Lecture 16: Intro to Tableau

Nick Huntington-Klein

A lot of this class has been spent going over ggplot2 in R
Today we’ll be going over Tableau
(not to mention all we leave out - PowerBI, matplotlib/Seaborn in Python, every stats package other than R, automated stuff like Azure, etc. etc.)
Why introduce these different tools, and what do they do?

It’s most important that we understand the underlying concepts of data communication
But also we have to know how to implement those concepts
And since so much of this is subjective, we have to practice, practice, practice. Working with the software is a requirement!
Plus, stuff like learning the grammar of graphics, baked into ggplot2, can help strengthen those concepts

R: High-power but with a learning curve. Great for any size data sets, great for summarizing, endless customization. Great for contrasting groups
If the data isn’t already in graphable format, R itself is crucial for cleaning/formatting it
Supports dashboards, notebooks, interactivity, and animation
Until you’re used to it, a lot of work to get going though!
Because there is direct access to the underlying pieces of the graph, there is no better tool for making one beautiful image
Requires additional work for working with enormous databases

Tableau: High-power, with less learning curve, but more constrained. Great for any size data sets, great for summarizing, great for contrating groups
Great at making a host of graphs at once to explore a data set thoroughly
It can smartly suggest for you what kind of graph to use
Built-in easy to work with tools for enormous databases and dashboards
Everything is interactive by default
Easier than R and also less work but deep customization a bit more difficult

It is extremely common in software to have to choose between a deeper, more customizable option (such as coding something in ggplot2) or a more canned, one-click-does-it option (like Tableau or Excel)
Until you’re very familiar with the coding option, the canned option will be faster and easier…
IF the thing you want to do is something the canned-software authors anticipated and decided to make easy
If it’s not, it will usually be much harder than just coding the thing up

Tableau is a very cool tool and I encourage you to use it (as are other more canned systems like PowerBI, Azure ,etc.)
But don’t be surprised when you start thinking “hmm but what if my visualization was different in THIS nonstandard way” and start Googling, only to find it’s a fourteen-step process that doesn’t actually work
Using Tableau for the simple stuff, and immediately shifting to code for anything that’s not stock-standard, is a decent strategy!
That said, let’s explore Tableau

As opposed to (the way we’ve used) R, Tableau uses database logic - it sends the calculation to the data rather than bringing the data to the calculation
So we’ll have “connections” to data rather than bringing data in; also, if data updates, so will Tableau
And the output we have will include underlying data, but only our calculations of the data.
Let’s walk through some usage of Tableau

A connection can be “Live” (database-like) or an “Extract” (bring it in!)
Specify how variables should be read by clicking on them
Tableau will guess as much as possible
Unlike R we do filtering, selection, and recoding at THIS step
Strongly recommended to do all data-cleaning in R BEFORE importing to Tableau - Tableau can clean data but it’s a real drag

NAs can make it think it’s a string, let’s tell it earnings is a number
Let’s group pred_degree_awarded_ipeds: 1 = less-than-two-year college, 2 = two-year, 3 = four-year+
Could also do a filter here, or only select certain variables
Or create variables calculated from others - let’s make an employment rate

Dimensions (categorical) and measures (continuous numeric) on the left. Tableau really cares about discrete vs. continuous.
Columns and Rows areas to drag to
(single image export: Worksheet \(\rightarrow\) Copy image)
“Show me” default graphs on the right. Decent for initial exploration but generally not what you want to work with

To start making a graph, think about what you want on the x-axis (column) or y-axis (row)
How it treats a variable can then be manipulated: discrete vs. continuous and measure vs. dimension. Measures get aggregated/summarized in some specified way, dimensions don’t!
Plenty of aggregation options for continuous variables

Also click the variables to do things like sort the order or turn them “dual axis” (two variables on the same axis)
In general, like Excel, click what you want to edit, like axis titles
Note with a graph we can find the data that goes into it
Some variables like “Measure Names” can become their own “shelf” like column/row for more complex diagrams

If something’s not working, check whether your variable has the correct continuous/discrete measure/dimension settings
Note that sometimes “grouped” variables don’t work the same as regular categorical variables - make life easier by just making them strings grouped as you like in R and import them
Did you try Googling to see how you can do (nonstandard thing X)?