Recently, three Danish ministries, Danish Regions, and Local Government Denmark published a report, Indblik i sundhedsvæsenets resultater 2015 on Danish healthcare results.

While the report contains a wealth of valuable information on Danish healthcare, it is also a source of endless joy for a silly graphs enthusiast like me.

I will limit this post to one especially silly graph, figure 4 on page 23, displaying life expectancy in 20 European countries in a seriously misleading manner.

fig1

What is wrong with this graph? Correct, it breaks bar graph rule number one: Always begin a bar graph at zero. From a quick glance one may conclude that life expectancy in Denmark is less than half that of Iceland. The true difference is just 3.4% or 2.8 years.

Bars communicate quantitative information through their length and are useful for nominal, ordinal, or proportional comparisons. Thus, when bars do not begin at zero their lengths are no longer representatives of the values they are meant to display.

The next graph correctly conveys the information. Note how small the relative difference between life expectancy is between European countries.

Sometimes, however, one is interested in absolute rather than relative differences between numbers. And, especially if the numbers are large, bar graphs may not be the most effective way to communicate this. Actually, bar graphs are rarely the most effective way to communicate anything.

The Cleveland dotplot was invented by William S. Cleveland as an alternative to bar graphs and pie charts (note: Never, never, ever use pie charts!). Everything that can be done with a bar graph can be done more effectively with a dotplot. Dotplots communicate numbers through position rather than length and are able to display a lot of data on very limited space and have a very high data-ink ratio. Dotplots may or may not start at zero depending on the message to communicate.

If we were interested in the absolute differences in life expectancy between countries, this dotplot would tell the story.

If, on the other hand, we were interested in the relative difference (as in the bar graph), we would begin the dotplot at zero.

Depending on the message we wish to communicate, one or the other (or both) dotplots tells the story effectively.