# COVID-19 New Case Trends Cartogram

The challenge this time was set by Luke using the data being collated via Tableau’s Covid-19 data hub.

This viz is essentially equivalent to a small multiple display where the charts for a specific dimension (in this case State) get displayed across numerous rows and columns. The difference here, is the row and column for the State to be displayed in, is specifically defined, rather than just sequentially based on how the data is being sorted.

Luke very kindly provided the logic to determine the rows and columns.

Building out the data

Once again, I’m going to start by putting all my data into a table so I can check my calculations, especially since this challenge does involve table calculations (there’s a hint on the Latest Challenges page)

Data Source Filter

Although not explicitly mentioned in the requirements, the information we need to present is based on the dates from 1st March 2020 to 31st July 2020. I messaged Luke to check this, before I then realised it was stated in the title of the viz – doh!

To make things easier, I therefore added a data source filter to remove all the other dates, setting the Report Date to range from 01 March 2020 to 31 July 2020 (right click on the data source -> add data source filter

I then added the basic fields I needed to a table

• Province State Name to Rows
• Report Date discrete, exact date (blue pill) to Rows, custom formatted to mmmm, dd
• People Positive New Cases to Text

I excluded the fields where Province State Name = Null

We need to calculate the 7 day rolling average per Province State Name which we can do by using the UI to create a Moving Average quick table calculation against the People Positive New Cases pill, and then editing to compute over the previous 6 records (+ the current record makes 7 days). But I want to be able to reference this field, so I’m going to ‘bake it’ into the data model by creating a specific calculated field

7 day moving Avg

WINDOW_AVG(SUM([People Positive New Cases Count]), -6, 0)

Add this into the table, and edit the table calculation to compute by Report Date only and set the Null if not enough values checkbox

Do a basic sense check that the averages are correct by summing up 7 sequential rows and working out the average by dividing by 7.

So it looks like we’ve got the core measure to be plotted, but we’re going to need some additional fields in the presentation.

Again it’s not explicitly stated in the requirements, but it is in the title, that the values plotted need to be normalised. This is to ensure the data for each state is visible; if a state with a relatively low number of cases is positioned in the same row as one with a very high number of cases, it will be hard to see the data for the state with the low cases, because while the axis can be set to be independent, this will only work against a row and not an individual instance of a chart.

To normalise, we need to understand the maximum 7 day rolling average for each state.

Max Avg Per State

WINDOW_MAX([7 day moving Avg])

Add this to the table, setting both the nested table calcs to compute by Report Date.

In normalising, what we’re essentially going to do is determine the 7 day moving Avg as a proportion of the Max Avg Per State, so every value to plot will be on a range of 0-1.

Normalised Value

[7 day moving Avg]/[Max Avg Per State]

Format this to 2 dp, and add to chart, remembering to check the table calcs continue to compute by Report Date only.

In the chart displayed, each State is titled by the name of the State. In a small multiple grid of rows & columns, we can’t use a dimension field for this, as it won’t appear where we want it. Instead we’re going to achieve this using dual axis, and plotting a mark at the centre point. For this we need to determine the centre date

Centre Date

FLOOR(DATEDIFF(‘day’,{FIXED:MIN([Report Date])},{FIXED:MAX([Report Date])})/2)
,
{FIXED:MIN([Report Date])}
)

This looks a bit complex, so I’ll break it down. What we’re doing is finding the number of days between the minimum date in the data set and the maximum date in the data set. This is

DATEDIFF(‘day’,{FIXED:MIN([Report Date])},{FIXED:MAX([Report Date])})

We’re then halving this value ( /2) and rounding it down to the nearest whole number (FLOOR).

We then add this number of whole days to the minimum date in the data set (DATEADD), to get our central date – 16 May 2020.

Now we need a point where we can plot a mark against that date. It needs to be above the maximum value in the main chart (which is 1). After a bit of trial and error, I decided 1.75 worked

Plot State

IF [Report Date] = [Centre Date] THEN 1.75 END

Finally we need to create our Rows and Columns fields which provides the co-ordinates to plot each state. The calculations for these were just lifted straight out of the requirements – thanks Luke!

Building the Viz

Start by adding the Rows and Columns fields to their respective shelf. Set them to be discrete dimensions (blue pills). You should immediately see a ‘map’ type layout of the US States.

Exclude the Null Rows value.

Now add Report Date as a continuous exact date (green pill) to Columns and Normalised Value to Rows, remembering to set the table calc to compute by Report Date only for all nested calculations. Change the mark type to Area.

Add 7 day moving Avg to Label and set the label to display the max value only and adjust the font size – I ended up at 7pt. Then add Province State Name & People Positive New Cases Count to Tooltip. Format the tooltip to match.

Remove all column/row lines and grid lines, zero lines etc.

There is a requirement to ‘add a line underneath each of the area trends’.

For this I added a 0 constant reference line formatted to be a solid black line.

But you’ll notice that for the charts that sit directly side by side, the line seems to be continuous, but I want to break it up. I re-added the column divider line to be a thick white line to get the desired effect.

Right, now lets get the State label added.

Add Plot State to Rows before Normalised Value and change the aggregation from SUM to MIN.

Change the Mark Type to Text and move the Province State Name field from Tooltip to the Text shelf. Adjust the text label to remove any other fields that are displaying, and resize the font – again I used 7. Clear the Tooltip for the this mark, so nothing displays on hover.

Make the chart dual axis and synchronise axis. Remove the Measure Names pill from the Colour shelf on both marks cards which will have automatically been added.

And now all you need to do is remove all the headers (uncheck Show Header) against Rows, Columns, Report Date & Plot Value, then right click on the >8k nulls label at the bottom right and select Hide Indicator.

You’re all done – you just need to add to a dashboard now. My published version is here.

I really enjoyed this challenge – a nice mix of calculations & format complexity but not overly cumbersome, which meant this blog didn’t take so many hours to write this week ðŸ™‚

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 ðŸ™‚

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, “Â …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