Can you use Tableau to estimate Aaron Judge’s home run trajectories?

It’s community month still for #WOW2022, and this week saw Samuel Epley set this challenge to visualise the home run trajectories of Aaron Judge.

I had a little mini-break to Rome this week, so was hoping I was going to be able to get this week’s challenge done and dusted on the Tuesday evening if it landed early enough, as I wasn’t going to be around.

It did land on the Tuesday for me, but wow! it was not going to be easy! I managed to build the KPIs & the scatter plots on the Tuesday evening, and knowing I didn’t have much time, just chose to use the Home Runs stats data set only. I knew these charts weren’t going to need any data densification, so found this approach simpler.

I’m afraid I’m still constrained by time at the moment, so this post isn’t going to be the detailed walkthrough you might usually expect – sorry! I’m just going to try to pull out key points from each chart.

KPIs

I built this on a single sheet, using Measure Names and Measure Values.

I used aliases on the Measure Names (right click -> Aliases) to change the label you can see displayed ie the Distance pill is aliased to ‘Average Distance’

I also custom formatted the various numbers and applied suffixes to display the unit of measure

Note – to To get the degree symbol, I typed Alt+ 0176

Scatter Plots

I built the Exit Velocity by Distance scatter plot first, and completed all the formatting & tooltips. Then I duplicated the sheet to form the basis of the other scatter plots, and just swapped the relevant pills as needed.

For the ball shape, I loaded the provided images as custom shapes into my shapes repository. I then just created the following calculated field to use as a discrete dimension I could add to the Shape shelf

Ball Shape

[HR Number]%9

It’s not as completely randomised as perhaps it should be, but it looks random enough on the display.

The Pitcher in the data is in the format <Surname>, <Forename>, but on the tooltip it needs to display as <Forename> <Surname>, so I just used a transformation on the Pitcher field to split the field based on the comma (right click Pitcher -> Transform -> Split). This automatically created 2 fields I could use on the Tooltip.

I also noticed a very subtle wording change in the tooltip based on whether the match was Home or Away. If Home, the tooltip read ‘New York Yankees vs. <Opposition>’ otherwise it read ‘New York Yankees at <Opposition>’. I used a calculated field for this logic

TOOLTIP: vs or at

IIF([Location]=’Home’,’vs.’, ‘at’)

The Trajectory Plot

OK, so this was the hardest part of this challenge, and mainly due to getting your head round the physics involved, as so many of the calculations are dependent on each other.

I’m generally pretty confident with my maths, but this was complex, especially with the force calculations for the y-axis. Samuel stated that both gravity and drag impacted the Y-axis calcs, but it wasn’t clear to me how both these forces should be applied (a bit of trial and error and I ended up adding them within the formula).

By the time I came to tackle this challenge, Samuel had already posted a video walkthrough, which can be viewed here and is another reason why I’m not going down to the nth degree in this post.

My suggestion is to watch Samuel’s video and/or feel free to download my workbook. I built my workbook independent of Samuel’s video, so there may be steps/calculations that differ.

However, I have tried to number my calculations in the order in which I created them, so you can hopefully follow the thought process. I have also left a CHK:Data sheet in the workbook, which I used to sense check what I was doing.

All the table calculations in the CHK:Data sheet are just set to the default ‘table down’ as I have filtered the sheet to a specific Home Run (HR Number = 1) only (ie I didn’t change any of the table calc settings as I added the pills to the sheet).

However, when you build the main trajectory chart, you have multiple HR Numbers in the view, so all the table calculations must be set so that calculations are only working for each HR Number. This means that any table calc (and any nested calculations) need to have all the fields except HR Number checked

When using the Pages shelf, which isn’t something I’ve ever really had to do before, you need to Show History and adjust the various settings to get the trail lines to show

To rotate the ball (the bonus option), you need another field to use on the Shape shelf. I had lost the will to live a bit by this point, so used the formula from my friend Rosario Gauna’s solution.

Rotation Shape

STR(IIF([14-Start Position Y m] <= 0, 0,
(MIN([Time Interval]) * 1000 / 25) % 9))

Note – when you add this to the Shape shelf, and select your baseball palette, just then use the Assign Palette button to automatically assign a ball to a number – this will get them into the correct order, without you having to do it one by one.

Finally, when adding the reference average lines, be sure to set the scope to per pane rather than table, otherwise you’ll end up with the wrong figures.

I think I’ve pretty much covered all the ‘little’ points that I came across that may trip you up, aside from all the tricky calcs of course!

My published workbook is here. I hope what I’ve written is enough for you to build it yourself. I think I’d still be here next year if I tried to do anything more fully! I’m off for a lie down now!

Happy vizzin’!

Donna

Advertisement

Can you track headcount?

Luke decided to set us a challenge this week based on Human Resources (HR) data, using a dataset provided as part of the #RWFD Real World Fake Data project managed by Mark Bradbourne.

The challenge focussed on reporting monthly headcount, using a dataset that contained 1 row per employee with a column relating to the start date (Hire Date) and a column relating to the date the employee left (Termdate), which could have been a future date or NULL/blank.

I’ve worked on headcount reporting before in my day job, but have used a snapshotted data source, which captures the active employees as at midnight on the 1st day of every month. This method makes counting the headcount each month very simple, as it’s just counting the number of rows per snapshot date.

Obviously, this wasn’t an option in this case. Luke gave very little clues as to what approach would be required, apart from saying ‘you will also need to manipulate the data’….

A word about the data…

Just a little side note here…. I ended up in discussions with my fellow #WOW participant Rosario Gauna as after attempting one method, I wasn’t getting as high numbers as the solution originally suggested. There were rows in the data set that had NULL values for most of the data except job title, but I think this was only around 225 rows, so even including them I didn’t get close to the figures. Rosario however, was somehow managing to see approx 15,000 rows of null data. I ended up building and publishing my initial solutions based on a version of the data sent to me directly by Rosario.

Since then, it seems others were also confused, and so the solution was adjusted to ignore all the null rows. As a result I reworked one of my solutions to use the original data source I downloaded from the site, and it’s this I will focus on in the blog.

A word about my solutions…

With little instruction, it took a bit of head-scratching to figure out how to work through the problem. I initially decided to use a scaffold dataset to help. I built a simple file in Excel which contained 1 column listing 1 date per month which went from 01 Jan 2017 to 01 Jan 2021. I then joined this to the HR data in Tableau Desktop using the physical data layer (ie I did not use relations), using join calculations as below (note, this is using the amended copy of data I received where Termdate was already defined as a date field.

The solution using this method is published here.

However, I didn’t feel completely comfortable with this approach. Luke had suggested the difficulty of this was 9/10 and referred to nested tableau calculations… the above solution just used quick table calcs. So I decided to see if I could come up with an alternative, just using the provided data, and that’s what I’ll document below.

Setting up the data

I downloaded the Human Resources.csv file from the link provided, and connected to it in Tableau Desktop. I found the Termdate field presented as a string field, so I converted it to a date, just by clicking on the ABC symbol of the data type against the field. I also added a data source filter to exclude all records where the Id field was NULL.

I then added a new data source, and connected to the same file again. This meant I had 2 instances of the Human Resources data source listed in my Data pane. I renamed the 2nd instance to be Leavers Only, changed the data type of the Termdate field again and added a data source filter so that only records with non-Null Termdate values were retained.

Note – What I will be doing here is using the Human Resources data source to manage the months we’re reporting headcount over. This method only works if there is at least 1 new starter every month, which in this case there is.

Building the calculations

The requirements state that a new starter doesn’t count in the monthly headcount figures until the following month, so the date field we need to use in the output, needs to be created as

Report Date

DATE(DATETRUNC(‘month’,DATEADD(‘month’,1, [Hire Date])))

This basically shifts the date the new starter joined to the 1st day of the following month.

To make the calculations easier to read, I also created

Starters

COUNT([2022_04_06_WW14_Human Resources.csv])

which is literally a reference to the field that automatically gets added as part of the connection – it’s just a bit of a mouthful as its named according the data connection.

Let’s pop these out in a table as below – I’m deliberately using the dateparts of the Report Date field as they’ll be used in the viz like this.

So this is just showing the new starters against the month at which they will count for headcount. We need to get details of leavers now, and we’ll do this by blending.

When I blend data sources, I tend to create specific fields so it’s clear to me that they’re being used for blending, rather than defining blend relationships between differently named fields. So in the Human Resources data source I’ve created

BLEND – Date

[Report Date]

and then in the Leavers data source, I’ve created

BLEND – Date

DATE(DATETRUNC(‘month’,DATEADD(‘month’,1,[Termdate])))

which is shifting the Termdate forward to the the 1st of the next month, since leavers should also only be recorded in the following month.

In the Human Resources data source I then created

Leavers Per Month

ZN(COUNT([2022_04_06_WW14_Human Resources – Leavers Only].[2022_04_06_WW14_Human Resources.csv]))

which is just referencing the automatically generated ‘count’ field from the Leavers data source. I’ve wrapped it in ZN so 0 is reported in the event no match is found.

Add Leavers Per Month into the table, and ensure the linking field from the secondary Leaver data source is connected on the BLEND – Date field

So now we know how many starters and leavers per month, we now need to total these up by generating a cumulative running sum of the starters, but subtracting any leavers along the way.

Official Headcount

([Starters] + PREVIOUS_VALUE(0)) – [Leavers Per Month]

Takes the value of the Starters in the current row, adds it to the value of Official Headcount from the previous row, then subtracts any Leavers recorded against the current row. Add this onto the table, and as its a table calculation, edit it so it is explicitly computing my the month & year of Report Date

For the annual change, we’re looking to compare the difference of the Official Headcount value for the current month eg Jan 2021, with the Official Headcount value for the same month 1 year ago eg Jan 2020.

Annual Change

(ZN([Official Headcount]) – LOOKUP(ZN([Official Headcount]), -12)) / ABS(LOOKUP(ZN([Official Headcount]), -12))

Take the Official Headcount from the current row, subtract the Official Headcount from 12 rows before, then divide the result by the Official Headcount from 12 rows before.

Format this to a percentage at 1 dp. (For some reason my numbers here don’t seem to match Luke’s revised solution…).

Again add to the sheet and set the table calcs of both the nested calcs to compute by month and year

The final measure was a bit of a strangely named one IMO. I felt I should be doing some rolling calculation, but ultimately, it just seemed to be reporting the difference between the headcount now and that 12 months ago ie the numerator of the calculation above.

Net Rolling 12 Change

(ZN([Official Headcount]) – LOOKUP(ZN([Official Headcount]), -12))

Pop this into the view, adjust the table calc settings again

Building the viz

On a new sheet, add Report Date at the month level, but then set to be continuous, to Columns and Report Date at the year level to Colour. Add Official Headcount to Rows (adding the linking field for the blend when prompted), and adjust the table calc settings to compute over both month & year, ensuring the Year of Report Date is listed first.

Then in the legend, select all the years from 2000-2017 and ‘hide’ (this is probably a cheat way, but the quickest – we need to retain the data from the previous years, so can’t just ‘filter’).

Edit the axis to not start from zero. Adjust colours to suit.

Add Annual Change and Net Rolling 12 Change to the Rows shelf, adjust both of the table calc settings, so all nested calcs are computing by Year & then Month.

Edit the axis of both these additional measures to also exclude 0. On the All marks card, click the Label button and tick the Show mark labels option.

Final steps ….

…on the All marks card, click the Tooltip button and uncheck Show tooltips so no tooltips display on hover.

Right click on the Month axis and format the axis so the dates in the Scale section are displayed as abbreviated dates

Edit the same axis and remove the title, then format the whole chart to

  • remove all row and column dividers
  • to set the row banding with a band size of 1
  • adjust the colour of the row and column gridlines to be a slightly darker shade of grey

And you’re done 🙂 My published viz is here

Had a bit of all sorts this week… I’ll be intrigued to see how Luke solved it!

Happy vizzin’!

Donna

Filtering Table Calcs

It was Sean Miller’s turn this week to set the challenge, which had an added twist – to complete only using the web authoring feature in Tableau Public, something I’d never tried before.

Most challenges use a version of Superstore Sales which I usually already have on my laptop, due to the multiple instances of Tableau Desktop installed, but this time, I had to get the data from data.world. I saved this locally then logged into Tableau Public online and started the process to build a viz online, by clicking the Create a Viz (Beta) option.

When prompted I then chose to upload from computer, the data source file I’d downloaded, and once done I was presented with the online canvas to start building

Unsure as to whether there was any auto save facility online, I decided to save the workbook pretty much immediately, although I was first prompted to navigate to the Data Source tab and create an Extract before I could save.

Now I was ready to start building out the requirements.

I started with the trend chart. The grain of the chart was at the month level, so I first created a specific field to store the Order Date at this particular level

Order Date Month

DATE(DATETRUNC(‘month’,[Order Date]))

I then added this to the Columns shelf as a continuous (green) exact date, and added Sales to the Rows shelf. I changed the mark type to Area, set the Colour to a blue, dropping the opacity to 50% in an attempt to match Sean’s colouring and adjusted the tooltip.

I then added another instance of Sales to the Rows shelf next to the existing one, changing the mark type of this instance to Line and setting the Colour to a darker blue at 100% opacity. I then used the context menu on this pill to set a Quick Table Calculation of type Moving Average.

By default this sets the average to the 3-month rolling average as per the requirement. I adjusted the tooltip, made the chart dual axis and synchronised the axis, then removed Measure Names from the Colour shelf that had automatically been added.

My next step at this point was to work on the filtering requirement. To help with this, the first thing I did was to Duplicate as crosstab (right click on the chart tab). This created a tabular view of the data so I could work on getting the calculations I need for the filtering. I flipped the columns and rows, so my starting point was as below.

I chose to use parameters to capture the min and max dates that the user selects on the dashboard.

pMinDate

Date parameter defaulted to 01 Jan 1900

And I also created pMaxDate exactly the same way.

I then needed fields to store the relevant dates depending on whether a selection had been made or not

Min Date Selected

IF [pMinDate]=#1900-01-01# THEN {FIXED : MIN([Order Date Month])}
ELSE [pMinDate]
END

The FIXED statement sets the date to the minimum month in the data set, if the parameter is set to its default. The below doed similar to get the maximum date in the data set.

Max Date Selected

IF [pMaxDate]=#1900-01-01# THEN {FIXED : MAX([Order Date Month])}
ELSE [pMaxDate]
END

Using these dates, I then created a field to determine whether the month was within the min & max dates

Is Month Selected?

[Order Date Month]>=[Min Date Selected] AND
[Order Date Month]<= [Max Date Selected]

Pop this field onto the table view, and show the parameters and set them to some other values eg Min Date = 01 Jun 2018 and Max Date = 01 Sept 2018. You’ll also need to adjust the table calc setting of the Moving Average Sales pill to compute by Is Month Selected as well.

You’ll see True is stamped against the rows which match the date range. Now, if you make a note of the Moving Average values against these rows, and then add Is Month Selected? = True to the Filter shelf, you should get just the True rows displaying, BUT the Moving Average values are now different. This is because, with this type of field as a filter, the filter is being applied BEFORE the table calculation is being applied – the table calculation essentially has ‘no knowledge’ of the rest of the rows of data as they have been filtered out. Solving this issue is the crux of this challenge.

Remove Is Month Selected? from the filter shelf. We’re going to create a new field to filter by instead

FILTER

LOOKUP(MIN([Is Month Selected?]),0)

This field uses the LOOKUP table calculation, and is basically looking at the value of the Is Month Selected? field in it’s ‘own row’ (denoted by the 0 parameter).

Pop this onto the table view, and you can see that it’s essentially a duplicate of the Is Month Selected? field.

Now add the FILTER =True field to the Filter shelf instead, and this time the rows will once again filter as required, but the Moving Average values will be the same as those in the unfiltered list, which is what we need.

This works because the field we’re filtering by is a table calculation. When added to the filter shelf, these types of fields are applied at the ‘end’ of the processing; ie the table calculation is computed against all the rows in the data AND THEN the filter gets applied. As a result the Moving Average values are retained.

This is all part of Tableau’s Order Of Operations and is a fundamental concept when working with filters.

Armed with this knowledge, we can now apply the FILTER=True to the Filter shelf of the chart we built.

Now we can start to work on the bar charts, which will show Sales along with Average Monthly Sales. For this 2nd measure, I need to work out the number of months captured between the min & max dates.

No. Months

DATEDIFF(‘month’, [Min Date Selected],[Max Date Selected])+1

And with this I can now create

Avg Monthly Sales

SUM([Sales])/MIN([No. Months])

The bar chart viz can then be quickly created by adding Sub-Category to Rows and both Sales and Avg Monthly Sales to Columns, and sorting by Sales descending. The colours of each bar are adjusted to suit.

The bars also need to react to the dates selected, but we can simply add the Is Month Selected? = True to the Filter shelf for this, as none of the values displayed on this chart are reliant on table calculations.

To remove the gridlines from both charts, select Format > Workbook from the menu, scroll down the Format dialog displayed on the right and set Gridlines to Off

Finally, title both sheets appropriately, and then add to a dashboard. We need to add 3 actions to this dashboard – 2 parameter actions and 1 filter action.

Create a parameter action which will on Select, set the pMinDate parameter by passing the Minimum Order Date Month, and will reset back to 01 Jan 1900 when unselected.

Then create another instance, which sets the pMaxDate parameter instead, passing the Maximum Order Date Month.

Finally, create a Filter Action which on Select of the bar chart, filters the trend chart

And with that, the challenge should be complete. My published viz is here.

Happy vizzin’! Stay Safe!

Donna

How many consecutive starts?

Another table calculation related challenge this week, set by Luke, visualising cumulative starts for NFL Quarterbacks per team from 2006.

Luke provided the data within a Tableau workbook template on Tableau Public, so I started by downloading the workbook and understanding the data structure.

The challenge talks about teams playing over 17 weeks, but the data showed some data associated to weeks 28-32. So I excluded these weeks by filtering them out.

I then started to build out the data in tabular form, so I could start to build up what was required. I added Team, Season, Week and Player ID to Rows, and just to reduce the amount of data I was working with while I built up the calcs, chose to filter to the Teams ARZ, ATL & BLT.

What we’re looking to do is examine each Player ID and work out whether it is the first record for the Team or whether it differs from the previous row’s data. If so then we’re at the ‘start’ of a run, so we record a ‘counter’ value of 1. If not, the values match, so we need to increment the counter.

We’ll do this in stages.

Firstly, let’s get the previous Player ID value.

Prev Player ID

LOOKUP(MIN([Player Id]),-1)

This ‘looks up’ the record in the previous (-1) row. Change this field to be discrete and add to the Rows. Set the table calculation to compute by all fields except Team.

Each Prev Player ID value matches the Player ID from the row before, unless its the first row for a new Team in which case the value is Null.

Then we can create a field to check if the values match

Match Prev Player ID

MIN([Player Id])=[Prev Player ID]

Add this to the view and set the table calc as above, and the data shows True, False or NULL

Now we can work out the consecutive streak values

Consecutive Streak

IF (NOT([Match Prev Player ID])) OR ISNULL([Match Prev Player ID]) THEN 1
ELSE 1+PREVIOUS_VALUE(-1)
END

If we don’t match the previous value or we’re at the start of a new team (as value is NULL), then start the streak by setting the counter to 1, otherwise increment the counter. Add this to the view and set the table calc for both the nested calculations as per the settings described above.

Next we need to identify the last value of the Consecutive Streak for each Team.

Current Streak

WINDOW_MAX(IF LAST()=0 THEN [Consecutive Streak] END)

The inner IF statement, will return the value of Consecutive Streak stored against the last row for the Team. All other rows will be Null/blank. The WINDOW_MAX() statement then ‘spreads’ this value across all the rows for the Team.

Add this onto the view, and set the table calc for all the nested calcs.

Finally, we need one more bit of data. The chart essentially plots values from 2006 week 1 through to 2020 week 17. We need to ‘index’ these values, so we have a continuous week number value from the 1st week. We can use the Index table calculation for this

Index

INDEX()

Add this field to the view, set it to be discrete (blue pill) and position after the Week field on the Rows. Set the table calc as usual, so the Index restarts at each Team.

Now we’ve got all the data points we need, we can build the viz. I did this by duplicating the tabular view and then

  • Remove Prev Player ID and Match Prev Player ID
  • Move Season, Week and Player ID from Rows to Detail
  • Move Current Streak from Text to Rows and change to be discrete (blue)
  • Move Index from Rows to Columns and change to be continuous (green)
  • Move Consecutive Streak from Text to Rows
  • Change mark type to bar, and set to fit width to expand the view.
  • Change Size to be Fixed, width size 1 and aligned right
  • Set the border on the Colour shelf to be None.
  • Remove the Team filter and adjust the row height

All that’s left now is to set the tooltip (add Player to the Tooltip shelf to help this), and then apply the formatting. You can use workbook formatting (Format -> Workbook menu) to set all the font to Times New Roman.

Hopefully this is enough to get you to the end 🙂 My published viz is here.

Happy vizzin’! Stay Safe!

Donna

Can you build a Customer Lifetime Value Matrix?

Ann Jackson challenged us this week to build this matrix depicting the average worth of customer cohorts during their lifetime.

This challenge involves a mix of LoDs (Level of Detail calculations) and table calculations.

First up , we need to define our customer cohorts (ie group the customers), which for this challenge is based on identifying the quarter they placed their first order in. This will involve an LoD calculation. For a good introduction to LoDs with some worked examples (including a similar cohort analysis example), check out this Tableau blog post.

ACQUISITION QUARTER

DATE(DATETRUNC(‘quarter’,{FIXED [Customer ID] : MIN([Order Date])}))

The 2nd part of the formula in the { … } returns the earliest Order Date associated to the Customer ID, which is then truncated to the 1st day of the quarter that date falls in ie 23 Feb 2019 is truncated to 01 Jan 2019.

For the ‘quarters since birth’ field, we need to calculate the difference in quarters, between the ACQUISITION QUARTER and the ‘quarter’ associated to the Order Date of each order in the dataset.

QUARTERS SINCE BIRTH

DATEDIFF(‘quarter’,[ACQUISITION QUARTER],DATETRUNC(‘quarter’,[Order Date]))

Drag this field into the ‘dimensions’ area of the left hand data pane (above the line if you’re using later versions of Tableau).

Lets sense check what this looks like, by adding

  • ACQUISITION QUARTER to Rows (Discrete, Exact Date)
  • ORDER DATE to Rows, set to Quarter (quarter year ie May 2015 format which will make a green pill), then set to discrete to change to blue
  • QUARTERS SINCE BIRTH to Rows

You can see that while the first row against each cohort starts with a different quarter, the QUARTERS SINCE BIRTH always starts at 0 and counts sequentially down the table.

Next we want to count the number of distinct customers in each cohort, and we’ll use another LOD for this.

CUSTOMERS

{FIXED [ACQUISITION QUARTER]: COUNTD([Customer ID])}

Once again move this field into the Dimensions section of the data pane.

Add this onto the Rows of the above data table, and you should get every row for the same cohort displaying the same number

Add Sales onto Text to get the value of sales made by the customer in each cohort in each quarter. The ‘customer lifetime value’ we need is defined as the total sales so far / number of customers in the cohort.

Remove the QUARTER(Order Date) field from the table, as we’re not going to need this for the display, and it’ll affect the next steps if it’s left.

To get the cumulative sales, we need a Running Total Quick Table Calculation. Click on the Sales pill on the Text shelf and select Quick Table Calculation -> Running Total. The click again and Compute By -> QUARTERS SINCE BIRTH. Add Sales back into the table, so you can see the quarterly Sales value and how it’s cumulating until it reaches the next cohort.

We’ve now got the building blocks we need for the CLTV value we need to plot

Avg Lifetime Value

RUNNING_SUM(SUM([Sales])) / SUM([CUSTOMERS])

Note – I purposefully haven’t called this field what you might expect, as I’m going to ‘fill in the gaps’ that Ann describes in the requirements, and I’ll use that name then.

Pop this field into the table above, again setting the table calculation to compute by QUARTERS SINCE BIRTH

You can now use the data table above to validate the calculation is what you expected.

Now let’s build the viz out.

On a new sheet

  • QUARTERS SINCE BIRTH to Columns
  • ACQUISITION QUARTER (exact date, discrete blue pill) to Rows
  • Avg Lifetime Value to Text, setting the table calculation to Compute By QUARTERS SINCE BIRTH

From this basic text table, you can see the ‘blank’ fields, Ann mentioned. In the data table view, it’s not so obvious. The blank is there because there are no sales in those quarters for those cohorts. To fix we need another table calculation

CUSTOMER LIFETIME VALUE (CLTV)

IF ISNULL([Avg Lifetime Value]) AND NOT ISNULL(LOOKUP([Avg Lifetime Value],-1)) AND NOT ISNULL(LOOKUP([Avg Lifetime Value],1))

THEN LOOKUP([Avg Lifetime Value],-1) ELSE [Avg Lifetime Value] END

This says, if the Avg Lifetime Value field is NULL but neither the previous or the subsequent values are NULL, then use the Avg Lifetime Value value from the previous column (LOOKUP).

Replace the Avg Lifetime Value with the CUSTOMER LIFETIME VALUE (CLTV) field (setting the Compute By again), and the empty spaces have disappeared.

If you hover over the cells in the lower right hand side of the view, you’ll see tooltips showing, indicating that a mark has been drawn on the viz with Null data. To fix this, add CUSTOMER LIFETIME VALUE (CLTV) to the Filter shelf and specify non-null values only to show via the Special tab.

Now if you hover over that area you don’t get any tooltips displaying, as there aren’t any marks there.

Now it’s just a case of formatting the viz a bit more

  • Add CUSTOMERS to Rows
  • Add CUSTOMER LIFETIME VALUE (CLTV) to the Colour shelf by holding down the Ctrl key then clicking on the field that’s already on the Text shelf, and dragging via the mouse onto the Colour shelf. Using Ctrl in this way has the effect of copying the field including the table calculation settings, so you don’t need to apply them again. This will change the colour of the Text.
  • Then change the mark type to Square, which will then fill out the background based on the colour.
  • Then edit the colour legend to the relevant palette (which you may need to install via Ann’s link).
  • Set the border of the mark via the Colour shelf to white
  • Remove the row & column dividers
  • Set the row Axis Ruler to a dark black/grey line
  • Format the 1st 2 columns so the font is the same and centred. Widen the columns if required.
  • Update the tooltip

And then you should be ready to add the viz to your dashboard. My published version is here.

This blog is a bit more detailed that my recent posts, but I’m also conscious I’ve skipped over some bits that if you’re brand new to Tableau, you may not be sure how to do. Feel free to add comments if you need help!

Happy vizzin’! Stay Safe!

Donna

Can you compare a 3-day vs 14-day moving average and describe the latest trend?

This week for #WOW2020, Ann provided a table calculation feast of a challenge! This certainly is not for the faint-hearted! As well as cracking all the table calcs, the challenge features multiple views, measure swapping, parameters, BANs, filtering, sorting …. it’s got it all going on!

Ann hinted you’d probably want to start with the table, and even if there hadn’t been a table output in the display, this is what I would have done. If you’ve read enough of my blogs, you’ll know I often like to build up a ‘check data’ sheet, which just contains the data I need in tabular form as a quick reference. When working with table calculations this is an absolute must have!

So let’s build out that Check Data table to start with. I have a feeling this is going to be a lengthy blog 🙂

Initial Set up

First up, the requirements stated that the latest date would be 7 June, but I found records with a 8 June date. All the associated info for this date was null though, so I set a data source filter to exclude this. This means I wouldn’t get any issues if I needed to store the max date in a FIXED LoD calculation at any point.

I also found it easier to rename a couple of the measures provided to match the output, so rename PEOPLE_POSITIVE_NEW_CASES_COUNT to New Cases and PEOPLE_POSITIVE_CASES_COUNT to Reported Cases. I’ll refer to these renamed fields going forward.

Building all the Calculated Fields

To build out the table, we’re just going to focus on one State & County, as there’s a lot of data. So add Province State Name = Tennessee and County = Davidson to the Filter shelf.

Add Report Date (discrete exact date – blue pill) and New Cases & Reported to Rows. As you scroll down, you’ll see data starting to come in on 8 March.

We want to create our moving average calculations

3 Day Moving Avg

WINDOW_AVG(SUM([New Cases]), -2, 0)

14 Day Moving Avg

WINDOW_AVG(SUM([New Cases]), -13, 0)

Notice the number of rows to average over is 1 less than you might expect, as the current row is included, so the calculation is saying ‘current row’ and 2 | 13 previous rows.

Add these to the table, and adjust the table calculation so it is explicitly calculating by Report Date. This would have happened automatically, as the calculation would have been computing ‘down’ the table, but it’s best to fix the computation, so it doesn’t matter where the pill gets moved to in the view.

We now need to work out whether there is an increase or not between the 3-day and 14-day average.

Is Increase?

IF [3 Day Moving Avg] > [14 Day Moving Avg] THEN 1 ELSE 0 END

Is Decrease?

IF [3 Day Moving Avg] <= [14 Day Moving Avg] THEN 1 ELSE 0 END

I’m using 1s and 0s as it’s going to help with a later calculation.

NOTE – I’m assuming that if there is ‘no change’ it’ll be recorded as a decrease. This is how I interpreted the requirement, “ …whether it is an increase or a decrease (or no change)” and it wasn’t easy to find any matches anyway.

I also need some text to indicate the increase or decrease

Increase | Decrease

UPPER(IF [Is Increase?]=1 THEN ‘Increase’ ELSE ‘Decrease’ END)

The UPPER is used as that’s part of the tooltip formatting.

Let’s get these onto the view, always making sure the table calculations are set to Report Date.

We need to calculate the number of days that has been reported INCREASE in succession, and the number of days where successive DECREASE has been reported.

So first, let’s identify which rows match the previous row.

Match Prev Value?

LOOKUP([Is Increase?],-1) = [Is Increase?]

If the value of the Is Increase? field in the previous (-1) row is the same as the Is Increase? field in the current row, then this is true, else false.

Add to the view, and verify the table calculation for itself and all nested calculations being referenced, is set to Report Date.

We now have all the information we need to help us work out the number of days in the increase/decrease ‘trend’.

Days in Trend

IF (FIRST()=0) OR(NOT([Match Prev Value?])) THEN 1
ELSEIF [Increase | Decrease] = ‘INCREASE’ THEN ([Is Increase?]+PREVIOUS_VALUE([Is Increase?]))
ELSEIF [Increase | Decrease] = ‘DECREASE’ THEN ([Is Decrease? ]+PREVIOUS_VALUE([Is Decrease? ]))
END

If the row in the table is the very first entry (so there’s nothing previous to compare against), or the row in the table didn’t match it’s predecessor (ie there was a change), then we’re starting a new ‘trend run’, which obviously starts at 1.

Otherwise, if the current row we’re on indicates an increase, then we’ll add the value of the Is Increase? field (which is 1) to the previous value (which is also 1). PREVIOUS_VALUE works recursively though, so it essentially builds up a running sum, which gives our trend.

We ultimately do the same thing using the Is Decrease? column. This is why using 1 & 0s in the earlier calculation help.

Adding into the view, and setting the table calculation correctly, you should get something similar to this…

Finally, there’s one key field we need to add; something to help identify the latest row as we will need it for filtering in the table that’s displayed on the dashboard. Simply applying a standard ‘quick filter’ won’t work, as the table requires we show the 3-day & 14-day moving averages. A ‘quick filter’ to limit the data to the latest date (7th June), will show the wrong values, as the data related to the other days will be filtered out, so the table calc won’t have the information to correctly compute over.

We need to create another table calculation that we can use as a filter, and that due to Tableau’s ‘order of operations’ will apply later in the filtering process than a traditional quick filter.

Max Date

{FIXED : MAX([Report Date])}

The latest date in the whole data set.

Show Data for Latest Date

LOOKUP(MIN([Report Date]),0) = MIN([Max Date])

If the Report Date of the current row is the same as the maximum date in the whole data set, then return true.

We’ve now got all the core data components we need to create the various charts.

In the interest of time (my time in writing this out), I’m going to attempt not to describe the building of all the charts in too much detail, but just call out the useful bits you might need. If you’re attempting this challenge with the table calcs above, I’m assuming you know Tableau enough to not need everything defined to the lowest level.

The whole report is driven off a parameter which the user must enter a State – County combo.

You’ll need a calculated field to store the combo

State – County

[Province State Name] + ‘ – ‘ + [County]

and then create a parameter (State – County Parameter) off of this (right click, Create -> Parameter) which will create a string parameter with all the permutations.

When displaying on the dashboard, set this to be of type Type In

BAN

The BAN is a basic summary of the latest trend for the entered state county.

We need to filter the sheet to the value entered in the parameter

Is Selected State County?

[State – County Parameter] = [State – County]

Add this to the Filter shelf as true, along with the Show Data for Latest Date.

Add the relevant fields to the Text shelf to display the required text. The Report Date needs to be custom formatted to ddd, mmm d to get the Sun, Jun 7 display

Map

For the map, as well as filtering the latest date, we’re also going to need to filter just to the state only (not state & county) as above. So I created

Is Selected State?

LEFT([State – County Parameter], FIND([State – County Parameter],’-‘)-2) = [Province State Name]

This is unpicking the State – County combined string stored in the parameter, to just find the State part and compare to the Province State Name.

Build a filled map based on County and filter to the latest date and the selected state. I set the Map Layers to that below, which seems to match up

You’ll need to set both the Is Selected State County? and Increase|Decrease fields to the Colour shelf.

Bar & Line Chart

You’re going to need a few more calculated fields for this.

Moving Avg Selector

for the user to choose what the line should display. I’ve set it to an intger parameter that displays text

We then need a field to show on the display depending on what’s been selected in the parameter

Moving Avg to Display

If [Moving Avg Selector] = 3 THEN [3 Day Moving Avg] ELSE [14 Day Moving Avg] END

You’ll need a Dual Axis chart plotting New Cases and Moving Avg to Display against Report Date (continuous exact date)

The data only starts from 8th March, so I added Report Date to filter to start from 8th March. 8th March is also added as a constant reference line.

Table

Based on the State / County entered, the table is filtered to show the data for the latest date for all the counties in the state entered. Although not stated in the requirements, the first row is the county selected, with the rest ordered by Reported Cases.

You can get the selected county to the top, by adding Is Selected State County as a hidden field to the Rows, and moving ‘True’ to the top.

And that should be everything you need to build the dashboard, which is pretty much just stacking all the sheets one on top of each other in a single column.

My published viz is here.

Happy vizzin’! Stay Safe!

Donna

Can you create a chart with a dynamic week start?

This week Ann set the challenge and suggested she had ‘distilled it down a bit’ so it wouldn’t take as long to solve….hmmmm… not sure if that really worked out for me…. I certainly managed to crack through it to begin with, but the requirement to ‘show the data when there were missing values’ seemed to end up being trickier than I thought it would be, and the trying to display a date on the tooltip for those days…. well, that was just so freakin’ hard! Given this was ‘distilled down’, I thought I must have been missing something pretty obvious to be struggling so much. I played round with so many different table calcs to try to get this to work, but ended up admitting defeat and peaked at Ann’s solution.

The missing dates bit… that ended up being a bit odd. I rebuilt my chart multiple times, starting with a table of data as a ‘check’ and then carefully converting that into the required Viz. Eventually it just seemed to work, which left me very puzzled, as I was pretty certain what I ended up with, I’d had previously… I can only assume there is something about the order of the steps I took that may have caused the problem.

As I write this blog out, I’ll be rebuilding as I go, to generate the various screen shots, so fingers crossed, if you follow step by step, you’ll get a working solution 🙂

Right let’s get onto it…

Building out the data we need

As I’ve done before, I’m going to build up the fields I need into a tabular format, so I can ratify I’ve got everything correct before I build up the viz.

First off, we need a couple of parameters to drive the inputs for the chart :

Week Ending On

A date parameter set to 24th Oct 2019, that allows all values

Include X Prior Weeks

an integer parameter set to 10 by default

From the requirements, we need to display data for the 7 days of the week up to the Week Ending On date, along with data for each of the 7 days in the previous number of weeks as stated in the Include X Prior Weeks parameter.

So first up we need to work out when the ‘latest’ week starts

Start of Selected Week

DATEADD(‘day’,-6,[1a.Week Ending On])

Then I want to group all the Order Dates into their respective weeks, where each week needs to start on the same day of the week as the Start of Selected Week field. eg if Start of Selected Week is Friday 18th Oct, then I want all the dates of the previous x weeks to be grouped to the Friday of the week they’re in. So all the dates from Friday 11th Oct to Thurs 17th Oct are grouped as Fri 11th Oct, dates from Friday 4th Oct to Thurs 10th Oct are grouped as Fri 4th Oct, and so on.

Now the ideal way I think I should be able to do this is use a DATETRUNC function passing in the name of the day of the week, the week should start, which is based on day associated to the Start of Selected Week field. So I wanted to write something like this:

DATETRUNC(‘week’, [Order Date],DATENAME(‘weekday’,[Start of Selected Week]))

but Tableau doesn’t like the fact the 3rd parameter is a calculation rather than simply the text ‘Friday’.

So instead I had to build up the logic as follows

Order Date Week

CASE DATEPART(‘weekday’, [Start of Selected Week])
WHEN 1 THEN DATETRUNC(‘week’, [Order Date], ‘sunday’)
WHEN 2 THEN DATETRUNC(‘week’, [Order Date], ‘monday’)
WHEN 3 THEN DATETRUNC(‘week’, [Order Date], ‘tuesday’)
WHEN 4 THEN DATETRUNC(‘week’, [Order Date], ‘wednesday’)
WHEN 5 THEN DATETRUNC(‘week’, [Order Date], ‘thursday’)
WHEN 6 THEN DATETRUNC(‘week’, [Order Date], ‘friday’)
WHEN 7 THEN DATETRUNC(‘week’, [Order Date], ‘saturday’)
END

This Order Date Week is essentially the field that represents each line on the final viz.

I also need a field that I can used to represent the x-axis on the final viz, which is labelled based on the days of the week, but (as per the challenge), is starting at a different day.

To do this, I chose to ‘baseline’ all the dates in each week, to be aligned to the ‘current’ week. What I mean by this is that I want to create a field whereby if the Order Date is a Friday, the field will store the date of the Friday of the ‘latest’ week, in this case Friday 18th Oct. Any Order Date on a Saturday will be ‘aligned’ to Saturday 19th Oct and so on.

Order Date Baseline

DATE(DATEADD(‘day’, 7 * DATEDIFF(‘week’,[Order Date Week],[Start of Selected Week]), [Order Date]))

This finds the number of weeks between the week the order date is in and the start of the ‘latest’ week, which is then multiplied by 7 to convert to a days difference, which is then added onto the Order Date itself.

So from below, you can see both Fri 13th Sept & Fri 20th Sept map to 18th Oct; Tue 17th Sept and Tue 24th Sept both map to 22nd Oct etc.

I now need another field to restrict the dates being displayed to those within the timeframe dictated by the Include X Prior Weeks parameter.

Dates to Include

[Order Date]>=DATEADD(‘week’,-1 * [Include X Prior Weeks], [Start of Selected Week])
AND [Order Date] <= [Week Ending On]

So let’s start building the check table of data as follows:

  • Order Date Week on Rows as discrete, exact date
  • Order Date Baseline on Columns as discrete, exact date
  • Dates To Include on Filter shelf, set to True
  • Sales on Text

You can see we have some gaps where there are no Sales, we need these to display 0, which we do using

Inc Null Sales

IFNULL(LOOKUP(SUM([Sales]),0),0)

This is a simple table calc that is ‘looking up’ its own value, and if it’s null, sets itself to 0 otherwise uses it’s own value.

So let’s put this on the Text shelf instead

We’re also going to need a field to define the ‘latest week’, as this is coloured differently on the viz

Is Latest Week

[Order Date Week]=[Start of Selected Week]

Bring this into the table

We’ve now got the core things we need to build the majority of the chart.

Building the chart

Firstly, duplicate the table above, then move the pills round as follows:

  • Move Is Latest Week to Colour shelf, and adjust colours accordingly
  • Move Order Date Week to Detail shelf
  • Move Inc Null Sales to Columns shelf
  • Swap the Colour Legend so True is first (makes the latest line to the front)
  • Format the Order Date Baseline field to custom formatting of dddd which will just the display the day of the week rather than the full date (sneaky huh?)

Then tidy up the formatting to remove gridlines, get the axis in the right format, change the font of the axis labels etc.

And that’s the main viz… now we need to sort the tooltips.

Tooltips

On the tooltip we need to display

  • The order date
  • The sales value or ‘no sales’ if there are no sales
  • The start and end day of the week

Let’s start backwards.

Day of Week Start

DATENAME(‘weekday’,[Start of Selected Week])

Day of Week End

DATENAME(‘weekday’, [Week Ending On])

Put these on the Detail shelf (as we’re going to need them for the title too).

Add the Sales onto Tooltip and format it to $ with 0dp. This will display the Sales value when there is one, and ‘nothing’ if there are no sales (as opposed to the 0 that the Inc Null Sales field will show.

To get the ‘no sales’ text we need

Tooltip: No Sales

IF IFNULL(LOOKUP(SUM([Sales]),0),0)=0 THEN ‘no sales’ END

Note – this is what I did and is in my published solution, but I could have just checked whether Inc Null Sales = 0 – doh!

Add this onto the Tooltip and position directly adjacent to the SUM([Sales]) field, so either one will be displayed (they can never both display).

Right, now we need to get the date.

The problem we have is that like the no sales, if there’s no orders on a date there’s no date to display either, but unlike the ‘no sales’ value, we haven’t got a single default value like 0 to use. Instead we need to work out what the missing date should be.

I tried various things for this, but finally had to look at Ann’s solution to get this, which is :

Tooltip: Order Date

IFNULL(ATTR([Order Date]),
IFNULL(DATEADD(‘day’,1,LOOKUP(ATTR([Order Date]),-1)),DATEADD(‘day’,-1,LOOKUP(ATTR([Order Date]),1))))

Urgh! Horrid right!… so what is this saying…

  • If the current order date doesn’t exist, then
  • lookup the previous order date and add 1 day to it, but if that is also null then
  • lookup the next order date and take 1 day off it

So this is only working on the assumption that there are not 2 days in a row with no orders.

Add this to the Tooltip and format accordingly to get the layout required.

Finally amend the sheet title to match the title displayed. The title can reference the Day of Week Start and Day of Week End fields along with the Week Ending On and the Include X Prior Weeks parameters.

When you then add onto the dashboard, make the parameters floating and position them top right.

Hopefully, you’ve ended up with a working solution. I know I approached mine slightly differently from Ann’s, so it’s worth checking hers out too.

My published viz is here.

Happy vizzin’! Stay Safe!

Donna