How We Track Crime in Trinidad & Tobago
Crime Hotspots exists because no single source in Trinidad & Tobago gives you a complete, up-to-date, independent picture of crime. Official statistics arrive years late. Mainstream media covers only what it chooses to cover. Political figures have an interest in the numbers looking a particular way. This platform was built to fill that gap.
Trusted by readers across the Caribbean
monthly readers
monthly impressions
incidents tracked
areas monitored
Why Our Count Is Higher Than Other Sources
Our murder count is consistently higher than tallies from any single external source. That gap is not an error — it reflects the difference in methodology.
Official stats are delayed
The Central Statistical Office (CSO) publishes crime figures that are still marked provisional for years 2021–2024 as of 2026. Annual reports routinely arrive 1–3 years after the fact.
Media doesn't cover everything
Many crimes are reported to police but never appear in a newspaper or TV broadcast. Crimes in less-covered communities, crimes without dramatic details, crimes that happen at the wrong time of day — all get missed.
Tallies aren't published consistently
Major outlets have increasingly stepped back from publishing running murder counts. No single public source maintains a daily independent tally.
We monitor multiple source layers simultaneously. Because we draw from more of the landscape than any single outlet, our count reflects a more complete picture — not a different one.
Where Our Data Comes From
Mainstream media — automated
RSS feeds from Trinidad & Tobago's primary news organisations are checked continuously throughout the day:
Trinidad Express · Newsday · Trinidad Guardian · CNC3 · TT Crime News
When a new article is detected, it enters our AI extraction pipeline automatically.
Community and social sources — monitored
High-traffic Facebook pages that report on crime in Trinidad often publish details before mainstream media — or publish incidents mainstream never covers at all:
DJ Sheriff · Ian Alleyne Network
These sources are monitored manually and submitted through a structured intake process. Where the same incident appears in both a media report and a community post, we keep one record — the more complete version.
Direct community reports — reviewed
Residents can submit anonymous reports of incidents they witnessed or know of. Every submission is reviewed before it enters the dataset. We verify it is specific and credible, does not identify individuals, and has not already been captured from another source.
One record per crime. If the same incident is reported by multiple sources, we record it once — using the most complete version available. We do not tag records by which source they came from, because the goal is the most accurate single count, not a source breakdown.
How We Process Data
AI Extraction
Each article or post is fed into Claude AI (Haiku), which reads the unstructured narrative and converts it into structured data: crime type, location, area, date of incident (not publication date), victim count, and source URL. The model is tuned specifically for T&T crime reporting patterns.
Human Review
High-severity records and any entry with ambiguous location or crime type data are reviewed against the source before publishing. Every record links back to its original article or post so the context can always be verified independently.
Deduplication
Semantic similarity checks run on every incoming record to catch duplicate reports of the same incident arriving through different sources. A record that matches an existing entry above a similarity threshold is discarded rather than added as a second count.
How We Count Crimes
Primary and Related Crime Types
A single incident can involve multiple crime types — a murder that also involved a shooting, for example. We record a primary crime type (Murder) and related crime types (Shooting). Both count toward totals. An incident with one primary and two related types contributes 3 to the overall crime count.
Victim Count Rule
Victim count applies only to the primary crime type. An incident with 2 murder victims contributes 2 to the murder count. Related crime types always count as 1 regardless of victims. This prevents double-counting across crime type totals.
Regional Risk Ranking
The dashboard ranks regions by weighted crime severity, not raw incident volume. Murder and kidnapping contribute more to a region's score than theft or burglary. Rankings are relative to the national total for the selected period and update automatically as new data is added.
How We Calculate People Affected
Raw crime counts tell you how many incidents occurred. They do not tell you how many people's lives were disrupted. Our weekly newsletter includes an estimated "people affected" figure that adds the immediate human cost beyond the direct victim — family members, partners, and household members whose lives are meaningfully changed by what happened.
What "Affected" Means
We define "affected" as people with a direct personal relationship to the victim — immediate family, cohabitants, and close household members. We do not claim to count community-level fear, workplace impact, or broader social disruption. The real number is almost certainly higher than what we publish.
Schema by Crime Type
Each crime type carries a secondary-affected multiplier applied per victim or per incident. The affected total is calculated from the primary crime type only — related crime types on the same incident are excluded to avoid double-counting (e.g. a murder that also carries Shooting as a related type counts the family once, not twice).
Per-victim crimes — multiplier applies to each victim
| Crime Type | Victim counted? | Secondary per victim | Total per victim |
|---|---|---|---|
| Murder | No (deceased) | +8 | 8 |
| Manslaughter | No (deceased) | +8 | 8 |
| Attempted Murder | Yes | +5 | 6 |
| Kidnapping | Yes | +8 | 9 |
| Sexual Assault | Yes | +4 | 5 |
| Shooting | Yes | +1 | 2 |
| Assault | Yes | +3 | 4 |
| Carjacking | Yes | +3 | 4 |
| Domestic Violence | Yes | +0 | 1 |
| Robbery | Yes | +2 | 3 |
| Extortion | Yes | +2 | 3 |
| Fraud | Yes | +1 | 2 |
Per-incident crimes — multiplier applies once per incident
| Crime Type | Secondary per incident | Total per incident | Note |
|---|---|---|---|
| Arson | +3 | 4 | Household + immediate neighbours |
| Burglary | +2 | 3 | Household members |
| Home Invasion | +0 | 1 | Context type — Burglary always implied; counted there |
| Theft | +1 | 2 | Lower household disruption |
| Seizures | +0 | 0 | Enforcement incident — excluded entirely |
Murder and Manslaughter. Because the victim does not survive, they are not counted as an "affected" survivor. Only the secondary figure (8 family members) is applied. This is based on academic literature reviews (Gale Academic / SciELO) that place the number of immediately bereaved family members per homicide at 7–10. We use a conservative 8.
Shooting carries a lower multiplier than other violent crimes. Shooting frequently appears as a related crime type alongside Murder or Attempted Murder. Applying a full secondary multiplier to both would count the same family twice. A conservative secondary of 1 is applied to avoid this inflation.
What we don't claim. No Caribbean-specific secondary victim research exists. Our multipliers use published academic figures where available and conservative estimates elsewhere. The real number of people affected by crime in Trinidad & Tobago is higher than what we publish — these figures represent a deliberate undercount, not an overcount.
What This Data Cannot Tell You
Being honest about limitations is part of being a credible source. Here is what our data cannot capture:
- —Unreported crimes. If an incident was never reported to police, never published in media, and no community member submitted it, it does not appear in our database.
- —Geographic coverage bias. Areas with less media attention — rural communities, less-connected neighborhoods — are likely underrepresented relative to areas that receive consistent coverage.
- —Crimes with no public trace. Some crime types — fraud, domestic violence, extortion — are less likely to surface in the sources we monitor regardless of how common they are.
Our data is more complete than any single source. It is not complete. The actual crime picture in Trinidad & Tobago is higher than what we show — just as it is higher than what official statistics show.
Privacy & the Public Record
Where a source article includes names, addresses, or vehicle details, we retain that information as part of the public record established by the original publisher. We do not generate new personally identifiable information, and we do not use crime data for individual targeting or commercial purposes. Every record links back to its source so users can read the original context in full.
Safety Tips
The Safety Tips section is grounded in incident patterns from our dataset — not generic advice. When our AI model identifies a preventable tactic in a crime report, it flags the incident for editorial review. Approved tips are published with a link back to the originating report.
Have questions about our methodology?
If you need more information or have specific questions about how we collect and verify data, check out our FAQ page.