The term “Join” comes from the database world, and I’ll admit that I’ve struggled with understanding it... especially when you combine it with some other keywords.  So this week I thought it would be good to explore it in a bit more detail, specially to show the different results when we merge tables using outer joins.

Different types of Outer Joins

There are actually three flavours of Outer Join you could come across, and each work in different ways.  Those three flavours are:

• Left Outer join
• Right Outer join
• Full Outer join

Which… to an Excel person… mean very little.  To confuse matters more, there are also Inner joins, and Anti joins.  (We’ll look at those next week though.)

Sample Data

To illustrate the different join types, we are going to work with the set of data shown below (which you can download here.)

So two tables of data, one called Transactions, and one called ChartOfAccounts.

Now, the key piece you want to watch here is that we need the Account-Dept combination to exist in both tables in order to make a perfect join between them.  Think VLOOKUP for a second… if you tried to make a VLOOKUP for account 10045 from the Transactions table against the ChartOfAccounts table, what would you get?  Of course, you’d get #N/A since 10045 doesn’t exist in the ChartOfAccounts table.

In this case we have items in both tables that don’t exist in the other.  (The yellow records in the Transactions table don’t have a match in the ChartOfAccounts table, and the red records in ChartOfAccounts don’t have a match in the Transactions table.)  With these differences we can test how each of the first three join types available to us behave when we try to merge the data in both tables together.

Groundwork

Of course, the first thing we need is a pointer to each table for Power Query to work.  So let’s set that up first.

• Click in the Transactions table –> New Query –> From Table
• Go to Home –> Close & Load –> Close & Load To… –> Only Create Connection
• Click in the COA Table –> New Query –> From Table
• Go to Home –> Close & Load –> Close & Load To… –> Only Create Connection

I also right clicked each query in the Workbook Queries pane and choose to move them to a group I called Source, just to keep them organized:

This will work nicely.  We have a pointer to both tables, but didn’t cause any data duplication by loading a new copy to a worksheet.

Merge Tables using Outer Joins – Left Outer Join

Okay, let’s get started.  We need to join these based on the combination of the Account and Dept number.  Fortunately we can use the trick discussed in this post to do this with creating a concatenated key manually first:

• Right click the Transactions query and choose Merge
• Select ChartOfAccounts for the bottom table
• For the top query (Transactions) select Account, hold down CTRL and select Dept
• For the bottom query (ChartOfAccounts) select Account, hold down CTRL and select Dept

Your join selections should now look like this:

Notice that the Join Kind in the bottom is “Left Outer (all from first, matching from second).  Let’s click OK and see what that means.

When you get into the Power Query editor:

• Right click the NewColumn column –> Rename –> COA
• Click the Expand icon on the top right of the COA column
• Leave all the defaults and click OK

The resulting query should look as follows (barring the colour of course):

So this is a “Left Outer Join” – the default choice for Power Query.  It returns all entries in the left table (or top in the case of the power query editor) and returns the matching values it finds based on the lookup column.  Essentially this is the same as VLOOKUP.  It returns a match, except where it can’t find a matching record.  (The main difference between VLOOKUP and Power Query is that if Power Query found multiple matching records, it would return all of them, not just one.)

Again, the key point here is that every value from the Left table is returned, whether there is a match or not in the Right table.  The yellow rows here match the yellow rows in the original Transaction table shown at the beginning of the post.

But… notice also that accounts 10040 and 11000 (the red accounts in the COA table) do not shown in the listing at all.  This is your Left Outer join in action.  It pulls all records from the left table, any matches from the right (or null if no records on the right match).  It never looks at the right side at all to see if records exist there that don’t exist in the left hand table.

Okay, so now we can see what’s happening here, let’s finish it off:

• Change the name of the query to LeftOuter
• Go to Home –> Close & Load –> Close & Load To… –> Only Create Connection

Merge Tables using Outer Joins – Right Outer Join

Rather than discuss the difference here, let’s just demo it.  One caveat… it’s important to get a clear understanding that you start with same base table in order to make parallel comparisons!

• Right click the Transactions query and choose Merge
• Select ChartOfAccounts for the bottom table
• For the top query (Transactions) select Account, hold down CTRL and select Dept
• For the bottom query (ChartOfAccounts) select Account, hold down CTRL and select Dept
• Change the Join type to “Right Outer (all from second, matching from first)”
• Click OK

And then we’ll do the same thing we did before so that we can compare the results:

• Right click the NewColumn column –> Rename –> COA
• Click the Expand icon on the top right of the COA column
• Leave all the defaults and click OK

If you followed along correctly, your result should look like this:

Notice the big difference here?  This time the values from the Right table (ChartOfAccounts) show whether there is a match in the Left (Transactions) table or not.  The red rows (containing 10040 and 11000 from our original table) are now present, where they weren’t in the previous scneario.

But the (yellow) items that were in the Left (Transaction) table which don’t have a match in the Right (ChartOfAccounts) table?  They’re nowhere to be seen!

Where I now know I struggled with this when writing SQL code from scratch is that no one ever explained to me which table was Left and which was Right.  Lacking that knowledge there really isn’t anything to explain what these joins are truly doing.  But now that we can see that the table we start our merge from is the Left table, and the one we are joining to it is the Right table… it starts to make a LOT more sense.  The only kicker we have with Power Query is that the Left table is the top in our merge dialog, and the Right is the bottom.  But you can certainly see why the UI was designed this way (imagine trying to fit it on your screen if it was side by side?)

At any rate, we can now compare and contrast those two joins.  Let’s finish this one off and look at the final join we’ll examine today.

• Change the name of the query to RightOuter
• Go to Home –> Close & Load –> Close & Load To… –> Only Create Connection

Merge Tables using Outer Joins – Full Outer Join

By now, you know the drill…

• Right click the Transactions query and choose Merge
• Select ChartOfAccounts for the bottom table
• For the top query (Transactions) select Account, hold down CTRL and select Dept
• For the bottom query (ChartOfAccounts) select Account, hold down CTRL and select Dept
• Change the Join type to “Full Outer (all from second, matching from first)”
• Click OK

And now modify the query:

• Right click the NewColumn column –> Rename –> COA
• Click the Expand icon on the top right of the COA column
• Leave all the defaults and click OK

And what do we have?

Aha!  So the Full Outer join makes sure we’ve got all items from both sides.  I can see this being SUPER useful for trying to compare to lists of transactions like in a bank reconciliation.  Transactions that match would get lined up nicely, and any that needed attention would have a bunch of nulls beside them on either side.  Very cool.  (I’ll have to do a post on that some day!)

Wrapping Up

This shows how to merge tables using Outer Joins… If you'd like to learn about the three remaining join types shown below, you can do so at this article:

• Inner Join
• Left Anti Join
• Right Anti Join

Some time ago I got an email from Alex asking me if there was a way to identify duplicates using Power Query, but without removing non-duplicate records in the process.  This post explores how to do that.

Background

Suppose someone has given you a list like the one shown below (which you can download here if you’d like to follow along):

While multiple brands are okay here, we need a list that shows only unique SKU numbers.  While the list provided to you was supposed to be duplicate free, you’re not 100% sure that it actually is.  While it would be easy to just hit the SKU column with the Remove Duplicates function, you don’t want to do that.  Instead you’d like to indentify which records have duplicate entries in the list.

So how do we do this?

Naturally, there will be a few different ways to do this.  I’m carving off one method that is the easiest to replicate via the user interface…

Step 1:  Link to the Data

Of course we’ll start by pulling the data in to Power Query

• Click anywhere in the Products Table
• Create a new query –> From Table

The data will be loaded in to Power Query, and you’ll see two steps in the Applied Steps window:

• Source (pointing to your source data)
• Changed Type (setting the data types for the columns)

This might seem like an odd step right now, but we’re going to add a Index column to this table as well.  The reason will become apparent later, but for now:

• To to Add Column –> Add Index Column –> From 0

Your data should now look like this:

Now we need to figure out how to flag any repeating SKU as a duplicate.

Step 2: Indentify Duplicates via Grouping Rows

The trick here is to use the Group By feature in Power Query, while preserving the relevant matching records.

NOTE:  We cover the Grouping feature in Chapter 14 of M is for Data Monkey.

Here’s how we do this:

• Go to Transform –> Group By
• Set your Group By Options as follows:
  • Group By:   SKU Number
  • New column name:  Duplicates –> Count Rows

Next, click the + to the right of the “New Column Name” section to add another detail row.  Set it up as follows:

• New column name:  Duplicates –> All Rows

When you’re done, the dialog should look like this:

And upon clicking OK, the results will show that there are, indeed, items that show up more than once:

Let’s tweak this a bit, and subtract 1 from each value.  That would give us a truer representation as to how many duplicates there are.

• Select the Duplicates column –> Transform –> Subtract –> 1

Resulting in the following:

Much better.  We’re now seeing that SKU 510010 appears to have 1 duplicate entry in the data set.

But there is still an issue here.  When we grouped our records, we lost both the Brand names column, but also any duplicate records.  Since the whole point of this exercise was to Indentify Duplicates but not remove the duplicate records, we’re still not in a good place.

Step 3: Identify Duplicates and Show Duplicate Records

Let’s fix this.  Remember how we added a new step to show “All Rows” for the ProductDetail column?  That step gave us the ability to do something pretty cool… it gave us the ability to get back all the lost records and product detail information we’re currently missing.

• Click the Expand button at the top right of the ProductDetail column
• Uncheck the SKU Number option (as we already have it)
• Uncheck the option to “Use original column name as prefix”

As you can see, this will bring back all the details we lost earlier.

Step 4: Final Cleanup

But hang on a second.  Let’s look at this output a bit more closely…

Notice, that it re-sorted the data.  That’s not exactly a desirable outcome, as we are trying to flag duplicates for a reason. Maybe we want to know where they exist in an inventory count or we have some other reason for wanting to preserve the original sort order of our data.  It’s for this reason that we added the Index column earlier.  That came through with the All Rows step, so let’s put our data back into its original order.

• Click the drop down arrow on the Index column –> Sort Ascending
• Right click the Index column –> Remove

And we can now finalize the query:

• Rename the query to ShowDuplicates
• Go to Home –> Close & Load

Step 5: Make the Duplicates Obvious

With the data now in an Excel table, we can make the duplicates even more obvious by applying some conditional formatting to the table.  To do this:

• Select all the values in the Duplicates column of the table
• Go to Home –> Conditional Formatting –> Data Bars –> Choose a colour

I chose blue data bars, which makes the data look like this:

Conclusion

Our goal is now complete.  We were able to identify duplicates and flag them without removing non-duplicate items.  In addition, we have preserved the original order of the data in case that was important to us for any reason.