Question

I have following query in sql server 2005 with a table

select t1.id, CONVERT(VARCHAR,t1.dt,103) date_1, CONVERT(VARCHAR,t2.dt,103) date_2, t1.hotel,
    t1.price price_1, t2.price price_2, t2.price - t1.price difference, ((t2.price - t1.price)/t1.price)*100 as Diff_percentage
from test t1
inner join
(
    select *
    from test
) t2
    on t1.hotel = t2.hotel
    and t1.dt < t2.dt and t2.dt=(SELECT MAX(dt) from TEST) and t1.dt=(SELECT MAX(dt-1) from TEST)

I want to use count if function within this query. Based on difference column. So that i can count “How many increased, How many decreased, How many same, How many unavailable”

1. COUNT IF Difference>0 //How many increased
2. COUNT IF Difference<0 //How many decreased
3. COUNT IF Difference=0 //How many same
4. COUNT IF Difference=”” //How many unavailable –Difference is blank.

DEMO : http://sqlfiddle.com/#!3/b6f37/29

Answer 1

If I wanted your existing query result I would rewrite the query as:

select t1.id, 
       CONVERT(VARCHAR,t1.dt,103) date_1, 
       CONVERT(VARCHAR,t2.dt,103) date_2, 
       t1.hotel,
       t1.price price_1, 
       t2.price price_2, 
       t2.price - t1.price difference, 
       ((t2.price - t1.price)/t1.price)*100 as Diff_percentage
  from test t1
  join (select max(dt) maxDt from test) d
    on t1.dt = d.maxDt-1
  join test t2
    on t2.hotel = t1.hotel
   and t2.dt = d.maxDt

To extend your original query to include hotels with missing rows:

select t1.id, 
       CONVERT(VARCHAR,t1.dt,103) date_1, 
       CONVERT(VARCHAR,t2.dt,103) date_2, 
       h.hotel,
       t1.price price_1, 
       t2.price price_2, 
       t2.price - t1.price difference, 
       ((t2.price - t1.price)/t1.price)*100 as Diff_percentage
  from (select distinct hotel from test) h
  cross join (select max(dt) maxDt from test) d
  left join test t1
    on t1.hotel = h.hotel
   and t1.dt = d.maxDt-1
  left join test t2
    on t2.hotel = h.hotel
   and t2.dt = d.maxDt

The above query would be much more efficient with a normalized HOTEL table (1 row per hotel) to replace the SELECT DISTINCT subquery.

To get your requested result, I would use:

select count(case when (t2.price-t1.price) < 0 then 1 end) decrease_count,
       count(case when (t2.price-t1.price) > 0 then 1 end) increase_count,
       count(case when (t2.price-t1.price) = 0 then 1 end) same_count,
       count(distinct t1.hotel) - count(case when (t2.price-t1.price) is not null then 1 end) unavailable_count
  from test t1
  left join (select max(dt) maxDt from test) d
    on t1.dt = d.maxDt-1
  left join test t2
    on t2.hotel = t1.hotel
   and t2.dt = d.maxDt

Again the above would be more efficient with a normalized HOTEL table. I would restructure the query more like the previous one: select from HOTEL cross joined to the MAX date query, then outer join to the TEST table twice for the data for the 2 dates. The unavailable count could then be measured more directly, counting the number of rows where the difference computation is NULL.

Here is the SQL Fiddle for all the queries, along with some extended test data.