# The Y-Axis Shouldn’t Always Start at Zero?

19 Nov, 2015

Vox visualizes a lot of data as part of its storytelling. And Vox says whenever it features a chart with a y-axis that doesn’t start at zero, angry readers send emails.

“You see, an old book called How to Lie With Statistics has convinced people that truncated axes are a devilish tool of deception,” says Vox.

But Vox is of the thought that sometimes you have to start the y-axis somewhere other than zero to “show appropriate context.”

Here’s the full explanation:

What say you? Is it “long past time to say no to y-axis fundamentalism,” as Vox puts it? Tell us your thoughts in the comments below.

## Comentários

Submitted by Jonathan H. (não verificado) on

I'd say it's okay for line charts, but not for bar charts.

Submitted by Jeffrey Shaffer (não verificado) on

There is a major point missing here. Yes, there are instances where breaking an axis works. If the values are far from zero and have small variance, then it becomes necessary to break the axis to show those difference. However, this should only be done when using position to encode the value. You should never break an axis when length is being used. All of the examples in the video from Vox were line charts and breaking the axis works in those cases because the reader is comparing position of the line from one point to the next. However, in the Fox news example, they are using a bar chart, which encodes the value in the length of the bar. If encoding the value using length, such as a bar chart, then it should always start at zero. If it doesn't start at zero then the length is meaningless and potentially misleading. If using position to encode the value, for example, a dot plot or line chart then breaking an axis can be done. This is a very important distinction that is not discussed in the video. Also, even in the case of the line chart, there has to be some caution as to how far to go. When breaking the axis, it magnifies the difference, which is the whole point. The visual slope is at the discretion of the creator and that is what many people take issue with, including "How to lie with statistics". This can easily be taken to far. This is the "people" aspect that is talked about in the video. I prefer to think that people aren't intentionally misleading in the majority of these cases, but rather, they just don't know any better. When presenting for 30 minutes at a conference on data visualization, this can translate to "don't ever break the axis" and may sound like dogma. It's only when studying the subject deeper can we explore all of the exceptions within these data visualization "rules".

Submitted by gregory kaminski (não verificado) on

Yeah, the video is spot on. Before I watched it I had already thought about the example regarding percentages. If you're a business and your inventory normally falls in a range of 70%-75% there's no reason to view a graph with a 0 Y-Axis. You put the axis at 65-80% and monitor the small fluctuations that may be occurring.

Submitted by Colin (não verificado) on

In this case, you wouldn't use a bar chart which you would want to start at zero, or even a line graph. You might, instead, use a box and whiskers plot which would be a better visualization for tracking the variance around a given point for different items in the inventory.

Submitted by Shawn Wallwork (não verificado) on

Quibble all you want, but this was an excellent video explaining why Zero isn't the end-all-be-all axis anchor.

Submitted by Ludo K. on

Interesting video, I agree with the points being made. It's the context that matters and the value that it represents. I liked the Celsius/Fahrenheit/Kelvin example: sometimes 0 is just number someone made up.

And even if the zero line is relevant within the context, why not give a user the option to zoom in and see the details? Tableau is a dynamic tool for a reason. Most importantly, the viewer should'nt be misled by the representation of the data, the aim always should be to provide insight.

http://www.infotopics.nl/infotopics-tableau-blog/entry/best-practice-dat...

And here's a possible way to let users decide for themselves what's relevant:
https://public.tableau.com/profile/ludo.koelman#!/vizhome/Hidingzeroonan...

Thanks for sharing!

Submitted by Jeffrey Shaffer (não verificado) on

Shawn,

It is a good video, but I don't think my comment is a quibble. As I mentioned, there is a very important distinction that was missed in the video. Just as we don't want a dogmatic rule of "never break the axis", I think it's important to understand that the guidance offered in this video relates to values encoded using "position" and not "length". That's at the root of why the Fox News bar chart example doesn't work and their other examples using line charts do.

Jeff

Submitted by Darren McGurran (não verificado) on

The comment in this video about whether you use Fahrenheit, Celsius or Kelvin is a little specious as the measures here do not affect the relative value of each point to each other or to the x-axis but the absolute location line on the chart.

Using a Fahrenheit scale that starts at zero when talking about temperature does not add much value when discussing humans and a chart that starts at 80 degree F is more than likely not going to be misleading. Of course if you are showing something about recorded temperature where someone has recovered from hypothermia (55 or so degree) then having a zero point on the chart is the best to get an scale of variance from the norm.

In the example given about current employment level, the line chart is meant to again show variance from a nominal time in the past. However, a bar chart showing current value difference from that point would present this information more clearly and concisely.

The problem seems to me is that using a line or bar chart to indicate items other than trend (line) or scale (bar) is completely open to manipulation by altering the expect base value and that base value should be generally zero or a generally accepted base value. In any other case, there are better visualizations that present the information more clearly and minimise incorrect interpretation.

Darren

Submitted by Kristopher Erickson (não verificado) on

I think the video is completely wrong. First of all they fail to apply their own principles of context to their graph of employment. In reality the ~2.5% decline in employment is not "a big deal" as the presenters claim it to be. This chart is not anchored at zero and it still doesn't support their claim: https://research.stlouisfed.org/fred2/graph/?g=2EQQ Here's the chart starting at zero: https://research.stlouisfed.org/fred2/graph/?g=2ER0

In both charts you CAN see the decline, it's just not a significant decline after decades of rise.

I don't care if the chart were anchored at 100% though. But an absolute of some type has to exist to anchor the line charts. Otherwise the line chart is just floating in the wind.