Interesting stuff, but definitely a 101-level course. A more in depth treatment would discuss the trade off between non-data ink and embellishment, which is one of the most difficult problems in good data viz.
Some more specific thoughts:
ALWAYS represent more than one data series if possible: otherwise you're wasting space. But make sure your thesis is still sensible. It's very helpful, for instance, to plot a series of interest against a backdrop of the greater population.
A histogram is NOT the best way to show the average value. A box and whiskers plot shows the average as well as the bulk of the population in a one-dimensional form that can be extended simply for comparison across categories or time.
Also always remember that the data is king. Line smoothing is only a good idea if you're not using it to give the impression that you have more data points than you really do.
All in all their visualizations provide great examples but they could do more to share the nuance that went into designing them.
Indeed it's a 101. Thanks for your comments. The only one I would really disagree with is "always represent more than one data series". That's really tough to represent multiple series on one single chart and manage to keep it immediatly readable.
Certainly a chart should be readable, but I think it's okay to ask for a few moments of careful study. I'm much more interested in a chart that provides some context or basis for comparison, if not with several series than with small multiples or a representation of the general distribution.
>use a pie chart or a donut. We know it's very controversial (data visualization mullahs tend to yell after anyone using a pie chart) but it's actually pretty usefull in a business environment.
Translation: it's for people who (you think) want pretty plots instead of an actual understanding of the data. The article's repudiation of "mullah" (?!) opinion is a little anti-intellectual IMO - these people tend to have a lot of experience and therefore understand common pitfalls better than a lay person's intuition.
Pie charts do not work as intended in general, but may work in specific cases like the one the article shows. Pie charts have low perceived precision because people are bad at estimating radial distances. If in the article's example potatoes were 60 and rice 58, the pie chart would not convey that rice is lower than potatoes - they would look the same. A bar chart or stacked chart does not lose this information because humans perceive location with very high precision. There are several other pie chart failure scenarios, like too many categories, or very imbalanced categories, or very balanced categories. Overall, it's a very non-robust visualization method. It may look good in your presentation now, but when you get updated numbers you may find it becomes a mess.
Maybe pie charts are not robust, but as I'm saying in the post, it's actually the best - if not the only - method to represent the contribution of the Top N contributors to a global measure. And only this. It's not meant to let the reader compare the relative measures of different items. If you wish to do so, I fully agree with your comment: use a bar chart instead.
>it's actually the best - if not the only - method to represent the contribution of the Top N contributors to a global measure
Yet it's still not great at that. Stacked bar chart (basically an unrolled donut chart) is better - again, because people perceive linear distances with much higher precision than radial:
Treemaps are very powerful during the data crunching phase, but they are not ok for a final deliverable, as they ask way too much to the reader's brain to get understood.
Makes sense. I'm curious though, do you mean that even with the same number of categories for each it's too difficult? ie. instead of using hierarchical groupings and all categories no matter how small, use only the top ~10 and bucket the rest into "Other" similarly to a pie chart or bar.
12 comments
[ 3.2 ms ] story [ 50.8 ms ] threadSome more specific thoughts:
ALWAYS represent more than one data series if possible: otherwise you're wasting space. But make sure your thesis is still sensible. It's very helpful, for instance, to plot a series of interest against a backdrop of the greater population.
A histogram is NOT the best way to show the average value. A box and whiskers plot shows the average as well as the bulk of the population in a one-dimensional form that can be extended simply for comparison across categories or time.
Also always remember that the data is king. Line smoothing is only a good idea if you're not using it to give the impression that you have more data points than you really do.
All in all their visualizations provide great examples but they could do more to share the nuance that went into designing them.
Personally, I found this very useful:
http://complexdiagrams.com/properties
* this is not a shameless plug!
Translation: it's for people who (you think) want pretty plots instead of an actual understanding of the data. The article's repudiation of "mullah" (?!) opinion is a little anti-intellectual IMO - these people tend to have a lot of experience and therefore understand common pitfalls better than a lay person's intuition.
Pie charts do not work as intended in general, but may work in specific cases like the one the article shows. Pie charts have low perceived precision because people are bad at estimating radial distances. If in the article's example potatoes were 60 and rice 58, the pie chart would not convey that rice is lower than potatoes - they would look the same. A bar chart or stacked chart does not lose this information because humans perceive location with very high precision. There are several other pie chart failure scenarios, like too many categories, or very imbalanced categories, or very balanced categories. Overall, it's a very non-robust visualization method. It may look good in your presentation now, but when you get updated numbers you may find it becomes a mess.
Yet it's still not great at that. Stacked bar chart (basically an unrolled donut chart) is better - again, because people perceive linear distances with much higher precision than radial:
http://i1.wp.com/flowingdata.com/wp-content/uploads/2008/08/...