How to read your percentile
A percentile counts only the submissions below you. How that differs from the average and the median, and what selection bias does to your position.
Your percentile is one thing only: the share of stored submissions that scored below you. It is not a grade, not your rank among people your age, and not a statement about the general public. It also moves when the mix of people taking the test changes, even if your own answers never do.
That last property is the one worth understanding, because it is what separates a percentile that means something from one that only looks like it does.
Three numbers that answer three different questions
The average (mean) adds every score and divides by the number of submissions. It answers "what is the total, spread evenly?" and it is dragged around by extremes: a run of deliberate zeroes pulls it down hard even though it barely changes what a typical result looks like.
The median is the middle value once every score is lined up in order — half sit above it, half below. It answers "what does a typical submission look like?" and it shrugs off extremes, because moving the lowest score from 12 to 0 does not change which value sits in the middle.
Your percentile is your own position in that line-up. It answers "how many people are below me?" and nothing else.
A worked example with ten scores
These ten numbers are invented purely to show the arithmetic — they are not our data. Say ten people score:
12, 41, 55, 58, 61, 63, 66, 72, 88, 100
The mean is 616 divided by 10, or 61.6. The median sits between the fifth and sixth values, 61 and 63, so it is 62. Someone who scored 66 has six of the ten below them, putting them at the 60th percentile.
Now add three joke entries of 10 — three people who clicked their way down the column for a screenshot. The list becomes thirteen scores, and here is what happens to each number:
| Statistic | Before | After 3 joke entries | Change |
|---|---|---|---|
| Mean | 61.6 | 49.7 | −11.9 points |
| Median | 62 | 58 | −4 points |
| Percentile of a 66 | 60th | 69th | +9 percentile points |
Three bad rows dragged the average down by almost twelve points, a fifth of the useful range, while the median moved four. Meanwhile the person who scored 66 rose nine percentile points without answering a single question differently.
That table is the whole argument of this article. The average is fragile, the median is sturdier, and your percentile is a statement about the crowd as much as about you — which is why our statistics page shows the full distribution instead of leading with one headline number.
The distribution is shaped like a bell, and that means less than it looks
Plot the submissions in five-point buckets and you get a hump: few scores at the very top, few at the very bottom, most piled in the middle. It is tempting to read that as a natural bell curve of human experience. It is not one.
A bell curve gains its meaning from being a sample of a defined population — you draw people at random from a group you can name, and the shape tells you about that group. Nobody drew this sample. It is composed entirely of people who searched for a purity test, found this site, decided to answer a hundred personal questions, reached the end, and pressed submit. Every one of those steps filters who ends up in the data, and none of them filters at random.
Four filters between the public and the graph
Who arrives. Searching for "rice purity test" is itself a behaviour with a demographic. The current audience skews young and skews toward people who saw the test on social media, which is not the age distribution of any country.
Who finishes. A hundred questions takes several minutes. People who abandon partway are not a random slice — someone startled by the explicit section in the back half is more likely to quit there, which quietly removes a particular kind of respondent.
Who submits. Seeing your score and choosing to record it is a separate decision from taking the test, and people with scores they find embarrassing in either direction make that decision differently from everyone else.
Who is joking. Straight-down-the-column zeroes and untouched hundreds both accumulate at the edges. Repeat submissions from the same person add to it: one curious visitor can take the test four times in an evening, and the data cannot tell that apart from four people.
The direction of the combined bias is not even predictable. Some filters push the recorded scores up, others push them down. What is predictable is that the result is a picture of submissions, not of people.
What our percentile figure claims, precisely
It claims: of the submissions stored for this specific test, this percentage recorded a score strictly lower than yours.
It does not claim: anything about the population of your country, your age group, your campus or your generation; that respondents told the truth; that each row is a distinct person; or that the same figure would appear if the same people took a differently worded version — see purity test versus innocence test for why scales do not transfer.
One detail of the arithmetic matters. The percentile counts submissions strictly below your score, so everyone who tied with you is excluded from it. That is why we show the number of people who scored exactly the same as you as a separate figure rather than folding it in — at a common score, ties can be a large group, and burying them inside a percentile would make your position look more precise than it is.
Using it without being misled
Read the direction, not the digits. "Lower than most submissions" is supportable; "the 63rd percentile, up from the 61st last month" is reading noise as signal.
Use the buckets. The five-point bands on the distribution are roughly the resolution the data can carry. Differences inside a band are not real.
Compare yourself to yourself. The one comparison free of selection bias is your own score across time, since the sample is one person and the question set is identical. Your saved history keeps those results locally on your device for exactly that.
Ignore a two-point gap with a friend. It is one ambiguous question answered on a different day.
If you have not got a number yet, take the test and the comparison figures appear with your result. If you have one and want to know what it says about which sections you checked, every score has its own page, and what counts as a good score explains why the ranking instinct is the wrong one to bring to it.