The assault on the pie chart | Hacker News Reader

The assault on the pie chart

priceonomics.com

55 pointssonabinu25 comments

It seems to me many of these tests are asking pie charts to do something they're really not meant for. Sure, you can create a synthetic problem and dataset where you need to order four different categories, all comprising around quarter of the whole, and there a bar chart would excel.

But that's not actually that common a use case: a lot of the time, I'd only be interested that they are all roughly a quarter of the whole, and not care beyond that (providing a ranking of small differences can actually be pretty misleading, depending on the problem you're trying to answer).

I can just as easily say, I want to know roughly what proportion of the whole a category comprises. And with widely varying category sizes, the pie chart excels, while with a bar chart I'd have to sum the bar sizes in my head, and divide that total by the category in question. That's at least as taxing as having to differentiate between small differences in pie slice sizes.

The answer isn't "always use pie charts" or "never use pie charts": it's to use what's best for a particular situation and context.

There's no rule that a bar chart cannot mention the proportion of the category above each bar, and there's also no rule that a bar chart has to be for absolute and not relative numbers.

After having worked with a lot of data visualization I came to the conclusion that the purpose visualizing most data is to show proportion, trend or significance not details.

I.e. it only has value if it clearly visualizes the point trying to be made. Otherwise you might as well show it on a list or it might come up as an alert.

The challenge is not to be able to read and understand graphs but to know what to do with them. This is why most dashboards are purely made to make the product look sexy not to actually provide any valuable insights.

Agreed. Charts are narrative.

They are the visualization equivalent of TLDR. It captures the presenter's gist of the data but doesn't tell the whole story.

Here's a (brain) dead simple rule about pie charts: if you're using them to demonstrate a difference or change, so that you end up showing two, or more, pies, and if the aggregate amount represented by each pie isn't equal, then you're up to no good. Even if you try to compensate by making one of the pies have a larger radius than the other this is bad, as area comparisons will always be under perceived. If the case distinctions have the same summed aggregate, and you're just showing how the apportioning of the groups have changed then you're ok. But then graphically why not just have two bar charts which are visually more easy to interpret?

I build BI products for a living and used to get into a lot of pedantic arguments about whether a pie chart or worse, the dreaded 3d bar or line plot, should ever be used. Eventually I had to give it up though as it's not a productive use of time to try and pry beloved bad visualizations away from business users, plus I sensed I was starting to become a stephen fewesque curmudgen. They haven't reasoned their way into liking them so you can't reason them out of it most of the time.

Instead I'm focusing on building tools that show data in better ways first, but let you get to other options if you really want to giving viewers you share views with a way to quickly switch back to something easier to grok. Even initially graphically purist tools like Tableau have had to go this route and you can now find nightmarish packed bubble plots and the like in there. In modern visualization there are so many worse visualizations that have crept in that arguing about the pie chart seems almost quaint to me at this point. For example check out this flying 3d globe with bar chart where the bars point directly up (at the camera) https://www.microsoft.com/en-us/powerBI/power-map.aspx - A while back someone from Microsoft demonstrated this to me and after I gave him a hard time about it being an improper way plot geographic data he retorted "The world is round you know!"

edit: formatting

I have no problem with pie charts. The area chart on the other hand are pure evil. For example here, how is the percentage of ios over time? Has market share grown or shrunk from Q3'11 to Q1'14? It's extremly hard to answer this kind of question except for the lucky ones getting the top or bottom slot. http://static1.businessinsider.com/image/537b72f169bedd1e043...

The big, big problem with such charts is that the vertical thickness perpendicular to the x axis of the line is the actual proportion, but your brain will look at the thickness perpendicular to the gradient.

To put it visually: http://i.imgur.com/vZ7uTu1.png

The pink bars are the percentage market share as your brain reads it. The green bars are the percentage market share the graph is supposed to show. (The burning sensation in your eyes is because I made a poor choice of colours.)

If you were to naively look at the thickness of the black area (shown with the pink bars), it'd look like it was constant. But look at the vertical height (green bars): whatever this proportion is, the black has started small, grown massively, then contracted.

Area charts are valid only if the area is meaningful. There certainly are cases where it can make sense to use a visualization based on filled, stacked areas. Look at, for example, Weatherspark.com's colortemp visualization for historical weather trends, which shows 'percentage of time when temperature is in a particular band' over a time axis. That's a good visualization, because the quantity makes physical sense - integrating 'time in band per day' over 'days' gives you overall 'time in band'. Integrating 'market share' over 'months' gives you 'market share months', which is a meaningless value.

I've seen pie charts used particularly effectively in two areas of personal finance:

1) In budgeting apps, to show the distribution of monthly expenses across categories.

2) In investment tutorials and reports, to show a distribution across asset classes.

In the case of budgeting, it's true that a pie doesn't capture all variables, like total expense from month to month. But if you're trying to optimize a metric, such as savings (which can be included as an expense category to visualize savings rate), it makes it intuitive to see which budget categories are throwing you off. Suddenly it's not about reducing expenses, but rather shifting them between categories of a whole. At 50%, there is a tipping point where the pie slice becomes concave. If you're trying to push yourself to a 50% savings rate, it's an interesting visual benchmark.

For portfolios, here are a few examples. These are 3d pie charts, perhaps the most-mocked visualization type of all. But they communicate the information effectively enough. Repeated pie charts emphasize to the reader the importance of asset allocation.

https://www.bogleheads.org/wiki/Lazy_portfolios

What makes pie charts effective is that they are simple and approachable. There are better graphs for showing a portfolio's size and allocation over time-- but not for initially communicating the idea of asset allocation and giving a few examples. Bogleheads is one of the best resources on passive index investing, and I think it's not an oversight how often pie charts show up on the wiki.

If anything, I would say visualizing a portfolio as a line chart can be more deceptive. This is a common technique used to sell mutual funds that have recently outperformed to naive investors. Line charts can implicitly promise a future trend which may not materialize.

Pie charts aren't perfect but are useful when applied correctly.

Two great things about pie charts:

1. From the shape you can tell you are looking portions of a whole. For example, a pie chart of market shares shows all businesses in the category.

2. You can easily guess the proportion of the segments relative to each other. For example, in this chart http://www.liv.ac.uk/~cll/lskills/WN/pie%20chart1.gif you can tell that football and rugby make up half the total and cricket, tennis, athletics, and other make up the other half.

Works for that chart. But its depends on the author to draw the components that way, next to each other, and orient the chart so its obvious. And still, what other combinations add up to half? Can't tell from that chart. One with values side-by-side and a scale would allow you to do the math.

Pie charts are not serious graphs. They are useful for arguing points, where you (deliberately?) don't want to confuse folks with numbers.

That encourages you to make a spurious connection between football and rugby. Football and Other also make up a half, but that wasn't what you immediately spotted.

Well, they're both winter sports while the others (apart from "Other") are summer sports.

But I take the point -- you only notice that connection because of the way they happen to be arranged on the graph.

One warning about pie charts in many visualization applications today is you are often looking at how the terms in the chart relate to each other. Not over the set of all terms.

If I want to draw a pie chart of the top-5 terms for some field and the chart represents just those terms, then the chart represents how they compare against each other - not necessarily how those top-5 terms compare against the entire set.

I know, sounds obvious but I've seen data misinterpreted because of this. The pie chart must represent the entire set and if you're plotting a subset of the data in the chart and ignoring the rest then your visualization may not say what you think it says.

I can only assume this was covered 60 years ago:

    https://en.wikipedia.org/wiki/How_to_Lie_with_Statistics

Misleading charts have long been an important and valuable career skill.

> In Cleveland’s paradigm, the perceptual task associated with a bar chart is judging position on a scale, and the task associated with a pie chart is judging angle. They found that the length judgments for the bar chart were “1.96 times as accurate as angle judgments” for a pie chart.

In pie charts, I find it much easier to estimate a percentage than estimate the angle, perhaps just based on value range alone; 100 vs. 360.

I also find estimating percentage to be harder with a bar chart since there is no reference to the entire pool of data; the encompassing circle in a pie chart.

Pie charts are good if you want to say e.g. "Hey! A large parr of our problems are caused by X!" That's actually what Florence Nightingale used them for when she popularized them, she showed the British military leaders that "Hey! Most of our troop deaths are caused by infections in soldiers with minor injuries! We can fix that!"

The problem with pie charts is that it's too easy to use them in a wrong way. I would say the only scenario when a pie chart might be a right form of data visualization is when you need to display shares of two, maximum three, components of a total. It means 2-3 data points only. Trying to use pie charts for anything else will quickly degrade readability. More data points, day-over-day comparisons, variations with the circle size, grids of pie charts -- all makes things worse.

When I give demos of our product, I often show the pie chart and tell people, half jokingly never to use it, this always produces a good laugh.

Pie charts have merit, especially when used in a dynamic dashboard that changes based on either selection, filtering, or the passage of time. They may be over used, but they are not without merit entirely.

Rarely yes but only if those changes are noticeable which they most often aren't.

Most dashboards are purely made because they look good. The actual purpose of a Dashboard is to give a skilled person an overview of the entire system and see if anything is out of the ordinary and if not whats wrong.

The way most dashboards are used today they are more working like glorified lists and gives nothing of value to the user.

I love pie charts. Good luck making sense of scatter and bubble charts unless you're already intimate with the data.

And that's something of the challenge with visualizations. Obviously some visualizations are just bad. But others are merely complex because they include a lot of information. This latter sort of visualization won't be appropriate everywhere--like accompanying a typical newspaper story--but it might be just the thing to really dive down into data.

A great example of this is actually the Napoleon army graphic that's famous for being in Tufte's first book. It's a great visualization of what happened as Napoleon's army went off to Russia and returned. But what it's showing isn't really immediately obvious (and probably wouldn't be even it if were redesigned in modern graphical style).

fuck pie charts

Reply on news.ycombinator.com