From CS294-10 Visualization Fa07
 Good Visualization
This is a 2D histogram of cloud top pressure and cloud optical thickness. These are commonly used in atmospheric sciences and are known as "pc-tau" graphs. This particular plot compares cloud coverage predicted by 6 different models over the same region of the southern great plains (SGP). For example, high values in bottom left hand corner designate low thin clounds while red in upper right indicates high dense clouds. It is an effective way to view differences in models at a glance. Also, the vertical pressure axis corresponds to our notion of high/low. And I like the colors cause I picked them.
Source: own work for NASA/Goddard Institute for Space Studies; using IDL
The data model for each of the six sub-plots is some version of a latitude/longitude grid holding tau and cloud top pressure information. These monthly total arrays are limited to a location over central Oklahoma and are processed out of daily large binary files holding much more data. The data is sorted into a two dimensional grid of tau and ctopp values, which is quantitative-interval (not continuous). The resulting quantitative-ratio variable of "% of total points" is displayed on a continuous 0-10% scale.
The image model uses small multiples for intercomparison of the six different climate models. The color scheme is picked to highlight high and low areas of clouds. The main point of each graph is to show "what kind" of clouds were present for March 2000 over this geographic area. The eye distinguishes the bright patterns and intercomparison is inevitable. At the same time, each of the plots is interesting on its own as it gives insight into what kinds of clouds a given model predicted.
 Bad (but entertaining) Visualization
This is a zoom in on a much much larger timeline from book "The Wall Chart of World History". This is the 1800's to early 1900's part for Europe. Lots of fun and a good thing to look at while drunk or bored, but shall we say not a consistent historic record. Events to highlight, such as "visits london" for Nicholas of Russia were chosen somewhat inconsistently. Note 'alaska' branch below Alexander - how does it relate to rulers and what happens to it after 1881? But good fun.
The overarching data model must be a table linking monarchs and start, end dates of reign, number of children, historical curiosities. However an accompanying data set was a random collection of facts about the period itself, as well as a spotty tables of territory ownership dates, famous people birth/death dates, and a few others. Being time, the data is quantitative-interval.
The image model randomly (?) selects colors for each sequential time period. Inevitably, unconnected pieces of the chart appear to be connected. Tree like branches are used to represent both country rulers, famous peoples, as well as territory ownership. The width of the branches probably correlates to size or importance of a country, though perhaps only in the author's eyes. To overcome the confusion created by branches representing more than one thing, and use of color, the image relies on a lot of annotations.
This design aimed to clean up they display by separating the graph into two panels - top for events/people and bottom for countries/rulers. One of the most useful aspects of the original graph was the relative 'sizes' of the branches representing the countries. I tried to represent this 'importance' factor on the vertical scale (we can pretend it's some combination of GDP and area). Whenever a territory is split off from a country, it is represented by a dashed line of the corresponding color. When country switch area/GDP importance, they are taken up or down by a solid line of the corresponding color. Colors highlight contries' existence in time, though information about the rulers is written in black.
Though the idea of the new layout is not a bad one, it is tough to execute and there are many subjective editorial decisions to make. Of course, the chart looses much of its entertainment appeal too.