Going Back to Cali: Dashboard Week Day 4

by Collin Smith

Curveball! Today, Andy asked us to download and use Tableau 8.3 for our dashboard-building. This version of Tableau, released in 2015, has much of the functionality we have come to know and love in The Data School, but there were a couple potholes to discover as well. The data we used today was on historical registrations in California universities.

Getting the Data

Thankfully, the data acquisition today was far easier than Tuesday or Wednesday. The file contained 750,000 rows of data and 16 fields, a measly 44 MB. Each row represented a student at a California University for one year. A student who attended a university for four years would have four records. Included was information on degree, gender, and their place of origin.

Data Preparation

The data was fairly simple to manipulate in Alteryx. My primary task was matching up codes for a student’s degree with the actual name of that degree, which was provided in a word documentation file. It let me use a nifty little bit of text to columns.

Turn this into data!
Match and output

I also used a new file type which I’ve never used before, a Tableau Data Extract (.tde). These extract files have been mostly mothballed since the switch to .hyper, but Tableau 8.3 is in the pre-hyper era.

Data investigation and design

After opening up my brand new old version of Tableau, I was pleased to find that the interface felt fairly familiar. I brought in my .tde and got down to business looking at the who, where, when of the data. I struggled to get the mapping functions to work and I knew the data conflated country, city, and state into one field depending on the student’s origin. I decided to leave the where well alone.

I was feeling a little uninspired and comfortable with my timing, so I broke out the Big Book of Dashboards by Steve Wexler, Jeffrey Shaffer, and Andy Cogreave. I figured that Wexler’s ideas worked well for me yesterday, why not try another of his designs? I landed on one that provides a sort of executive summary of year over year changes, comparing a Now value to a Then value.

In retrospect, I’m not sure this dashboard was the right choice for this dataset. The dashboard is built to highlight changes in year over year percentages that are statistically significant. The percentages it shows are all in a fairly small range. In a historic dataset like mine which covers about 50 years of registrations, a dashboard looking at year to year differences is probably too granular. I discovered this while building out the viz though, and decided against fully pivoting to a different kind of dashboard.

My adaptation in sketch form

Building the viz

The preliminary construction of my viz went well, but I ran into some major issues when trying to line up the charts in the viz. I couldn’t seem to get the same values for ‘degree’ to show up across all three charts. Introducing filters to one would change the others in unexpected ways. I spent hours this afternoon trying to troubleshoot the issue and I still haven’t cracked the nut. I included a highlight action that helps the user visually associate the correct sparkline to the correct bar chart, however I’m not satisfied with this result at the end of the day.

To combat the overly granular charts, I decided to add a line chart at the top that shows the total enrolments by school at the top of the dashboard.

Here’s how the highlighting action looks in practice

As you’ll notice, the highlighting does not line up across the rows as it should. I think that to fix this issue, I would need to re-work how the sparkline is created somehow.

Tomorrow is another day, and the last day DS17’s dashboard week. Here’s to finishing strong!