I gave a talk last week about “Data to Dashboard” and I wanted to share it here, too. There is a lot of discussion in the analytics space about dashboards and how to make them look good but less about how to get to that point. This is my take on the subject – I hope you enjoy it.
For our inaugural duel – Jacob created a data set based on 538’s NBA Predictions. He’ll create a deep dive into the mechanics of the model and how to leverage Excel’s data table function for no-code simulations in a later post. The data is available at the bottom of this post.
Jacob is a native Excel user and has created similar models for his Fantasy NBA league. He was able to take those models and dress them up for this duel – albeit in a format that was native to PowerBI and Excel. More on how that impacted the Tableau side of the analysis below.
Since we were using the 538 data set, we decided the first part of the challenge should be to replicate the view above in PowerBI & Tableau.
Some of the data weren’t readily available, i.e. projected point differentials and team logos. For the purpose of the commentary below, we will be ignoring these facts.
As a phase 2 / stretch goal for this challenge, we also set out to create our own, novel visualization of the scenario combinations. This helped us to answer questions like “When the Bucks make the finals, who are their most likely opponents?” or “What are the paths for the Celtics to the conference finals?”.
This is section is written in first-person by Jacob.
Part 1: 538’s visualization
Where PowerBI succeeded: Getting the calculations out of the attached data set was fairly easy once I sorted out the data model in my head. While the data wasn’t perfectly formed, it was quite easy to shape it using PowerQuery to get what I needed.
I added a couple of measures on top of it and the table working pretty quickly. Getting the conditional formatting to match was fairly easy too, although to get an exact color match I used the “color dropper” from powerpoint on a screenshot of the website (gross).
Where PowerBI struggled: I couldn’t quite figure out how to get the sorting to work when I replaced certain values with “icons”, i.e. >99% or the “checkmark” icon. PowerBI treats the field as a string and therefore does a character-based sort. This means that apply a single sort on the outcome of the model doesn’t really work! Instead, you have to sort by ELO rating or by Projected standings to get a cohesive sort.
After I wrote this initially – I did find a workaround for this sorting issue, sort of. This video from Guy In A Cube explains the “hack” – but it is indeed just a hack.
Part 2: Scenario Modeling
I am pretty satisfied with how this visual turned out – but the sorting on probability fields continued to plague me. Also, the mental model for this data was effectively recursive, and I am not sure how to accomplish this in PowerBI, so I imported the same table twice. See the image on the leftfor how this was accomplished.
After fighting with DAX on and off for a few days, I was able to get a “base scenario” calculation using the ALL Filter. This meant that when you selected a Team from “series_winners” you could calculate the odds of that scenario versus the “base” scenario. This surfaces really neat scenarios in the modeler, such as an OKC win in the second round which double’s Milwalkee’s championship odds.
You can find the DAX for stripping the filters from the “series winner” table, below.
All Scenario Win Pct = CALCULATE ( COUNT ( series_detail[TeamID] )/ DISTINCTCOUNT ( series_winners[ScenarioID] ), ALL(series_winners) )
This section is written in first-person by Nate.
Part 1: Data Prepping
Where Tableau succeeded: Tableau handled the data really well once I completed a lot of trial-and-error to get the data into the right format. The data model I put together involved two tables in there twice, so it’s likely sub-optimal but is functional. Specifically, Tableau is consistently improving how data can be loaded & prepared (See the recent changes just launched in 2020.2) but my unfamiliarity with those new features meant I didn’t have time to give them a go on this analysis.
Where Tableau struggled: My experience getting data into Tableau nearly always involves a connection to SQL – either a direct connection to a table or a very clean CSV output of a SQL query. Since the data model created by Jacob is in PowerBI/Excel, I had to do some manual adjustments to the tables to get them in the format I needed, such as creating long tables (just a few columns) out of wide tables. This resulted in several more hours of work as I did trial & error between modifying data and trying to visualize it in Tableau.
Part 2: 538’s visualization
Where Tableau succeeded: Getting this table created was very simple once I finalized the data model. Sorting works well across all columns and the Tableau method of dragging dimensions & measures around to get colors & formats worked well.
Where Tableau struggled: I could not get some of the nifty 538 features in the table – such as a checkmark at 100%, and a string for “<1%”. Instead, Tableau displays 0% for situations that round down to 0%. I tried adding in a decimal place, but that just cluttered up the view. As well, Tableau does not have strong conditional formatting capabilities for the background of cells. It’s possible (see the KB article here) but I found the saved development time for other work by instead just coloring the numbers and shifting to a darker cell background.
Part 3: Scenario Modeling
Where Tableau succeeded: After several rounds of arm wrestling, pleading and bribing – Tableau finally assented to the view I wanted which included:
(a) The original odds
(b) New odds based on selection
(c) Visualization demonstrating change in odds
There’s much more I wish I could have done but in the interest of time (the playoffs are going now!) it was time to hit ‘Publish’
Where Tableau struggled: I spent multiple hours trying to get the FIXED LOD formula in Tableau to work which would allow me to compare the odds from both the unfiltered view and the filtered scenario view at the same time. Tableau can be frustrating to use when trying to visualize multiple levels of detail in the same view, and likely my chosen data model contributed to the struggles.
I asked the best Tableau user I know for some advice as I was getting this viz prepared and his advice sums up the struggle well: “When dealing with LODs, I usually just try every permutation until something somehow works.” Turns out he was right in this instance, too.
Overall Winner: PowerBI
538 Duplication: PowerBI – but really neither piece of software allows the customization that the web package used by 538 has. Still, we gave it a decent go. PowerBI does tables a bit better, so it wins here.
Scenario Modeling: While Tableau is very snappy and honestly more “discoverable” (good job with the tooltips, Nate), the PowerBI “tournament visual” is very intuitive for sports fans. Additionally, the analysis at the bottom of the chart is more comprehensive and more understandable than the tableau bar charts of the same info. We give the edge to PowerBI.
A note on data prep (not scored): The in-app data prep with PowerQuery is a no-contest when compared to Tableau. This will pretty much always be true and can be both good and bad. Good, because it allows a lot of control at the fingertips of the analyst, and bad, because a lot of code, either in M or DAX, gets added to PowerBI instead of database engine of your choice.
PowerBI Commentary from Tableau User
Where PowerBI succeeded: PowerBI is more equipped than Tableau to display data in a tabular format such as the one on 538 and that shows in the final product. The ability to put many small views of data into a single dashboard also proved to be powerful in the final scenario modeling output.
Where PowerBI struggled: PowerBI depended too strongly on its native table functionality, resulting in lots of details but a lack of bold & clear visuals. Sorting also turned out to be tough as you may notice when using the final interactive version.
Tableau Commentary from a PowerBI User
Where Tableau succeeded: The data model is much easier to grasp even to technical users. It’s much faster to interact with, and the tool-tips make it very easy to understand. Additionally, clever use of the NBA logo immediately contextualizes the user.
Where Tableau struggled: As Nate mentioned, the data prep took a significant amount of time. The conditional formatting inside tables is not very finely tuned, especially compared to PowerBI. Hilariously, sorting inside a table has its own set of issues (sort on “Win Championship”).
Let us know your thoughts below! A list of the files can be found after the jump.