Lying with Charts 102: Deceptions of stack

It’s time to lie with charts!

A few months ago I posted a FriendFeed item about how deceptions of scale can be used to make misleading charts.

Earlier this month, Wired argued in an article that “the web is dead” and tried to prove their point with a rather curious chart: A stacked area chart with the changing percentage of the Web’s share of total Internet traffic between 1990 and 2010.

Peer-to-peer and video take up a large share of total traffic in 2010 — naturally, because these files are huge, and web pages are relatively modest in size. A more honest chart would look at total volume of use (eyeballs/visits). If you read a tweet on twitter.com and then watch a video on Netflix, it’s not at all fair to assume the video was 3,571,429 times more important to you than the tweet.

But even setting aside the stupidity of conflating the importance of contents with its file size, a stacked percentage area chart is a visual deception. There are many ways to lie with charts, and stacking is one of the most frequent.

Here’s a simple example. Suppose your company sells three products, Wobjects, Dooders, and Flozzels. You have sales data for a year. It turns out that Wobjects are growing fast, and Flozzels are selling pretty well, but Dooders are sucking wind with sales volume decreasing. If you want to hide how poorly Dooders are doing, just use a stacked chart.

First, the pretend data:

Product Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
Wobjects 67 69 72 76 80 82 82 88 88 90 94 103
Dooders 70 70 69 68 68 64 67 63 66 66 66 62
Flozzels 42 43 43 44 46 51 54 58 57 58 60 61
Total 179 182 184 188 194 197 203 209 211 214 220 226

The most deceptive chart first:

Example 1: Stacked bar

A stacked bar chart serves to obscure individual trends

Certainly if you know what you’re looking for you can eyeball the decrease in sales from Dooders. But for most people only the overall growth would jump out.

Almost as deceptive is a stacked area chart:

Example 2: Stacked area chart

A stacked area chart is almost as good as a bar chart for obscuring individual trends

This is slightly less deceptive because the eye is slightly better able to see that the middle area is decreasing in size over time.

To be more straightforward, use a line chart with a total:

Example 3: Line chart with totals

A line chart is not very deceptive for individual trends. Including the total makes the scale larger, hiding the amount of increase/decrease for individual series, but is valid if the overall total is the most important detail to convey.

The least deceptive approach would be a simple line chart with no totals:

Example 4: Line chart

The most honest approach for showing the performance of individual product lines is a straightforward line chart.

And to make a total mess of things, use a stacked percentage area chart, with gratuitous 3-D:

Example 5: Stacked percentage area chart

Stacked percentage charts are usually disastrous, because the relative share of three product lines is often totally irrelevant compared to the actual change in value. And 3-D is just a distortion.

3 Responses to “Lying with Charts 102: Deceptions of stack”

  1. Howard Says:

    Tufte would be proud.

  2. Davis Freeberg Says:

    With skills that are this impressive you should be working at an investment bank instead of a tech outfit.

  3. DIGIVORE Says:

    clever…

Leave a Reply

AVATAR: Sign up for a free avatar with Gravatar.
CLICK FOR COMMENT XHTML TAG HELP