Need to find the max term_code value for each person in my table

Question

I have a list of people with multiple term_code values. I need to find the max for each person that has a 201930 or 201940 record. I need to take the 201930 if there is both such as the case with Bob. I then need to return other fields for each person with that term. Only the red records will be returned. Fred should not show up in the output.

Here is the query I currently have, but it grabs the 201940 record for Bob. The total number of records is correct with it, but it gets some incorrect values.

SELECT userid, term_code, race, gender
FROM mytable a JOIN (
                  SELECT userid, MAX(term_code) AS term_code
                  FROM mytable
                  WHERE term_code <= '201940'  
                  GROUP BY userid
                ) b ON (a.userid = b.userid and a.term_code = b.term_code)
WHERE term_code IN ('201930', '201940');

Using this line seems logical to me and it gets the right value for Bob, but it cuts my results by about 30%.

WHERE term_code <= COALESCE ('201930','201940') 

Any suggestions?

Answer 1

With NOT EXISTS :

select m.* from mytable m
where m.term_code = (
  case when not exists (select 1 from mytable where userid = m.userid and term_code = 201930) 
    then 201940
    else 201930
  end
)

Or if you only want the userid and term_code then you can do it with simple aggregation:

select userid, min(term_code) term_code
from mytable 
where term_code in (201930, 201940)
group by userid

If you want the full row from the table then you can join to the table:

select m.*
from mytable m inner join (
  select userid, min(term_code) term_code
  from mytable 
  where term_code in (201930, 201940)
  group by userid
) t on t.userid = m.userid and t.term_code = m.term_code

Or with ROW_NUMBER() window function:

select t.userid, t.term_code, t.race, t.gender
from (
  select m.*,
    row_number() over (partition by userid order by term_code) rn  
  from mytable m
  where m.term_code in (201930, 201940)
) t 
where t.rn = 1

See the demo .
Results:

> USERID | TERM_CODE | RACE | GENDER
> :----- | --------: | :--- | :-----
> Bob    |    201930 | null | null  
> Tim    |    201940 | null | null

Answer 2

with t  (USERID,   term_code ) as (
  select 'Bob',   201601 from dual union all 
  select 'Bob',   201605 from dual union all   
  select 'Bob',   201609  from dual union all         
  select 'Bob',   202930 from dual union all          
  select 'Bob',   202940 from dual union all          
  select 'Bob',   202950 from dual union all  

  select 'Tom',   202940  from dual union all         
  select 'Tom',   201605 from dual union all          
  select 'Tom',   201609  from dual union all  

  select 'Mac',   201601 from dual union all          
  select 'Mac',   201605 from dual union all          
  select 'Mac',   201609 from dual 
)
select userid, term_code from
(
SELECT t.*
, sum(case when term_code in (202930, 202940) then 1 end) over (partition by userid order by term_code) rnk
FROM t
)
where rnk = 1 

USE  TERM_CODE
--- ----------
Bob     202930
Tom     202940

Please note the term_code values are not the same except for the ones that you are interested in. For each USERID the term_code is ranked based on your condition using a SUM() analytic function. Once that is worked out, the outer query simply filters out the 1st ranked row produced in the inner query.