Theory and Methods

Many well-known personality type tests sort people into one box or the other. Yet when you actually collect personality scores, most axes form a gentle hill with the most people near the middle. Cutting that hill into black and white puts nearly identical people, standing just either side of the line, into different types.

So this test first estimates where you sit on each axis as a continuous value, a coordinate, and only then maps it to a type for readability. The type is the doorway; the coordinate is the substance.

For each axis we calculate how confident we are that you lean toward one pole. If that confidence is below 75%, we do not force a call: the axis is shown as a boundary, and a "·" replaces its letter in your code.

Having many people on the boundary is natural, because a hill-shaped distribution puts the most people near the middle. A boundary axis can be seen as a wide one, where either side may show up depending on the situation.

• A1 S or M: Sadist (top) vs. Masochist (bottom)
• A2 Turn-on: Atmosphere (mind) vs. Direct (body)
• A3 Temperature: Honey (sweet) vs. Lash (cold)
• A4 Pace: Grooming (ritual) vs. Wild (impulse)
• A5 Brakes: Volatile (runaway) vs. Keel (in control) (inner face only)

Every question is an ordinary personality question: whether you recharge around people or alone, think abstractly or concretely, weigh feelings or consistency, plan ahead or improvise, and how much your feelings sway. We reread those leanings as S or M, turn-on, temperature, pace and brakes. The rereading rests on playful associations, such as people who run the room tending to take the lead; it is entertainment, not a research-backed link between personality and sexual preference.

The first four axes are measured for both the outer face (the play style others see) and the inner face (the desires you feel when you fantasize alone). The fifth, the brakes, is measured only by the inner-face questions. Putting outer and inner faces on the same axes lets you compare directly where they drift apart, and by how much.

The two faces are told apart by the viewpoint of each question. Outer-face questions ask how you act and come across when other people are around, such as "When I meet a lot of new people, I am usually the one who starts the conversation" or "I let people know my plans well in advance." Inner-face questions ask about feelings that linger afterward and how things feel or what runs through your mind on your own, such as "After I turn someone down, it stays on my mind for a while" or "On a day off with nothing planned, I like to go wherever my mood takes me."

During the test, the two kinds are mixed into one continuous set at a fixed rhythm, and no question is labeled as one or the other (the quick test is 12 outer and 8 inner questions, 20 in all). When people brace for a "who am I really?" question, ideals and self-image tend to creep in. We sort your natural answers into outer and inner item by item, estimate each, and only split them into two faces on the result page.

The idea of a second self that stays out of sight may remind you of the psychologist Carl Jung's notion of the shadow. This test does not adopt that theory as such, though; it simply measures two situations with the same question format and compares them.

• S or M
• Turn-on
• Temperature
• Pace
• Brakes

To estimate your coordinates from your answers, we use the graded response model from item response theory (IRT). For each question, it models which parts of the axis each of the seven answer options tends to come from, and Bayesian estimation then gives the most likely position along with its margin of error.

The next question is the one that adds the most information to the current estimate. Each axis stops once its error drops below 0.35 or it reaches 10 questions (with a minimum of 6), so the outer face takes between 24 and 40 questions depending on the person. The inner face uses the same 20 questions for everyone.

The calculations, including the question order, follow fixed steps, so the same answers always produce the same questions and results. We test them against a separate reference implementation.

For each axis we calculate the probability that your direction differs between outer and inner faces. If the directions differ and that probability is 75% or higher, the axis flips. If the direction holds but your position moves by 1 or more, by more than measurement error would explain, the axis sways.

Those results combine into eleven patterns: eight where a single axis flips, one where two or more flip, one with sways only, and a consistent pattern where little changes.

• See the gap patterns

We measure how far your coordinates sit from the center of everyone's distribution using a distance that accounts for correlations between axes (the Mahalanobis distance). With four axes, the square of that distance follows a chi-square distribution with four degrees of freedom, so we can calculate what share of people sit farther from the center than you do.

High dimensions have a counterintuitive property. Only about 0.7% of people sit very close to the center (a distance under 0.5). If the axes were independent, each axis would give you a 50% chance of landing in its middle half, yet only 0.5 to the fourth power, about 6%, would land in the middle on all four. The other 94% or so are on the minority side of at least one axis. Someone who is average on every axis is actually the rare one.

The SM Test's rarity shows this property as it is, without inventing numbers. Until enough real data has been collected (around 1,000 responses), the figures are based on a theoretical distribution with mean 0 and spread 1, and they come labeled as theoretical along with the number of responses behind them.

If some of your axes sit near the boundary, a single type cannot describe you fully. So we multiply the confidence on each axis to get your share of each of the 16 types, and show the second and third largest as near types, together with the situations where those sides tend to appear.

Types with a share below 5% are not shown. Listing every faint possibility would only blur the picture of who you are.

• "It splits people in two, but real scores cluster in the middle." We measure continuous values and label axes that could go either way as boundaries.
• "Many people get a different type when they retake it." We show the margin of error and the strength of each lean, and we do not hide the axes that wobble.
• "It ignores how easily feelings sway." We include a brakes axis, built on how much your feelings sway and measured by the inner-face questions.
• "The descriptions would fit anyone." The text is assembled from blocks tied to your axis values, and axes near the boundary get more cautious wording.
• "Compatibility claims have no basis." We do not predict whether a relationship will work; we only show how two people's coordinates relate.

The questions are adapted from public-domain items in the IPIP (International Personality Item Pool), rewritten so they read naturally. We record the source item for every question.

• IPIP (International Personality Item Pool)