Thread Closed 
Week 4: Ebola Virus Rapidly Spreading in West Africa
09-12-2014, 03:26 PM (This post was last modified: 09-16-2014 10:05 PM by jdawson.)
Post: #1
Week 4: Ebola Virus Rapidly Spreading in West Africa
   

Source Visualization and Article: http://www.usatoday.com/story/news/natio...2qa2sm6yw0

Source Image Only: http://i.usatoday.net/_common/_notches/3...0905_1.png

This visualization depicts a data set that has one categorical attribute for the country and two quantitative attributes for the number of cases and the number of deaths due to the Ebola Virus outbreak in Africa. I imagine that there would need to be an (x, y) coordinate, lat/long, or some other form of spatial coordinate per data item in order to position the marks on the map.

The cases and deaths attributes are encoded with an area mark utilizing the area magnitude channel and the spatial identity channel is used to layout the areas according to the geography of Africa.

The use of color in this visualization works. I verified the image in the Coblis - Colorblindness Simulator and there were no identified issues. The blue fill highlighting the countries with data provides a good pop-out effect from the rest of the map; however, the hatch pattern fill that represent the water off the coast of Africa is confusingly extended into those same countries.

Overall this is not an effective visualization. As we know, the effectiveness of area channels is not very good and by overlaying the deaths over the cases we allow for a relative difference illusion. If you look at the 494 circle area in Guinea and compare it to the 476 circle in Sierra Leone, it appears considerably larger even though it is technically only one pixel larger in diameter. The difference in size of the surrounding pink circles creates the illusion. Overall using area makes it very hard to compare/contrast the data and you end up ignoring the area marks and focus on the actual data values instead. It took me a while to notice that there is a 2x2 pixel dot representing the one case that Senegal had.

If I were recreating this visualization and wanted to truly relay the data to the scientific community I would title it "Ebola Outbreak Cases and Deaths as of August 31, 2014." I would have a grouped bar chart with value on the y-axis and deaths and cases by country on the x-axis. I would change the title because the visualization does not actually show us how the Ebola virus is rapidly spreading. It is just the data as of August 31, 2014. In order for the visualization to show how the virus is rapidly spreading it would need to have a temporal attribute as well. Then we could show either an animation over a fixed time scale or a series of small multiples to actually visualize the rapidly spreading virus.

If the geography of the spreading virus is truly important and the time and location (lat/long) of every case/death is in the data set you could create a very good animation to show how the virus is rapidly spreading.

As an idea see: http://healthmap.org/ebola/
Find all posts by this user
09-15-2014, 03:59 PM
Post: #2
RE: Week 4: Ebola Virus Rapidly Spreading in West Africa
The use of stacked area encoding creates the illusion you mentioned with circle sizes. Scaled pie charts probably would have been a better choice for those circles. That way the angle and size could work together.

I also agree that a temporal aspect was implied in the title but not delivered in the graphic.
Find all posts by this user
09-17-2014, 09:55 PM
Post: #3
RE: Week 4: Ebola Virus Rapidly Spreading in West Africa
I agree with the breakdown of attributes and channels used. One additional channel that I would argue this visualization makes use of is the color saturation magnitude channel to differentiate between cases and deaths.

I very much agree regarding the crosshatching within countries and don't understand what this corresponds to.

Again, I think this critique is spot on regarding the use of the area magnitude channel. The illusions that you mention regarding relative size make it quite difficult to compare data between countries. I think this acts as a good example of two channels (area and color saturation) that interfere with one another and hinder the intended perception of the data.

Given that we are looking at essentially 10 data points, the data density is incredibly low. However, I feel this decision was made to make for a more interesting and memorable visualization rather than generating a Tufte-optimized visualization.

One of the largest improvements that I feel could be made is to better correlate the quantitative attributes (Cases and Deaths) with their respective categorical attribute (Country). I felt that this was not intuitive and admit to opening a map to determine where Liberia was.

Expanding on the suggestion to use a grouped bar chart, the color hue identity channel could be used to visually encode each country. I feel this would offer sufficient discriminability as there are only 5 bins that need to be represented. In addition to using the vertical spatial position channel for the quantitative attributes, I feel that the color saturation magnitude channel could drastically help improve ones ability to locate the country that the data corresponds to.
Find all posts by this user
09-18-2014, 12:53 AM
Post: #4
RE: Week 4: Ebola Virus Rapidly Spreading in West Africa
I also feel that color saturation channel has been used to differentiate between cases and deaths.

Like others, I also do not understand why the crosshatching background is there on the countries affected. My guess is it's a coding glitch, the designer/developer mistakenly used the blue color with some alpha.

I feel that here the use of area magnitude channel is right. The reason is that it tries to bring forth the ratio of death vs cases for each country. And I would argue that any visual representation of quantity for multiple instances without a fixed reference becomes difficult. For example, here the circles are scattered over a 2d map instead of a fixed axis.

And I observe another subtlety here is how they have explicitly written the text for '# deaths' and '# cases' for smaller bubbles for visual symmetry that comforts our eyes.

I believe showing the small map of continent Africa as an overlay and showing the area under focus inside a blue rectangle is relevant in the context of erratic Ebola spread.
Find all posts by this user
09-19-2014, 09:31 AM
Post: #5
RE: Week 4: Ebola Virus Rapidly Spreading in West Africa
I agree with you that spatial coordinate per data item is needed in order to position the marks on the map. The most confusing thing about this visualization is the location of the circles: Those are not reflecting the actual location of the countries on the map. If it's for identifying locations of outbreaks, what's the point of using the map?
Find all posts by this user
09-19-2014, 07:22 PM (This post was last modified: 09-19-2014 07:24 PM by mcarter.)
Post: #6
RE: Week 4: Ebola Virus Rapidly Spreading in West Africa
So I slightly disagree with the general consensus. I think it is important to keep in mind that this graphic is in a daily newspaper with general readership. I think the main point of the graphic is to engage readers, and give a geographic context to the data being presented (where in Africa, roughly, is the outbreak?). I would argue that this graphic is NOT intended to convey the progress of the Ebola outbreak over time (I admit, the graphic is poorly titled), nor is it to show the relation between number of cases and the geographic spatial dimension (if this is what was trying to be conveyed, it would need to be more granular than by country).

As has been mentioned, there are only 10 data points. This small amount of data could easily be included in a data table, but this is kind of boring and disengaging. Instead, it is delivered semi-graphically, and results in an engaging way for me to learn which countries are involved, and how hard they've been hit (in number of cases).

I would also argue that a visual area channel is NOT the main mode for encoding the magnitude. Sure, it does attempt to do this (and as has been noted, does so poorly...). However, I think the numeric value labels are the main "encoding" for the magnitudes. When I looked at this graphic, I only used the area channel to rank the countries: Liberia has more than Sierra Leone which has more than Guinea, etc. I feel the area channel does a reasonable job at this, for such a small data set.

With regards to the complaints about positioning of the area marks, again, for a small data set, I don't see the problem. I assumed that the relative positioning of the marks on the left side reflects the relative positioning of the countries (Liberia is on the bottom, Sierra Leone is directly above/to the left of Liberia, etc).

In general, I think that consideration of audience and the message to be conveyed is just as important in visualizations as it is with writing - to convey a simple idea to a general audience, engagement is as much a goal as precision. This graphic tells me where in the world the Ebola outbreak is occurring, and which countries are most impacted.
Find all posts by this user
09-21-2014, 10:46 AM
Post: #7
RE: Week 4: Ebola Virus Rapidly Spreading in West Africa
I believe this is not the best visualization and it does not show why ebola virus is rapidly spreading in West Africa.

I agree that overlaying deaths over the cases allows a relative difference illusion. It makes it difficult to compare countries. A bar chart showing the cases and deaths with respect to countries would have been a better visualization.
Find all posts by this user
09-21-2014, 08:11 PM
Post: #8
RE: Week 4: Ebola Virus Rapidly Spreading in West Africa
Verifying the colors used for the visualization via Coblis - Colorblindness Simulator is an excellent use of a tool available to help ensure the maximum potential of a visualization. To a degree, I do agree that the difference between Africa & the Atlantic's hues with the crosshatch pattern does not help to differentiate between both. It appears the author attempted to ensure that the affected countries represented would be the focal point of the graphic, though the homogenousness of everything else (non-affected countries & Atlantic) did require a double take. This might not have been the best balance to ensure the pop-out affect.

I agree that the the overall effectiveness of the visualization is not effective. The area of the cases & deaths per country do not seem to correlate to anything size-wise, though there is a vast difference in area per country. At first the actual location of the cases & deaths circles for Sierre Leone & Liberia is a bit confusing to immediately realize that the cases & deaths circles are actually offset from their respective countries due to the area that both sets of circles cover. Given the data present, especially with Senegal's deaths being 0 & only 1 case Ebola represented, the choice of the using the area circles for deaths & cases reported seems silly for a total of five countries reported. In this, I also concur that the use of area marks is not effective in comparing nor contrasting the relevant countries.

The suggestion for having a temporal component is also a good one. This snapshot of cases & deaths does not include a mechanism to ascertain the actual spread of Ebola. Though, following the link you included, it does look like that USA Today is updating the visualization (see attachment) even though checking the web site to see what cases & deaths have changed is not very effective in seeing the overall growth in Ebola for these pasts of Africa.


Attached File(s) Image(s)
   
Find all posts by this user
09-21-2014, 08:14 PM
Post: #9
RE: Week 4: Ebola Virus Rapidly Spreading in West Africa
I did not notice the size illusion difference between the inner circles when I first glanced at the visualization. This factor alone to me makes it a bad visualization. Avoiding combinations that cause illusions seems to be an important one. Also I agree with your idea about using a bar chart if presenting this to the scientific community. This visualizations intended audience seems to be be the general public. Even so due to the above criticisms it is ok at best. I does successfully do what I think its most important job is. That is to associate the data with the different land masses. Due to the illusion though it seems to be bad for comparing.
Find all posts by this user
09-21-2014, 09:12 PM
Post: #10
RE: Week 4: Ebola Virus Rapidly Spreading in West Africa
I feel that this visualization is misleading with respect to the title it has.

Even though i agree with mcarter that it may be designed for general reader to realize the impact and fast spreading nature of ebola virus. But then the title of the article should be different as jdawson mentioned.

Use of area channel for number of deaths and number of cases isnt a very wise choice. First of all the senegal case is very hard to notice and can be easily missed. Secondly all the cases should have been represented on the same scale. This way the ambiguity due to relative size could have been avoided.

The best part of this viz is its color scheme and works well for the overall visualization.
Find all posts by this user
Thread Closed 


Forum Jump:


User(s) browsing this thread: 1 Guest(s)