how to get better at embracing unknowns - scientific...

9/4/2019 How to Get Better at Embracing Unknowns - Scientific American

https://www.scientificamerican.com/article/how-to-get-better-at-embracing-unknowns/?print=true 1/15

B E H A V I O R & S O C I E T Y

How to Get Better at Embracing Unknowns

Interpreting uncertainty through data visualizations

By Jessica Hullman on September 1, 2019

Credit: Jessica Hullman and Jen Christiansen

When tracking a hurricane, forecasters often show a map depicting a “cone ofuncertainty.” It starts as a point—the hurricane's current position—and widens into aswath of territory the storm might cross in the upcoming days. The most likely path isalong the centerline of the cone, with the probability falling off toward the edges. Theproblem: many people misinterpret the cone as the size of the future storm.

Researchers have found that the misunderstanding can be prevented if forecastersinstead show a number of possible paths. Yet this approach can also introducemisunderstanding: lots of people think the probability of damage is greater whereeach path intersects land and less likely between the lines (maps).



Credit: Tiffany Farrant-Gonzalez; Sources: National Hurricane Center (cone of uncertainty);

“Visualizing Uncertain Tropical Cyclone Predictions Using Representative Samples from Ensembles

of Forecast Tracks,” by Le Liu et al., in IEEE Transactions on Visualization and Computer Graphics, Vol.

25; August 20, 2018 (multiple storm paths)

Uncertainty pervades the data that scientists and all kinds of organizations use toinform decisions. Visual depictions of information can help clarify the uncertainty—or compound confusion. Ideally, visualizations help us make judgments, analyticallyand emotionally, about the probability of different outcomes. Abundant evidence onhuman reasoning suggests, however, that when people are asked to make judgmentsinvolving probability, they often discount uncertainty. As society increasingly relieson data, graphics designers are grappling with how best to show uncertainty clearly.



What follows is a gallery of visualization techniques for displaying uncertainty,organized roughly from less effective to more effective. Seeing how differentapproaches are chosen and implemented can help us become more savvy consumersof data and the uncertainty involved.

The least effective way to present uncertainty is to not show it at all. Sometimes datadesigners try to compensate for a lack of specified uncertainty by choosing atechnique that implies a level of imprecision but does not quantify it. For example, adesigner might map data to a visual variable that is hard for people to define, such asa circle floating in space rather than a dot on a graph that has x and y axes. Thisapproach makes the reader’s interpretation more error-prone. Alter natively adesigner might use a program that creates a hand-drawn or “sketchy” feel. Bothapproaches are risky.

A D V E R T I S E M E N T

N O Q U A N T I F I C A T I O N




Intervals may be the most common representations of quantified uncertainty. Errorbars and confidence envelopes are widely recognized, but even though they seemexact and straightforward, they are notoriously hard to interpret properly. Researchshows they are often misunderstood, even by scientists.

I N T E R V A L S




Designers can map uncertainty directly to a visual property of the visualization. Forexample, a gradient plot can shift from dark color (high probability) at the center tolighter color (low probability) at the edges. In a violin plot, wider points mean greaterprobability. Mapping probability density to a visual variable displays uncertainty ingreater detail than interval methods (error bars and confidence envelopes), but itseffectiveness depends on how well readers can perceive differences in shading, heightor other visual properties.

P R O B A B I L I T Y D E N S I T Y M A P S




A R R A Y S O F I C O N S



Reframing a probability such as 30 percent as a frequency—three out of 10—canmake it easier for people to understand uncertainty and consequently use suchinformation appropriately. People may better understand discrete probabilitiesbecause they run into them in everyday experiences.





Plotting of multiple samples in space can be used to show probability in a discreteformat for one or more variable quantities. One example of this approach is a quantiledot plot. It shows a number of distinct cases from the quantiles of the datadistribution, so that the number of dots (such as two dots high or five dots high, inthe example below) conveys probability. When there is uncertainty about parameter

M U L T I P L E S A M P L E S I N S P A C E



values from which estimates are drawn, such as initial conditions, samples can begenerated that vary these parameters and can be shown in a single visualization.


Plotting multiple possible outcomes as frames in an animation makes uncertaintyvisceral and much harder to ignore. This technique, called hypothetical outcomeplots, can be used for simple and complex visualizations. Perceptual studies indicatethat people are surprisingly adept at inferring the distribution of data from thefrequency of occurrences: we do not necessarily need to count the number of times anevent occurs to estimate its probability. One important factor is the speed of events,which must be fast enough so that people can see a sufficient number of samples yetslow enough for them to consciously register what they saw.

M U L T I P L E S A M P L E S I N T I M E



Credit: Tiffany Farrant-Gonzalez (election needle), Jessica Hullman and Jen Christiansen (graph)

Designers can create effective uncertainty visualizations by combining differenttechniques rather than choosing a standard chart “type.” One example is a fan chart,made famous by the Bank of England. It depicts data up to the present, thenprojections into the future; uncertainty about the past is an important component inassessing uncertainty about the future. The fan chart presents probability from higherchance to lesser chance in multiple bands that represent different levels ofconfidence, which the reader can choose from. Readers can perceive the informationthrough the position of the edges of the bands, as well as lightness versus darkness.Some modern software packages for statistical graphics and modeling make it easy tocombine uncertainty visualization approaches.

Sign up for Scientific American’s free newsletters.

H Y B R I D A P P R O A C H E S

Sign Up



Credit: Jen Christiansen Source: Inflation Report, Bank of England, February 2010



A B O U T T H E A U T H O R ( S )

Jessica Hullman

Jessica Hullman is a professor of computer science and journalism at

Northwestern University. She and her research group develop and evaluate

data-visualization and data-interaction techniques to enhance reasoning about

uncertainty.

Credit: Nick Higgins

M O R E T O E X P L O R E




Scientific American is part of Springer Nature, which owns or has commercial relations with thousands of scientific publications

(many of them can be found at www.springernature.com/us). Scientific American maintains a strict policy of editorial

independence in reporting developments in science to our readers.

© 2019 SCIENTIFIC AMERICAN, A DIVISION OF SPRINGER NATURE AMERICA, INC.

ALL RIGHTS RESERVED.

Picturing the Uncertain World: How to Understand, Communicate, andControl Uncertainty through Graphical Display. Harold Wainer. PrincetonUniversity Press, 2009.

Visualizing Uncertainty. Claus O. Wilke in Fundamentals of Data Visualization.O'Reilly Media, 2019.

Uncertainty + Visualization, Explained. Blog series by Jessica Hullman andMatthew Kay. https://medium.com/multiple-views-visualization-research-explained

F R O M O U R A R C H I V E S

Saving Big Data from Itself. Alex “Sandy” Pentland; August 2014.

how to get better at embracing unknowns - scientific...

Documents