We’ve all faced this problem: Duplicate rows in our data. It messes up our reports and causes general confusion and consternation for people we supply data to. The good news is you can take care of this problem fairly easily in an SQL query using a handy function called LISTAGG. I discovered this almost by accident just this week, and I’m delighted to share.
Duplicate rows are a result of what we choose to include in our database query. If we are expecting to get only one row for each constituent ID, then we should be safe in asking for Gender, or Name Prefix, or Preferred Address. There should be only one of each of these things for each ID, and therefore there won’t be duplicate rows. These are examples of one-to-one relationships.
We get into trouble when we ask for something that can have multiple values for a single ID. That’s a one-to-many relationship. Examples: Any address (not just Preferred), Degree (some alumni will have more than one), Category Code (a person can be an alum and a parent and a staff member, all at the same time) … and so on. Below, the first constituent is duplicated on Category, and the second constituent is duplicated on Degree.
That’s not the worst of it; when one ID is duplicated on two or more elements, things get really messy. If A0001 above had three Category codes and two degrees, it would be appear six times (3 x 2) in the result set. Notice that we are really talking about duplicate IDs — not duplicate rows. Each row is, in fact, unique. This is perfectly correct behaviour on the part of the database and query. Try explaining that to your boss, however! You’ve got to do something to clean this up.
What isn’t going to work is pivoting. In SQL, the PIVOT function creates a new column for each unique category of some data element, plus an aggregation of the values for that category. I’ve written about this here: Really swell SQL: Why you must know PIVOT and WITH. Using PIVOT is going to give you something like the result below. (I’ve put in 1’s and nulls to indicate the presence or absence of data.)
What we really want is THIS:
The answer is this: String aggregation. You may already be familiar with aggregating quantitative values, usually by counting or summing. When we query on a donor’s giving, we usually don’t want a row for each gift — we want the sum of all giving. Easy. For strings — categorical data such as Degree or Category Code — it’s a different situation. Sure, we can aggregate by using counting. For example, I can ask for a count of Degree, and that will limit the data to one row per ID. But for most reporting scenarios, we want to display the data itself. We want to know what the degrees are, not just how many of them there are.
The following simplified SQL will get you the result you see above for ID and Category — one row per ID, and a series of category codes separated by commas. This code works in Oracle 11g — I can’t speak to other systems’ implementations of SQL. (For other Oracle-based methods, check out this page. To accomplish the same thing in a non-Oracle variant of SQL called T-SQL, scroll down to the Postscript below.)
Obviously you will need to replace the sample text with your own schema name, table names, and field identifiers.
SELECT SCHEMA.ENTITY.ID, LISTAGG ( SCHEMA.ENTITY.CATG_CODE, ', ' ) WITHIN GROUP ( ORDER BY SCHEMA.ENTITY.CATG_CODE ) AS CATEGORY FROM SCHEMA.ENTITY GROUP BY SCHEMA.ENTITY.ID
The two arguments given to LISTAGG are the field CATG_CODE and a string consisting of a comma and a space. The string, which can be anything you want, is inserted between each Category Code. On the next line, ORDER BY sorts the category codes in alphabetical order.
LISTAGG accepts only two arguments, but if you want to include two or more fields and/or strings in the aggregation, you can just concatenate them inside LISTAGG using “pipe” characters (||). The following is an example using Degree Code and Degree Code Description, with a space between them. To get even fancier, I have replaced the comma separator with the character code for “new line,” which puts each degree on its own line — handy for readability in a list or report. (This works fine when used in Custom SQL in Tableau. For display in Excel, you may have to use a carriage return — char(13) — instead of a line feed.) You will also notice that I have specified a join to a table containing the Degree data.
SELECT SCHEMA.ENTITY.ID, LISTAGG ( SCHEMA.DEGREE.DEGC_CODE || ' ' || SCHEMA.DEGREE.DEGC_DESC, chr(10) ) WITHIN GROUP ( ORDER BY SCHEMA.DEGREE.DEGC_CODE, SCHEMA.DEGREE.DEGC_DESC ) AS DEGREES FROM SCHEMA.ENTITY INNER JOIN SCHEMA.DEGREE ON ( SCHEMA.ENTITY.ID = SCHEMA.DEGREE.ID ) GROUP BY SCHEMA.ENTITY.ID
Before I discovered this capability, the only other automated way I knew how to handle it was in Python. All the previous methods I’ve used were at least partially manual or just very cumbersome. I don’t know how long string aggregation in Oracle has been around, but I’m grateful that a data problem I have struggled with for years has finally been dispensed with. Hurrah!
After publishing this I had an email response from David Logan, Director of Philanthropic Analytics at Children’s Mercy in Kansas City, Missouri. He uses TSQL, a proprietary procedural language used by Microsoft in SQL Server. Some readers might find this helpful, so I will quote from his email:
I was reading your post this morning about using the LISTAGG function in Oracle SQL to concatenate values from duplicate rows into a list. A couple of months ago I was grappling with this very problem so that I could create a query of Constituents that included a list of all their phone numbers in a single field. LISTAGG is not available for those of us using T-SQL … the solution for T-SQL is to use XML PATH().
Here’s a link to a source that demonstrates how to use it: http://sqlandme.com/2011/04/27/tsql-concatenate-rows-using-for-xml-path/.
One key difference in this method from LISTAGG is that the delimiter character (such as the comma—or characters as in your example where you have comma space) is added before each value returned so the raw string returned will have this character before the first value. You can use the STUFF() function to eliminate this first delimiter.
Here’s my solution using XML PATH() to return a list of phone numbers from our donor database:
SELECT CA.ID, STUFF ( ( SELECT '; ' + PH.NUM + ' [' + re7_ReportingTools.dbo.GetLongDescription(PH.PHONETYPEID) + ']' FROM re7_db.dbo.CONSTIT_ADDRESS_PHONES CAP INNER JOIN re7_db.dbo.PHONES PH ON CAP.PHONESID=PH.PHONESID WHERE CAP.CONSTITADDRESSID=CA.ID AND PH.DO_NOT_CALL=0 ORDER BY CAP.SEQUENCE FOR XML PATH('') ),1,2,'' ) PHONENUMS FROM re7_db.dbo.CONSTIT_ADDRESS CA GROUP BY CA.ID
I concatenated my list using a semi-colon followed by a space ‘; ‘. I used the STUFF() function to remove the leading semi-colon and space.
Here are a few sample rows that are returned (note I’ve “greeked” the prefixes for donor privacy):