VizWiz

Table calcs have ALWAYS been my biggest weakness in Tableau. I find them incredibly difficult to understand and finding a plain language way of explaining them has always been tough. If I heard the words addressing and partitioning, they might as well have been speaking Swahili.

I'll have a more detailed blog coming soon on how I've learned to explain table calculations; I had a serious lightbulb go off Friday morning while I was teaching. In the meantime, I thought I'd look back at this week's Workout Wednesday and explain what the calculation for the columns is actually doing.

First up, here's the final viz:

The trickiest piece for me was understanding how the width of the bars worked. Here's the calculation we'll review:

Notice that I named the calculation "Columns". I do this when I know I need a calculation to be placed on a specific shelf. It's a way that I remind myself. When I choose "Edit Table Calculation" I'm presented with this view.


This setup tells me that for each Sub-type (i.e., gender in this case), calculate the Columns formula by Job Type. Ok, got it? Got it! Let's look back at the formula and explain what is actually being calculated.

Note: The Job Types are sorted by the % of female in descending order.


Hopefully this helps explain a bit about how the calcs work to get the bar widths correct in a Merimekko. I know it helped me think through it.

How might you explain it differently?

Wow! That this was a tough workout! I've never created a Merimekko chart before and I'm very, very thankful that Emma provided a link to a great how-to blog by Jonathan Drummey.

I read through Jonathan's post once and tried to create the Merimekko on my own, but that didn't work out so well. Apparently I was 99% of the way there, however, the order that I placed my dimensions in the view is what tripped me up. I decided to follow Jonathan's blog step-by-step and built a table first like he did to verify the calculations before changing the table into a chart.

This is a good practice in general when working with calculations, especially table calcs and LOD calcs. Build a table to verify the calculations are working as expected, then create your chart. It's much easier to identify problems.

Emma's rules were pretty straight forward:

• Show the 50% point of equality as a grid line 
• Set the dashboard size to 660 * 650 pixels 
• Add on the direction of less equality 
• Label the columns of the Marimekko to match hers 
• Label the first two segments Male and Female 
• Make sure the tooltips match hers

The backlog of videos that I want to create is quite extensive and including min/max indicators on sparklines has been sitting there for a while. This week's #MakeoverMonday by Ann Jackson kicked me into gear.  Thanks Ann!

First, let's have a quick look at Ann's viz:


Notice the sparklines with the indicators on the right. In this week's tip, I'll walk you through how I would go about creating those. The video is under five minutes and easily replicable. Enjoy!

Well, this certainly was a data set I never thought I'd see. Leave it to Eva to surprise us again. I'm really enjoying how she's mixing things up and allowing me to participate like everyone else. I also need to thank her for sending me her great color palettes again.

This week, we looked at the EU potato sector. Seriously! We're creating vizzes about potato production. The original website has things kind of all over the place. First there's this table:


Then there are a few donuts chart, most of them look more or less like this one:


They also included a few bars charts and a line chart. All in all, it's quite colorful.

What works well?

• Donut charts are sorted
• Tables are good for looking up specific values
• Line chart provides context by comparing to an index of 100 to make yearly change easier to understand

For Workout Wednesday week 7, I challenge you to build this dynamic trellis chart. Special thanks to Chris Love for explaining to me why the trellis works the way it does. He has a great blog post series over on The Information Lab blog.

In this example, I'm using the same Superstore Sales I've used in previous Workout Wednesdays. Here are the guidelines:

1. Match my colors.
2. The user should be able to choose the level of detail they want for the date.
3. The date axis format should not change, irrespective of the date level chosen.
4. The user should be able to pick from the list of dimensions shown.
5. The dimensions should be sorted from upper left to lower right based on the sales in the most recent time period.
6. Match my tooltips. Note that they change based on the options the user selects. Pay attention to the date formats in particular.
7. The title should update dynamically based on the date level and dimension selected.
8. The end of each line should be labeled to the right of the last point. 
9. There should be a little circle on the end of each line.
10. Each section of the trellis should include a label for the value of the dimension for that section. E.g., California should be on the upper left when you selected quarters by state.
11. The dimension labels should be centered in each section.
12. There should be no gridlines, but the zero line should be included.
13. Each row should should have a light divider between them.
14. My final view is 900x700.

This week is a double tip. First, I take you through adding a buffer to the end of sparklines so that you can have your labels next to the end of the lines. Second, I show you how to create a dynamic date selector, which I then use to create a dynamic buffer for the labels.

Because I'm tend to forget Valentine's Day (I consider it a Hallmark Holiday), it almost passed me that this week included Valentine's Day and I had intended to use another data set for Makeover Monday. Fortunately we have people on Twitter to keep us straight and this tweet changed the theme for this week.

Maybe some love or Valentine's Day dataset for #makeovermonday week 7? We love data. It's our day too. @TriMyData @VizWizBI

— Staticum (@staticum) February 11, 2017

This meant spending my Sunday morning find a new viz and data set. A quick google search turned up this infographic from KarBel Multimedia:


What I like:

• Color choices that match the theme
• Simple title that tells me what I'm about to see
• Proper sourcing
• Nice description that include a question that explains what the viz is about
• Donut chart works well here as it's only 2 slices
• Clear labeling

Another fun Workout Wednesday challenge from Emma this week. She asked us to build a butterfly chart, something I'd never done before.

Conceptually, I knew exactly what I needed to do:

1. Create separate measures for males and females
2. Create additional measures to make them percentages by dividing by the population estimate
3. Create an LOD for the national average for males and females
4. Create additional measure to make them percentages
5. Make the male axis reversed
6. Set all of the axes scales to be equivalent (Emma didn't do this, but I think the axes should be the same on each side of the center)
7. Create the calcs for the rows and columns of the trellis
8. Add a filter for local authority and add that to the title
9. Throw is all together in a dashboard size 1000x800

Aggregated extracts are an undervalued and underused feature in Tableau. 

Week 6 of #MakeoverMonday allowed us to work with 105M Chicago taxi trips from an Exasol data source. This is fantastic, until you need to publish the viz to Tableau Public which has a 15M record limit. The way to work around the 15M limit is by creating an aggregated extract, for which Tableau has created a great quick start guide here. In the video below, I show you how I created an aggregated extract with the Chicago taxi data.

Keep in mind, though, that your data has to be aggregatable. For example, if you do cohort analysis, you likely won't be able to aggregate the data and maintain the cohorts.

Why aggregated extracts?

1. Smaller extracts
2. Better performance
3. Only contains the necessary dimensions
4. Makes extract refreshes faster
5. Reduces resource burden on your Tableau Server

1. Hide unused dimensions
2. Add extract filters (optional)
3. Aggregate the data for the visible dimensions
4. Create the extract

Now THIS was a fun week! Eva and Exasol set us up with 105M records to play with. Often time, this can feel like an incredibly daunting exercise, but I took a very methodical approach. More about that in a minute. First, let's take a look at the viz we want to makeover:



What works well?

• Using a line chart to portray a time series rarely is a bad choice
• Nice simple labeling of the axes
• Title tells me what I'm looking at
• Red line on the grey background works well
• The y-axis is truncated, but the chart maintains a nice 3x2 ratio so that the line trends aren't too distorted.

I started by connecting to all 105M records live because I wasn't sure which dimensions I would end up using. When I finally finished (this took well over an hour), I created an aggregated extract. The trick to this is to hide the dimensions you aren't using before you create the extract and to rollup to the lowest date level in the view (month for my viz). This took the viz from 105M to 299K records. Incredibly, the extract was ready in less than 10 seconds. Exasol is crazy fast!

Overall, another fun week. Tonight I get to introduce this to a bunch of new Tableau users at the #MakeoverMonday Live session at Tableau HQ in London. Can't wait to see what they come up with!

Last week on my Data Viz Done Right site, I wrote about a distribution visualisation created by Harry Enten that shows the range of dates for snowfall at select U.S. cities. It's Super Bowl week, so I decided to recreated the style of Harry's viz in Tableau with the same NFL data that Emma used last week. Your challenge this week is to re-create my viz.

Below is the visualisation that I created. If you're reading this on a phone, tap on the image for the interactive version. Some requirements to keep in mind that are intentionally designed to make this tougher and to make you learn:

1. All of the elements must be floating on a dashboard sized 650x650.
2. You cannot use the Player dimension anywhere in the view.
3. Match my colors including the background
4. Create the legend (HINT: It's not an image)
5. Match the tooltip (Note the stats that are displayed in the tooltip. This will be a bit tricky. Essentially you need to count the number of players that are contained within each band.)
6. The viz should update based on the stat selected. The user should be able to choose between: Attempts, Completions, Interceptions, Touchdowns, and Yards
7. The title should update dynamically based on the stat the user selects.
8. Optional: Use Montserrat font (you can download it from Google fonts)

After the complexity of last week's data set, it's kind of nice to work on something simpler with the objective of telling a clearer story. Business Insider posted this visualisation about the growth of the economies in G-7 countries:


What works well?

• Good to add a note about rounding
• Citing the data source

What could be improved?

• A clearer title that makes the message more evident
• Countries should be sorted in descending order
• Labels should include the %
• Pie chart is a bad choice for this many categories
• Too many colors
• Chart titles are confusing (to me at least)

Inspired by the visualisation by Harry Enten that I highlighted today on my sister site DataVizDoneRight, I decide to look at the New Zealand tourism data again and see if I could build a similar view. After all, no data visualisation is ever “complete”. I really like how this turned out (and thank you to Eva Murray for feedback).

I incorporated a legend on the upper right to make the bars easier to interpret. Basically the grey bar shows the 25th to 75th percentile of all of the regions and the red dot indicates the median of all regions. I’ve removed the total region from the view.

Nice challenge from Emma this week! She’s a massive NFL fan and since the Super Bowl is upon us, she decide to challenge us to create a common baseline chart that shows the passing yards for QBs in the NFL over the course of their careers. Go to her blog for the full challenge details.

First requirement was to filter to QBs that had played at least 3 seasons and had at least 2000 total passing yards. I did this by adding a data source filter. The benefit of doing this is that my Player list will now only include those that meet the criteria and I won’t need filters elsewhere.

Next, I created a LOD calc to get the first year for each QB.

I built upon that calculation with this calculation that gives me the number of seasons played per QB. This goes onto the Columns shelf.

The cumulative passing yards is merely a running total table calc set at the Year level. This goes on the Rows shelf.

I put Player on the detail shelf to get a line per QB. I also put Year and Yds on the Detail shelf since I need those for the tooltip.

Next was a parameter to pick a QB and use that to highlight the QB chosen. I then created a simple calculation that check the Player again the parameter and put that on the color shelf.

Last was the dot on the end of each line. To do that, I created a calculation that checks if it’s the end of the line and the player selected and, if so, return the cumulative passing yards. Since this is a nested table calc, it’s important to set both table calcs to compute using Year.

Some tidying up, adding the footnotes and I was done. I decided to float all elements on the dashboard to ensure they would render exactly as I wanted them to. Another fun week of learning something new! Thanks Emma!!