A1-AmandaAlvarez
From CS294-10 Visualization Fa07
Contents |
[edit] Good Visualization
From Kandell, Schwartz, Jessell: Principles of Neural Science, Fourth Edition (2000), p. 541.
[edit] Explanation
This image shows an important construct in neuroscience: the cortical hypercolumn. The whole brain on the left provides the context for the schematic on the right: we are looking at a volume of brain extending from the surface inwards. Some of the elementary ideas that are taught about the visual system are that the brain segregates information based on which eye it is coming from, or what orientation the pattern has, or whether the stimulus is color or motion, etc. This picture shows that the brain has modules that process all these things together in a little volume, while still keeping different 'streams' apart. These streams of different features are orientation columns or pinwheels, blobs, and ocular dominance columns (input from the C[ontralateral] or I[psilateral] eye).
Someone unfamiliar with the subject can recognize that we are dealing with connections from the LGN (shown on both the brain on the left, and on the bottom right) to somewhere else in the brain. Further, we see that this bit of brain has layers (numbered 1-6), and that different parts of the LGN project to different layers in the cortex. These lines are of two different colors, and go to areas of corresponding lightness in the cortex (labeled C and I). There are also dark blob-shaped areas, which are in fact called blobs. Lastly, the wavy white lines on the surface represent areas that are sensitive to different spatial orientations; they are organized in little pinwheels. Once someone has been told that the brain segregates different elements of the visual world, it is helpful to see how it all comes together in the form of a hypercolumn. This image could have been improved by an addition showing how an array of these hypercolumns covers the entire visual field (a fact that is mentioned in the caption of the image). On the whole I think the use of labels, subtle colors, and context in this image does a good job of conveying the complex ideas of three-dimensionality, modularity, and parallel processing of the brain.
[edit] Deconstruction
The underlying (implied) data set is very large and was probably gathered over decades. It contains positional neuroanatomical information (quantitative, xyz), connection patterns (ordinal, which area is upstream or downstream), functional information (nominal, for ocular dominance; quantitative for orientation columns; ordinal for blobs). There are labels (nominal strings). The image model is largely positional (spatial). There are actually three sub-diagrams in the image, only one of which has three axes (the central column); the other two are 2-2.5 D (ordinal for the bottom LGN figure, interval for the brain; not quantitative as they are not to scale and only act as accessories for the larger central figure). In neuroscience nothing is absolute, so the data model is best thought of as an aggregation of statistics along with a generous amount of simplification. The conceptual model for this image is a feedforward network.
The three axes x,y,z are encoding length (quantitative), though the image is not to scale. These axes do not apply to the bottom of the graph (layers of LGN). There are both dark (layers 2, 3, 5) and lighter (layers 1, 4, 6) lines coming out of the LGN at the bottom, the dark lines going to the 'I' column, and the light to the 'C' column. Value encodes eye of origin (nominal; this would be more useful had the corresponding layers in the LGN also been colored). We also have shape, which here acts to provide an outline, a boundary context (eg. the shape of the volume of brain is a cube). Size is also used, to a somewhat misleading effect, as is position. It is true that the piece of brain popped out on the top left is much larger that it would be in actuality (about 1 cubic millimeter), and that the location it pops out of is not anatomically correct.
Shading on the top surface of the cube (bounded by black lines) indicates ocular dominance columns (which eye the input came from, contralateral or ipsilateral). This same shading and lines continue down the depth of the cube. Dark blobs in layers 2/3 and 5/6 indicate the presence of color processing areas (blobs). Oriented lines on the surface of the cortex indicate orientation columns (neurons selective for a certain orientation). These columns form pinwheel patterns at the surface. It is interesting to note that (apparently) orientation cannot encode a quantitative variable, although the data underlying the pinwheel pattern is undeniably quantitative (0-360 degrees).
There is a lot of redundant encoding in the image. All the features presented are cyclical and repeated across the brain. Ocular dominace is actually not just endoded by value (color), but also by position, and it also has an attached label. Orientation is encoded by line orientation and position.
[edit] Bad Visualization
Untitled Hand drawn Diagram, from a Memorandum to the President, May 8, 1970. (from: Henry A. Kissinger, subject: Seabeds Convention to Determine Continental Shelf Boundary and Seaward Regime). pdf of the memo
[edit] Explanation
This is the only diagram accompanying this official memorandum to the President. It is meant to represent the proposed area of national sovereignty on the continental shelf, as well as the status of the continental margin beyond this area. The first flaw in this diagram is that, although some values are displayed, it lacks axes. The x-axis should be labeled 'Distance from land', and the y-axis should be 'Depth'. Another fact that complicates the labeling of the axes is the use of mixed units; depth values are shown in meters, whereas distance from the shore is given in miles. This leads to a third point, which is that the drawing is not to scale. A location near the water's surface is labeled 200 meters, and the bottom of the picture is labeled 4000 meters (at a spot where one might expect the depth to be 2500 meters max). There is also mislabeling of the chart: there are two areas marked 'continental margin', one of which should probably be 'continental slope'. The entire diagram is also untitled (an appropriate title might be Diagram of the Continental Shelf).
Besides all these errors, this diagram fails to serve its purpose, which is to illustrate the seabed topology to allow the president to decide on the US position on the seabed convention. The memo discusses the positions of four government branches on the boundaries between national (which could be called 'ours') and international (everyone's) sovereignty. This diagram would have greatly benefited from a simple line showing where each proposal thinks the ours/everyone's boundary should go. It would also have been possible to add more features, eg. showing where strategic resources exist in this scheme. The use of color might also have been beneficial.
[edit] Deconstruction
The underlying data consists of ocean depths and distance from land (all 2D floats, quantitative). The axes are not shown. There are instead a number of labels/annotations (nominal strings). These labels are actually tied to quantitative ranges of distances and depths; unfortunately the values are not shown. The image model is positional. Texture is used to indicate an important area (nominal). This is bivariate data that also serves as an illustrative map of geography.
[edit] Redesign
This redesign of the continental shelf diagram was done in Illustrator. It strips away a lot of the misleading labels, and streamlines the image. Axes with labels and values have been added; the image is now to scale, with consistent units being used. Unnecessary labels that are never referred to in the memo have been removed (eg. continental terrace). The mislabeling of continental margin has been corrected (to continental slope). Labels that are self-explanatory (land, ocean) have been removed. The four key areas that were previously identified with labels and texture (hatching) are now identified by color, and there is a legend. The legend also briefly summarizes which areas the four different proposals agree on (green and red), and which areas are under contention (yellow and orange). The depth of 200 meters is highlighted with a dashed line, as this is referred to often in the memorandum.
I have created another small diagram. It displays the four different proposals (from the Departments of State, Defense, and Interior, and from Undersecretary Richardson), and the positions they advocate along the continuum from full national control to full international control over the disputed areas. The two bars convey the same information, but one uses color and the other grayscale. The positions of the labels show the proportion of national/international control each proposal recommends. Each label corresponds to one color: this is the color that proposal would like to paint the disputed (yellow+orange) areas. Defense, for instance, would like to change the yellow+orange areas red, so that the area of international control would be much larger. The colored bar might be confusing, as it (intentionally) uses the same colors as the plot; in that case one can refer to the grayscale bar.
[edit] Original Bad Visualization
Oxford Rowing Bumps Chart, from Anu Dudhia. This particular example is here. These originally appeared annually in the London Times, but the archive as a whole only exists online.
This chart shows the progress of three different women's rowing crews (shown in red on the far right) in Bumps races from 1980 to 2006. Time runs along the top, and the crews are listed on the sides. This graph really provides none of the context that would be necessary for its interpretation. Particularly interesting events, eg. where a number of lines cross at once, could be annotated (perhaps a spectacular crash of boats occurred at that point, for instance). There is no explanation for the discontinuity of some lines from one year to the next, especially in the lower half of the graph. The presence of dots (eg. in the column of 1993) is a mystery, unless you know that the river was swollen that year and competition was suspended. In general, following the progression of any particular crew is very hard, especially if you try to read all the way from left to right. The fact that many lines on the left are unlabelled is unhelpful; the information about which crew each line is is contained in the graph, but is not explicit (you would have to trace backwards from the right). The graph is also not helped by the fact that the underlying data are very sproradic; some crews did not participate at all in some years, or river conditions prevented competition. The only redeeming feature of this graph is the highlighting of a few lines in red. As is the case for any time series, only very general trends can be observed, but interesting details are completely obscured.
This image is very similar to the train schedule praised by Tufte (p. 31 in the Visual Display of Quantitative Information). Both the bumps chart and the train schedule tell a story about a large collection of numbers; both, however, could benefit from having an alternative presentation of the data (a table perhaps, or just a different graphical representation) with greater detail.
Someone who feels that bumps charts are "the perfect alignment of form and function" has posted their bumps chart analyses here and here.
Tufte seems to be of the opinion that these kinds of charts are excellent examples of sparklines. (Also p. 111 of Envisioning Information)
