Inspired by a recent post on Kristan Wheaton’s blog Sources and Methods highlighting the neat Cymbolism project website, I felt inclined to submit various words that are relevant to my interest in risk and uncertainty analysis.  After all, who wouldn’t want to know what colors people associate with certain words?  For those readers unfamiliar with the Cymbolism project, basically it is an Internet-wide survey that presents participants with a word and asks them to select one of 19 colors that best matches the mood or visual images inspired by the word.  The goal of this project is to:

attempt to quantify the association between colors and words, making it simple for designers to choose the best colors for the desired emotional effect (see the blog for more details)

I just checked the site, and to my surprise, all of the words I submitted are now taking opinions from web passer-bys.  Unfortunately, the site doesn’t offer a feature to dynamically link to the color scheme for a word, at least not yet.  So, instead below are the links to the various words of risk and uncertainty that might be of interest for risk and uncertainty students, researchers, and practitioners.

• benign: (very mild looking)
• confident: (very bold colors here)
• consequence: (I see lots of yellow and red, but no apparent agreement)
• certain: (very similar to confident)
• dangerous: (VERY RED, much more than consequence)
• hazardous: (VERY RED, then yellow)
• ignorant: (no apparent agreement on colors, which doesn’t surprise me)
• improbable: (the array of colors feels gloomy)
• invulnerable: (very dark)
• insecure: (oddly mustard yellow)
• impossible: (very bold and dark in a manner similar to confident and certain, but with more black.  Very interesting, especially given that “impossible” is a high confidence or certainty word)
• likely: (dominated by a lot of green and blue)
• possible: (way more green than likely)
• probable: (no apparent agreement in colors)
• risk: (very surprisingly red, much like hazardous and consequence)
• secure: (no apparent agreement)
• safe: (dark greens and blues, very dull and relaxed)
• threat: (super red, just like risk, dangerous, and hazardous)
• trust: (deep greens and blues, nice and bold)
• uncertain: (very gray, subdued)
• unlikely: (also gray, but less so than uncertain)
• unsafe: (very red, but less so than risk, threat, etc.)
• vulnerable: (no apparent agreement)
• vulnerability: (oddly purply-pink)

Observations

I can’t help but feel partly inspired to develop several questions that might warrant future experimentation.  For example, can the intensity of colors be pegged to varying degrees confidence?  Based on what I have seen so far, more confident words were bolder (brighter, more intense, e.g., “certain”) than less confident words that were more subdued (bland e.g., “uncertain”).  Does this hold in general?  I submitted several other words to test this hypothesis (albeit in a very loose way), including “distrust,” “high confidence,” “moderate confidence,” “low confidence,” “no confidence,” “surprise,” and “chance.”  These phrases have yet to be posted.  The results from this general experiment might be of interest to the intelligence community; after all, there was a time when colored text was used to communicate levels of confidence.  Maybe we should focus on changing the intensity of the colors to express confidence rather than changing the colors themselves?  That is, instead of making low confidence red and high confidence green, what if we made low confidence light gray and high confidence black?  Can you imagine what an NIE that followed such a scheme would look like?  Just for grins and illustration, I took the unclassified key judgments of the November 2007 NIE entitled “Iran: Nuclear Intentions and Capabilities” and applied shades of gray to complement judgments (see below).  Overall, it doesn’t look that bad.

Also, notice that the words “likely” and “probable,” while similar in meaning, have different color distributions (you have to see for yourself).  It seems that there is general agreement on what “likely” means “cymbolically,” yet no apparent agreement on what “probable” means.  Perhaps this result suggests that “likely” is a better word stem to use than “probable” when communicating degrees of likeliness given its stronger cymbolic agreement.  Of course, whether cymbolic agreement matters depends on whether it serves as a proxy for conceptual agreement.  I have yet to do a literature review to see how academia might address this line of thought, or if remains an open question.

What I found really interesting was the fact that the words “risk,” “threat,” “hazardous,” “dangerous,” and “unsafe” are all cymbolically similar (also very red).  I will add the words “hazard,” “danger,” “peril,” to see just how red they get, but I suspect they will yield the same results.  In practice, many risk scholars get quite frustrated when practitioners equate the concepts of risk with threat.  Having seen the Cymbolism results, it seems now that the equality happens at perhaps a much deeper level.  Maybe one strategy to break people of this equality is to push different colors for the terms risk and threat (it is ok if threat and hazard are taken to be the same), such as through colored text definitions, colored images, marketing materials, etc.  The question I have then is whether color re-conditioning assists in getting people to dissociate previously accepted word equalities?  Or in simpler speak, if I condition a group of people to start to see threat as cymbolically purple, will they also stop equating the words risk and threat?  If so, would complementary colors work better than more similar colors?  A very interesting experiment indeed.

Another question to ask is how an adverb such as “very” or “somewhat” affects the color schemes for a word.  From the results so far, likely appears dominated by greens and blues.  What if we say “very likely” instead?  Does the color scheme become more green, more blue, less intense, more intense, or what?  Does the phrase “somewhat likely” have the opposite effect?  Again, more questions worth testing.  I will try to submit these words too.

Finally, I am curious as to whether the definitions offered for a word actually have any affect on the survey responses.  An interesting study might, at random, decide to show or not show the definition, keeping track of the responses in both cases.  My current hypothesis is that the average results would be the same regardless of whether the definition is shown.  If I am wrong, then what effect do differing usage contexts have on the color scheme?  An extension to this experiment might test the impact of incorrect definitions on colors.  So many things can be done with this project!

Proposed Enhancements to the Website

As you might have guessed, I am VERY intrigued by this study.  But I feel that more data post-processing can and should be done to learn discover new and interesting things about the relationships between words and colors.  Here are some suggestions that, hopefully, make their way to the developers at Cymbolism:

• Make it so that users can download the data, say, in an Excel spreadsheet or data file.  The image below shows how I would like to see the raw data (and I am almost certain it is stored in the database this way).  Having the data available would help us in the academic community better understand the relationship between colors and words (and in my case, between colors and notions of risk or degrees of confidence), and whether such a relationship is worth further study.

• Calculate the level of “conflict” in the responses. Basically, what is the entropy of the color scheme?  Maximum entropy occurs when there is complete disagreement among colors for a word.  For example, a word for which there are equal number of votes for each of the nineteen color has maximum entropy.  In contrast, zero entropy occurs when everyone agrees that a particular word is associated with a particular color.  Such a metric, as straightforward as it would be to implement, would help me understand the extent of disagreement without having to count the number of colors represented and measure the relative proportions using a soft-ruler (e.g., line segment in MS Visio).  Given the fractions p_i = N_i/N_Total for each of the 19 colors i with data inputs N_i (N_Total is total number of responses), entropy can be assessed as follows:

S = -Σp_ilog(p_i) (summation over all colors i = 1, 2, … 19)

• Calculate a word similarity measure. Envision a matrix such as is shown below, where column and row headings each are filled with all words for which there is data.  A cell in this matrix would provide a measure of similarity in “cymbolic” meaning, where, say, 100% means the two words are “cymbolically” equivalent and 0% corresponds to completely inequivalence.  Such a measure would allow me to make more defensible statements of the extent to which “risk” and “threat” are cymbolically the same.

• Enable embedding of the color scheme in personal blogs, wikis, etc. This is just something I, personally, want to have for my blog.  As you can see from my static blog page header image, I currently have to take a snapshot of the webpage, open it in paint, cut out the color bar, copy and past it into Visio, measure it out and replicate it using rectangular boxes, and then save as a .gif or .jpg so I can post it to my webpage.  That is a lot of work!  Embedding the image would have been far easier.

Send article as PDF to