Comments on: Use Dot Plots for Better Categorical Comparisons

By: Jeff Weir

Jeff Weir — Sun, 21 Feb 2010 20:48:18 +0000

Just found another blog that covers some of this ground: Eagereyes. Check out

http://eagereyes.org/criticism/cost-of-a-sick-chart

By: DaleW

DaleW — Thu, 04 Feb 2010 03:25:15 +0000

I agree with Ed Ferarro and especially Jeff Weir here. While either a dot plot or a bar chart is a clear improvement over the original pie chart, none of them convey the full story in this data.

The oft-discussed loss of jobs in male-dominated manufacturing actually seems to be hurting women more than men in this NZ data, as Jeff pointed out by his supplemental chart of jobs lost/gained by sector and gender. It is only the big gain of jobs by women (but not men!) in the Health field that allows the recession overall to impact the employment of women less than men in NZ. A bar/tornado chart is a very effective way to convey such nuances.

When zero matters, and there are only a few (1-3?) series per cluster, a bar chart seems easier to grasp than a dot plot — even for those familiar with dot plots.

By: Jeff Weir

Jeff Weir — Mon, 01 Feb 2010 18:22:57 +0000

Yikes…labor pies everywhere. http://www.bls.gov/opub/ted/

By: Excel Links of the Week – now even more downloads edition | Pointy Haired Dilbert: Charting & Excel Tips - Chandoo.org

Mon, 01 Feb 2010 09:21:13 +0000

[...] Use dot plots instead of pies to compare categories [...]

By: derek

derek — Fri, 29 Jan 2010 22:08:11 +0000

Here is what I had in mind. I thought of another advantage of ordering by increasing difference between two series: the two series are by definition guaranteed to have at most one crossing point, minimizing confusing clutter. I'm a convert to the value of lines between the points in nominal category series, thanks to the work of Alfred Inselberg. BTW, this is my lazy dot plot method-- I just make an Excel line chart and rotate it 90 degrees when its time to save the image using my favorite photo app.

By: Jeff Weir

Jeff Weir — Fri, 29 Jan 2010 21:18:58 +0000

I think Chip’s right that the bars tend to work for blue-on-blue comparisons at the expense of blue-on-red, and that it’s harder to see at a glance who is leading who across all categories.

So technically I think dot plots are a better medium for people that understand charts well. But are they the best medium for the majority of people more familiar with bar charts? So something we should factor in is the ‘reading age’ of the publication. This source material is public interest, not an article in a scientific periodical. So perhaps we need to err on the side of ‘understanding of the masses’.

While the lines on bar charts add ‘clutter’, I think that the upside is that this clutter helps draw the eye back to an axis. Whereas with the dot plots, you look at the dots and then have to decide ‘what do I have to connect these things to?’ or ‘where should I look next?’

Bar charts may be for sissies, but if that’s what most of our readers are, then perhaps bar charts it is, then.
Regardless, even more important than chart style is chart content. If this publication is about the effect of the recession across sexes and sectors, then plotting how things have changed over a year is a better choice than plotting how things look right now… as I’ve done in my 2nd bar chart above.

Content is still king. Or to paraphrase Tufte, is this the best data we can present to illustrate this story?

By: Chris P

Chris P — Fri, 29 Jan 2010 21:15:29 +0000

I am with Naomi on not serially connecting items that are randomly ordered. yes, it is nice to see the two series, but the angles created by the lines between categories are NOT data–in fact they are artifacts. The segmentation of male/female and by category provide dimensions that you feel are important to understand, but the within gender or between gender hypothesis could drive two different sets of plots. Your last dot plot introduces exogenous data (the change over time) to the graph–but we do not see this dynamic in the graph. A table sorted by the % of men in the industry then showing the change in unemployment over the period would help establish the claim better than the current dot plot.

By: chip

chip — Fri, 29 Jan 2010 18:32:09 +0000

Interesting discussion in the comments. Naomi made the comment that I came here to make: that the lines are connecting unrelated data and lead the eye to follow a pattern that’s not really there. I think your revised charts work perfectly well. I like them better than Jeff’s bar charts, because I think the bars make me more likely to look at the patterns of the blue bars in relation to each other rather than in relation to the red bar next to them. With the dot-plot I can also quickly discern how many categories have men “leading” vs. women, while the bars seem to make my brain work harder at getting to that.

By: Jon Peltier

Jon Peltier — Fri, 29 Jan 2010 18:15:57 +0000

Naomi - I was not thinking of you when I cited the No Lines commandment: your approach has always seems to be utilitarian, not totalitarian. Your guidelines for when to use lines are helpful. One of the most useful cases of lines in a categorical chart is in a parallel coordinates plot. Without the lines, the chart would be useless, yet I've had the discussion that I shouldn't use lines. Here are my dot plots, stripped of their lines. Does anyone have a comment?

By: derek

derek — Fri, 29 Jan 2010 17:03:44 +0000

Sorry, I meant instead of ordering from industries men are more likely to be in to industries men are less likely to be in, I’d order from industries men are more likely to be in compared to women to industries men are less likely to be in compared to women (as determined by subtracting the two percentages).

In practice, I see this actually only means switching Professional and Retail in this case. I was also a bit confused by the “Other” category, which I didn’t find useful since it’s just the bin that makes the whole thing up to 100% for each sex. I’d drop that category and just present the sequence C, M, P, R, E, H.