Search notes:

DBMS_STATS.GATHER_xxx_STATS: parameter ESTIMATE_PERCENT

Create table with one million records:

create table tq84_estimate_percent_test (id number, txt varchar2(10));

insert into  tq84_estimate_percent_test
select
   level,
   dbms_random.string('A', 10)
from
   dual connect by level <= 1 * 1000 * 1000;
  
commit;

Gather statistics for the table using approximately 10 percent (value of estimate_percent) of the data in the table:

begin
   dbms_stats.gather_table_stats (
      ownname          =>    user,
      tabname          =>   'TQ84_ESTIMATE_PERCENT_TEST',
      estimate_percent => 10
   );
end;
/

It turns out, statistics were calculated from a basis (sample size) of 99873 records, a little less than 10 percent of the million records in the table.

Because the sample size is a bit too little, the estimated number of records (num_rows) in the table is also reported a bit too little.

select
   sta.num_rows,
   sta.sample_size
from
   user_tab_statistics sta
where
   sta.table_name = 'TQ84_ESTIMATE_PERCENT_TEST';
--
--   NUM_ROWS SAMPLE_SIZE
-- ---------- -----------
--    998730       99873

drop table tq84_estimate_percent_test;

DBMS_STATS.GATHER_xxx_STATS: parameter ESTIMATE_PERCENT

See also