Raw mean vs shrunk Vårdbetyg
A simple ranking sorts health centres straight by their average in the National Patient Survey 2025. Then a unit where only a handful of patients responded can end up on top by chance. Here the same health centres are ranked both ways, side by side, so you can see what happens once the number of responses and the waiting times count.
Two rankings of the same units
- Raw mean – the survey’s equally weighted average straight up, ignoring how many responded.
- Vårdbetyg score – the same average pulled toward the national average by how many responded (Bayesian shrinkage), and combined with the waiting times (accessibility).
The gap between a unit’s two positions comes from two things: (1) shrinkage toward the number of responses – few responses pull toward the national average – and (2) the accessibility weight – long waiting times pull the final score down. The cards below say nothingabout a single health centre being good or bad, only how much the position changes when those two things count.
The units that move the most
Top 25 on the raw mean – and where they land
The table lists the 25 health centres that rank highest on the raw mean, with their Vårdbetyg position alongside. The Responses column explains only a small part of the move. The heaviest factor is accessibility (waiting times), which is part of the Vårdbetyg but not of the raw mean.
↓ = the unit ranks lower on the Vårdbetyg score than on the raw mean (shrinkage or low accessibility pulled it down). ↑ = the opposite. At 30 responses a unit’s own result and the national average carry exactly equal weight.
In a raw mean a unit with a high average on a few responses can end up on top. The Vårdbetyg score takes the number of responses into account.
What the score does not capture
Shrinkage handles chance, that is the risk that a small basis lands wrong. It does not correct systematic bias. A health centre in an affluent area with healthy, confident patients tends to receive higher ratings than one serving a population with heavier care needs, for reasons that say nothing about how well the clinic does its job. The Vårdbetyg score does not adjust for such differences in patient mix (case-mix), any more than a raw mean does.
Two more things worth remembering. The National Patient Survey measures patient experience, not medical outcomes, and it is answered by fewer than half of those asked (45.2 per cent in 2025). Those who respond are not necessarily representative, and neither the shrinkage nor anything else in the score corrects for that. The score is therefore best read as a stable measure of how patients experience a clinic, not as a fair verdict on the quality of care independent of who the patients are. Compare clinics in similar areas where you can, and weigh in what you know yourself.
The whole method, with the formulas, is on the methodology page.
What the score never does
There are four things the Vårdbetyg score never does. Each line is a promise that holds even when the method is revised, with a date and the page where you can check it. The lines were checked against the code on 14 October 2026.
We never fill in a missing measurement with a guess. If part of the basis is missing, the score is calculated on what exists. Without waiting-time data the score rests entirely on the patient survey, and if too few patients answered we set no score at all. If the number of cases behind a waiting-time share is missing, we use a cautious standard basis for how far the share is drawn toward the national average, never an invented share.
Popularity and search volume are never a ranking factor. How many people visit a page, click through or search for a health centre never affects the score or the order of a list. Lists are sorted by the Vårdbetyg score, and the site search by how well the name matches what you typed.
Google reviews and providers' own statements are never weighed in. Google reviews are shown as context and always marked as an external source, and we do not accept submitted reviews. What the provider states itself, such as opening hours on 1177 or its own improvement actions in our portal, never affects the score or the order. Opening hours only decide who appears on a page such as evening opening. Nor does a seal, the paid licence to show an award as a badge on the health centre's own pages, affect the score.
We never publish lists of the worst health centres. The top lists show the highest ranked, and the municipality lists the whole municipality with the best first. We do not hand out such lists on request either. To see how things stand where you live, look at the municipal measure for the care guarantee: how many of the municipality’s health centres meet the target, without singling anyone out.
The municipal care-guarantee measure is on the municipality and county pages where at least three health centres have figures. How the municipal measure is calculated.
How to read the margin of error
The patient survey asks a sample of patients, not all of them. Had other patients happened to answer, the figure would have come out a little different. The margin of error (the confidence interval) is the range where the true value most likely lies. The National Patient Survey reports it for every area with 95 percent certainty.
An example. A score of 82 with a margin of error of 6 means the true value most likely lies between 76 and 88. If the national average is 79, it falls inside the range, and the difference cannot be told apart from chance. If the national average is 74, the whole range lies above it, and the difference is reliable.
Few answers give a wide margin of error, many answers a narrow one. That is why a health centre can be a few points above the national average without it being possible to say that it really is.
Health centre pages with a score can carry one of two labels: “Reliably above the national average” when the whole margin of error for patients’ answers lies above the national average, and “Cannot be told apart from the national average” when the margin of error reaches it. If the whole range lies below the national average there is no label at all, because we never set a label saying that a single health centre is below the national average.
The label concerns patients’ answers in the survey areas, compared with the national average for health centres in the same areas. The areas build on the same patients’ answers, so we use the wide margin of error. Better too few “reliably above” than a verdict that does not hold.
Right now 134 of 1,192 rated health centres (11%) are reliably above the national average.
Municipality and county pages show three states, above, in line with or below the national average, because they concern the average of at least three health centres and not a single clinic.
All health centres in one picture
Each dot is a health centre with a score, 1,192 in all. Further right means more answers in the patient survey. Higher up means patients’ answers are above the national average for health centres, lower down that they are below. No names appear in the picture.
Between the inner lines the margin of error reaches the national average. There the difference may come down to which patients happened to answer, and that is where most health centres are. The band is wide where few have answered and narrow where many have.
A dot above the upper inner line is reliably above the national average. It is the same rule as the label “Reliably above the national average” on the health centre’s page, so the picture and the label always say the same thing. The outer lines are stricter, 99.8 percent certainty instead of 95. Outside them the difference is so large that chance can almost be ruled out, upwards and downwards.
The lines say how certain a difference is, not how good the care is. The patient survey measures how patients experience their care.
The large green dot is the health centre the link marks.
Right now 134 health centres lie above the upper inner line, and 49 of them above the outer line as well. 986 lie between the inner lines. 72 lie below the lower inner line, and 8 of them below the outer line as well.
The picture shows patients’ answers before the shrinkage toward the national average, because that is where the margin of error is. The dots furthest to the left spread out the most. That is why the Vårdbetyg score pulls health centres with few answers toward the national average.
New awards from 2027
From 2027 there are two new awards. They are given for the first time on the National Patient Survey results published in spring 2027, and the rules are the same for every health centre.
Like today’s awards they are free and visible to everyone. No one can buy an award and no licence can switch it off. The awards never affect the score or the ranking. The only thing a health centre can pay for is a seal. A seal is a paid licence that lets a health centre with an award show it as a badge on its website or in the waiting room, with a link back here where anyone can check it.
Reliably above the national average in one area
The patient survey reports six areas for health centres. They are care and treatment, involvement, respect and empathy, continuity, information and accessibility. A health centre receives the award for an area when three things hold in the same year.
- At least 30 patients answered the questions in the area.
- The whole margin of error lies above the national average for health centres in the same area. It is the same rule as the “Reliably above the national average” label above, but for each area on its own.
- The health centre meets the basic requirements for all our awards. It has a score, at least 30 patients answered the survey, and it did not open during the measurement year.
Accessibility here means patients’ answers in the survey, not the waiting times.
The award carries the year after the measurement, like our other awards. A measurement from 2026 therefore gives awards marked 2027, for example “Reliably above the national average in continuity 2027”.
Two and three measurements in a row
If a health centre has been reliably above the national average in the same area in two or three measurements in a row, the award says so. One example is “Reliably above the national average in respect and empathy 2028, three measurements in a row”. The measurement years are always written out next to it.
We count measurements and not calendar years, because some regions only take part in the patient survey every other year. A year when the region was not measured does not count and does not break the run either. A year when the health centre was measured but did not make it breaks the run.
The run only counts within the same measurement. The patient survey got new questions and answer scales in 2025, so results from before 2025 are never counted. If the National Patient Survey restarts the measurement again, the count starts over from the new start. Results from 2025 do count already in 2027, since the rule is the same both years.
Top-rated three measurements in a row
The streak award goes to a health centre that has been “Top-rated in [municipality]” in three measurements in a row. It is counted the same way as the runs above, and only from 2025. It can therefore be given in 2028 at the earliest, once the measurements from 2025, 2026 and 2027 exist. The label then reads, for example, “Top-rated in [municipality] 2028, three measurements in a row”.
There are no awards in the other direction. We never award anything because a health centre is below the national average, and a broken run is never shown. Only the years that count are written out.
The rules apply from award year 2027. If they change, we will say so here, with a date, before the new rules take effect.
When a value is missing
We never fill in missing values with guesses. When a figure is missing we say why, with a code that describes the source and never the health centre. The same codes appear in the open files.
The score weighs patient responses at 70 percent and waiting times at 30 percent. If waiting times are missing for a health centre, the score is based on patient responses alone. The weights are then spread over the source that has data, so the missing source neither lowers nor raises the score. We call this a rating that rests on 1 of 2 sources, and it is shown next to the score on the unit page and in the coverage line under the leaderboards. If too few patients responded we set no score at all, whatever the waiting times.
| Code | Meaning | Source |
|---|---|---|
| F01 | Too few responses. The unit was part of the National Patient Survey, but too few patients responded for a rating to be shown. | Nationell Patientenkät |
| F02 | Not reported, not in the survey. The unit, or the whole region, was not part of the National Patient Survey sample for the year. | Nationell Patientenkät |
| F03 | Not reported, the region is changing record systems. The National Board of Health and Welfare has no waiting-time figures for the unit, and states that the region is changing its patient record system, which can affect reporting. | Socialstyrelsen |
| F04 | Not reported. The National Board of Health and Welfare has no waiting-time figures for the unit, and the source gives no reason. | Socialstyrelsen |
| F05 | Not applicable. The unit has closed, is a branch or has merged with another, so current figures do not apply. | Vårdbetygs register över nedlagda mottagningar |
| F06 | The rating rests on 1 of 2 sources. Waiting times are missing, so the rating is based on patient responses alone. The weights are spread over the source that has data. No missing value is filled in with a guess. | Vårdbetyg |
The codes appear in the columns vardbetyg_fotnot and tillganglighet_fotnot in the open file of health centres and in each health centre’s data.json. The full code list is its own file, fotnoter.csv. A code never changes meaning. The export schema is version 3.1.
- Version 3.1 (27 October 2026). Municipality code and county code: new columns kommun_kod (the 4-digit SCB municipality code as text with a leading zero, e.g. 0117) and lan_kod (the 2-digit SCB county code) last among the fixed columns of units.csv, kommuner.csv and historik.csv. Municipality names in historik.csv now follow the SCB spelling: four misspellings from older survey years (Upplands Bro, Upplands-Väsby, Malung-sälen, Solentuna) are corrected, so the file has 290 municipalities, not 294. Region Sörmland did not take part in the survey in 2025, so Sörmland health centres have rows in historik.csv only up to 2023.
- Version 3.0 (27 October 2026). Who runs the health centre in units.csv: huvudman (region, private or unknown, from the National Patient Survey), kedja (the brand in the name) and vardgivare, vardgivare_orgnr and vardgivare_kalla where an open source names the company. New file historik.csv with one row per health centre and survey year from 2015. The fields are descriptive; Vårdbetyg does not compare ownership types.
- Version 2.0 (14 October 2026). Footnote codes for missing values (F01 to F06) in units.csv (columns vardbetyg_fotnot and tillganglighet_fotnot) and in each unit’s data.json, the code list as its own file (fotnoter.csv) and the schema_version column.
- Version 1.0 (10 July 2026). units.csv and kommuner.csv on the site. fragor.csv (question level) was added on 2026-07-30 without a new version.
More promises about missing values are listed under What the score never does.
Source: National Patient Survey 2025 (SKR). Open data at resultat.patientenkat.se.