Atlas / Free reports / Researchers

Six Axes: the open codebook

The full coding scheme, the anchor definitions, the evidence standard, and an honest account of three tests the instrument failed.

What this is, and what it is not

The Political Atlas codes 319 political organisations — parties, movements, regimes and doctrinal schools from 508 BCE to the present — on six independent axes, with a rationale and at least one citable artefact behind every one of 1,914 scored cells. The whole dataset is free to download, reuse and redistribute with attribution.

It is not peer reviewed, and it is not the output of the Chapel Hill Expert Survey, the Manifesto Project or V-Dem. It is a synthesis. Where you need an authoritative figure, cite theirs.

Where it may be useful is coverage. Party-based instruments do not record movements without electoral ambitions, doctrinal schools, or regimes that ended before the survey began — and those are precisely what most political arguments compare. This codes them on the same scale as contemporary parties.

The six axes and the independence test

An axis earns a place on one criterion, which is empirical rather than conceptual:

Organisations that sit together on every other axis can still be found at opposite ends of this one.

The practical form is to find a signature pair: two organisations a five-axis model places in the same neighbourhood and a six-axis model separates. Each axis has one, and they are named in the codebook. Candidates that failed the test were excluded, including religiosity (a mixture of social order and membership), nationalism (a composite of membership, scale and often social order) and environmental commitment, which distributes across the existing axes rather than forming one.

The anchor definitions

A number means nothing unless its scale is fixed. Each axis is anchored at five points with a description and real cases that sit there.

EC · Economic organisation

Who owns productive assets and what decides their use. The negative pole abolishes private ownership of the means of production and allocates by plan or by need; the positive pole leaves ownership private and lets prices allocate. The midpoint is a genuine mixed economy, not an absence of opinion.

CU · Social and cultural order

How society should be ordered below the level of the state: family, religion, gender, sexuality, national culture, and whether inherited norms carry authority. Political science measures a version of this as the GAL–TAN dimension, and since roughly 2010 it has been the stronger predictor of party competition in Europe than economics.

AU · Organisation of power

Whether authority is divided among institutions that can genuinely stop each other — courts, legislatures, press, opposition parties, federal units — or gathered into a single centre that claims to embody the whole. This is a question about method of rule, and it is logically independent of every policy question.

SC · Scale of decision

At what level binding decisions are taken: the commune, the province, the nation, or a body above the nation. Distinct from concentration, because a decision can be taken centrally by an accountable parliament or locally by an unaccountable strongman.

ID · Rule of membership

Who counts as a full member of the political community, and on what grounds. The negative pole admits on residence, consent or common humanity; the positive pole admits on birth, blood, faith, language or lineage.

CH · Method and tempo

How the programme is to be achieved and how fast: by election, argument and amendment, or by rupture, decree and force. It combines the classic reform-versus-revolution question with the question of whether political violence is legitimate.

The test any scheme should survive

Four organisations at the corners of the economic and authority axes. Any scheme that cannot express this configuration is not measuring what it claims to measure, and the check takes a minute.

Anarcho-communism and anarcho-capitalism sit nineteen points apart on economics and at the same point on authority. Stalinism and the Pinochet regime sit nineteen points apart on economics and at the same point on authority. A one-dimensional model cannot represent either pair, let alone both.

The evidence standard

Every scored cell carries at least one artefact, tagged by kind because the kinds are not interchangeable.

• Programme — a clause of a manifesto, constitution or founding document. Evidence of stated intention.
• Record — a documented act: a statute passed, an industry nationalised, an organisation banned, an election cancelled. Evidence of what was done.
• Scholarship — a settled finding in the comparative literature, cited to a specific work.
• Dataset — a published measurement from an existing instrument.

Where programme and record diverge, the record wins. A programme is published at the moment of maximum incentive to be attractive and costs nothing; a statute has to survive a legislature and a signature. The exception is an organisation that has never held power, for which the programme is all there is — and those codings carry a lower confidence rating for exactly that reason.

Three findings that weaken the instrument

These are published because an instrument that only ever reports findings in its own favour is not measuring anything.

Two axes correlate at 0.86. Social and cultural order and the rule of membership carry largely the same information most of the time. The defence is that the correlation is empirical rather than conceptual, and that the corpus contains cases which come apart — secular ethnonationalisms permissive on personal conduct and closed on descent, religious universalisms that prescribe heavily and admit anyone who converts. The defence is real and it is not conclusive.

The merge test returns 0.981. Collapsing those two axes and recomputing all 41,328 pairwise distances produces a matrix correlating with the original at 0.981. For almost every comparison anyone would actually make, a simpler model returns the same answer.

Leave-one-out: no axis is load-bearing. Dropping any single axis in turn leaves a distance matrix correlating with the full model between 0.959 and 0.977. Six tests, six results above 0.95.

The missing measurement

One coder, therefore no inter-coder reliability figure. There is no way to self-report that, and no amount of auditing substitutes for it, because the auditing is done by the same person.

Two things stand in its place. A check that the codebook’s named anchor cases actually sit at their anchors, which failed on ten cases the first time it ran. And a blind coding instrument that shows any reader a cell’s evidence with the score withheld and asks them to assign one.

A single reader agreeing with a single coder is not reliability. It is the only honest way available to find out whether the codebook is precise enough to produce the same score in someone else’s hands — and if the answer turns out to be no, that is the most useful finding this project could produce.

Using and citing the dataset

• Format. JSON, free, no registration. Every score, every rationale, every artefact, every logged revision. Download.
• Licence. Free to use, quote, redistribute and build upon with attribution, including commercially.
• Cite it as a synthesis, and cite the underlying professional datasets for any figure that needs to be authoritative.
• Version it. The dataset carries a version number and a revision log; cite the version you used, because scores change on evidence.
• Challenging a score requires naming the cell, naming the evidence, and saying what the score should be instead. Thirty have been changed that way.

The other three reportsMore for this role