What: % of all respondents in a country which represent this type. Country = 100% , but wider bars indicate, that this % of this type is among the top 3 highest in all listed countries. If you’re looking for this type – it is best to look in this country. When: Data was gathered in 2018. Where: Only the countries with representative samples are portrayed in the chart. Source:MBTI Manual Global Supplements Series | The Myers-Briggs Company (themyersbriggs.com)
What: Annual inflation of different items. “Energy” is an aggregate compiled from “Transport” and “Housing” items, so it does overlap with them. When: 1997-2021 Where: European Union – from 15 countries in 1996, to 27 countries in 2021. Source: Eurostat database
Tableau is famous for following the best of the best practices in the DataViz world – it has no 3D effect, the pie chart is marginalized, there are no curved lines, no dashed lines, and the colour palettes are near perfect. One will question whether dashed lines are really a bad practice, another one will easily recreate curved lines with the “data-densification” technique, and here I will argue that there are still plenty of ways to make misleading charts in Tableau.
May this article draw your attention to what the dark practices might be, and whether to use them or to recognize and fight them is the decision to make for each reader.
This is the simplest and probably most used dark practice in real life – simply cutting the axis above zero. Whether it is a good or dark practice depends a lot on the context. Line chart allows to do this “legally”, but here I’m adding some more distortion by changing colours to a more dramatic hue and additionally encoding the size of a bubble to a variable.
Not so easy comparison
Tableau allows for bubble charts that look cool, but are rarely useful to communicate changes in values until your aim is to hide that subtle change. If there is a KPI on your dashboard you don’t really want anyone to see changing – use bubbles. Leaving out the printed numbers would render this visualization practically useless.
However, the same bubbles could actually emphasize the difference if placed one above the other – this would be more like an unnecessary overdesigning rather than dark practice.
Not so easy comparison again
Bubbles are very effective to obscure comparisons between categories. The same could be achieved also with colour, and Tableau allows it to happen so easily! Of course, the top and bottom values are easily seen, but the middle is quite muddy, it would be very difficult to sort them without printed numbers.
Both dark practices shown here would be difficult to reproduce on Excel.
Not so easy comparison by stacking
I’ve seen such stacking more as a result of trying to make a cooler chart rather than a dark practice (it was a donut chart by the way). But it could be used to conceal the true difference in numbers. Additionally – make the part you want to look bigger much more saturated, and put the border of the same colour around both bars. No smart reader would be fooled by such arrangements, but if there are more charts to digest on the page, the message might slip unnoticed.
Hiding a relation
Scatterplots are not that easy to read and get for the untrained audience and that is why they are not used often, however they are perfect to show a relation between two variables. Line charts might be used to show the relation, but they could be used to hide that relation – removing lines and making markers a bit oversized effectively obscures the actual direction of data.
Disguised negative number
This was the actual problem I faced in my job – how to explicitly show profit numbers and somehow make the loss at one period be not that very visible. First of all, we could make the profit and loss to be the same colour, then move the number of loss above – as those of profit, then remove the reference line and finally make the bars thinner. Again – acute reader would quickly recognize dark practices used against him, but only if he is not overwhelmed with charts.
3D still possible
And finally the trick as old as Excel – 3D pie charts are notorious for distorting reality and more often than not they distort it unintenionally, however clever masters of dark practices utilize this feature to their benefit.
With smart preparation of data, dynamic 3D charts are still possible in Tableau. The one seen here is drawn “by hand”, feeding Tableau exact coordinate of each point. A more saturated colour and annotation further serve exaggeration of our company’s market share.
Click the tabs in the visualization above to see all the interactive versions of charts in this article.
Dark practice is not making bad decisions about data vizualizations. It’s making smart intentional decisions to distort viewers’ perception of the data.
Dark practice is not faking the data, it is making the data to appear showing what it might not be showing.
Dark practice is not lying, it is not telling the truth.
The question is it easy to replicate the default settings of one charting software in another charting software bothered me for some time. Are the default settings more universal or less universal? Do different vendors have different attitudes towards what should be the default setting?
I chose to work with a line chart because different software interprets differently how to arrange multiple series in a bar chart – some tools stack them, some not. By adjusting this arrangement I would lose the defaultness, while without any corrections the charts would be less comparable. I made all charts squared, so they fit better in the grid.
Insights about the defaults:
All have horizontal gridlines, ggplot2 and Tableau have even vertical ones.
Only Google Data Studio and Tableau have highlighted the zero line, although Tableau highlight is barely noticeable.
Blue and red or orange are within the first three choices in every palette.
ggplot2 looks exceptional with its grey panel.
The default settings of Tableau make the least sense because they are configured for more charts with more legends. One chart with one legend looks a bit weird.
Grey squares at the top right of Google Data Studio charts are how the control buttons are rendered as an image.
Insights about the comparisons:
Of course, ggplot2 manages to replicate even the most complex cases. The biggest challenge was using Google font from Google Data Studio because the library”showtext” which seemingly allows achieving this does not work well with ggplot2.
Settings of ggplot2 itself were the most difficult to replicate.
Tableau was the only software that could not replicate the exact colours of lines, because a colour must be chosen from a predefined palette there.
It was quite annoying that Power BI and Google Data Studio could only export to PDF, however they are not meant to make pretty pictures after all.
Somehow square charts from Excel lost the squareness after saved as images.
Google Data Studio insisted on a black line indicating zero and refused to show vertical gridlines. Maybe I just don’t know this tool well enough or maybe these are the limits.
Adjusting the limits of the x-axis was always a challenge, the y-axis is often allowed for way more freedom.
Adjusting legends was always the most difficult part. Legend is what distinguishes one tool from another.
I believe this exercise is of little use, but it was fun to do it!
Different freedom indices measure different fields of freedom – press, economy or general human freedom. Are they consistent within a country?
As we see from the 2nd segment of this chart – no, freedoms are not consistent. Economic freedom index is quite often much higher than indices of other freedoms, especially in those countries at the low end. I did my best to adjust the indices so their ranges are uniform, but their averages are still quite different – most countries love economic freedom, while many of them do not care much about democracy or moral freedom.
What: Freedom indices recalculated to fit the range from 0 to 1, where 1 means the best index and 0 means the worst (in 3 out of 5 cases – it’s North Korea). Country average is a simple average of all 5 indices. Index, When, Source: Democracy Index, 2019, EIU Human Freedom Index, 2017, The Human Freedom Index 2018: A Global Measurement of Personal, Civil, and Economic Freedom Economic Freedom Index, 2020, The Heritage Foundation Moral Freedom Index, 2020, The Foundation for the Advancement of Liberty Press Freedom Index, 2020, Reporters Without Borders Where: 172 countries were ranked on at least 3 of these indices.
I’ve found five freedom indices measuring various fields of freedom. The question is whether all those measures are consistent within countries, or do they vary a lot?
The answer is in the chart below the chart below – more often they’re consistent than not. Exceptions are in some Muslim countries which do not like moral freedom or democracy – those ratings are low, but they want high economic freedom – so this one particular rating is often high.
What: Freedom indices recalculated to fit the range from 0 to 1, where 1 means best index and 0 means the worst (in 3 out of 5 cases – it’s North Korea) Index, When, Source: Democracy Index, 2019, EIU Human Freedom Index, 2017, The Human Freedom Index 2018: A Global Measurement of Personal, Civil, and Economic Freedom Economic Freedom Index, 2020, The Heritage Foundation Moral Freedom Index, 2020, The Foundation for the Advancement of Liberty Press Freedom Index, 2020, Reporters Without Borders Where: 172 countries were ranked on at least 3 of these indices.
The Stacked Bar chart is one of my favourites, I even made the same stacked bar chart with 9 online tools – but it has one major weakness, it’s difficult to compare changes of its segments over time. I tried finding a way to improve it and here let me introduce the Comparative Stacked Bars:
Triangles show the absolute increase or decrease of each segment. They are colour-coded to make it even easier to read.
Too many is too many – if there are too many categories the triangles will make the chart look messy and difficult to read. But even more difficult it would be without triangles.
Too small is too small – if the change is too small the triangle might become invisible. But without them, the changes would get invisible much sooner – just observe the top segment in the above chart.
The example above was made in R, and the example below was made in Tableau. Unfortunately I have no not-overly-complicated solution for Excel. If anyone knows how to implement it properly, please let me know!
The pie chart faces a tremendous amount of criticism for attempts to show part to the whole relations. Of course – it is easily the single most misused chart! However more and more data visualizations practitioners are writing articles to defend it with Robert Kosara being the most thorough and methodical in my opinion.
Here I will offer an alternative which is something like a mix of square pie chart, marimekko and packed bars. Let me introduce the Cake Chart:
Always a regular square, total area = 100%.
Each segment is a bar for easy comparison, but its area represents the percentage.
The height of the chart is distributed evenly among bars.
Only the selected number of largest categories are shown separately.
All other categories are aggregated into the irregular steppy gray area.
If there are 5 categories shown separately, then the largest cannot be larger than 100% / 5 = 20%. For X categories to be shown – the maximum value cannot be larger than 100% / X.
Aggregated other categories are difficult to compare to bars.
There are limitations how long the longest bar can be without destroying the squariness of the square.
Square is not as intuitive to be 100% as circle.
Bars are super easy to compare.
It’s still a regular shape.
It’s quite obvious how to make it – make the panel square, make its background grey. However, making the Others category selectable in Tableau is a challenge and here is the result: