What does a top-grade percentage leave out?

What the current example shows

A-level grades at A*: SPGS’s recorded result was 7.5 percentage points higher than St Paul’s in 2025.

A-level grades at A or above: SPGS’s recorded result was 3.8 percentage points lower than St Paul’s in 2025.

A-level grades at B or above: SPGS’s recorded result was 2.7 percentage points lower than St Paul’s in 2025.

The direction changes when the grade threshold changes. The top-grade comparison alone therefore leaves out a meaningful difference in the broader recorded distribution. This describes the held examination entries; it does not identify why the distributions differ.

Move from the apex to the distribution

A* asks about the highest grade. A*–A and A*–B ask how much of the distribution lies above progressively broader thresholds. A school’s position on one threshold need not tell the whole story on another. The worked figures keep these cumulative bands separate rather than collapsing them into an overall score.

A cumulative A*–A band includes A*. It must not be read as the isolated percentage awarded A. A fine-grained distribution requires compatible grade counts or explicitly defined bands; it cannot be inferred from attractive chart shapes.

Keep concentration and scale side by side

A percentage can be high in a small examination population or lower in a larger one. Entry totals help the reader understand scale without turning entries into pupils. Where a source publishes rounded percentages, multiplying them by entries would at best yield an approximate count; this article does not promote that estimate into an exact number of awards.

The SPGS example includes its published entry total. An absent denominator in another record remains absent. Comparing a count with a percentage, or borrowing an unrelated cohort size to complete the display, would answer a different question.

Ask what the comparison cannot identify

Subject mix, prior attainment, selection and entry policy may all affect aggregate grade distributions. A broad upper band does not establish an individual attainment floor, just as a high top-grade share does not establish a future pupil’s probability of receiving that grade.

GCSE and A-level results carrying the same year do not normally track the same pupils through time. Their percentage gap is not a value-added measure. A claim about progress would require linked cohorts, compatible outcomes and a suitable method.

Try another threshold and period

Use the linked comparison for A*, then change to A*–A or A*–B while keeping the selected schools and period. Inspect the included years and any field-specific exclusions before describing a pattern. A historical article’s scope stays fixed even when you choose a new question in the comparison.

Supporting evidence and limits

Open a record to inspect its full definition, publication references and the scope of any checks. A source reference alone does not verify a figure.

1. St Paul’s Girls’ School · 2025 · Published examination results · 2025 · A-level

   Held native A-level grade share; provisional publication status remains in the evidence. Held native A-level grade share; provisional publication status remains in the evidence. Held native A-level grade share; provisional publication status remains in the evidence. The scale of subject awards; not a pupil denominator.

   Original-publication checks are recorded for 4 fields. This does not verify every field or every interpretation of the record; inspect the field-specific checks below.

2. St Paul’s School · 2025 · A-level results · 2025

   Held native A-level grade share on the same threshold. Held native A-level grade share on the same threshold. Held native A-level grade share on the same threshold.

   Original-publication checks are recorded for 1 field. This does not verify every field or every interpretation of the record; inspect the field-specific checks below.

• The article does not create a league table or a composite school score.
• The separate Bayesian and regression calculations in the older analysis are not imported as current site results.