A1-NicholasKong
From CS294-10 Visualization Fa08
"Good" visualization
Source: Kintisch, E. Turbulent Times for Climate Model. 22 August 2008. Science. Vol. 321, no. 5892, pp. 1032-1034.
Description:
This visualization displays the relative accuracy of an old and new climate model in predicting temperatures during El Niño. All the plots are clearly demarcated; the top plot is of the actual temperatures, the lower left of the old model's predicted temperatures, and the lower right the new model's predicted temperatures. Colors are used effectively to display the temperature, especially as they align with the common metaphors of red for hot and blue for cold. In addition, the proximity of the plots allows for easy comparison; it is clear that the new model is much more efficacious than the old, and we can also simply identify where the model fails. I am not sure why the real-world temperature display is slightly larger than the model displays, but the slight difference is not overly detrimental to the visualization.
Deconstruction:
The data can be represented in a statistical data model, where at each latitude and longitude the temperature during El Niño is measured and a quantitative difference between that temperature and the regular temperature is stored. The latitude and longitude, both quantitative-interval data types, are represented implicitly by the maps and are thus encoded using x and y position. The temperature difference, a quantitative-ratio value, has been encoded using color hues. To encourage comparison the three charts are placed in close proximity, a small-scale example of Tufte's small multiples.
Critique:
While at first I defended the choice to encode the temperature difference in hue, upon reflection it appears to be a suboptimal choice. Beyond the fact that color does not encode quantitative values accurately, although I still feel this is somewhat mitigated by the use of the standard color spectrum, the colors do not represent absolute temperatures. Rather, they represent relative temperatures. The exact difference that each color represents is not stated and the scale vaguely refers to the extremes as "warmer than normal during El Niño" and "colder than normal during El Niño". Encoding the exact difference using color intensity instead of hue would have been a better choice; while exact values would not be read as easily using intensity, it would have represented the data more precisely while still enabling easy comparison between the two models and between the models and the real differences in temperature. In addition, the ticks surrounding the border of all three charts seem superfluous and would best be removed. Despite these flaws, I feel that the visualization serves its purpose well, in that the aforementioned comparisons are painless and instructive.
"Poor" visualization
Source: Fehr, E., Bernhard, H., and Rockenbach, B. 28 August 2008. Egalitarianism in young children. Nature 454, 1079-1083.
Description:
Before describing the purpose of the visualization some introduction to the source paper is necessary. The purpose of this paper was to investigate the development of "egalitarian" behaviour in children aged 3-8. Egalitarian behavior refers to behaviour that results in equality to all participating parties. In this experiment, each child acted as a decision maker. The child would make an egalitarian (fair) choice in lieu of either a generous or selfish choice; the partner that was affected was not physically present.
In performing these three tasks, children were paired up by (a) gender, and (b) ingroup/outgroup. Ingroup pairs were pairs who attended the same playschool, kindergarten, or school. Outgroup pairs were pairs who had attended different institutions. Ingroup/outgroup was determined by showing the children group pictures of their own institution or a different institution, respectively. They were told that their partner would come from one of those groups.
In both graphs, the solid lines represent ingroup pairs and the broken lines represent outgroup pairs. The top plot displays data for boys and the bottom for girls. The percentage of egalitarian choices relative to all of the choices all the children made is plotted against the age ranges of the children in question; this is in order to show that egalitarian behaviour is much more common in the ingroup for male pairs, but the ingroup/outgroup distinction does not play a factor for the female pairs.
This visualization has a number of strikes against it:
- The x-axis for the upper plot has no label.
- Both plots share the same axes, so space could be minimized and they could easily be combined. While the authors' intent was to clearly delineate the difference between the male and female trials, colour could also be used effectively to show this.
- Perhaps most egregiously, there is no legend or title to indicate the meaning of the various marks. The reader must read the caption to discover that the solid lines represent ingroup trials, or that the upper plot shows male trials. Without the caption the visualization is impossible to make sense of.
Deconstruction:
A statistical data model underlies this chart; each point represents a single measurement upon three different conditions: male or female, ingroup or outgroup, and age. The data set is relatively small, containing nine points. What results is the mapping of four data variables onto four visual variables:
- The age-ranges can be modeled as an ordinal variable. This is displayed in x.
- The percentage of egalitarian choices is a quantitative variable and is displayed in y.
- The group represents a nominal variable that can take two values, ingroup or outgroup. This has been mapped to both texture and shape. The solid lines represent ingroup and the dashed lines represent outgroup. In addition, the data points change shape: the circles are for ingroup, the diamonds for outgroup.
- The gender, another nominal variable with two values (male or female), is mapped using y. The upper chart is for male pairings, the bottom for female pairings.
Redesign:
I sought to make my redesign more compact and readable than the original; in the process I encountered a few more issues. While constructing my redesign, I tried to recover the data from the graphic in order to start a design from scratch. I found it difficult to read the exact value of each data point since each point takes up a significant amount of area. In addition, comparisons between the two graphs were hindered because they were stacked on top of each other and therefore made necessary a mental translation operation. Putting the graphs side-by-side would have aligned their y-axes and so disposed of that translation.
I decided to retain the line-graph design because the data is somewhat time-related. The lines help the viewer to observe the trend of increasing egalitarian behavior with increasing age. I've combined the two original charts into one and mapped the gender to color: blue for male, red for female. In addition, the group variable is now only mapped to one visual variable, the intensity of color. I deemed the group-to-shape mapping unnecessary as it was already being adequately represented.
The light gray background and grid lines are based on Tufte's recommendation to eliminate stark, high-contrast backgrounds. The grid is in place to help retrieve a more accurate estimation of the values represented by the data points. It is light gray to avoid distracting the viewer from the overall trend in the chart. I've also rotated the y-axis label to read horizontally, acting on Tufte's opinion that the graph becomes "friendlier". I agree with this assessment: we should not design charts that require the viewer to do more work than necessary, and in this case the space savings over the original charts more than justify the additional horizontal space required.
The legend is laid out in such a way to suggest a separate comparison between male ingroup/outgroup and female ingroup/outgroup, as in the original chart. However, placing all the data in a single chart reveals other interesting trends. For example, boys and girls had roughly equal percentages of ingroup egalitarian choices in the 3-4 and 5-6 age ranges. This changes in the 7-8 range, where ingroup drives even more egalitarian behavior in boys but levels out in girls. This conclusion could be drawn from the original charts, but I feel with the redesign it is much easier to find.
