Models
Political Compass vs Political Typology — How the Models Differ
One model plots you in continuous space; the other sorts you into a group. They answer different questions, and mixing them up is where most confusion starts.
If you have taken more than one political self-assessment, you have probably noticed that they do not agree — and not in the small way where one says 62 and the other says 58. They disagree about what kind of answer they are giving you.
That is because two fundamentally different modelling traditions sit underneath them. One places you as a point in a continuous space. The other assigns you to a group of people who answer similarly. Both are legitimate. They are built for different questions, they fail in different ways, and a great deal of confusion about political tests dissolves once you can tell which one you are looking at.
The dimensional model: you are a set of coordinates
The compass family is dimensional. It assumes a small number of continuous underlying dimensions, estimates your position on each, and reports the coordinates. The familiar version uses two — an economic axis and an authority axis — and draws you as a dot in a square.
The appeal is that it gives you a location rather than a category, and locations support comparison in an intuitive way. You can be slightly one way, strongly another, or genuinely on the line. Movement over time is easy to express: the dot moves. Two people who received the same label under a grouping model can be visibly far apart under a dimensional one.
The dimensional model also makes a falsifiable commitment. It claims the dimensions are approximately independent — that knowing your economic position tells you little about your authority position. That claim can be checked against real response data, and when it fails, the model is wrong in a way you can detect and fix. Models that make checkable claims are better than models that do not.
Where it strains: a dimensional model only reports what its axes can express. If a meaningful pattern in the population runs at an angle no axis captures, the model quietly flattens it. Two axes are famously too few for this reason, which is why our own test uses four — economic, social, state and world — as described in what a political spectrum test measures. Four is better. Four is also not a claim that four is the true number; it is a claim that four captures most of the structure that recurs reliably, at a length a person will actually finish.
The typological model: you are a member of a cluster
The typology family works from the other end. Rather than assuming dimensions in advance, it takes a large survey sample, runs a clustering procedure over the response patterns, and finds groups of people whose answers resemble each other more than they resemble anyone else’s. Then it describes each group and tells you which one you fall into.
This has a real advantage: it finds structure nobody specified. If a substantial group holds a combination that no axis would have predicted — strongly one way on economics, strongly the other way on social questions, and unusual on something else entirely — the clustering will surface it as its own type, because the data said so. A dimensional model would render that same group as an unremarkable dot in a quadrant.
Typologies also describe better. “You are in the group that tends to weigh obligation heavily and distrust concentrated power in any form” is more informative to most people than “you are at −34 and +12.”
Where it strains, in three specific ways:
The boundaries are softer than they look. Cluster membership is usually probabilistic. A great many people sit near an edge and could plausibly belong to either of two types. Presented as a clean assignment, that uncertainty disappears from view, and a 51/49 split reads with the same confidence as a 95/5 one.
It is a snapshot of a population, not a map of value space. The clusters come from who was surveyed, when, and where. Run the same procedure on a different population and the types change — sometimes substantially. That is not a defect, but it does mean a typology describes the sample’s structure at a moment, not a permanent feature of how humans hold values.
It struggles with movement. If your views shift moderately, a dimensional model shows a dot moving a moderate distance. A typology either keeps you in the same box, in which case it registered nothing, or flips you to a different box, which overstates what happened.
The comparison in one pass
| Dimensional (compass) | Typological (clusters) | |
|---|---|---|
| Output | Coordinates on fixed axes | Membership in a named group |
| Assumes | The axes, chosen in advance | Only that patterns exist |
| Best at | Comparison, nuance, tracking change | Description, capturing odd combinations |
| Fails by | Flattening structure no axis captures | Hiding boundary uncertainty; drifting with the sample |
| Answers | Where do I sit relative to others? | Who answers the way I do? |
What a good test does with both
Our test is dimensional at its core and typological at the surface, in that order, and the order is the whole point.
The four axis scores are the measurement. They are computed first, they are what the statements were written to estimate, and they are what you should treat as the finding. The named profile is derived afterwards by finding which region of that four-dimensional space you sit closest to. The profile is a summary of the coordinates, never a replacement for them.
This is also why the result carries a match percentage. It is the answer to the question a typology usually suppresses: how well does this label actually fit? A match in the high eighties means the description will feel written for you. A match in the low sixties means you are between descriptions, and reading only the named one will mislead you about yourself. Publishing that number is a small thing that prevents a large error.
Reading either kind without being taken in
A few habits travel well across both families.
Ask what the axes are made of. An axis is only as good as the statements feeding it. If a test will not show you its construct definitions, the axis label is decoration. Ours are set out in the methodology, including which statements are reverse-keyed and why.
Check whether the dimensions were assumed or found. Both are fine; they mean different things. Assumed dimensions are a theory being applied. Discovered ones are a pattern being reported. A test that presents assumed axes as if they were discovered is overselling.
Treat a type name as a compression, not a diagnosis. Any name that fits millions of people is lossy by construction. The useful question is not “is this label right?” but “which parts of this description land, and which parts do not?” The parts that do not are where the model lost you, and that is usually the most informative part of the whole result.
Be suspicious of a model that never puts you near an edge. Real distributions are dense in the middle. A test that hands nearly everyone a confident, distinctive, flattering position is optimising for how the result feels rather than for whether it is true. That tendency, and how to resist it in your own reading, is the subject of how to read your results without fooling yourself.
The upshot
A compass answers where am I relative to everyone else. A typology answers who else answers the way I do. Neither question is more legitimate; they are simply different questions, and a test that pretends to answer both with one number is answering neither well.
The combination that holds up is coordinates first, label second, with the fit between them stated out loud. The vertical axis in particular — the one running between personal liberty and collective authority — is where the two traditions most often disagree, and it repays a closer look on its own terms in the vertical axis explained.