I recently presented to a group of Business Analytics Masters students at the University of Illinois Chicago. As new users of Tableau, they wanted to understand how to begin building when touching Tableau for the first time.
Tableau Public is an awesome tool. It's free and allows anyone to begin digging into their data.
Getting the Data
Not only does Tableau Public provide the tool, they also provide starting datasets. So, together, we looked at two datasets:
We first x-rayed the datasets by opening them in Excel. This allowed us to see what the data looked like in its "raw" format. The Netflix dataset is nicely normalized and optimized for Tableau relationships. It has a main table showing titles and additional tables each for cast, directors, categories, and countries. For comparison, we looked at the Pokemon dataset and realized some of the linkages would be harder to make between the pokemon and their spells and other attributes.
In viewing the data, we also noted roughly what one row represents (the grain) and that each tab could have a 1-to-many relationship back to the primary titles table. Now, on to the fun part, right?
While we had a dataset, we were still missing important data - information around how we should design. This data set comes from Netflix and we have to capture some of Netflix brand. That should be simple, right? Netflix is red and black. We instead took a deeper look at how Netflix displayed information.
Despite the iconicity of Netflix being red on black, their broader style guide uses a dark grey behind white to soften the reading experience. Pure white on black is a harder reading experience, particularly if you have an astigmatism (this is why we soften our text to a fuzzier grey on the Versalytix site). In addition, the red is only used for buttons and calling out key information, so, very sparingly. This additional data, the visual style guide, helps us decide how we should present our information before we even open Tableau.
Analyzing the Data
Before we design a dashboard, we have to have a conversation with the data ourselves. We've already briefly x-rayed the data in Excel. As we pull the data into Tableau, relationships or the noodle will play a pivotal role in how we connect all the linkages. We think through the types of questions we might ask and compare that to the tables themselves. Now, some actors are also directors and regardless of their role, we want to easily find Andy Serkis or Angelina Jolie. We union both the director and actor tables. Their names don't line up yet, but a simple IFNULL calc will fix that. Our students have limited time, so we opted to keep all data shaping activities in Tableau Desktop rather than popping to Tableau Prep. Our final source in Tableau created 3 relationships and one union.
Once all the data is extracted into Tableau, we start with our first chart, a simple bar chart giving us a rough idea of how many titles we're analyzing. This exercise helps center our analysis, and gives us a grain. While we're on that chart, we format it to our rough style guide. We opted to put a light grey mark on a darker grey background to mimic what we saw with the FAQ.
We exclude any excess fields, such as the keys that allow us to make relationships to secondary tables and the director and actor fields, replacing those instead with our calculated person field and using the sheet name from the union to create the role field. We also explore the max date of the dataset, finding this data was last updated before the COVID-19 pandemic - so, a whole lifetime ago. We create a Level of Detail calc establishing the max date with {MAX([Date Added])}, a more elegant way to express {FIXED: MAX([Date Added])}.
We spent several minutes making charts. In Tableau, charts are cheap and we can always delete them. We didn't even name them during this process, instead focusing on variety and finding information of interest. We discuss our goals with this analysis. I love Korean shows and so we started by looking at the international category. As titles could have multiple categories, any part-to-whole comparison wasn't a great fit. We opted for the word cloud to address our primary need to find any genre quickly with minimal sacrifice on space. Once we had this, we needed to easily get to South Korea, so we made a map and explored formatting options. We noted some key ways to format the information to keep the data in front, rather than pushing it back.
We hit a point where we had enough charts for the time we had allotted. We went to the dashboard to start the real fun.
Communicating to others
Dashboards help our users have a structured conversations with data. We explored the data through the lens of finding content under specific categories. We explored the effects of building tiled versus floating and opted to float. We then added sheets to the dashboard, moved them around loosely, and played with some of the automatic filtering to think through flow. We were not finalizing, but thinking through the message we wanted to craft. The image below replicates some of our early rough thinking.
We started shaping up our analysis and finalizing formatting. We opted for a flow that started with genre, then moved to countries, and let us see titles. We could also get insight into who starred or directed the movie. We also added click icons to highlight some of the interactivity and played around with different ways to filter and highlight the data.
This dashboard is very much a meal starter. We can certain spend more time elevating it by adding design elements that further help guide our user. We may alter the colors a bit more or take different slices of the data. If you want, feel free to play with it.