Methodology

How our statistics work

What we collect, what the measures mean, who is in the sample, and what the figures cannot tell you.

Figures updated 23 September 2026

Everything on these pages comes from people who took a CoreOpinion test. This page explains exactly what we store, what each measure means, who the people in the data are, and the limits you should keep in mind before quoting any of it.

What we collect

Before showing a result, the test asks a few optional questions: age band, gender, nationality, and for country tests, the party you feel closest to. Below them is a research consent box. The box is ticked when the screen loads, and a result is saved for research only if it is still ticked when you continue. Untick it and your answers are not saved for research, and you still see your full result.

A saved result contains the test you took, your position on the two compass axes, your framing bias score, your party alignment scores, and whichever of the optional answers you gave. It does not contain your name, and we do not store IP addresses with results. If you choose to email yourself a link to your result, a separate copy is stored with that address so the link keeps working, as our privacy policy explains. Those emailed copies are left out of these statistics entirely, so nobody is counted twice and nobody who unticked the research box is counted at all.

What the measures mean

Framing bias. Key positions in every test are asked twice: once described in terms of what they achieve, and once in terms of what they cost. Someone with perfectly fixed views would give mirror-image answers to the two versions. The framing bias score measures how far each person departs from that, on a scale from 0 to 100. A score of 0 means the wording made no difference at all. Higher scores mean the answer moved more with the wording.

Best-matching party. Each country test scores your answers against our reading of each main party's published platform and voting record, and reports the closest one. This is the party you are compared against in our party analysis. It measures agreement on policy only, not on leadership, record in office, or any of the other reasons people support a party.

Self-reported party. The optional question asking which party you feel closest to. It asks about affinity, not about how you voted.

Who is in the data

As of 23 September 2026, the data holds 8,284 saved results collected since 26 June 2026. 7,248 include an age band, 6,802 a gender, and 2,558 a party the test scores.

Results by testTests with at least 100 results. A further 367 results come from smaller tests.
Show the data
TestResultsShare of allAged under 25
Finland3,37440.7%85%
United Kingdom1,39316.8%66%
General test86110.4%79%
Australia5516.7%60%
United States5386.5%73%
Germany2973.6%63%
Canada2723.3%64%
Austria2332.8%65%
New Zealand2262.7%37%
Sweden1722.1%43%

One test dominates. Finland accounts for 41% of all results, and 85% of Finnish respondents who gave an age are under 25. Most of it arrived within a few weeks, which looks far more like use in classrooms than ordinary search traffic. We check the main findings with Finland removed, and say so where the result changes.

Results by monthSplit between the Finnish test and every other test.
FinlandAll other tests
June 2026110
July 2026844
August 20262,405
September 20264,925
Show the data
MonthAll resultsFinlandOther tests
June 20261100110
July 202684434810
August 20262,4059351,470
September 20264,9252,4052,520
Results by ageSplit between the Finnish test and every other test.
FinlandAll other tests
Under 182,914
18 to 242,465
25 to 34779
35 to 44399
45 to 54311
55 to 64227
65 or older153
Show the data
AgeAll resultsFinlandOther tests
Under 182,9141,4461,468
18 to 242,4651,1411,324
25 to 34779241538
35 to 4439984315
45 to 5431177234
55 to 6422728199
65 or older15321132

What the data cannot tell you

How we report

Small groups are never shown. A country, age band, gender or other group appears only if it contains at least 100 people. A party within a country appears only with at least 50. This protects privacy and stops small, noisy groups from producing striking but meaningless numbers.

Uncertainty is shown, not hidden. Averages and proportions carry 95% intervals. When we say one group differs from another, the interval on the difference excludes zero. When it does not, we say that too.

Figures update, analysis is written. The numbers on these pages are recalculated from the database every week. The analysis around them is written and reviewed by the CoreOpinion editorial team. Each article's key claims are checked automatically against the new figures on every update, and any claim that stops holding is flagged for us to rewrite rather than left in place.

Only aggregates leave the database. The statistics pages are built from counts and averages alone. No individual result is ever published here.

Questions about the data, or spotted something that looks wrong? Get in touch.