Registered: 1182202224 Posts: 100
Reply with quote #1
For the October/November/December 2017
Visual Business Intelligence Newsletter article, titled Heatmaps: to Bin or Not to Bin?, Stephen explores if and when we should use binned color scales on heatmaps—as opposed to continuous color scales—now that the technological limitations that led to the invention of binned scales are no longer relevant.
Do you agree with Stephen's conclusions in the article? We invite you to post your comments here.
Registered: 1325855753 Posts: 9
Reply with quote #2
Interesting article about an area I've never thought deeply about! It's convincing that you can see the difference on maps, though. And so, why misrepresent potentially important information using bins?
However, even though you can see the difference, I wonder if it helps for some kinds of interpretation. So, for example, I might have a scatterplot with a third variable shown using color. If I use a continuous color scale, I think it's harder to see (and show to others!) the pattern in the data, than if the color is binned. Same with heatmap tables. Of course, by drawing attention to this attribute to make interpretation easier, I might be forcing a bias...
Registered: 1135986598 Posts: 838
Reply with quote #3
Please provide us with an example of an heatmap matrix or a scatterplot that would benefit from binned colors. Unlike you, I cannot imagine a case when binned colors would offer an advantage. Regarding scatterplots in particular, typically heatmap colors cannot be effectively used in scatterplots because the data points are too small. To discern and discriminate colors, the objects must be of a sufficient size, larger than they typically are in scatterplots. __________________ Stephen Few
Registered: 1136226912 Posts: 247
Reply with quote #4
I assume the bins in your example are using simple equal-interval bins(?)
What if you use quantile binning instead? For example, if you have 5 quantile bins, then each bin would contain approximately 1/5 (ie, 20%) of the countries. Quantile binning provides somewhat 'meaningful' bin ranges, without the software or user having to know anything about the data. I guess my main complaint about continuous legends is that it's difficult to look at a color of a country in the map, and know where it falls in the continuous legend range (whereas this is easier with a binned legend). I guess the best choice depends on what's the purpose of the map! :)
Registered: 1135986598 Posts: 838
Reply with quote #5
Quantile binning is actually a good example of a binning approach that is meaningful. I used the example of student grades as a potentially meaningful binning approach in the article, but quantile binning is probably a more useful example. The bins are meaningful in that they are not merely a set of equal ranges into which the full range has been divided, but instead vary in range in some meaningful way. Quantile bins serve a different purpose than bins that merely divide a continuous quantitative range into equally sized subranges. Regarding your complaint, as I mentioned in the article, when we want to decode specific values, heatmaps are not a good solution whether we use bins or not. We typically use heatmaps to make rapid comparisons and to discern patterns, not to look up specific values. A table of numbers will always trump a graph when looking up specific values is our primary task. __________________ Stephen Few