Hacker News Reader

Comment by colinbartlett | Hacker News Reader

Pie chart: The world's worst way to display time-series data.

http://o.aolcdn.com/hss/storage/adam/2c59e6862a532820a9854bd...

For the statisticians and data scientists in the room : do pie charts ever have a valid use case (as defined by being superior to another form of visualization)?

Of course. Pie charts look natural and easy to comprehend when visualizing things like market share of certain products, for example browser market share (http://www.netmarketshare.com/browser-market-share.aspx?qpri...). In fact I'd argue that that's a much better representation than a bar chart, like http://www.w3counter.com/globalstats.php, because the former makes it easier to comprehend roughly what proportion of the total market each browser represents, which is not as straightforward to see in a bar chart (in bar chart it's only easy to see how different browser market shares compare to each other). So yes, pie charts may be appropriate at times, although I agree that it's a poor choice in this particular case, as the root comment explains.

How about a stacked single-bar chart? I'd suggest that it has all of the benefits of a pie chart without introducing perception issues of angles and areas.

Well, I guess you could make a fat bar and still run into problems... Ah, visualisation!

This pie chart -- http://www.youthblog.org/pie%20chart2.jpg -- is accurate and presents the data better than any alternative.

(so no one has to click blindly, it's a photo of a pie showing how much of it was eaten or not)

Whenever the data to be displayed is most easily conceived of as a proportion of a whole. If all of your data points added together equals the population you're sampling, then a pie chart might be appropriate.

It's natural to ask: "Of all R&D expenditures, what percentage came from each year?"

A time-series series (cumulative sum) would also answer the question, and could actually be plotted on the same chart as the original series.

On a side note, time series is not a great phrase for the original data set ("time sequence" would have made more sense, since the numbers shown are not cumulative)

Yes, a cumulative sum would have done a much better job than that pie chart at illustrating their point (from the text: "Apple over the last four years spent more on R&D than it did during the preceding 15 years combined")

Also note that 'time series' is a well-defined and widely-used term that exactly describes the dataset presented here (http://en.wikipedia.org/wiki/Time_series).

But if that is displayed as a line chart, that clearly displays the year-over-year differences. So if someone wants to ask "what is the overall trend of spending over the past five years?" that is nearly impossible with the pie chart.

Pie charts are by far one of the worst forms of graphs ever devised and should be killed off as a concept for good, as in, please remove this from Excel.

Everyone is this thread is conveniently ignoring that the data is initially presented as a bar graph. The pie chart is secondary for a very narrow purpose, and it's far from terrible in this application.

I disagree, that was a very effective use of a pie chart. By looking at it for a couple of seconds you can tell that the last 4 years take up over half of the total.

I suspect that many people will need to be told that. So the graph is a useful demonstration, but not useful to discover that information.

And the first bar chart which is not zero-based! (groan)

Reply on news.ycombinator.com