Methodology

The illustro personality type test is a Big Five instrument. It asks 36 questions, scores five traits, and names the pattern. This page shows every question, where it came from, and exactly how the numbers on your result were worked out. Nothing here is proprietary and nothing is withheld.

What the test measures

The five traits are the ones used in mainstream personality research: Openness, Conscientiousness, Extraversion, Agreeableness, and Emotional Range. Emotional Range is the trait psychologists usually call Neuroticism. We use the plainer name because the research label reads as a judgement, and the trait is not one.

The archetype name is a summary, not a separate finding. It is decided by four of the five traits, each split at the median of a large research sample. Four traits with two levels each gives sixteen patterns and sixteen names. Emotional Range is shown as a fifth bar and never forms part of the archetype, because a five-way split would produce thirty-two names and split the traits far more finely than a 36-question test can support.

Where the questions come from

Every statement is a public-domain item from the International Personality Item Pool (IPIP), assembled by Lewis R. Goldberg and free for anyone to use. We take the 36 from the IPIP-NEO-120, John A. Johnson's 120-item public-domain inventory, and we use his published scoring key for the trait and facet each item belongs to and the direction it is keyed.

The IPIP publishes items as short phrases, such as "Worry about things." We show each one as a first-person sentence by adding "I" and lowercasing the first letter: "I worry about things." Nothing else is reworded. The table below prints the item exactly as we show it, along with its IPIP-NEO-120 item number so you can check it against Johnson's published item list and scoring key at https://osf.io/ycvdk/.

Seven items measure each of Openness, Conscientiousness, Extraversion, and Agreeableness. Eight measure Emotional Range. The extra item is there for keying balance: it makes 18 forward-keyed and 18 reverse-keyed statements across the test, so agreeing with everything cannot push every bar in the same direction. It sits on Emotional Range because Emotional Range is the one trait that does not decide the archetype, so lengthening it cannot shift anyone into a different type. Within each trait the split is as even as the item pool allows: four forward and three reversed, or the reverse, and four of each on Emotional Range.

We left some IPIP-NEO-120 items out on purpose. Nothing about politics, religion, drinking, or self-harm appears in this test. Those items are perfectly good psychometrics, and they are not a fair thing to put in front of someone who arrived from an advert.

The 36 items

Forward means agreeing scores higher on the trait. Reversed means agreeing scores lower, and the score is flipped before it is added.

How your answers are scored

Each answer is a number from 1 for Strongly disagree to 5 for Strongly agree. Forward-keyed items keep that number. Reverse-keyed items are flipped, so a 5 becomes a 1 and a 2 becomes a 4. The flipped and unflipped numbers are then added up within each trait. Four traits run from 7 to 35. Emotional Range, with its eight items, runs from 8 to 40.

A higher Emotional Range score means you report reacting more strongly and more often. A lower one means you report reacting less. Neither end is the good end, which is why we do not use the older name.

Scoring happens in your browser. Your individual answers are not sent anywhere.

The norm sample

Percentiles and rarity come from a published research dataset, not from illustro takers. We use the IPIP-NEO-120 response file from John A. Johnson's IPIP-NEO data repository on the Open Science Framework.

• Citation: Johnson, J. A. (2014). Measuring thirty facets of the Five Factor Model with a 120-item public domain inventory: Development of the IPIP-NEO-120. Journal of Research in Personality, 51, 78-89.
• Data: IPIP120.dat, downloaded on 24 August 2026 from https://osf.io/tbmh5/
• Records in the file: 619,150
• Records used: 530,370
• Records dropped: 88,780

The 2014 paper reports an earlier sample of 307,313 cases from the same collection. Collection carried on after the paper was published, so the file is larger now, and the copy we downloaded on 24 August 2026 holds 619,150 records. After the completeness filter below, our norms rest on 530,370 of them.

Other screens on the site describe this as a sample of 300,000 or more. That is a floor we hold to, not a count, and the sample we use is well above it. The exact figure is the one printed here, and it is the one every number on this page was computed from.

We kept every record that answered all 36 of our items with a value from 1 to 5. The file codes an unanswered item as 0, and we dropped a record if any of our 36 was unanswered. That is the only filter. We did not drop anyone by age, country, or response pattern, and we did not weight the sample.

Only our 36 items are used. The distributions below are the sums of these 36 items in that sample, so a percentile compares you against people answering the same questions, not against a longer test rescaled.

One honest difference. In the research sample people answered these statements on a five-point accuracy scale, from Very Inaccurate to Very Accurate. illustro shows a five-point agreement scale, from Strongly disagree to Strongly agree. Both scales have five ordered steps and a neutral middle, and we treat step 1 as step 1 and step 5 as step 5. It is a real difference in wording and it is worth knowing about when you read a percentile. We have not measured how much difference it makes, so read a percentile as a good guide to where you sit rather than as an exact place in a queue.

How percentiles are computed

For each trait we counted how many people in the norm sample scored at or below every achievable sum, from the lowest to the highest. That gives a lookup table, one entry per possible score.

Your percentile is the share of the sample scoring below your sum, plus half the share scoring exactly your sum. Splitting the tie in half is standard practice and stops a common score from claiming everyone who shares it. A percentile of 70 means roughly 70 out of 100 people in that sample scored lower than you did.

We report percentiles as whole numbers from 1 to 99. A sum at the very bottom or the very top of the table would otherwise round to 0 or 100, and nobody is behind everyone or ahead of everyone, so those are reported as 1 and 99. It affects only the sums at the extreme ends of each table, and on some traits only the bottom end, which is between 0.5% and 1.1% of the norm sample, depending on the trait.

Sample medians on our 36 items:

How the archetype is decided

Openness, Conscientiousness, Extraversion, and Agreeableness are each compared with the sample median in the table above. That is the median of the sample, not the middle of the scale: half the sample scored above it and half below. A score at or above the median counts as high, below it counts as low. The four highs and lows in that order give one of sixteen patterns, and each pattern has a name.

A median split is a blunt instrument, and we would rather say so than pretend otherwise. Someone one point above the median and someone one point below get different letters and different names, even though their trait bars look nearly identical. The bars and the percentiles carry the detail. The name is the headline.

How rarity is computed

We applied the same four median splits to all 530,370 people in the norm sample and counted how many landed in each of the sixteen cells. That count is the rarity figure. When the result screen says "6 in 100 people", it means 6 in 100 people in this research sample share your pattern. It does not mean 6 percent of illustro takers, and we never present taker counts we do not have.

The sixteen cells are not equally sized, and no honest Big Five test would produce sixteen equal cells. The traits are mildly correlated in the population, so people who score high on one tend to score a little higher on the others. That is why the all-high pattern is the most common one and the all-low pattern is the third most common. Percentages are rounded to one decimal place and add up to 100 give or take rounding.

Limits worth knowing

• The norm sample is people who chose to take a personality inventory online. It is large, it is not a random sample of the population, and it skews younger and more English-speaking than the world does.
• The test is self-report. It measures how you describe yourself today, not how you behave, and not how you will describe yourself in five years.
• Seven or eight items per trait is short. A 36-question test is less precise than a 120-question one, which is the trade we make for a test you can finish in one sitting.
• Results are indicative only. This is not a clinical assessment and it does not diagnose anything.

Reproducing our numbers

Everything on this page comes from two scripts that read the raw data file and write out the tables the site uses. They are published with the site at scripts/norms/, and scripts/norms/README.md gives the dataset link, the exact commands, and what each one should print. Download IPIP120.dat from https://osf.io/tbmh5/, run scripts/norms/compute_norms.py over it, and you should get the same sample size, the same medians, and the same sixteen percentages. If you get something different, write to support@illustro.tech and tell us what you found.