Method note: Grade deviation

The grade deviation — also known as the grade inflation score — is a blunt measure with known systematic weaknesses. This note explains when and why it can be misleading, particularly for schools with a high share of pupils studying Swedish as a second language (SVA) or pupils who have recently arrived in Sweden.

What does the grade deviation measure?

Grade deviation compares the school's grade point per subject with the school's average result on national tests in Swedish, maths and English. Both metrics are expressed on the 0–20 scale where A=20, B=17.5, C=15, D=12.5, E=10 and F=0.

Grade deviation = (Merit rating ÷ 17) − NP average (Swedish/maths/English)

Positive value (e.g. +2.0) — final grades are higher than the NP results suggest.
Near zero — final grades and NP results match.
Negative value — final grades are lower than NP results, possibly indicating stricter grading.

How this differs from Skolverket's standard definition: Skolverket normally reports the "deviation between test grade and subject grade" subject by subject in the same three subjects (Swedish, maths, English). Skolkoll's formula instead compares a 17-subject average (merit rating divided by 17, i.e. 16 core subjects plus modern languages) against a 3-subject average (NP average in sv/ma/en). The reason is pragmatic — complete subject-level NP and grade data per school is not available in Skolverket's open data — but it means that grading culture and teacher staffing in subjects beyond sv/ma/en also affect the metric. It is a broader proxy and should be read as such.

A positive value does not necessarily mean that the school awards unwarranted grades. It measures a gap between two data points computed on different subject samples, and that gap can arise for several other reasons.

When is the measure unreliable?

The measure assumes that the same pupils are being compared on both sides — final grades and national tests — in a comparable way. That assumption breaks down systematically on four points:

1. Swedish vs Swedish as a second language (SVA)

The Swedish National Agency for Education's own statistics for spring 2025 show that 93 % of pupils passed the NP in Swedish, compared to only 74 % in the NP for Swedish as a second language — a 19-percentage-point gap. At schools with a high share of SVA pupils the NP average in "Swedish" is dragged down by SVA results, while the final grades in SVA are set according to the separate SVA syllabus. The gap between NP average and grades therefore ends up systematically overestimated.

2. English has no SVA equivalent

All pupils sit the same national test in English regardless of their Swedish language skills. Newly arrived and multilingual pupils with a short time in the Swedish school system rarely pass the NP in English — but still receive a subject grade based on formative assessment throughout the year. The result is an artificial gap between NP average and final grades.

3. The NPs in Swedish, maths and English are language-intensive

Even the NP in maths requires strong reading comprehension to solve word problems. Pupils with weaker language skills systematically perform below their actual subject knowledge on the NP, which pulls down the school's NP average while affecting the subject grade less — because the grade is based on multiple assessment occasions and oral communication.

4. Final grades rest on broader evidence than one NP

Subject grades are set based on formative assessment, oral components, several written tasks and submissions across the full school year. A single NP is a point measurement at one occasion, under time pressure and with limited support. For pupils with weaker language skills, the gap between what they can demonstrate on a written NP and what they actually know in the subject is largest.

Sum of the asymmetries: a school with a high share of SVA or newly arrived pupils can look like it has grade inflation, where a substantial share of the gap actually reflects an asymmetric pupil selection between the NP and the final grades. In those cases the measure often says as much about pupil composition as about the correctness of grade-setting — without ruling out that there may also be a real difference in grading practice on top of that.

Case: Skårbyskolan North and South (Kungsbacka)

Two parallel units in the same building, with the same principal team and in many cases the same teachers, but with different pupil composition. The case is pedagogically clear precisely because what differs is primarily the pupil population.

Comparison of key indicators between Skårbyskolan North and Skårbyskolan South
Indicator	Skårbyskolan North	Skårbyskolan South
Merit rating year 9	249	218
SALSA residual	+8	−3
Grade deviation	—	+1.3
Eligibility for upper secondary	—	73 % (national: 86 %)
Share of newly arrived pupils	—	11 %

A 31-point difference in merit rating between two parallel units points to pupil composition being the dominant variable. The SALSA residual −3 and the grade deviation +1.3 point in opposite directions: SALSA says the school underperforms relative to its pupil population, the grade deviation suggests grades that are generous relative to NP. One plausible explanation is that NP results are asymmetrically pulled down by pupils with weaker language skills, while final grades capture subject knowledge through broader evidence — but this is not the only possible explanation. See alternative explanations below.

Caveat: The case is a single data point (N=1). The observation about pupil composition comes from Håkan Kollberg Winsnes, development lead at Kungsbacka municipality (April 2026) — i.e. from the school operator responsible for the school in the example. The observation is worth considering but does not replace independent validation: a methodological conclusion about the size of the measurement error at the population level requires aggregated analysis across the full dataset, not a single case.

Sources: Skolverket’s open data (merit rating, SALSA, year 9 eligibility).

Alternative explanations to consider

Asymmetric language-based pupil selection is a plausible hypothesis in the case above, but several other explanations can produce the same combination of numbers — or interact with each other:

Real grade inflation in parallel with a challenging pupil cohort. SALSA and grade deviation measure different things against different reference points. A school can both underperform relative to its socioeconomic prediction and award generous grades relative to NP. Both can be true at the same time — there is no logical contradiction.
SALSA does not capture all background factors. The SALSA model uses parental education, gender and recent-immigrant status — not the share of SVA pupils or language proficiency directly. For schools whose linguistic complexity SALSA underweights, the residual can be "too negative" and therefore not a fair benchmark.
Selective NP participation. If pupils with weaker language skills are more often absent or exempt from the NP, the NP average is affected — the direction depends on how Skolverket counts pupils without a test grade (missing value vs. zero points). This is a known uncertainty in aggregate NP results.
Random variation in small cohorts. With fewer than 30 year-9 pupils, +1.3 may sit within noise. This is also what triggers Skolkoll's small-cohort warning flag.
Different teachers despite the same principal. Two parallel units with "the same principal team and in many cases the same teachers" nonetheless do not have an identical teacher composition. Different teachers mean different grading culture even with the same leadership.

The conclusion is that a high grade deviation can be due to asymmetric pupil selection, but it can also be due to actual differences in grading practice, statistical noise, selection bias in NP participation, or a combination. The measure always requires complementary analysis before it carries interpretive weight at the unit level.

How the grade deviation should be interpreted

To use the grade deviation responsibly — in journalism, school inspection, political debate or school choice — the following interpretation rules apply:

Useful in aggregate, unreliable at unit level. Research shows that grade deviation correlates with other indicators of generous grading at the operator and system level. The measure is therefore meaningful for analyses of e.g. for-profit school chains or the Swedish school system as a whole — but comparing or auditing individual schools requires deeper analysis.
A high grade deviation is an indicator, not proof. The measure should never be cited as decisive on its own. It requires deeper analysis of pupil composition, share of SVA pupils, newly arrived pupils, cohort size and NP participation.
The measure is especially unreliable when the school has a high share of SVA pupils, a high share of newly arrived pupils, or small cohorts where individual pupils' results weigh heavily on the average.
Always combine with the SALSA residual. When SALSA and grade deviation point in opposite directions, that is a strong signal that the measure is capturing pupil composition rather than grading practice. See the SALSA method.
Do not compare units with different pupil populations directly. Skårbyskolan North and South illustrate why: same school leadership, different pupils — very different numbers.
Never use the measure to single out individual teachers or units without additional investigation. The risk of incorrect conclusions is significant.

References and further reading

Skolverket (2025). Test grades and subject grades — results from national tests year 9, spring term 2025. Share with passing test grade: Swedish 93 %, Swedish as a second language 74 %. skolverket.se/statistik
Skolverket. Deviation between test grade and subject grade. Skolverket's standard definition is computed subject by subject in the same three subjects (sv/ma/en). Skolkoll's formula is a broader proxy as described in the What does the grade deviation measure? section above.
Vlachos, J. (2018). Trust-based evaluation in a market-oriented school system. IFN Working Paper 1217, Research Institute of Industrial Economics. — Research on grade inflation and the effect of competition on grading practice in Swedish schools.
Henrekson, M. & Wennström, J. (2022). Dumbing Down: The Crisis of Quality and Equity in a Once-Great School System—and How to Reverse the Trend. Palgrave Macmillan. — Overview of the research on grade inflation and the incentives of the Swedish school market.
Hansson, Å. & Gustafsson, J.-E. (2016). Pedagogical segregation: Teacher competence in the Swedish compulsory school from an equity perspective. — Background on how language background affects NP performance.
Skolinspektionen (2018). Rättvis och likvärdig betygssättning i grundskolan? [Fair and equivalent grading in compulsory school?] (case no. 2016:11436). Quality review of 20 compulsory schools by the Swedish Schools Inspectorate, discussing variation in grade-setting and the limitations of comparing individual schools using grade deviation. skolinspektionen.se
Grade inflation 2026 — which schools inflate the most? — Skolkoll's analysis where the grade inflation score is used systematically. This method note should be read alongside it.
The SALSA method — Skolverket's regression model that adjusts merit rating for background factors. Used as a complement when grade deviation is discussed.
Method policy — How Skolkoll documents methods and handles methodological changes.

This method note was added following a methodological point raised by Håkan Kollberg Winsnes (development lead, Early Years & Compulsory School, Kungsbacka municipality) in April 2026. Because the input comes from the school operator whose school appears in the case study, the arguments have been weighed specifically against independent sources and alternative explanations (see the section above). Improvements and corrections are welcome via info@skolkoll.se, see the corrections policy.