DubaiSchools.ai parent data guide

Published 9 November 2026 · Pre-registered analysis of KHDA inspection and fee data

What a KHDA Rating Tells You: Six Questions We Tested Before Looking

Before running a single model, we wrote down six questions, what we expected to find, and what result would prove us wrong. Then we tested them against KHDA’s data. Two expectations held cleanly, one held with a twist we did not predict, one failed outright, one we could not test at all, and one is a forecast still waiting to be scored. This is what each result means for a parent reading a rating.

First, a test of ourselves

An analysis of KHDA ratings is only as good as its copy of KHDA’s ratings, so the first question was whether we could reproduce KHDA’s own published count for the 2023-24 inspection cycle: 23 Outstanding, 48 Very good, 85 Good, 51 Acceptable and 2 Weak, from 209 schools.

Not at first. Counting every rated school whose inspection is dated in that cycle gave us 206, 3 short, and the first pass was further out still, because one of our own files was missing schools and our date parser misread an abbreviated month. Closing the gap took school-by-school work against KHDA’s own list. 10 schools appear in KHDA’s list under the names they were registered with at the time, and have since been renamed. Matching them by hand worked only because each pair also carried the same rating on both sides. One school is in our index without a dated inspection. The 2 schools KHDA rated Weak that year appear in no later KHDA register we hold.

The exercise also turned up a mistake of ours. One school was shown on DubaiSchools.ai as Good. Its inspection report, KHDA’s open-data workbook and KHDA’s own 2023-24 list all say Acceptable. We have corrected it. KHDA is the authority on ratings, and we would rather find our own errors in public than have a parent find them for us.

Ratings are sticky, and mostly move up

Across 1,252 pairs of consecutive inspections of the same school, the rating stayed the same 85% of the time. It went up 11% of the time and down 4%. In other words, for every school that slipped, roughly 3 improved.

Two things could have distorted this picture, and we checked both before pooling the years. In 2015-16 KHDA moved from a four-point scale to a six-point one, adding Very good and Weak. A school that went from Good to Very good across that line may not have changed at all; the ruler did. We set those transitions aside. The second was the long pause in full inspections after 2019-20. The pause did not visibly shift the ruler: 14% of schools moved up across it, against 11% in a typical year.

For a parent, stickiness cuts both ways. A rating is a reliable description of a school’s recent past. It is a weak signal of a recent change, because most schools do not have one.

Fees and ratings: the link is stronger than it looks

We expected more expensive schools to be rated higher, and they are. In the raw data, each doubling of fees goes with 2.7 times the odds of a higher rating (176 schools with verified fees and complete data). We also expected that holding curriculum, school age and intake constant would shrink that link. It did the opposite. Among schools of the same curriculum, age and intake, a doubling of fees goes with 11.6 times the odds of a better rating (95% bootstrap interval 7.0 to 29.3).

The reason is two groups of schools pulling against each other. Newer schools charge more, but older schools are rated higher: every extra decade of operation goes with 2.6 times the odds of a better rating. And Indian-curriculum schools charge a median of AED 13,066 a year, against AED 49,636 for British-curriculum schools, yet 77% of them are rated Good or better. Both patterns flatten the raw line between fees and ratings. Take them out, and the line gets steep.

The relationship is not the same everywhere. Within British-curriculum schools, fee rank and rating rank move together strongly (Spearman’s rho 0.67). Within Indian-curriculum schools the link is also clear (0.51). Within American and IB schools it is weak (0.19 and 0.08) and statistically indistinguishable from no link at all. A high fee at an American or IB school is weaker evidence of quality than the same fee at a British one.

Does a better rating buy a bigger fee rise? We could not see it

Under KHDA’s fee framework, a school that improves its rating is allowed a larger fee increase than one that holds steady. If that rule bites, it should show up in actual fees. We compared schools that had just improved with schools that had held their rating, in years when the framework applied. As a placebo, we ran the same comparison in years when fees were frozen, where no gap should exist.

The result was negative. In framework years, improvers’ fees rose 0.9 points more than holders’ on average, with an interval running from −0.2 points to 1.9 points. That crosses zero, and it falls short of the 1-point gap we had set as the bar in advance. The placebo years, which should show nothing, produced a gap of −1.6 points: larger in size than the effect we were looking for. In these data the noise is louder than the signal, so we report this test as failed.

One year is worth mentioning, with the warning that we did not pre-register it. In 2024/25, schools holding their rating raised fees by 2.5% on average, close to the 2.6% cap that applied to them, while the 23 improvers raised them by 4.2%. That looks like the rules working. A single year picked after the fact is a hypothesis for the next cycle, not a finding.

Who is missing from every rating statistic

24 of the 241 schools we index have no KHDA rating yet, so every rating statistic, ours included, describes 217 schools, not 241. The unrated schools are not a random slice. Of the 11 with a known opening year, the median opened in 2024, against 2006 for rated schools. Their median fee is AED 58,836, against AED 33,864. The newest and most expensive schools are exactly the ones a rating cannot yet speak for. Given the age effect above, it would be a mistake to assume they will start where their fees suggest.

The question we could not answer

The most useful thing a KHDA report says is not the headline rating. It is whether children make good progress, as distinct from reaching high attainment. Attainment mostly reflects who a school admits. Progress is closer to what the school adds. We planned to test whether fees buy attainment more than progress.

We could not, and we would rather say so than paper over it. The sub-judgements in our extracted copy of the reports failed the integrity checks we set in advance. Not one of the 208 extracted reports carries a progress judgement. In 10 reports, judgements appear for a sixth form at a school that does not have one. In 50 of the 183 reports with more than one phase, every phase carries identical judgements, which points to a parsing fault rather than to uniform schools. Answering the question means going back to the report PDFs, which KHDA does not currently serve to automated requests. Until then, we publish no claim about progress.

A forecast, on the record

KHDA has said full inspections will not resume in 2025-26. When they do, our model, built from how ratings have moved in typical years, expects about 27 Outstanding schools (90% range 21 to 32) and 53 Very good (44 to 61) among the 217 schools rated today, up from 23 and 48. We recorded this on 9 November 2026 and will score it when the next cycle is published.

What this means when you choose

Read a rating as a well-measured description of a school’s recent past, which rarely changes. Read a fee alongside the school’s age and curriculum, not on its own: a young, expensive school has not yet had the chance to earn its rating, and a modest fee at an established school can sit next to a very good one. And when a school has no rating, that tells you about its age, not its quality.

Sources, method and limits

Ratings come from KHDA’s open-data workbook (inspection history from 2008/09), KHDA’s 2023-24 Inspection Key Findings list and the KHDA directory. Fees are KHDA fee records by grade and year; only verified fee schedules are used. The 2023-24 benchmark is KHDA’s figure as published in the government press release of June 2024. The six questions, expectations and falsifying results were registered on 9 November 2026 before any model was run. Later changes were logged with reasons: the 2015-16 scale change, the additional fee-freeze years, a school-name reconciliation table, and a correction to the forecast’s prior, which had allowed rating jumps that never occur. Ratings are modelled with a proportional-odds ordered logit, checked against an independent reference implementation; intervals come from 2,000 bootstrap resamples, with Benjamini-Hochberg correction across curricula. The intake measure is the Emirati share of enrolment, the only one available, and is a coarse proxy. All figures are associations in observational data. See the methodology.