How to use Explain Data in Tableau Desktop 2019.3 and newer.

0:00One of the exciting new features in 2019.3

0:05is a feature called explain data. In order

0:09to set the context, I'm going to walk you

0:10through some ad hoc analysis that you would

0:12do to arrive at a very similar conclusion.

0:16In the scatterplot, you can see sales and

0:18profit on columns and rows. And this chart

0:21is broken down by cities. Now immediately,

0:24you can see that there's some outliers over

0:26here on the right hand side. If I was to

0:29select

0:29this top one, one of the things I'd

0:32normally want to do is try and understand

0:34why is this

0:35value an outlier. There's multiple ways I

0:38could do that. I could view the table of

0:41data,

0:41go into full data, and then start doing

0:44this analysis in the probably the least

0:47ideal format,

0:48just essentially analyzing the profit, rows

0:50, ratios, and trying to see if I can analyze

0:53anything from this table, it's actually

0:56quite hard to do. The other option, if I'm

0:58a desktop

0:59author of a Tableau dashboard is I'd

1:01probably select this item, keep just that

1:04item in my

1:05data set by pretty much just creating a set

1:08, you'll see that it's created an inclusion

1:10set over here in the filters pane. And then

1:13breaking this view down to its most gran

1:16ular

1:16level. In this particular case, it's going

1:19to be order ID and the product name because

1:22a customer can order multiple products in a

1:25single order. And then what we can now see

1:28is the spread of data that was originally

1:30on that one data point. And you can very

1:32clearly

1:33see that there's an outlier over here, that

1:35's probably skewing our data. You've also

1:38got

1:39a large sort of center of mass over here on

1:40the left. So actually, that one data point

1:43is being driven heavily by some of these

1:46outliers out here. Now, if I just go back a

1:49few steps,

1:52that's an analysis that's very hard to do

1:55just from this one data point. So one of

1:58the

1:58nice new features in 2019.3 is I can click

2:01on this option. And you'll now see this new

2:04option, which is a circle just on this

2:07command tool pane. I'll just highlight it

2:10here for

2:11you. This is the feature that allows you to

2:14explain the data. When I click on this

2:17table

2:17is actually running a range of machine

2:19learning models, you'll see it's already

2:21finished doing

2:22them. And it will highlight the data point

2:24I selected across the top and show me the

2:27options that I saw in the tooltip. But you

2:30'll also see that it's done some analysis on

2:33that

2:33particular data point. And the first thing

2:35it's realized is that this mark has a

2:37higher

2:37than expected number of records. So this

2:40particular one data point has a higher than

2:42normal number

2:43of records sitting behind it. That means

2:46that its position could be skewed simply by

2:48the

2:48number of records sitting behind it. The

2:51second thing you'll see here is that Table

2:53au has

2:53actually identified that there is an

2:56extreme value in our data set. And it's

2:58generated

2:59this chart to show you that value. It's

3:01also highlighting the particular line and

3:03the particular

3:04customer in this particular case. And if I

3:07scroll down, you'll notice that it actually

3:10draws a view that excludes that extreme

3:13value. What I can do is I can click on this

3:16icon

3:16here on the top right to open as a new

3:18worksheet. When I do that, it actually

3:21brings out that

3:22worksheet onto view. And I can move this

3:24annotation over. And it now shows me my new

3:27data point

3:28with that extreme value removed. The great

3:32thing is that this window is modal. So I

3:34can

3:34still see this analysis whilst I'm working

3:37with this view, I can head back to the tab

3:39that originally activated this view. And I

3:42can keep doing my analysis. I'll switch

3:45over

3:45to the sum of cells. You'll notice that it

3:47's pulled out these two measures because

3:50they're

3:50fundamentally the measures that are on the

3:52columns and rows. If I added a third

3:54measure

3:55onto my marks pane, let's say quantity, you

3:58'll notice that Tableau warns you when its

4:01analysis

4:01is no longer valid. Essentially, it's

4:03noticed that I've added a third measure

4:05onto my marks

4:06pane. And it gives you a warning to let you

4:08know that the visualization has changed, I

4:10need to rerun this analysis. So if I go

4:13back, click on that same data point, and

4:17then rerun

4:17the analysis, you'll see that I now get a

4:21new option for the quantity which I added

4:24onto the size. If I head over to the

4:26quantity, I'm now able to interrogate that.

4:30And the

4:30only stand up item here is the number of

4:32records sitting behind this data point.

4:35That's a consistent

4:36feature across all the measures. Because it

4:38refers to that one data point, it's going

4:40to be a consistent item. You'll also see

4:43that in terms of profit, this data point

4:45also has

4:46another extreme value related to profit. It

4:49's again, the same customer. So that one

4:51customer

4:52from a sales and profit perspective is an

4:56outlier. That's just one example of

4:59explained

5:00data. In a different example, I'm going to

5:03show you how this works in a more real life

5:06situation. Let me switch over to another

5:09visualization that I have. This chart

5:13prepared by Tableau

5:14shows the average duration of home and home

5:17ownership in London. You can see that the

5:20color of each borough represents the

5:22average duration of a home ownership. So

5:24you can see

5:25here that Newham has the lowest value of 4.

5:308 years, whereas Sutton and Bexley and

5:34nearly

5:35Hillingdon have some of the higher values.

5:38If I want to analyze some of that data, I

5:40can. Again, I just select the data point.

5:45Then hit explain data. You'll see that

5:49Tableau

5:50runs its machine learning algorithms. This

5:52time is taking a bit longer, the data set

5:54is a little bit larger. And you can see

5:56that Tableau has done analysis into that

5:58one particular

5:59item and it's generated some new insight

6:02for us. So the first thing to notice is

6:04that in

6:05blue, it represents the mark that we've

6:08selected. In this particular case, this is

6:11the Sutton

6:12region. And in gray is the rest of our data

6:16. And so what this is showing is that in

6:20Sutton

6:20in particular, there is a higher proportion

6:23of households in terraces and flats, nearly

6:2740%. And then you can see the remainder of

6:29properties that aren't over here on the

6:31right

6:32hand side in blue. So this blue percentage

6:34on the right, plus this on the left

6:36constitute

6:37100%. It also shows you the spread for all

6:40marks. So you can see here that for all

6:43other

6:43data, that value is much, much lower,

6:46probably just under 10%. If I go over to

6:50this next

6:51dimension, you can see that it's looked at

6:53one of the attributes and it's realized

6:55that

6:55in this particular borough, there's a

6:58higher percentage of young adults, less

7:01than 20%,

7:02and 20 to 30% are both almost exclusively

7:06the only values available in Sutton.

7:10However,

7:10if you compare to the spread in our data

7:12set, you'll see that the spread is a lot

7:14higher,

7:15and a lot more evenly spread out across all

7:18the items. This is some of the powerful

7:20analysis

7:21that explain data can do right off the bat

7:23without you having to create ad hoc

7:25analysis

7:25for the user to dive into. That's it.