Y-axis starting at zero is essential for bar charts, but not necessarily time series / line charts. Insisting on a zero axis can crunch the data and obscure the trends and is unnatural for some data sets (e.g. the mortality time series—if that trend line ever goes to 0...something has gone awry).

Quote from Tufte on the subject:

> ” In general, in a time-series, use a baseline that shows the data not the zero point. If the zero point reasonably occurs in plotting the data, fine. But don't spend a lot of empty vertical space trying to reach down to the zero point at the cost of hiding what is going on in the data line itself.”

https://www.edwardtufte.com/bboard/q-and-a-fetch-msg?msg_id=...

If you're looking for trends, sure. But this graph seems to be clearly designed to compare the overall US mortality rate across the last five years. It's hard to appreciate scale when comparing elements on an axis which does not begin with zero.

Imagine that this were a time series graph of your body temperature over the last week. With the Y-axis set at (min_temp, max_temp) or similar, the graph would clearly show a mean of about 98.6F and consistent trends based on the hour of the day. If you had a fever (>100F), it would be readily apparent in the graph.

A zero baseline for this chat is meaningless and would render the plot useless, unless the purpose of the graph was to show what happens to your body temperature as you turn into a White Walker.

Similarly, the axis chosen for the mortality chart is effective since it shows the typical range of values (and thus does convey the scale effectively), the consistent seasonal trends, and the anomaly (which is very significant in terms of standard deviations above the mean). Setting a zero y-axis would squeeze the graph into 1/3 of the vertical space, leaving 2/3 as useless white space for no benefit at all.

You are comparing the Y values for multiple points with different X values in the same series. You are looking for trends and fine differences.

I am comparing the Y values for multiple points with the same X values in different series. I'm only looking for course differences.

For my purpose, it makes sense for the scale to start at zero.

At a glance, the week of April 12, 2020 looks like it has four times as many deaths as any other year since 2015. In reality, it has roughly 1.5x as many. That's still a huge difference, but I had to check the scale and do some quick mental calculations to figure that out - which is more than I would have had to do if the graph's vertical scale started at zero.

I'm not arguing that all graphs must start their scales with zero. I'm not even saying that this is a bad graph. It's just not as effective for the kind of analysis I want to use it for.

Agreed. If you go up two graphs to "Excess mortality for all ages"[1], though, you can see the same stat as a percentage delta from the average of the five preceding years.

[1] https://ourworldindata.org/excess-mortality-covid#excess-mor...

Better, but they should also adjust for increase in population sizes.

Why? For a roughly stable population like the US, the "baseline" is that something like 1/79/52 of the population dies every week, since life expectancy is 79 years. Zero deaths is nonsensical.

The difference between the peak of this year and previous years is about 50%. The graph makes it look much larger.

That's not the point. It's very hard to appreciate the scale of a difference between two elements on a graph if the graph's scale does not start at zero.

That's completely wrong. And by that I mean "you just failed your statistics class" wrong.

Imagine you had a graph of earth's temperature and decided to set the y axis at absolute zero. What would that look like? Would it be in any way useful?

I was mildly curious what that'd look like:

https://ibb.co/qnx2Y4H

Or, another way, % change compared to the average:

https://ibb.co/5T5dMZk

The data is available for download so you can plot as you please.