Skip to main content
Search form
  • 00:01


  • 00:16

    SPEAKER: Color is one of the most visually exciting waysin which you can completely screw over your visualization.What can start off as a great diagramcan be absolutely ruined by lack of color judgment.That being said, once you've selected an encodingand shapes--this is something we've already talked about--color is a terrific way to add layers of information.

  • 00:39

    SPEAKER [continued]: Color forms a distinct visual channel.And as we've already seen, it canact as a powerful force for grouping.This grouping can come not just from shapes of the same color,but shapes of a similar color.This should make you ask, how do wecharacterize color similarity?Can we quantify this similarity?Can we say that one pair of colors

  • 01:01

    SPEAKER [continued]: is more similar than another pair, and by how much?Luckily, we can.Actually, one of the reasons white color canbe used so effectively in a visualizationis because it has a very quantitative characterization,which is based on how we actually perceive colorwith our retina.This is a very exciting prospect,since other senses don't have this mathematical model.

  • 01:25

    SPEAKER [continued]: Can I say that one pair of smellsis twice as different as another pair?No.Color is one of the things you see first, especiallyif it's used sparingly.It's extremely salient, moreso than size and shape.As such, it's extremely useful for establishing emphasis.If you want to draw the eye to something, color is terrific.

  • 01:48

    SPEAKER [continued]: But if everything is colorful, this kind of emphasisis actually impossible to achieve.There must be contrasts between aspects of a visualizationfor things to stand out.When it comes to color, this means contrast or differencewith respect to color properties that we can perceive.Overuse of color is rampant.Please don't be someone who adds to this problem.

  • 02:10

    SPEAKER [continued]: This graphic and many such like it exist in the literature.It suffers not merely from the fact that many different colorsare used, but more importantly from the fact that color isused to encode three dependent kinds of information--feedback mutation status, in the center,sample cross clusters, top rows, and gene clusters,

  • 02:30

    SPEAKER [continued]: right columns.Differences between colors within these groupsare large as between groups, and this makes the groups blendinto one another on the page.Some colors from one group, again, in the heat mapand third column in the gene cluster are very similar--reddish.And you wonder whether their meaning is related.

  • 02:52

    SPEAKER [continued]: Other color choices-- brown for mutation and gain--appear less salient then their component counterpart,green mutation, red again, even though the importanceof what they encode is greater.Now I just mentioned that color forms distinct visual channels.Just look how the red pops out here--how the patterns in the gain part of the heat map,

  • 03:13

    SPEAKER [continued]: RHC, if I suppress the use of all the other colors.Color is an excellent way to establish continuityin a narrative.By placing the color legends at the topof this multi-panel story, I've given the readerthe color key before they start looking at the plots.I'm sure that you would agree that it would be a bad decision

  • 03:33

    SPEAKER [continued]: to suddenly change the meaning of the blue colorin panel three, to, for example, encode lung cancer and orangeto mean colon cancer.Continuity would be lost, and comparisons would clash.Suddenly, we would be associating womento colon cancer because they were both representedby the same color.This kind of thing happens a lot in presentationsacross many slides.

  • 03:54

    SPEAKER [continued]: The same color is used for many different things.Avoid this, where possible.The use of spot color is uncannily effective.The orange nodes draw your attentionand very strongly communicate that theseare the most relevant parts of the diagram.Other elements are shown to provide context.This idea of showing patterns and relevant data

  • 04:15

    SPEAKER [continued]: in the context of background, noise, or the full data setis very appealing.It's not always easy to achieve.Balance of visibility and salience has to be struck.Elements must be visible and clear,but draw your attention to a different degree.Because color is so salient, it establishes a hierarchyof importance, whether you want it or not.

  • 04:35

    SPEAKER [continued]: So you better want it.The use of gray here for the control and orangefor the patient emphasizes that the control is the baseline.It would be borderline irrationalto use red for the control and great for the patient.Notice how this encourages the interpretation of the variationcount decreased in the control as opposedto the more sensible interpretation

  • 04:55

    SPEAKER [continued]: that they increased in the patient.Color is excellent at communicatingcategorical information.As long as no one color stands out more than others,the categories are seen as equal,which is the case for a nominal categorical variable.An example of a nominal category variable is gender.There is no intrinsic ordering here.

  • 05:19

    SPEAKER [continued]: If one of the colors does stand out, such as the yellow,it does so because it is much brighter than the others.The diagram can falsely suggest that this category is markedlydifferent than the other two.Also notice how the yellow has the effect of makingthe dark green and blue look similarand therefore themselves act as a group.

  • 05:39

    SPEAKER [continued]: Instead, one should use colors where this kind of groupingis minimized, such as in the bar plot below.Color is a kind of visual alarm.It really makes you pay attention.And you don't always need an alarm,nor do you always need color.Sometimes, just a little visual reminder or noticeis all that's needed.

  • 06:01

    SPEAKER [continued]: This can be achieved by tone--so just grays.Let's see how we can keep the attention of the guyand communicate categories and emphasis without color.Remember this plot?Do we need color here?No.Here, the dark gray is more salientthan the light gray and plays the same role as orange.

  • 06:25

    SPEAKER [continued]: There's obviously a limit to how many colorsyou can perceive, distinguish, and more importantly, remember.This figure is all kinds of wrong.And its generosity with color is one of its faults.Notice how some colors are so dark that they appearalmost black in the legend.The thick black outline doesn't help here.What can you do in this case?Well, when you're faced with a situation where

  • 06:46

    SPEAKER [continued]: you have too many categories, it might be betterto throw color out altogether.The graphic is really a table and best shown as such.The categories don't need to be identified by colors.They already have distinguishing labels.Where color might be useful here is if some categories wererelated in an important way.For example, all categories that relate to cell death

  • 07:09

    SPEAKER [continued]: might be colored dark red.Also, notice that in this form, the figure makes it obviousthat not all conditions appear in both columnsand makes it clear which conditions these are.This is an even worse use of colors.A large number of colors is used to encodea quantitative variable, whose values obviouslyhave an intrinsic order.

  • 07:30

    SPEAKER [continued]: But the colors have no apparent order to them.Moreover, it is very hard to see howthe positions of the points on the plotrelate to the changes in the value of this variable encodedby the color.The mixture of decimal and scientific notationin the legend is frustrating.By making an explicit connection between the linetraces and their corresponding alpha value,

  • 07:52

    SPEAKER [continued]: the color problem is completely avoided.There are times when a large number of colors,such as six to 10 or so, can be managed reasonably well.Other than some awkward design choices, this graphicisn't bad.The color palette is reasonably uniform,and no one single color captures your attention.

  • 08:13

    SPEAKER [continued]: There is one strategy that you can adopt in a case like this--start bar plots-- that helps you emphasize important patternsand the proportion of categories of crossbars.By making some categories draw from gray tones and othersfrom a color palette, the boundaries between categories,and importantly, the ranking of the categoriescan be quickly perceived.You would assign categories to grays and colors

  • 08:35

    SPEAKER [continued]: to focus on patterns that are relevant to a data set.Not all will be.This works particularly well if some categories are indeedmore interesting, from a results perspective-- for example,if changes in the proportion of [INAUDIBLE],, the light blue,are worthy of emphasis.Color can be also used to encode quantitative variables.

  • 08:56

    SPEAKER [continued]: I say can be and not should be because this process isfraught with danger.The purpose of this graphic is to demonstratethe location or locations of the minimum value of a variable.The color map has been created by controlling fluctuationsin the color's richness and perceived brightness,with hue being the main characteristic thatis used for the encoding.

  • 09:16

    SPEAKER [continued]: We'll return how to make good colorramps as one of the exercises.When it comes to color ramps, this is what not to do.As R gets more negative, the bluegets lighter for a bit and then darker.When R gets more positive, the redgets lighter, then darker, then later again.

  • 09:37

    SPEAKER [continued]: The blue ramp seems to vary in hue,and the difference in the last three bluesis much bigger than the difference in the last threereds.What can we do?The so-called brewer pallets simplify the selectionof color palettes.They provide color choices for categorical and quantitativevariables.Many options exist, and some are safe for colorblindness.

  • 09:59

    SPEAKER [continued]: We can replace the horrible red and blue color rampfrom the previous figure with a choicefrom the blues and red sequential palettes here.Notice how much more uniform and cohesive the encodingsbecome once brewer colors are used.One important point to bring up hereis that it's unlikely that a smooth color ramp is actuallyneeded.It might be enough to discretize the R interval from minus five

  • 10:22

    SPEAKER [continued]: to five into, for example, five segmentsand assign a color to each.This would give you three magnitudes, zero,low, and high, together with a positive and negativedirection.One of the reasons why most color ramps don't work wellis that they're not perceptually uniform.

  • 10:42

    SPEAKER [continued]: This means that the same change in the datais perceived differently dependingon the position of the baseline.There's only a small perceived differencehere between the two greens shown on the left,but a large one between the yellow and the red,even though they correspond to the same change in the data.Look what happens when we take a typical rainbow color ramp

  • 11:03

    SPEAKER [continued]: and encode two signals with it.Depending on the baseline, differencesare either invisible or huge.For example, if the background is 0.5,the 200% peak reaches into the redbecause it has a value of one.But the 150% peak is only at 0.75,which is a lime green and hard to detect above the background.

  • 11:25

    SPEAKER [continued]: But if the baseline is 0.6, both peaks reach into red,and both are obvious.None of this matters when gray is used.Notice how it's pretty easy to see both--that the relative brightness between the signalsis the same, or at least not hugely fluctuating,and that the background level is increasing.

  • 11:46

    SPEAKER [continued]: If dark text is drawn on a dark background, it's hard to read,and there's not enough luminance contrast.In this case, make the color lighter or dispense with colorand choose gray tones.You can get into serious trouble if you don't pay attentionto contrasts.If you combine pure colors, you get simultaneous contrast,which is murder on the eyes.

  • 12:08

    SPEAKER [continued]: OK, but at least the text is legible, if onlyfor short periods of time.But if you don't have enough luminance contrast,then the text actually may be illegible, especially whenprojected at small sizes.The gray-scale brewer palette is very usefulbecause like other pallets, the luminance progressionacross the colors is reasonably uniform.

  • 12:29

    SPEAKER [continued]: If you sample uniformly from RGB space,you don't get this effect.Look how much darker GAP and RAF are from the uniformly sampledRGB ladder than when the brewer palette is used.The moral here is that often color isn't even necessary.It can get in the way of legibility and clarity.

  • 12:51

    SPEAKER [continued]: When using color, ask yourself, do I need it?Try to work around color and use gray tonesfrom brewer palettes.If you succeed at doing this, you're in a perfect placeto use spot color sparingly, for emphasis.Now, color does make things exciting to the eye.But then what's your goal?

  • 13:11

    SPEAKER [continued]: To excite the eye or inform the brain?Often if you just do the first one,the brain checks out because it gets satiated early.I mean, when was the last time yousaw a movie that had a lot of great special effectsbut also had a great plot?Always be very critical of any kind of graphicsthat use a color ramp.

  • 13:32

    SPEAKER [continued]: If your doctor is looking at your brain scan,and it's using a rainbow color map, get another doctor.Above all, become familiar with alternative and more usefulways in which color is characterized.Read up on LCH and LAB color spaces.Use that luminance, the L coordinate,

  • 13:53

    SPEAKER [continued]: in the Photoshop color picker.[MUSIC PLAYING]

Video Info

Series Name: Essentials of Data Visualization

Publisher: University of Sydney & Canada's Michael Smith Genome Sciences Centre

Publication Year: 2017

Video Type:Tutorial

Methods: Data visualization

Keywords: color perception; color planning; communication aids; contrast perception; design information; encoding; luminescence; visual communication ... Show More

Segment Info

Segment Num.: 1

Persons Discussed:

Events Discussed:



Martin Krzywinski explains the roll of color and contrast in visual data. Krzywinski provides examples of color use and cautions to use grey tones if legibility and clarity are not achieved.

Looks like you do not have access to this content.


Martin Krzywinski explains the roll of color and contrast in visual data. Krzywinski provides examples of color use and cautions to use grey tones if legibility and clarity are not achieved.

Copy and paste the following HTML into your website