Tableau Playbook - Scatter Plot

We will start with an example chart to introduce the concept and characteristics of the scatter plot.
By analyzing real-life datasets, we will learn to build a scatter plot step by step. We will then optimize and polish the chart with advanced features.

Example

As a whole, we use the trend line to fit the linear relationship between experience and salary.
From a horizontal perspective, we can compare the age distribution of the developers who work in various languages. From a vertical perspective, we can compare the median salary distribution for developers who work in each language.
Additional visual elements, such as size and color, allow a scatter plot is able to convey more information. In this example, the size of the circles intuitively expresses the popularity of language.
With the help of the scatter plot, we can dig out useful information. For example, developers in Go, Clojure, and F# are being paid more even given how much experience they have. Developers using languages below the line, like PHP and Visual Basic 6, however, are paid less, even given years of experience.

Concept and Characteristics

A scatter plot is a type of plot using Cartesian coordinates to display values for typically two variables for a set of data. If the points are coded (color/shape/size), one additional variable can be displayed.

Scalability - can hold a large number of points: Scatter plots give us an option to display a lot of data in a small area with relatively low confusion rates.
Analyze correlation: A typical use of a scatter plot is to determine whether two measures are correlated. Tableau provides statistical variables such as the P-value and R-squared. But it's important to note that we need to treat correlation objectively. When two variables are correlated, it does not mean that one variable caused the other.
Observe data intuitively: In a scatter plot, you can visually observe outliers, data ranges, or specified areas. What's more, with the interactive operation provided by Tableau, we can further analyze these points in detail.

MEDV: Median value of owner-occupied homes in $1000's
RM: average number of rooms per dwelling
CRIM: per capita crime rate by town
LSTAT: % lower status of the population

Before drawing, we need to do some data preprocessing with the help of external tools such as Excel. To display each row as a point, we need to add ID to identify. The easiest way is to add the ID column in Excel. At present, it is hard to create unique identifiers in Tableau. If you insist on that, please refer to this post.

We can generate a basic chart automatically by using Show Me. This is the easiest way to build a scatter plot. Click on Show Me and you will see these instructions:

For scatter plots, try 0 or more Dimensions, 2 to 4 Measures.

Now we notice there is only one point in the chart. That is because all the records are aggregated together. Here we can split the data by ID, which we created before.

Convert ID into Dimension.
Drag ID into Marks - Detail.
Switch to Entire View for a nicer visualization.

On the top of the scatter plot, sixteen data points have a 1MEDV1 value of 50.0. They are outliers that have been clamped by the upper bound. For a more accurate analysis, we should remove these outliers. Multi-select them and click Exclude in the pop-up dialog. Tableau will exclude them in the Filters.

For a more attractive chart, edit visual elements such as shape and color:

Expand the Shape card in Marks and replace the empty circle with the solid circle or any other shape that makes sense for your readers.
To reduce the impact of the overlay, expand the Color card in Marks and slide the Opacity to semitransparent.

Add a trend line to identify the correlation between RM and MEDV.

Right-click on the chart and choose Trend Lines -> Show Trend Lines.
Right-click on the trend line and click Edit Trend Lines...
Choose Linear as the Model type.
Check Show Confidence Bands.

In the last step, let's polish this chart:



Edit title to "Relationship between Room Number and House Price".
Rename the x-axis as "Room Number" and y-axis as "House Price".

First, let's build a scatter plot as we did previously.

This time we'll build it manually. Drag LSTAT into Columns Shelf and MEDV into Rows Shelf.
Drag ID into Marks - Details.
Right-click and choose Hide Indicator for nulls values.
Multi-select the top outliers and click Exclude in the pop-up dialog.
Switch to Entire View for a better view.

Add more visual elements to convey information. Here we show measure CRIM by size.

Drag CRIM into Marks - Size.
Adjust size by expanding the Size card or the size legend on the right side.

The clustering technique is useful to analyze the characteristics of a scatter plot. Tableau has built-in clustering algorithms, such as k-mean. Let's try out Tableau's clustering capabilities to look for the common properties of points.

Switch to the Analytics pane. Drag Cluster into the view and drop it on the Create Clusters box that appears.

We can see it is calculated by the three measures we created before. We remove CRIM and see what's going on with only two axis variables.

Notice Tableau found four clusters. Each cluster reflects a potential class of houses from different prices and LSTAT ranges. We can dig more information from these clusters, but we will stop here and focus on the scatter plot.

Let's add another measure RM as color.

Drag RM into Marks - Color.
The default color configuration is not good enough. Let's make it better. Click the inverted triangle in color legend, then choose Edit Colors...
To distinguish the color more clearly, we change the single color to diverging colors. Choose Red-Green-Gold Diverging in Palette.
In order to group the points by number of rooms, we check Stepped Color and set Steps to 5.

Add more quantitative indicators to the scatter plot:

Switch to the Analytics tab, and drag Average Line into Table - Cell of SUM(LSTAT) and SUM(MEDV). Left click on them and change the type to Median.
Add a trend line as we did before. The only difference is that this time, we simulate the model type as Logarithmic.

Put on the finishing touches:

Edit title to "Factors which affect House Price in Boston".
Rename the x-axis "Lower Status of Population" and y-axis "House Price".
Rename the size legend "Crime Rate" and color legend "Room Number".
If you think these grid lines are too distracting, you can remove them to make the chart cleaner: navigate to Format -> Lines... and set Grid Lines to None in Lines.