Select top 10 percentage

Question

Hi,
I want to get the top 10% of salaries in employees table. But I got error:
SQL> select top 10 percent salary
  2  from employees
  3  order by salary desc;
ORA-00923: FROM keyword not found where expectedHow can I get the top 10% percent?
Thanks a lot

Answer 1

Hi,
998093 wrote:
... What if I want to get, for example, the employees who are in the top 12% of the salaries? here we cannot use deciles (maybe we can but it won't be very nice).Actually, you can. How nice it will be depends on your data and your exact requirements.
Earlier, we said that the top 10% was the 10th of 10 buckets.
We could say that the top 12% is the last 12 of 100 buckets:
WITH     got_tenth     AS
     SELECT     last_name, salary
     ,     NTILE (100) OVER (ORDER BY salary)     AS hundredth
     FROM     hr.employees
SELECT       last_name, salary
FROM       got_tenth
WHERE       hundredth     > 100 - 12
ORDER BY  salary     DESC
;I won't clutter up this message by showing the results for every query; I'll just report the total number of rows. The query I posted earlier (for the top 10%) produced 10 rows of output; the query immediately above produces 12. That's starting to bother me. There are 107 rows in hr.employees. (They all have a salary; don't forget to deal with NULLs in general.) The top 10% should contain 107 * .1 = 10.7 rows. Of course, this has to be rounded to an integer, so it would have been a little better if the top 10% showed 11 rows (since 11 is closer to 10.7 than 10 is), but it's not a real big problem. Likewise, the top 12% should have 107 * .12 = 12.84 rows, so 13 rows of output might be better, but 12 rows isn't too bad.
Now say we want to simplify the condition "WHERE     hundredth     > 100 - 12". Instead of taking the last 12 buckets in ascending order, let's take the first 12 buckets in descending order:
WITH     got_tenth     AS
     SELECT     last_name, salary
     ,     NTILE (100) OVER (ORDER BY salary DESC)     AS hundredth
     FROM     hr.employees
SELECT       last_name, salary
FROM       got_tenth
WHERE       hundredth     <= 12
ORDER BY  salary     DESC
;The result set now contains 19 rows! Why? Because NTILE puts extra items (when there are extras) in the lower-numbered buckets. When there were 10 buckets, then buckets 1 trough 7 had 11 items each, and buckets 8 through 10 had 10 items. We were only dispolaying bucket #10, so we only saw 10 items. Now that there are 100 buckets, buckets 1 through 7 will have 2 items each, and buckets 8 through 100 will have 1 item each. When we numbered the buckets in ascending order, and took the last 12 buckets, we wound up with 12 * 1 = 12 rows, but when we numbered the buckets in descending order and took the first 12 buckets, we got (7 * 2) + (5 * 1) = 19 rows.
When you have R rows and B buckets, and R/B doesn't happen to be an integer, then NTILE will distribute the rows as equally as possible, but some rows will have 1 item more than others. When the ratio R/B is high, that might not be very significant, but when R/B gets close to 1, then you have situations like the one above, where the bigger buckets, although they have only 1 more item than the smaller buckets, are twice as big. Even worse is the situation where R/B is less than 1; then the last buckets will have 0 items.
So, as you noticed, NTILE isn't a good solution for all occasions. Here's a more accurate way, using the analytic ROW_NUMBER and COUNT functions:
WITH     got_analytics     AS
     SELECT     last_name, salary
     ,     ROW_NUMBER () OVER (ORDER BY  salary  DESC)     AS r_num
     ,     COUNT (*)     OVER ()                                AS n_rows
     FROM     hr.employees
SELECT       last_name, salary
FROM       got_analytics
WHERE       r_num / n_rows <= 12 / 100
ORDER BY  salary     DESC
;This produces 12 rows.
If you want 13 rows (because 107 * .12 = 12.84 is closer to 13), then you can change the main WHERE clause like this:
WITH     got_analytics     AS
     SELECT     last_name, salary
     ,     ROW_NUMBER () OVER (ORDER BY  salary  DESC)     AS r_num
     ,     COUNT (*)     OVER ()                                AS n_rows
     FROM     hr.employees
SELECT       last_name, salary
FROM       got_analytics
WHERE       r_num     <= ROUND (n_rows * 12 / 100)
ORDER BY  salary     DESC
;