Follow the chart

For the analysts who are in the trenches crunching the numbers behind slides (often after 18:00).

Presentations of financial data often evolve. You start with a relatively naive model, create some slides and iterate the numbers. Slowly, your team starts understanding what actually matters and discovers with drivers to focus on.

Instead of the exact numbers in your spreadsheet, your manager asks you to group this, that, and that into one number, quickly offline. Then another scenario, put that number in, quickly off line. Then another one.

In each round, you re-run your model, take out a calculator, scribble the summarized numbers, and update your slides. This takes a lot of time and is prone to errors.

Instead, build a quick layer on top of your ‘old’ model that spits out the required numbers quickly. In fact, make it a habit that every number in your presentation is pulled directly out of a cell in a spreadsheet.

My financial models would usually have these layers:

1. Data dump: straight copy-paste of raw input data, or data entered straight from a financial report without thinking, make sure the total is correct at the bottom. You get a new set of data: simply overwrite the entire worksheet, or add a column.

2. Model engine, this one does the hard lifting and runs your analysis

3. Bridge: this worksheet pulls numbers out of the engine and produces the required numbers for the charts (relevant to the scenario I described above)

4. (Optional) Slides. A small box that matches exactly every page in your presentation, with the exact numbers that appear in each slide. Useful if you need to run periodical updates of your presentation (weekly, monthly, quarterly results for example).

De-cluttering axes

In scientific documents, there are chart making conventions that make sense, clearly labelled axes, titles, etc. etc. Use these charts in your article that you submit for publication in a prestigious paper. For an on-screen slide show however, you could deviate from this standard. Your objective is to communicate the findings as best as possible, referring to the paper for the details.

See the example below (source), lots of duplication in axis labels.

You can make the page a lot calmer be omitting some of these labels. I quickly cut and paste the elements in the image below. (This is not a makeover, just a super rough reshuffle to show you what I meant).

Statistics: vaccine effectiveness might seem higher than it is

I love digging into COVID-related statistics. Recently, this paper was published that shows how vaccine effectiveness in local communities can be a lot lower than at the national level. Seems counter intuitive, but this chart explains the math.

I have added this slide to the SlideMagic library, so you could use it in your own presentations as well.

Public corona data dashboards

BI (“Business Intelligence”) dashboards with data used to be a corporate thing. Firms such as my previous employer McKinsey would advice clients what metrics to put on them, and how to display them. This is tricky, there is an infinite amount of data to choose from, and even more options to slide and dice the figures.

The COVID outbreak has created many country-wide public dashboard with data. In Israel where I am based, a large tribe of “amateur” statisticians has emerged that runs and discusses analyses on Twitter. The other dashboard I had a look at is the Dutch one (part of my family still lives there).

• Dutch board

• Israeli board (use the translate option that is offered by the Google Chrome browser)

The approaches are different, and I prefer the Israeli one.

• The Dutch board looks very pretty, has lots of explanations in text, and has useful maps of regions with color coding. The problem is that it stretches out over many, many, pages, and priotises static data over time series.

• The Israeli one is just one page, with lots of time series graphs, so you can see things develop over time. And not for basic statistics such as overall cases, benchmarks can get very specific. Benchmarks are normalised so you compare apples with apples (i.e., cases / 100,000 by vaccination status). Also, government policy and benchmarks are tightly integrated. The government wants to encourage parents to vaccinate children, so there are statistics specifically aimed at that target segment. Another example: after discussions whether to close the airport or not, stats about airport tests were published (split by country, so citizens can make the call to travel somewhere or not based on their personal risk appetite).

The biggest advantage of the all-on-one-page approach is that people start to understand it, and come back to it very often to get the latest data, even venting anger when it is not updated on a day.

Data visualisation to involve the public in decision making and/or influence day to day behavior.

——-

PS. Israel does a PCR test for every single arrival at its airport, so the arrival statistics on the Israeli dashboard are probably one of the best global indicators of what is going on in a particular country.

To stack or not to stack?

Two charts about a new sub-Omikron (BA.2) variant in Denmark. This line graph shows 3 variants as a % of all sequenced samples in Denmark.

Source: https://twitter.com/DrEricDing/status/1485456877723045891

The chart below shows the total number of variants found in the samples. The stack approach does a much better job to give the full picture of what is actually going on,.

Source: https://www.covid19genomics.dk/statistics

With just one data series, showing a share of the total as a stack or line (column) is the same chart. As soon as you have more than one, pick a stack chart so the audience can see the data in context.

Bar versus column chart

The chart below could have been made a lot better using a bar chart. You can avoid the many legend labels, which have a 1-to-1 relationship to the columns

Image source

Even better than I did

This Venn diagram is a great visualization of why you still see vaccinated people in the hospital.

I gave it a go myself a while ago, but this visualization is better. Source of chart: RIVM, source of image. One improvement suggestion: switch the colors red and green.

The case for not rounding numbers

In 99% of slides, it is better to round financial data. $1.9m is easier to read than $1,898,456.34. Also the rounded number is more in line with a financial model that relies on rough assumptions. If you project your company sales in 10 year down to the dollar, you lose some credibility with your audience.

In some situations, the opposite approach can work. Look at this poster below of an Israeli anti-vax group who makes the argument that the money that is spent on encouraging hesitating Israelis to get a vaccine, could have been used better in a different way. (I leave pro and anti-vax debates out this blog, although you might guess in which camp I sit).

Here the big number actually works. Anyone looking at this big amount of money instantly starts comparing it to other lump sums you know: how much do you make as an individual in a year, how much does a car cost, how much does an apartment cost. Also, the precision and suggested accuracy of the number adds to the drama. This is a similar effect that National Debt Clocks try to convey.

The correct way to look at these numbers is to relate them somehow: $ per citizen, % of total corona-related cost, compared to other government advertising campaigns, etc. etc. After that, you might still conclude that it is high, but you used the correct metric.

Don't make them study the graph

A random chart on Twitter made me pause to see what is actually graphed. The chart title suggested a positive correlation, but the line is actually sloping down.

On closer inspection you see that the vertical axis is “low is good, high is bad”, and the horizontal axis is “left is bad, right is good”, also the horizontal axis talks about “decline” instead of “growth”, so a positive number is actually a decline.

To analyze data, it is OK to ponder and study a chart. In a presentation of final results, not.

McKinsey slide makeover

A saw a slide by my former employer coming by:

It has a very sophisticated image effect: look how the background of the bars in the chart are part of one image. Still, there is room for improvement. I quickly replicated the chart in SlideMagic with a few changes.

• I brought back the more traditional, very in-your-face alternate coloring of the bars, blue for 2021, grey for 2020 and a legend, instead of the repetitive text labels with the years.

• I increased the size of the industry sector labels

• By replacing 910b and 582b by 0.9 and 0.6, I could get rid of the “t” and “b” in the bar label.

But the analysis of the slide can be pushed further. The main point of the slide is how markets have bounced back over the past year, which is independent of the ranking of the market capitalizations of the sectors. As an alternative, I constructed the combined table/bar chart below, de-emphasizing the absolute value of the market capitalizations, and using the bar chart to highlight the % increase in market valuation. The inside here is that all sectors grew more or less the same over the last year (except fashion, probably reflecting less dressing up for work.

I have added the slide to the SlideMagic slide library, look for “COVID” and they will show up. Emails subscribers: if the slide images don’t show up in the email, please open the link to the full blog post.

Information hierarchy

I just returned from a short trip to Paris to show my son around some of the famous sites and restaurants. In 2021, that means a lot of health checks and tests. I was probably the only one in the airline terminal that looked at all the forms with the eye of a typographer.

I am not talking about elegance here, pure functionality. The people at check in desks are looking for “positive” or “negative”, the date the test was given, and whether the passport numbers match. On the test result form, the thing that is printed biggest is the name of the testing laboratory…

All this can be fixed easily with an adjustment of font sizes.

Rounding numbers in data charts

How to round numbers in a data chart? It depends. The chart below does not look very appealing

The numbers are hard to read. This chart can serve 2 purposes. Either show the trend in sales, or show the exact sales figures. To show a trend in sales, simply show the accounts in thousands, and round up to one decimal point:

If you need to provide the actual precise sales data (for accounting or tax purposes), put it in an appendix slide that does not even pretend to show a trend:

Scaling of data charts in SlideMagic

In SlideMagic, you do not have to worry about picking the right scale for your data chart. The entire chart adjusts itself to the numbers you type in. See the example below:

To make sure that a consistent scale is used for your entire chart, you need to place all your data points in one shape, instead of using multiple shapes for example for each month.

P.S. I have added this monthly sales comparison chart to the SlideMagic slide library so you can easily use it in your own presentations as well. Search in the app for ‘sales’ and it will pop up.

Leaving the math to your audience

It is raining COVID statistics in Israel as we are the first country in the world to deal with a post-vaccination outbreak. Below is one table that was released by the Ministry of Health (I found it here).

I have translated it in a quick SlideMagic chart (it always puts a big smile on my face to see how quickly this can be done).

But this data is horrendous to understand. Percent of what? What is 100%? The audience is left to do the math themselves. Compare the categories to the breakdown of the population, look at differences between 3 and 7 days ago, look at the ratio between mild to severe, etc. etc.

Using bars instead of numbers (another smile) makes things a bit clearer.

But in this case, it would have been clearer to release the data in absolute numbers and let people construct their own charts.

I have added the charts above to the SlideMagic library, search for COVID in the app and the slides will show up (see the search here).

Cheating with statistics

The chart below (source) is a good example of “axis”. The drop in life expectancy looks huge, but upon closer inspection, we see the the vertical axis starts only at 72.

There is another problem with the chart: “the sharpest since World War II” is not supported by the data.

One way to bring out the significance of the message, and support the WWII point is to show the annual change (not the absolute number) in life expectancy since 1940.

Trying to understand vaccine effectiveness

Here in Israel we are ahead of most other countries in terms of vaccination and the prevalence of the delta variant. After almost zero cases, the count is starting to creep up again. There is a lot of confusing data going around and it is surprising to me that the scientific community does not have a generic approach to evaluate the effectiveness of vaccines.

Last night the following table appeared on the TV news. Severe cases by age category and vaccination status. But these absolute numbers cannot be taken at face value.

Source: https://twitter.com/arad_nir/status/1416832997597265933/photo/1

“Open source” statisticians went to work and made some adjustments. The population categories are not equally big (there are more young people than old people), and the vaccination rate is not the same (older people vaccinate more). So the correct approach is to look at severe cases / million, split by vaccinated and unvaccinated. I put the results in the graph below and added the chart to the SlideMagic library.

I put the results in the graph below and added the chart to the SlideMagic library. Search for “vaccine’ in the SlideMagic app and the designs will pop up, either for use in a COVID-related presentation, or maybe something completely different that requires a similar layout.

Dashboard design

In my current (stealth) side project I need to build many dashboards to show information in different cuts and slices. For me, it is a very interesting experience as I can apply the full arsenal of my slide design experience, but now with dynamic data. I control the full stack of technology: what information to store, how to slice it, what information to show, and how to show it.

Each of the above usually reside in a different person. Management consultants spend time recutting and re-combining data manually in spreadsheets because systems can’t do it. So called “BI” applications take data from systems and spit out an endless amount of bar and pie charts in the hope that it will give some insight in where things are going. Traditional front-end web designers can make data look pretty, but don’t really understand what data is required.

The principles of a good dashboard and a good slide are completely the same. Every detail is important. What information to show, what rounding, what order, what sort of graph, what headings, bold, not bold, margins, right aligned, left aligned., how to group things, where to put subdivisions, etc. etc.

But once you get it right, it will work for a long time.

Photo by Cody Fitzgerald on Unsplash

Tiny data labels

This chart shows 2 interesting things. One, Finland was pretty happy under lock down. Two, an interesting way to put data labels on a stacked column chart. The small boxes are always a problem in a regular format. Here you get the combination of the visual effect of the size of the boxes, versus the table of the actual information. This could be inspiration for a future SlideMagic expansion.

I would do some things different though. That row of zeros at the top does not add much. The flags make the whole chart even more busy. And given that this is a comparison, I would have shown the data as a stacked bar chart.

If you were to use me as a bespoke designer, I would actually show this data on a map of Europe, color-coding different countries with maybe only the some of the 2 blue data series. The geographical clustering of the countries is interesting. In addition, I would combine it with one stat about the health impact of COVID in these countries.

If you do not have the software and/or the time to make a chart like this, the solution is easy, take off the data labels completely and make a straightforward stacked column chart.

I found this chart on Twitter, without quoting a source, the format looks like a page in some document by the European Union though.

Order of data series

Here is a (sad) chart from today’s Economist:

The Economist put the data series that carries the main message of the chart at the bottom, pushing up all the other data series. My preferred option is the other way around, put it on top. In that way you can see all other regions staying pretty much stable, while India grows strongly.

(Unrelated). India has a very large population, and you need to look at COVID in that perspective. In terms of caseload, it is still behind other regions (such as Europe). The problem is the quality of the healthcare system, and the availability of basics such as oxygen in emergency rooms. Europe could handle the load (more or less), India is in a far worse position. Also, the India stats are averages for the entire country. On a region-by-region basis, there are likely to be places with much bigger caseloads than Europe. Let’s hope that it gets better.

Legends can be confusing

Some perspectives by @clauswilke:

Slides on redundant coding.https://t.co/miqrLROikZ#rstats #dataviz pic.twitter.com/m1H34F2Zuk

— Claus Wilke (@ClausWilke) April 14, 2021

He is right. Legends can be confusing and are prone to errors, when possible, put the data label immediately next to the line.