Reading public data · Numbers behind a claim

Percent or percentage points? Recalculate the claim before quoting it

A report says a service’s failure rate rose from 4% to 6%. One headline calls that a two-point increase. Another says failures rose by 50%. Those statements might describe the same change in the rate, but the second becomes misleading if readers understand it as a claim about the number of failures.

The distinction is simple: percentage points measure the gap between two percentages; percentage change measures a difference relative to a starting value. An OSINT check needs one more thing: the population behind each percentage. Without that denominator, even correct arithmetic can tell the wrong story.

Keep three quantities on the page

When checking a percentage in a public report, write down the numerator, denominator and resulting rate. The numerator is what was counted; the denominator is the total against which it was measured. For a delivery failure rate, that might mean failed deliveries divided by all eligible delivery attempts.

“Eligible” matters. If one period includes returned items and the other excludes them, the rates may answer different questions. Also check whether the unit is attempts, orders, parcels or customers. One customer can place several orders, and one order can have several delivery attempts.

The Office for National Statistics explains percentage points as the difference between percentages. Applying that distinction to 4% and 6% gives:

• Gap: 6 − 4 = 2 percentage points.
• Relative increase in the rate: (6 − 4) ÷ 4 × 100 = 50%.
• Increase in the number of failures: cannot be calculated from those two percentages alone.

Both of the first two results can be correct. Choose the one that answers the question, and retain the starting and ending rates so readers can judge the scale.

The count can fall while the rate rises

Invented teaching case: a fictional parcel service publishes the following table. Assume the definitions, measurement method and period length are comparable. These numbers are not customer data or an evaluation of a real business.

First reproduce the rates: 200 ÷ 5,000 × 100 = 4%, and 180 ÷ 3,000 × 100 = 6%. Then calculate the count change: (180 − 200) ÷ 200 × 100 = −10%.

The service recorded fewer failed attempts, but failures represented a larger share of a smaller workload. “Failures surged 50%” leaves out the unit that makes the statement intelligible. A clearer sentence is: “The failure rate rose from 4% to 6%, while the number of failed attempts fell from 200 to 180.”

That finding still does not explain why the rate changed. The mix of routes, workload or reporting rules could matter. Nothing in this table establishes a causal effect or an individual customer’s experience.

Turn a screenshot into a reproducible check

1. Find the original table. Save its title, source URL, release date and table identifier. A cropped chart may omit a footnote defining the population.
2. Capture the definitions. Copy the unit and inclusion rules into your notes. Record unknowns instead of quietly assuming the two columns match.
3. Check the period and version. Compare equivalent periods from a consistent data series. An older screenshot may contain an estimate that has since been revised.
4. Reproduce both rates. Keep the underlying counts when they are available, along with any stated weighting or adjustment. A weighted survey percentage may not equal a simple count divided by the sample size.
5. Calculate the relevant changes. Label count change, percentage-point difference and relative rate change separately.
6. Rewrite the claim in one sentence. Include the starting value, ending value, unit and any limitation that changes the conclusion.

Use the PDF comparison workflow if the figures differ between two report versions. Do not combine a revised denominator from one release with an unrevised numerator from another unless the producer explicitly supports that calculation.

As a real example of why versions matter, the ONS population estimates release consulted on 17 October 2026 tells users to use its associated time series rather than append the newest estimate to figures from earlier releases. That is source-specific guidance; read the equivalent notes for your dataset.

Watch the denominator when combining groups

An average of two percentages is not necessarily the percentage for the combined group. Consider a second fictional example: one branch has 1 failed attempt out of 10, and another has 9 out of 90. Here both rates are 10%, so the combined rate is also 10%.

Now keep the totals but change the failures to 2 and 9. The rates become 20% and 10%. Their simple average is 15%, but the combined rate is (2 + 9) ÷ (10 + 90) × 100 = 11%. The larger branch contributes more attempts. This calculation applies when the groups are disjoint and their counts have compatible definitions; overlapping groups require a different treatment.

For a check of a published headline, ask whether it concerns an average branch, a typical customer or all attempts together. Those are different quantities, and a source should identify which one it reports.

When the calculation cannot support a neat percentage

A starting value of zero: the usual relative-change formula divides by the starting value. Moving from zero to five recorded cases has an absolute increase of five; it does not have a finite percentage increase under that formula. Do not replace the undefined result with “500%”.

Rounded figures: rates displayed as 4% and 6% may hide decimals. Calculating 50.000% from them adds precision the source did not provide. Use unrounded data when available and retain the original rounding note.

Different populations: a changed eligibility rule can break comparability. You can describe each reported figure while leaving the claimed improvement or deterioration unverified.

A sample estimate: arithmetic alone does not establish that a change is statistically meaningful. Read the producer’s uncertainty and comparability information before making a claim about the wider population.

A compact record another reader can audit

Claim as published:
Source / release / table:
Population and unit:
Earlier count / denominator / rate:
Later count / denominator / rate:
Percentage-point difference:
Relative change, with its starting value:
Version, rounding or comparability limits:
Corrected sentence:

This is a calculation check you can usually complete with the original table and a calculator. When the numbers sit inside a larger case involving competing reports and company claims, the OSINT Jet report template helps keep the calculation separate from the source’s interpretation. If that wider research needs a scoped review, describe the contradictory claims and supporting documents. Neither a tool nor a polished report should conceal a missing denominator.