Thanks for the link. I’ll repeat the warning about color-vision-deficient viewers. This is why I like to stick to a monochromatic color scale (white to fully saturated color) or a simple two-color scale using blue and orange as endpoints. Green and red in the same scale will cause problems.

Here’s an automated VBA version that doesn’t use conditional formatting for those on earlier versions of Excel.

http://ramblings.mcpher.com/Home/excelquirks/snippets/heatmap

Bruce

Thanks for your comments.

The first way to accommodate colorblindness is to stick to bar charts and to clearly labeled line charts. The same features that make a chart friendly for those with color vision deficiencies (limited rainbow effects, minimal clutter, not too many series, clear labeling, etc.) make the chart friendly for all viewers.

If you need to use a heat map, then a range of gradients of a single hue (e.g., white through gray to black) is colorblind-friendly, as well as photocopier- and fax-friendly. The use of patterns is problematic for human viewing and for mechanical reproduction.

The ColorBrewer palette generating utility, which I described in Excel Chart Color Update, has a colorblind-safe setting. there are numerous sites (e.g., VisCheck and Colorblind Web Page Filter) that will modify an uploaded image or entire web page to show how colors which are distinct to a person with normal color vision become indistinct to a person with color vision deficiency.

Tim Wilson wrote a detailed article about colorblind-friendly data visualization in Data Visualization that Is Colorblind-Friendly — Excel 2007?

Stephen Few touches on accommodating color vision in Dashboard Design for Real-Time Situation Awareness (pdf) and in Practical Rules for Using Color in Charts (pdf).

Thanks for taking this one up! I’m slipping on my RSS feeds in that I missed the post. For my purposes, the heatmap is a definite winner. Thanks!

Subtle differences are more easily picked-up on Bubble – however Heat Maps are also well suited when the plots are non-overlapping. (I realize that is not the case here, but on a map for instance, gradients relating to points make little sense and when overlapping provide a false condition.)

If the intent is to normalize the data so that there are only 4-5 different gradient levels, then it may be easier to discriminate between each gradient level, otherwise the differences are imperceptible.

Lastly, if the intent is to convey as much detail as possible while still providing a general pattern, I would opt for the area in the very first example. ;-)