Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
Why Should Engineers and Scientists Be Worried About Color? (ibm.com)
95 points by edw519 on Aug 19, 2009 | hide | past | favorite | 19 comments


A better title would talk about "Why Should Engineers and Scientists Be Worried About Color in Data Visualization"

I initially thought the article would talk about colours in the design sense, and in which case engineers shouldn't worry whatsoever. It's not their job.


Making a big deal out of color in the design sense in tech is probably a great way to discourage men from buying your product: http://www.youtube.com/watch?v=lcBpXYI1r3Q


Not necessarily, though using feminine colours probably is. I've never seen a man get annoyed by the large number of available of colours when buying a car, or even a shirt.


Depends. I might not be able to see all colors, but I still appreciate their thoughtful use.

Make colors support your design, but do not rely on them. And please take the more common forms of color blindness into account.


I've only read partway into this article, but I really don't understand what he's saying in the section "Misleading Use of Color in Your Data". I literally didn't have a single one of the problems he listed with the colored representations. I don't see color banding, and I don't have trouble distinguishing various shades of green-to-cyan. Also, I think the third grayscale image in Figure 2 is really misleading, since it's showing a 2D cross-section as if it were 3D.

Used properly, the colorscale (is that a word?) should give 3x the amount of data as grayscale, which I think is clearly shown in the fourth image.


Yes, used properly. He's showing ways that information can be misrepresented when using colorscales IMproperly. Also he's showing that there is no such thing as "the one true colorscale".


Read it again then. Because his final images are MUCH better at visualizing the data, so he's doing something right.


Sure, now that I've finished it, I think he makes some excellent points overall. I just don't agree with what he wrote in that section.


>the colorscale... should give 3x the amount of data as grayscale

The visualization can't contain more data than the data itself!


Well, of course :) I mean the potential max amount of data can be increased with color, since now you have three axes (RBG instead of L). Although now that you mention it, he was only using "hue", and holding luminance steady, so I guess in this case it's the same - just rotating around the very outside edge of the color wheel without going inside. I still think it has more visual contrast, though.


3x because of RGB? please learn about color spaces...


For those wondering, I believe the visualizations on this page were created with OpenDX (http://opendx.org/) or its predecessor from IBM.


You can register by email to get a copy, but the link they sent me is dead (looks like the website doesn't exist anymore). Someone put up a copy at http://pagesperso-orange.fr/pierre.baldensperger/OpenDX-4.4....


I think "the rule" should be very simple: if your data has just one parameter, don’t use colors! The gray scale will give you the only faithful representation of the data.


One definite advantage to colors, however, is expressing certain thresholds (as in the coast-line example). The only way to express multiple thresholds in gray-scale is to distort the scale, and then you no longer have a faithful representation either.


If you are doing this visualization for yourself, you can experiment with colors and thresholds to bring out the features that are important (in your view). And that’s OK because you still have all the data at hand. If however this visualization is for someone else, such as the MRI given to a doctor, anything but a faithful representation would be very dangerous. Basically, you would substitute your judgment for that of a specialist.


You would still have to distort it unless you encoded the height to the luminance channel and only used the color channel to define boundaries.


“The only faithful representation”? What?


In other words it's the same thing: data = image = table of numbers.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: