a longer-run recurrence baseline.
- Refresh the latest rolling window on a stated cadence.
- Flag recent entries and exits separately.
- Show the count and eligibility rule behind the label.
- Recheck property conditions before changing coverage.
Questions property teams ask
Does top decile mean a block is unsafe?
No. It means the block was in the highest-count group under this study definition for a specific rolling window.
Does stable mean the block will remain elevated?
No. Stability describes past window membership. It is useful baseline context, not a prediction.
Why use rolling 12-month windows?
A full year reduces seasonal imbalance while monthly steps show entry, exit, and persistence more clearly than isolated calendar years.
How often should the analysis be refreshed?
The appropriate cadence depends on data maturity and the decision, but a current-change layer should be refreshed regularly and never substituted with one frozen historical map.
Why might a block enter or leave the top group?
Counts can change because of actual reported activity, reporting and publication patterns, ties, neighboring activity represented by a blurred label, or the moving 12-month window.
Our methods, limits, and sources
How we calculated this
This is original 911 Sentinel research — we gathered the records, ran every calculation below, and published the aggregate dataset.
We counted accepted Seattle property-offense records by published block in 25 rolling 12-month windows and tracked membership in each window's highest-count decile.
- We required at least 12 records across the full study period for the primary eligible set.
- We calculated the elevated cutoff separately in every rolling window, retaining ties.
- We compared eligible elevated memberships 12 months apart.
- We measured each ever-elevated block's persistence and repeated under alternate top-share and stability thresholds.
What this analysis cannot establish
- Top decile and stable are analytical definitions, not SPD labels.
- Full-period eligibility can introduce look-ahead selection; contemporaneous eligibility should be used before prediction.
- Ties can enlarge an elevated set beyond exactly 10%.
- Block labels are imprecise and low-record blocks are excluded.
- Record persistence is not causal evidence or a forecast of future victimization.
- No exposure denominator controls for property count, foot traffic, population, or activity.
Sources
The raw records come from the sources below; the study design, analysis, charts, and conclusions are our own.
Related questions and practical guides
New coverage — intake open
Pair the baseline with current property conditions
Use the public-data pattern to prioritize a review, then check access, occupancy, lighting, cameras, deliveries, and surrounding activity at the property today.