Dynamite plot

Dynamite plot is a bar plot with error bars attached to the bars. I argue that it is rarely the best choice.

The first problem is that there are usually better alternatives. What a dynamite plot conveys is a point estimate with an error estimate. For this, we have a simpler way: just a dot plot with an error bar. This is simpler and does not suffer from various semantic mismatches when we are not dealing with a linear scale starting from zero.

For instance, sometimes, it doesn’t make sense to start the scale from 0. In that case, the bars are highly misleading because the semantic of bars is more “physical”, that there’s a box stacked from the “ground”. See this case where it is misleading to use bars. The points are free from that semantics.

Another issue is when the axis is not linear. For instance, when we plot an odds ratio, the axis is usually in a logscale. A bar doesn’t make sense in this case.

Almost always, just replacing the boxes with points make the plot simpler and neater.

Another issue is when they are used to convey the

• Open letter to journal editors: dynamite plots must die - Data visualization