Funnel Plots

As I continue to read through Stephen Few’s “Signal: Understanding What Matters in a World of Noise” there have been some new charts or techniques I’ve come across.

In an attempt to understand their purpose on a deeper level (and implement them in my professional life), I’m on a mission to recreate them in Tableau.

First up is a funnel plot. Stephen explains that funnel plots are good when we may need to adjust something before an accurate comparison can be made. In the example video, I adjust how we’re looking at the average profit per item on a given order compared to all of the orders.

What’s interesting is that in tandem with this exercise, I’m working on an quantitative analysis class for my MBA, so there was quite a bit of intersection. I actually quickly pulled the confidence interval calculation (in particular the standard error equation) from the coursework.

I find that overall statistical jargon is really sub-par in explaining what is going on, and all the resources I used left me oscillating between “oh I totally get this” and “I have no idea what this means.” To that end, I’m open to any comments or feedback to the verbiage used in the video or expert knowledge you’d like to share.

Link to full workbook on Tableau public for calculated fields: https://public.tableau.com/views/FunnelPlot10_2_16/Results?:embed=y&:display_count=yes

Thoughts on sorting in Tableau

Now with video ūüôā

Last week I ran into an interesting situation with Tableau. ¬†I wanted to sort dimensions within larger dimensions by a measure. ¬†After that sort, I wanted to encode an additional dimension on color. ¬†Here’s what that would look like using Superstore:

Sorting Figure 1

In the view I am looking at sub-categories by each segment, hoping to rank them by the sum of Sales. ¬†I’ve encoded an additional measure (discount) on color.

This could be a great visualization for understanding demographics within hierarchical type dimensions.  Like say the gender breakdown of who has diabetes at hospital A.

The issue is, getting to the view shown above is somewhat more complex than I had originally thought.  So let me walk you through what happened.

  1.  Set up my view of Customer Segment, Sub-category, by sum of sales
  2. Created initial rank calculation (index()) and then did the typical sorting.
  3. Table calculation is set as follows (Custom Sort = sort by Sum of Sales, descending order):

sorting-2

4. Gets me here:

sorting-3

5. Now when I add Discount to color, my whole viz breaks:

sorting-4

6. To correct this a few things have to be done.  Initial table calculation needs to be modified to ensure the Discount is taken into consideration, but not considered for the sorting:

sorting-5

Super important to notice here that Discount is at the lowest level, but we’re computing at the “Sub-Category” level. ¬†We’re still restarting every “Segment.”

That gives us this:

sorting-6

So now we have the sub-categories correct, we’re looking by region. ¬†But we’re back at that original point of our sort isn’t computed correctly. ¬†This is because we’re sorting by the highest sum of sales for a given discount in a segment. ¬†The first sub-category is found and grouped together. ¬†Then the next sub-category is found with the next highest (not 2nd, just ‘next’) sum of sales for a given discount. ¬†Check it out in comparison to the crosstab, the pink highlights how the index() is working:

sorting-7

To fix this last step, we need to let Tableau (the table calculation, the world!) know that we don’t care about discount for the sum of sales. ¬†We only care about sub-category and segment. ¬†To resolve let’s pull in a simple LOD:

sorting-8

Now finishing it all up, replace the measure used in the table calculation for sorting:

sorting-9

And we’re back at what we originally wanted:

Sorting Figure 1

Full workbook with story point walk-through here: https://public.tableau.com/views/LearningMoment-sortingwithtablecalculations/Learningmoment-tcalcs?:embed=y&:display_count=yes

#IronViz Entry – Mobile Design

Part of being involved in the Tableau community means publicly publishing visualizations to learn and grow. ¬†It’s also a great way to find inspiration.

As I’ve pushed myself to be more active within the local Phoenix Tableau community and social (Twitter) community, I knew it was time to “step up” and make an Iron Viz.

My design aesthetic¬†tends to be minimal, slightly formal, and geared toward (in my mind) elegance. ¬†I like to make dashboards and visualizations that highlight the data, but don’t jump to many conclusions. ¬†I’m very conclusion agnostic, so leading people too far down a path doesn’t always seem right.

All that being said – I wanted to make an Iron Viz entry, but it needed to be simple. ¬†The deadline is September 18 (today as I’m writing this). ¬†So I wanted to develop something relatively straightforward that got to the heart of mobile design.

The data inspiration for the viz actually came from an animated bar chart .gif showing the “Top 10 Economies” growth from 1969 to 2030 that I saw on Twitter. ¬†I thought it would make a nice bump chart or slope chart, and the conclusion of the data was already compelling.

First the data gathering process – relatively simple on this one. ¬†The .gif referenced the source, a quick Google search led me to the results. ¬†I’m going to loosely promise to publish the excel file at some point.

Next up was diving right in. ¬†I recently made a “micro viz” in Tableau 10, designing it exclusively for a very tiny space. ¬†I actually didn’t use the device designer for this one, instead opting to develop the whole thing with my intended sizing. ¬†With the sizing set, development was similar to what I’ve done in the past in v9.3 (version of Tableau I use at my job).

Transitioning to device specific design was different than I thought.  Since I knew the final product (in my mind) would need to have more emphasis placed on the mobile view.  It is after all a mobile design challenge!

Like I mentioned above, I had a pretty good idea of what I wanted the final viz to look like. ¬†I knew there needed to be a bump chart and I was going to call attention to China and India. ¬†What I didn’t realize is that the device designer is really geared toward creating a “master view” and then augmenting that master for the device. ¬†This makes sense to me as I rethink the way the feature was presented.

What this meant for the creative process? ¬†I wasn’t able to make visualizations (sheets) and quickly drop them on the mobile layout. ¬†For each new “potential viz” I had to first drop it on the overall dashboard and then bring it on to the mobile specific dash. ¬†It made the whole process kind of clunky.

I also struggled a bit with getting to formatting features quickly. ¬†I can’t double click on titles to adjust font sizes in device preview, gotta go back to the default layout. ¬†I’ll have to adjust my thought process next time and really think about starting from the default view and optimizing a mobile version.

I probably cheated the viz out of more depth that could have been added if I had truly started with the default dashboard and then made a mobile design.  It is fair to say that the default dashboard has real estate for more data insight and data depth.  I am still very pleased with the overall final results.