Go to Top

Visualization Considerations

Information visualization has become a big part of web analytics, with Google Analytics leading the pack when it comes to visualizing user interactions with a site, and a few new tools for social media analysis touting visualization capabilities as reasons to invest. With newer technologies making it easier than ever to create visualizations, as well as allowing more visually-engaging options than ever before as 3D becomes easier to process, a post on some of the basic guidelines that can be used when judging the effectiveness of a visualization seemed worthwhile.

Let’s start with an example. Below are two visualizations of a wave form from Stuart Card’s Using Vision to Think”:

Clearly, the bottom graphic is more effective. But why?

One of the fundamental requirements in information visualization concerns use of basic human perceptive abilities. Ideally, a visualization allows for fast interpretation by making use of automatic processing, the kind that allows forms to pop-out without any conscious attempt. Is how well a graph does this subjective to judge? To some degree, perhaps, but just as usability standards exist, so do general guidelines for designing graphs.

Now, for an example closer to SEM, here’s the map portion of the Map Overlay report in Google Analytics:

I am actually not a huge fan of this graph, though most GA graphs are fairly intuitive. Ideally, a visualization should convey distinction well without cognitive overload, even between similar values. Otherwise, we could just use tables for everything. In the map, by using a gradient, the colors on the left side of the scale appear more or less the same. When I see most of the map appearing the same color, one hardly discernible from the background, my first assumption is that these countries aren’t bringing any traffic, or are bringing maybe 1-5 people (based on the scale in the lower left). In actuality, several are bringing web traffic in the hundreds. The graph gives the impression of a pattern, as it should, but the error in the judgment it encourages is worth considering. As a fan of visualizations, rather than tables, I’d rather the surprise came from the former, but with this graph I need the table to get a real sense of the data.

Here’s another map example, Mark Newman (of UM)’s well-known graph of US Election Results in the 2006 Congressional Election:

I like how the borders are de-prioritized in favor of giving the viewer a better sense of the distribution of democrats and republicans, but the state outlines allow one to still note what state.

An easy evaluation measure is to consider whether the mapping of the data to the visualizing structure is expressive, meaning it represents all and only the data in the source table. The classic example of a violation is a bar graph being used to graph two non-ordinal variables (like countries versus food exports).

A final major consideration usually concerns “focus + context”, a term used often in HCI and interface design, referring to providing both the big picture and the detailed view. Zoom is a classic distortion example that demonstrates focus + context when done right. The best visualizations are those like the zoom capabilities of Google Maps, where details don’t come at the expense of the bigger picture:

facebooktwittergoogle_plusredditpinterestlinkedinmailby feather

One Response to "Visualization Considerations"

  • Daniel O'Neil
    July 16, 2008 - 4:32 pm Reply

    I’m not sure about Newman’s graph. I think it does a lousy job of depicting geographic data in a meaningful way. The distortions that come from showing so many dimensions actually make it very hard to extrapolate any interesting information about the overlay that was the starting point for the exercise, which is the geographic map of the United States. I can KIND OF see how Florida and Michigan are large population centers, but there is very little additional that I can meaningfully extrapolate from the map in terms of geographic data.

    This isn’t to say that the Google map numbers aren’t distorted too, but that is a fairly easy fix; they need to create a more informative spectrum and perhaps choose a consistent population scale.

Leave a Reply

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

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>