Count Response once in 30 days SQL - sql

Question

If I have a customer respond to the same survey in 30 days more than once, I only want to count it once. Can someone show me code to do that please?
create table #Something
(
CustID Char(10),
SurveyId char(5),
ResponseDate datetime
)
insert #Something
select 'Cust1', '100', '5/6/13' union all
select 'Cust1', '100', '5/13/13' union all
select 'Cust2', '100', '4/20/13' union all
select 'Cust2', '100', '5/22/13'
select distinct custid, SurveyId, Count(custid) as CountResponse from #Something
group by CustID, SurveyId
The above code only gives me the total count of Response, not sure how to code to count only once per 30 day period.
The output I'm looking for should be like this:
CustomerID SurveyId CountResponse
Cust1 100 1
Cust2 100 2

Answer 1

Going on the theory that you want your periods calculated as 30 days from the first time a survey is submitted, here is a (gross) solution.
declare #Something table
(
CustID Char(10),
SurveyId char(5),
ResponseDate datetime
)
insert #Something
select 'Cust1', '100', '5/6/13' union all
select 'Cust1', '100', '5/13/13' union all
select 'Cust1', '100', '7/13/13' union all
select 'Cust2', '100', '4/20/13' union all
select 'Cust2', '100', '5/22/13' union all
select 'Cust2', '100', '7/20/13' union all
select 'Cust2', '100', '7/24/13' union all
select 'Cust2', '100', '9/28/13'
--SELECT CustID,SurveyId,COUNT(*) FROM (
select a.CustID,a.SurveyId,b.ResponseStart,--CONVERT(int,a.ResponseDate-b.ResponseStart),
CASE
WHEN CONVERT(int,a.ResponseDate-b.ResponseStart) > 30
THEN ((CONVERT(int,a.ResponseDate-b.ResponseStart))-(CONVERT(int,a.ResponseDate-b.ResponseStart) % 30))/30+1
ELSE 1
END CustomPeriod -- defines periods 30 days out from first entry of survey
from #Something a
inner join
(select CustID,SurveyId,MIN(ResponseDate) ResponseStart
from #Something
group by CustID,SurveyId) b
on a.SurveyId=b.SurveyId
and a.CustID=b.CustID
group by a.CustID,a.SurveyId,b.ResponseStart,
CASE
WHEN CONVERT(int,a.ResponseDate-b.ResponseStart) > 30
THEN ((CONVERT(int,a.ResponseDate-b.ResponseStart))-(CONVERT(int,a.ResponseDate-b.ResponseStart) % 30))/30+1
ELSE 1
END
--) x GROUP BY CustID,SurveyId
At the very least you'd probably want to make the CASE statement a function so it reads a bit cleaner. Better would be defining explicit windows in a separate table. This may not be feasible if you want to avoid situations like surveys returned at the end of period one followed by another in period two a couple days later.
You should consider handling this on input if possible. For example, if you are identifying a customer in an online survey, reject attempts to fill out a survey. Or if someone is mailing these in, make the data entry person reject it if one has come within 30 days.
Or, along the same lines as "wild and crazy", add a bit and an INSERT trigger. Only turn the bit on if no surveys of that type for that customer found within the time period.
Overall, phrasing the issue a little more completely would be helpful. However I do appreciate the actual coded example.

Answer 2

I'm not a SQL Server guy, but in Oacle if you subtract integer values from a 'date', you're effectively subtracting "days," so something like this could work:
SELECT custid, surveyid
FROM Something a
WHERE NOT EXISTS (
SELECT 1
FROM Something b
WHERE a.custid = b.custid
AND a.surveyid = b.surveyid
AND b.responseDate between a.responseDate AND a.responseDate - 30
);
To get your counts (if I udnerstand what you're asking for):
-- Count of times custID returned surveyID, not counting same
-- survey within 30 day period.
SELECT custid, surveyid, count(*) countResponse
FROM Something a
WHERE NOT EXISTS (
SELECT 1
FROM Something b
WHERE a.custid = b.custid
AND a.surveyid = b.surveyid
AND b.responseDate between a.responseDate AND a.responseDate - 30
)
GROUP BY custid, surveyid
UPDATE: Per the case raised below, this actually wouldn't quite work. What you should probably do is iterate through your something table and insert the rows for the surveys you want to keep in a results table, then compare against the results table to see if there's already been a survey received in the last 30 days you want considered. I could show you how to do something like this in oracle PL/SQL, but I don't know the syntax off hand for SQL server. Maybe someone else who knows sql server wants to steal this strategy to code up an answer for you, or maybe this is enough for you to go on.

Answer 3

Call me wild and crazy, but I would solve this problem by storing more state with each survey. The approach I would take is to add a bit type column that indicates whether a particular survey should be counted (i.e., a Countable column). This solves the tracking of state problem inherent in solving this relationally.
I would set values in Countable to 1 upon insertion, if no survey with the same CustID/SurveyId can be found in the preceding 30 days with a Countable set to 1. I would set it to 0, otherwise.
Then the problem becomes trivially solvable. Just group by CustID/SurveyId and sum up the values in the Countable column.
One caveat of this approach is that it imposes that surveys must be added in chronological order and cannot be deleted without a recalculation of Countable values.

Answer 4

Here's one way to handle it I believe. I tested quickly, and it worked on the small sample of records so I'm hopeful it will help you out. Best of luck.
SELECT s.CustID, COUNT(s.SurveyID) AS SurveyCount
FROM #something s
INNER JOIN (SELECT CustID, SurveyId, ResponseDate
FROM (SELECT #Something.*,
ROW_NUMBER() OVER (PARTITION BY custid ORDER BY ResponseDate ASC) AS RN
FROM #something) AS t
WHERE RN = 1 ) f ON s.CustID = f.CustID
WHERE s.ResponseDate BETWEEN f.ResponseDate AND f.ResponseDate+30
GROUP BY s.CustID
HAVING COUNT(s.SurveyID) > 1

Answer 5

Your question is ambiguous, which may be the source of your difficulty.
insert #Something values
('Cust3', '100', '1/1/13'),
('Cust3', '100', '1/20/13'),
('Cust3', '100', '2/10/13')
Should the count for Cust3 be 1 or 2? Is the '2/10/13' response invalid because it was less than 30 days after the '1/20/13' response? Or is the '2/10/13' response valid because the '1/20/13' is invalidated by the '1/1/13' response and therefore more than 30 days after the previous valid response?

Answer 6

The code below is one approach which yields your example output. However, if you add a select 'Cust1', '100', '4/20/13', the result will still be Cust1 100 1 because they are all within 30 days of each prior survey response and so only the first one would be counted. Is this the desired behavior?
SELECT CustID, SurveyID, COUNT(*) AS CountResponse
FROM #SurveysTaken
WHERE (NOT EXISTS
(SELECT 1
FROM #SurveysTaken AS PriorSurveys
WHERE (CustID = #SurveysTaken.CustID)
AND (SurveyId = #SurveysTaken.SurveyId)
AND (ResponseDate >= DATEADD(d, - 30, #SurveysTaken.ResponseDate))
AND (ResponseDate < #SurveysTaken.ResponseDate)))
GROUP BY CustID, SurveyID
Alternatively, you could break the year into arbitrary 30 day periods, resetting with each new year.
SELECT CustID, SurveyID, COUNT(*) AS CountResponse
FROM (SELECT DISTINCT CustID, SurveyID, YEAR(ResponseDate) AS RepsonseYear,
DATEPART(DAYOFYEAR, ResponseDate) / 30 AS ThirtyDayPeriod
FROM #SurveysTaken) AS SurveysByPeriod
GROUP BY CustID, SurveyID
You could also just go by month.
SELECT CustID, SurveyID, COUNT(*) AS CountResponse
FROM (SELECT DISTINCT CustID, SurveyID, YEAR(ResponseDate) AS ResponseYear,
MONTH(ResponseDate) AS ResponseMonth
FROM #SurveysTaken) AS SurveysByMonth
GROUP BY CustID, SurveyID
You could use 30 day periods from an arbitrary epoch date. (Perhaps by pulling the date the survey was first created from another query?)
SELECT CustID, SurveyID, COUNT(*) AS CountResponse
FROM (SELECT DISTINCT CustID, SurveyID, DATEDIFF(D, '1/1/2013', ResponseDate) / 30 AS ThirtyDayPeriod
FROM #SurveysTaken) AS SurveysByPeriod
GROUP BY CustID, SurveyID
One final variation on arbitrary thirty periods is to base them on the first time the customer ever responded to the survey in question.
SELECT CustID, SurveyID, COUNT(*) AS CountResponse
FROM (SELECT DISTINCT CustID, SurveyID, DATEDIFF(DAY,
(SELECT MIN(ResponseDate)
FROM #SurveysTaken AS FirstSurvey
WHERE (CustID = #SurveysTaken.CustID)
AND (SurveyId = #SurveysTaken.SurveyId)), ResponseDate) / 30 AS ThirtyDayPeriod
FROM #SurveysTaken) AS SurveysByPeriod
GROUP BY CustID, SurveyID
There is one issue that you run into with the epoch/period trick which is that the counted surveys occur only once per period but aren't necessarily 30 days apart.