~/projects/polyaxis
[02 / 04] ● live politics · statistics · web

polyaxis

a political test that refuses to collapse three different questions into one line.

year
2026
role
design, statistics, front end
stack
javascript
items
230
scoring
local, in-browser
[ specimen ]
three axes, one coordinate. every item belongs to exactly one.

[01] three axes, not one

ownership asks who owns and directs productive capital. change asks how far inherited norms should be conserved or revised. authority asks how much concentrated, unaccountable power over the individual is acceptable.

a ministry-owned factory is not worker control, and a centrally directed economy that subordinates workers is not communal ownership either. every item belongs to exactly one axis, and the report gives three coordinates instead of a single left-right score.

[02] scoring

agree/disagree and true/false items use a graded response model. multiple choice items use a position-based model, where each option is itself a point on the axis. the estimate is the posterior mean under a standard-normal prior, reported with an interval of ±1.96 posterior standard deviations.

skipped questions are excluded. they are not counted as centrist.

one axis cannot hold ownership, change and authority without flattening at least two of them.

[03] what it does not know

the item parameters are provisional. calibration needs a large representative sample, and this instrument has never had one. the discrimination and threshold values encode an explicit judgement about what each statement measures, and they are published in the engine source.

  • 12 of 230 items sit in the extreme difficulty band
  • above roughly 80, the interval matters more than the number
  • figures on the site are editorial estimates, not measurements

[04] privacy

there are no accounts and no cookies. scoring, rendering and export all run in the browser. a shared result travels in the url fragment, which browsers never send to a server.

finishing a test adds each answer to an anonymous tally so the items can be improved. what is stored is a count per question per answer, not a record of a person. the test scores identically whether or not that tally is reachable.