I often observe (on the web or in my students works) bar charts used inappropriately. What often happens is that bar charts are used with a non-zero baseline, which implies telling a visual lie.
Here I report a very simple example of how the issue emerges when drawing a chart in Excel and a possible solution (using a dot chart).
Disclaimer: this is quick post, apologies for typos and lack of clarity.
Let’s start with a sample dataset: a few categories each of which has a value associated with it.
This graph has an importan drawback: human brain quickly intreprets length of a bar as its main feature.
Therefore, looking at the graph it would interpret the value associated with Beta as double the value associated with Gamma.
This first, natural, immediate, visual interpretation is wrong. This is a visual lie!
This is why bars should always start at zero!
If we modify the above char to have a zero-based axis the result is the following:
Now we have removed the lie and the lengths of the bars are proportional to the value they represent.
The unfortunate side effect is that the new graph makes it harder to appreciate the (relatively small) differences among the values.
Dot plots allow showing the values without tying the interpretation to the bar length:
In this case no lie is contained in the graph and the differences between values can be easily appreciated.
Unfortunately Excel does not allow having categories on the y axis (except for bar chars).
A possible solution is to use a “Line” chart without lines, then rotate everything 90° counterclockwise. Eventually you can copy your graph and rotate it in whatever document you want.
Symbols can be linked to the labels using error bars, that need to be styled to be as subtle and unobtrusive, as possible.
The original excel file with the sample data and the graphs can be found here: BarVsDot_v1