nth percentile calculations in postgresql

Question

I've been surprisingly unable to find an nth percentile function for postgresql.

I am using this via mondrian olap tool so i just need an aggregate function which returns a 95th percentile.

I did find this link:

http://www.postgresql.org/message-id/162867790907102334r71db0227jfa0e4bd96f48b8e4@mail.gmail.com

But for some reason the code in that percentile function is returning nulls in some cases with certain queries. I've checked the data and there's nothing odd in the data that would seem to cause that!

Have you tried `percent_rank()`, `cume_dist()` or `ntile()` ? Details [here](http://www.postgresql.org/docs/current/static/functions-window.html). — Ihor Romanchenko, Jan 14 '13 at 10:24
See this as well: http://stackoverflow.com/a/14309370/330315 — , Jan 14 '13 at 10:28
Ah, I had found that other stack overflow entry, but hadnt created the array sort function. It works now. So this question can be closed or deleted. — Codek, Jan 14 '13 at 10:38

alfonx · Answer 1 · 2020-10-15T21:36:46.430

With PostgreSQL 9.4 there is native support for percentiles now, implemented in Ordered-Set Aggregate Functions:

percentile_cont(fraction) WITHIN GROUP (ORDER BY sort_expression) 
continuous percentile: returns a value corresponding to the specified fraction in the ordering, interpolating between adjacent input items if needed
percentile_cont(fractions) WITHIN GROUP (ORDER BY sort_expression)
multiple continuous percentile: returns an array of results matching the shape of the fractions parameter, with each non-null element replaced by the value corresponding to that percentile

See the documentation for more details: http://www.postgresql.org/docs/current/static/functions-aggregate.html

and see here for some examples: https://github.com/michaelpq/michaelpq.github.io/blob/master/_posts/2014-02-27-postgres-9-4-feature-highlight-within-group.markdown

CREATE TABLE aa AS SELECT generate_series(1,20) AS a;
--SELECT 20

WITH subset AS (
    SELECT a AS val,
        ntile(4) OVER (ORDER BY a) AS tile
    FROM aa
)
SELECT tile, max(val)
FROM subset GROUP BY tile ORDER BY tile;

 tile | max
------+-----
    1 |   5
    2 |  10
    3 |  15
    4 |  20
(4 rows)

The documentation does not have an example. Would you mind updating your answer with a few simple examples? It's weird but Googling has only yielded some complex examples. — tinkerr, Jan 16 '16 at 18:52
@tinkerr I suggest https://leafo.net/guides/postgresql-calculating-percentile.html — Zachary Ryan Smith, Apr 20 '22 at 15:56

score 26 · Answer 2 · edited Mar 25 '18 at 09:42

The ntile function is very useful here. I have a table test_temp:

select * from test_temp

score
integer
3
5
2
10
4
8
7
12

select score, ntile(4) over (order by score) as quartile from test_temp;

score    quartile
integer  integer
2        1
3        1
4        2
5        2
7        3
8        3
10       4
12       4

ntile(4) over (order by score) orders the columns by score, splits it into four even groups (if the number divides evenly) and assigns the group number based on the order.

Since I have 8 numbers here, they represent the 0th, 12.5th, 25th, 37.5th, 50th, 62.5th, 75th and 87.5th percentiles. So if I only take the results where the quartile is 2, I'll have the 25th and 37.5th percentiles.

with ranked_test as (
    select score, ntile(4) over (order by score) as quartile from temp_test
)
select min(score) from ranked_test
where quartile = 2
group by quartile;

returns 4, the third highest number on the list of 8.

If you had a larger table and used ntile(100) the column you filter on would be the percentile, and you could use the same query as above.

nth percentile calculations in postgresql

2 Answers2

Linked