Bar Is the New Pie

Everyone is sick of talking about how pie charts suck. And whenever I do a charting post, it’s trouble. So this post is destined for failure. However, I was thinking about pie charts recently. I don’t have the disdain for pie charts that some do, but I generally agree with the majority of pie-haters that it is an overused and abused chart type. I think pie charts can be effective when you have two data points; three at most; never more than three.

Pie charts are difficult because they rely on area. When you go from one dimension to two, the one dimension is squared. It’s more difficult for humans to grok differences in area than differences in linear measurements. But let’s get to the pretty pictures. I stole some data from Seth Godin.

The purpose of this chart is to show that “Trolls” is big. And it does that. If your goal is to show how one data point relates to the whole, this works. It violates my “never more than three” rule, but the case could be made that there are only two data points here: Trolls and non-Trolls.

Doesn’t have quite the same impact, though. It appears that non-Trolls are more significant. If instead of comparing one data point to the whole, you want to compare data points to each other, the bar chart is the way to go.

Nothing wrong with that. In fact, I think it demonstrates Troll dominance even better than the pie. Back to my pie chart ruminations. What I was wondering why pie charts are round. That is, why do we show one-data-point-to-the-whole as a circle? It works just as well, or better, when we eliminate (or just fix) the second dimension.

I didn’t invent the square pie chart, but I can’t remember ever seeing it before. I wonder why it’s not a more common chart type. Someone once said, and I’m paraphrasing, “Pie charts are good when all of your data points are divisible by 25%”. Good point. I changed the above chart from values to percentages and added an axis with 25% as the major grid. People can relate to 25%, especially on a circle but even on a stacked bar.

So now you know what I think about. Scary, isn’t it? I’m going to start a movement to make the single-series-stacked-bar-in-lieu-of-pie chart more popular. But I need a name for it. Something that makes people feel as good as pie. Brownies are typically baked in square pans. Maybe I should call it a brownie chart.

Some details: To make a brownie chart, select stacked bar and on the Data Range tab, select the Rows option.

When you try to turn your data labels on end, the labels get cut off. I had to add four spaces to the front of the data and two to the end. I used this formula to transform the original:

=REPT(” “,4″)&A1&REPT(” “,2)

What do you think? Are brownie charts the wave of the future?

Posted in Uncategorized

16 thoughts on “Bar Is the New Pie

  1. Let’s call it the Kusleika chart. For some reason it sounds like some great mathematician invented it. :)

  2. I love how at the Microsoft Excel Team Blog at they use a pie chart to compare calculation performance gains in 2010 against 2007. They have 4 categories:
    1. On par with 2007
    2. 10 – 30% faster than 2007
    3. 30 – 50% faster than 2007
    4. More than twice as fast as 2007

    Seems a mighty strange way to present the data, given the non-equal sized ‘bins’…never mind the fact that strangely nothing falls in the 50% – 200% range.

  3. <>

    While a bubble chart is guilty as charged (bubble diameter is determined by the value, not bubble area), a pie chart is innocent on that account. This is because the radius is the same for one pie slice as it is for the next. Pie slice area, arc angle and circumference are all proportional to the percentage value.

  4. Dick –

    The unstacked bar chart is still a better chart. Put a scale on the horizontal axis if you want to show percentages.

    In the unstacked bar, the Gremlins are equal to the Orcs. In the stacked meatloaf chart (or whatever you want to call it), it looks like the Orcs are a larger portion. Without a common baseline for the bars, it is difficult to judge the relative sizes of the bars.

    Brad –

    It’s true that the values in the pie are proportional to the area, or arc length, or angle. The problem is human cognition. When judging linear dimensions (lengths of bars in a chart, for example), we perceive a change in value roughly proportional to the change in length. When judging areas, we perceive a change in value that is only about 70% of the change in area. Angles are similarly misjudged.

  5. David, over here we call them Popsicles.

    I find that pie charts are useless unless the percentages are shown. Unless the pie “slices approximate easily recognized simple fractions like 3/4, 2/3, 1/2, 1/3, 1/4, 1/6, 1/8, people will not come close to estimating anything correctly.

    The stacked bar charts confuse me. While I may be able to approximate the relative size of bars (I prefer nanaimo - – by the way), the stacking doesn’t provide any immediately meaningful information except the total when the long dimension has large numbers at tick marks. Again, approximating a number near the middle, means approximating those at the extremities and working in.

  6. I thought of calling it a push up chart, but I thought the reference was too obscure. I guess crazy minds think alike.

    “it is difficult to judge the relative sizes of the bars.” But I don’t want you to judge the relative size of the bars. I want you judge the size of the trolls relative to the total. I don’t think the unstacked bar chart does that as well as the pie chart.

  7. “But I don’t want you to judge the relative size of the bars. I want you judge the size of the trolls relative to the total.”

    Then stick to your pie with Trolls and Non-Trolls.

  8. I think the big advantage which pie charts have is that people instinctively know that it’s a proportion of a whole. With a bar I don’t think there’s that instinct.

    I’m not saying it’s right or wrong, I just think that people would know without thinking what it represented.

    But two or three segments is probably the way to go.

  9. An essential rule for both pie and loaf charts is that the data must be displayed sequentially, smallest to largest. It helps, also, if the colour scheme has some kind of sequence – progressively darker moving from smallest to largest, or a ‘thermal’ gradient from warm to cool if you want to draw the users into value judgements.

Leave a Reply

Your email address will not be published. Required fields are marked *