How did you develop and validate your tests?

We began by researching each type using the existing literature, including studies, books, forums, and more. Then we ran our own comprehensive testing alongside in-depth interviews with many people of every type, subtype, and instinct. We used the resulting dataset to conduct a feature analysis to determine the most predictive behaviors and statements for each type, iterating on thousands of questions in the process to discover which ones would provoke the most polarizing reactions (strong agreement or strong disagreement). Once we gathered a large enough dataset of responses and analyzed them alongside respondents whose types we were certain of, we applied our proprietary distributed clustering and logistic regression models to determine the probability that a person is a given type, subtype, and instinct.

In our test development and validation, we focused heavily on our tests’ ability to discern common mistypes and outliers, zeroing in on the core aspects of each type. For example, Twos and Social Nines can look similar in their behaviors despite having different underlying rationales for acting as they do. We iterated on thousands of test questions to isolate those that best identify the highest-signal differentiators for each type and wing, and between commonly mistyped pairs. Tactics like subtle word choices and framing statements as positives or negatives make significant differences in whether certain types will agree or disagree with test questions.


Our tests are dynamic, so the type and number of questions changes based on how you answer. If your responses narrow you down to two similar types, the questions will shift to delineate between them. The tests will end when the model reaches a set threshold of confidence in your type, wing, and/or instinct, which means that the number of questions you receive may not be the same as the number that someone else gets.