Sunday, March 8, 2020

Bad Charting Part 3: Axis Abuse

This is Part 3 in a series on avoiding the pitfalls of bad charting. Today we'll discuss how axis abuse can make a chart deceptive.

Axis Abuse: What Is It?

Axis abuse happens when your chart's axes (such as the X-axis or Y-axes for a bar chart) don't follow standard conventions. Those conventions include starting your axis at 0, using a proper aspect ratio, and using an expected axis range. Axis abuse is unbelievably common, and that's because it's one of the easiest and most effective ways to distort how a chart is understood.

Axes Should Start at Zero

When charts show amounts with a visual element (such as bar, column, line, area, and pie charts), there's an implicit assumption that we are being shown the whole thing. When it turns out we're only seeing part of the picture, we're being deceived. The way we're all taught to understand charts with axes is that the bottom left corner is zero. When a chart fails to uphold that, we get the wrong idea about what's being shown. Let's look at some examples.

Example 1: Emissions

Look at the chart below. Taking in what the visuals portray, how do you feel about Emission A vs. Emission B? At first glance, Emission B seems about five times worse than Emission A.


The above chart has a flaw: the Y axis starts at 30, not 0, which paints a deceptive picture. Below is the same data, this time with a correct zero baseline. It tells a very different story, doesn't it? Now we see the truth: Emissions B is more like 1.5 times Emissions A.


Example 2: London Times

Here's an example of a misleading chart in the London Times. The chart—and the whole premise of the article—is about how the Times is outselling the competition. The chart visuals communicate that the Times has more than twice the circulation of the Daily Telegraph. Now look at the Y axis: the chart axis begins at 420,000! If it started at zero, the difference between the Times (485K) and the Daily Telegraph (446K) would be hardly noticeable, a non-story.

The Times Has A Massive Lead in Circulation ..or Does It?
source: https://www.statisticshowto.datasciencecentral.com/misleading-graphs/


It is unbelievably common for axis abuse to happen in newspapers and magazines. These are organizations that should certainly know better. Be on your guard!

Example 3: Television Sports Chart

Television news networks and sports networks are just as guilty as print media with axis abuse. Look at the screenshot below from a sports program, where R.A. Dickey's Knuckleball Velocity is made to look like it's half of what it used to be. With an honest baseline of zero the drop from 77.3MPH to 75.3MPH would be barely noticeable.

R. A. Dickey's Knuckleball Isn't What It Used To Be...?

Example 4: Average Male Height Increase Over the Years

The chart below is supposed to show how average male height (aged 21) has increased from the 1870s to the 1970s. It looks pretty astonishing; the visuals convey that men have doubled in height over a century! Once again, we have a dishonest baseline that starts at 155cm instead of 0cm.

My, How You've Grown!

source: https://www.eea.europa.eu/data-and-maps/daviz/learn-more/chart-dos-and-donts


Example 5: Server Load

Which server load chart below gives you more concern? By now you realize these two charts are driven by the exact same data. The only difference is that the first chart's Y axis starts at 80 while the second starts at 0.


Which Load Chart Concerns You More?
source: https://grafana.com/blog/2019/06/10/how-not-to-fail-at-visualization/

Example 6: Microsoft Edge Performance

One last example to illustrate how common and audacious this practice is. In the gauge charts below, Microsoft is crowing how Microsoft Edge outperformed Chrome and Firefox in a particular benchmark. While the claim is true, these charts are highly misleading. When you consider that Edge (blue gauge) scored 31,786 while Chrome (green) scored 29,619 and Firefox (red) scored 26,876 it's quite clear these gauges do not start at zero. Even worse than the prior examples, they're not even labelled.

Misleading Browser Charts

If we plotted the benchmark data honestly, it would look like this on a column chart:

 A more Honest Comparison

If You Must Show A Partial Axis...

On occasion you may feel you have a valid reason to truncate an axis. If you feel strongly compelled to not show a complete axis, there are responsible ways to depict that. The chart below makes it clear that the bars are truncated.

A Responsible Way to Show a Partial Axis

Chart with Integrity

All sorts of excuses are made for not starting with a zero baseline. There's not enough space. We want to focus in on the interesting part of the chart. People sit too far from the TV to make out the detail. None of these excuses justify the deception, and it is deception, whether intentional or not. You form a conclusion from the visuals long before (or if) you digest the numbers on the chart.

The widespread practice of deceptive charting does not make it okay. Beware this lack of integrity when you view charts, and don't go down this road with your own charts. Always value honesty and accuracy over getting attention.

Don't Deceive With Ranges

Charts can also be deceptive when the amount of data selected is altered to favor a preferred story. Look at the two stock charts below from Yahoo. Which stock would you rather own, the one in red that's descending or the one in green that's on the way up?


Which Stock Would You Rather Own?


In point of fact, these are both AMZN stock charts take at the same time. The only difference is that the first chart was a 1 Day range of data while the second showed a 1 Year of data.

Never pull the wool over your audience's eyes by selecting an unexpected data range, and never leave out important information like what the interval is. There's a reason public companies have to abide by strict rules and report data over careful intervals like quarters.

Don't Omit Intervals

Another way a chart can be deceptive is when you don't have (or don't include) all of the data intervals. The creator of the chart below didn't have data for 2003 or 2004, so it's not on the chart. But that's deceptive, because it makes it look like the data is trending differently than it really is.



The chart below is better because it makes it clear that some of the data is missing. The viewer is less likely to draw an incorrect conclusion about trends.


Use Proper Aspect Ratios 

Yet another built-in assumption we have when we view charts is that they follow a 45-degree slope. When they don't, we can be fooled into seeing an attractive or scary rate change. Look at the charts below, which show the same data but with varying aspect ratios. They certainly convey different emotions, don't they?

Aspect Ratio Changes the Message
source: https://www.eea.europa.eu/data-and-maps/daviz/learn-more/chart-dos-and-donts

Conclusion

One of the easiest ways to change the message of a chart is by toying with its axes. Axes that don't start at zero mislead because we are seeing only part of amounts in the visuals. Unexpected ranges also mislead because we are seeing less or more of the data than we expect. Improper aspect ratios and missing intervals can deceive us about trends.

Chart makers have a great deal of power. There's a visual contact between chart maker and chart viewer, and as a responsible chart maker you need to be aware of that contract and uphold it.

In Part 4, we'll look at conventions, color, and the use of 3D in charts.