By the same logic we should print charts of heart rates and body temperature from starting from 0. Regional temperature over time plots should always start from 0 kelvin, Google's stock price should displayed from 0 to 3000, etc.
I don't know about you but I would be annoyed if my smart watch app displayed BPM starting at 0 making my changing heart rate an ignoreable fuzz.
The idea that "charts should start at 0" is an absurd dogma that is immediately recognizable as such by anyone that has ever worked with data visualization to make decisions.
A better guideline is to display a reasonable range that lets you see historically what has been possible with a bit of buffer. For relatively Guassian changes a few standard deviations is good.
In general your limits should not be outside of the bounds of what is reasonably possible. Coffee trade is not going to 0 anytime soon, and suggesting that visually is dishonest.
> By the same logic we should print charts of heart rates and body temperature from starting from 0
No. We do not want to prove anything here. The trend can be only within a range so that range should be used.
> Regional temperature over time plots should always start from 0 kelvin
If you discuss how wide the changes are - yes. But thi si not what you are looking at - you are looking at temperatures between -50C and +50C (taking the extremes) and this is also the reason why thermometers are scaled like that.
> The idea that "charts should start at 0" is an absurd dogma that is immediately recognizable as such by anyone that has ever worked with data visualization to make decisions.
Certainly. I believe that physicists and engineers who did actual research on actual data can be dismissed.
> A better guideline is to display a reasonable range that lets you see historically what has been possible with a bit of buffer.
It depends on what you want to show. If you want to show relative changes then plot relative changes. If you want to discuss absolute changes (which in the article do not make any sense) then you have to address the whole range. We are talking about millions of something, and that something can drop to zero.
> For relatively Gaussian changes a few standard deviations is good.
On what basis do you assume that you have a gaussian distribution? (I assume - of changes, form your comment)
> In general your limits should not be outside of the bounds of what is reasonably possible
A data scientist making a real analysis will look at what is reasonably possible, but possible full stop. This sets the ranges of observation. To take your first example, the only possible temperatures when dealing with a live body in ambient temperature is from, say, 17C to about 44C. Thi is the range that should be used when discussing "wild changes in temperature. And not a graph that goes from 36.5 to 37 and showing valleys and mountains and drawing conclusions from there.
> Coffee trade is not going to 0 anytime soon, and suggesting that visually is dishonest
What is visually dishonest is to show absolute changes and make any kind of comments on how important they are by not taking the possible values, especially when talking about values going down.
I don't know about you but I would be annoyed if my smart watch app displayed BPM starting at 0 making my changing heart rate an ignoreable fuzz.
The idea that "charts should start at 0" is an absurd dogma that is immediately recognizable as such by anyone that has ever worked with data visualization to make decisions.
A better guideline is to display a reasonable range that lets you see historically what has been possible with a bit of buffer. For relatively Guassian changes a few standard deviations is good.
In general your limits should not be outside of the bounds of what is reasonably possible. Coffee trade is not going to 0 anytime soon, and suggesting that visually is dishonest.