Upvoted. Several times I’ve seen recommendations to start graphs’ y axes at zero by default, but it’s a tip that’s starting to grate on me for several reasons.
Usually, when I look at a graph, the y values’ variation is at least as relevant as the values themselves. I want that variation to be clear & obvious; if someone’s going to represent it on a graph, I want it spread across the available space. Cramming it into a small range near the top is a waste.
Visually compressing variation can be just as misleading as visually expanding it. Which is more misleading is case-dependent.
Sometimes I want to read numbers off a graph as accurately as I can. If the plotter stretches the y axis because they think I’m too dumb to read labels, that makes my task harder.
If the y axis is on a log scale, you can’t make it go to zero without some distracting gimmick like making the axis discontinuous.
For me points 1 & 2 apply here. (Although, as it happens, I don’t like that figure 2. It’s too close to a dynamiteplot for comfort, and it’s a space-hungry way to show me two averages & two standard errors. You could communicate the same information with a small table, or even a line of prose. And Kindly’s right about the caption. But starting the y axis at $25k is the least of that chart’s problems.)
These are excellent points. Unfortunately, I’m a bit hampered by the fact that I stole the chart in question from the original study (pdf), and they used only “dynamite plots” in their paper. After reading your links on the topic, I can definitely see why this is bad. I’m appending a short note to this effect as an edit to my original article.
Thank you for bringing this stuff to my attention.
Upvoted. Several times I’ve seen recommendations to start graphs’ y axes at zero by default, but it’s a tip that’s starting to grate on me for several reasons.
Usually, when I look at a graph, the y values’ variation is at least as relevant as the values themselves. I want that variation to be clear & obvious; if someone’s going to represent it on a graph, I want it spread across the available space. Cramming it into a small range near the top is a waste.
Visually compressing variation can be just as misleading as visually expanding it. Which is more misleading is case-dependent.
Sometimes I want to read numbers off a graph as accurately as I can. If the plotter stretches the y axis because they think I’m too dumb to read labels, that makes my task harder.
If the y axis is on a log scale, you can’t make it go to zero without some distracting gimmick like making the axis discontinuous.
People can’t decide whether this rule applies to bar charts specifically or graphs in general.
For me points 1 & 2 apply here. (Although, as it happens, I don’t like that figure 2. It’s too close to a dynamite plot for comfort, and it’s a space-hungry way to show me two averages & two standard errors. You could communicate the same information with a small table, or even a line of prose. And Kindly’s right about the caption. But starting the y axis at $25k is the least of that chart’s problems.)
These are excellent points. Unfortunately, I’m a bit hampered by the fact that I stole the chart in question from the original study (pdf), and they used only “dynamite plots” in their paper. After reading your links on the topic, I can definitely see why this is bad. I’m appending a short note to this effect as an edit to my original article.
Thank you for bringing this stuff to my attention.