If you run a Kingston shop and are reviewing next month's security budget, you need more than a national headline. Ask what has happened to the risks facing your staff and customers, over a comparable period, and what evidence would justify a change. A fall in murders is important news. You cannot copy that percentage into an insurance or staffing calculation.
TL;DR
- The minister's 6 October statement reported 407 murders against 534 in the corresponding 2025 period. The rounded decline is 24%.
- A year-to-date total, a quarter and a single month describe different windows. Keep each count with its dates.
- Police divisions, parishes and individual business locations are different units. National counts do not measure a shop's probability of a break-in.
- Before using AI to summarise a briefing, prepare a source sheet and check every number and comparison yourself.
What the October statement says
The Jamaica Information Service report published on 7 October 2026 describes National Security and Peace Minister Horace Chang's statement to Parliament the day before. It gives 407 murders as at 6 October, compared with 534 murders over the corresponding period in 2025. The difference is 127; dividing that difference by 534 gives a decline of about 23.8%, which rounds to the reported 24%.
The same account reports 123 murders during July–September and 36 murders during September. These are nested periods, so adding them to the year-to-date count would count some deaths more than once. The report is a dated account of the minister's statement, not a live feed.
Your first task as an analyst is to preserve that context. Write the event date and publication date separately. Give your reader the comparison period beside the number. If a later police release revises the count, retain both versions and tell your colleagues which one the briefing uses.
Definitions matter as much as arithmetic
The Jamaica Constabulary Force statistics page warns that figures can change as investigations continue. When checked on 13 October, its heading described a report ending on 19 September. That is a different stated cut-off from the October ministerial statement. If you combine values from the two pages, you risk presenting them to your team as one unchanged series.
You should be able to follow the table you cite. If its heading, columns or downloaded values do not line up, ask the publisher to clarify them before you brief colleagues. Do not repair the table by guessing which date or label was intended. This article does not use that page to claim a current parish trend.
UNODC's intentional-homicide metadata shows why labels need care. Its definition combines a death caused by another person, an intention to kill or seriously injure, and unlawfulness. Its victim measure counts people, while other tables concern suspects or convicted persons. You cannot swap those units merely because each table concerns homicide.
The metadata uses a calendar-year base period and explains that investigation stages and national recording practices can affect the data. A running national murder count and an international annual homicide series therefore need a documented comparison. The international definition does not decide how any particular Jamaican case should be classified.
If you calculate a rate, you need a suitable population denominator. UNODC describes the total rate as victims divided by population, multiplied by 100,000. Check that your count and population cover the same place and suitable years. This article does not turn a partial-year count into a forecast for December or compare it with an annual rate from another year.
Shadow Literacy in a data briefing
Adrian Dunkley's published FAQ describes Shadow Literacy as the judgement involved in using AI well, including recognising an answer that sounds convincing but needs checking. Applying that idea to a crime-data briefing is this publisher's editorial extension.
An assistant may copy your number correctly and still mislead your reader. It might call a police division a parish, remove the reporting date, or turn a fall in recorded offences into an identical fall in an individual firm's risk. You would miss those errors if you checked only spelling and arithmetic. You need to understand what the figures describe.
Start with a short source sheet before you ask for a summary. Give each statistic a row with its original publisher, publication date, reporting period, geography, unit and any revision note. Add a column for the claim you can support with that number. If you leave a field blank, ask a colleague to help find the missing detail rather than invite the model to fill it.
Keep a copy of the public source you used. A link is useful, but a changing web page may show different figures next month. A dated copy lets another person reproduce the comparison and explain why an earlier briefing used a different value. Record corrections without erasing the history.
What a Jamaican business can reasonably infer
A restaurant, hotel and hardware shop may face different risks even on the same street. Your opening hours, deliveries, cash handling, physical access and reporting practices affect the evidence available. A national murder trend supplies context. It cannot determine the right guard rota, opening hours, insurance cover or premium for any of those firms.
Separate the offence types in your records too. Murder, robbery, break-in, theft and a false alarm are not interchangeable events. A business log that groups them under “incidents” may be useful for administration, but its total cannot be compared directly with the national murder percentage. Tell your team which question you need to answer before choosing the comparison.
Show your reader the counts behind a percentage. In a hypothetical log, an increase from two incidents to four is a 100% rise and an increase of two events. Report both to your colleagues. Neither description establishes a stable trend on its own. There is no universal incident-count threshold that makes a percentage reliable for every business or purpose.
Check how your business has changed too. A shop open for more days, or a hotel serving more guests, has changed the opportunity for events to occur. Ask your staff whether they improved the log, replaced an alarm or began reporting near misses. A change in a log can reflect several things. It does not prove a change in the underlying danger or identify its cause.
Suppose your manager asks you to brief a Hanover retailer on Friday. Before you open an AI tool, ask what they need to decide and who could be harmed by a mistake. If they want to change closing time, a national murder count alone will not answer their question. They may need advice about staff travel, recent local reports and the layout of the premises. Write those missing inputs into the briefing. Your job is to make the limits clear enough that the manager can obtain the right help.
A five-figure check for this month
The following is a proposed source-checking exercise for a business analyst. The dates are suggested planning dates, not statutory deadlines. It tests a briefing's traceability; it is not a security assessment or a rule for changing insurance.
- Choose five figures by 23 October. Take them from a draft public-data briefing. Write down the decision each figure is meant to inform. Exclude personal incident reports, names, addresses and details that could identify victims or staff.
- Build the source sheet. Locate the original public release for each figure and complete the date, period, unit and geography fields. Count how many of your five figures have a complete trail. Mark an unresolved definition clearly rather than treating it as a minor footnote.
- Check without the assistant first. Recalculate the stated changes and write a short summary yourself. Ask a colleague to challenge the comparisons. Keep the original count beside the percentage so a large change on a small base is visible.
- Test an approved AI tool, if useful. Provide only the public, checked material. Ask it for a brief summary with dates and source links, and instruct it to leave unresolved fields unresolved. Compare every factual sentence with the source sheet. Record omissions, unsupported claims and review time.
- Repeat by 31 October. Correct the briefing process and try five fresh public figures of similar complexity. Compare the number of traceable claims and the time required to check them. A better result tells you about this exercise. You still need a person to review each new briefing.
If a proposed operational change depends on an unresolved figure, refer it to the responsible manager and qualified security or insurance adviser. Keep immediate safety concerns on the normal reporting route. A spreadsheet exercise should not delay a response to a real threat.
Who bears the cost of a bad comparison?
Staff and customers can be harmed if a business cuts precautions on weak evidence. They can also bear unnecessary costs or restrictions when a dramatic percentage is treated as proof of rising danger. Require the person proposing a change to explain the local evidence, uncertainties and alternatives. Give a named manager responsibility for the decision and a way for staff to challenge it.
Careless reporting can stigmatise a community. A police-division total does not describe every resident or business within it, and it says nothing about an individual's conduct. Do not use the article's examples to label people or automate decisions about customers, workers or neighbourhoods.
Incident logs can contain sensitive information about victims, allegations and staff. Limit who can read your log, keep only what you need and follow the privacy duties that apply to your work. Use invented cases for training and public statistics for this exercise. Do not paste private reports into a general-purpose assistant to obtain a cleaner chart.
The hardest question for you and your team may be whether the available data fits the decision at all. This article has not validated a business-level risk model, reconciled every national series or measured the effect of a policing policy. The exercise helps reveal those gaps; it does not resolve them by itself.
Questions readers ask
What should a source sheet record when a number changes?
Keep the previous value, its source and access date, then add the revised value and the publisher's explanation where available. Mark which version the briefing uses. A revision is not automatically an error or evidence of manipulation; its meaning depends on what changed and why.
Can a household survey validate a police count?
Only if the measures are designed to be comparable. Check the survey's questions, population, recall period and sampling method before drawing that conclusion. A general living-conditions survey is not automatically a crime-victimisation survey, and neither can be treated as a direct replacement for an administrative count.
What if two figures appear to cover the same period?
Check whether they count victims, incidents, reports or cases, and whether one is provisional or revised. Look for differences in geography and recording rules. If the discrepancy remains, state it and seek clarification from the original publisher instead of selecting the figure that best supports the argument.
Does a lower national count settle an insurance decision?
No. Cover and premiums depend on the policy, insured exposure and other relevant evidence. Discuss the circumstances with a qualified adviser and the provider. This article supplies a way to check the data in a briefing, not a recommendation to reduce cover or a formula for pricing it.
Publisher disclosure
StarApple Analytics is part of StarApple AI, founded by Adrian Dunkley. We provide data and research services, so this article comes from a commercial provider. The source-checking steps are editorial recommendations. For related work, see our discussion of base rates and the Caribbean AI Association.