I think a fundamental point is somehow missing, though -- chose the type of graph as a function of the nature of the data you have. You are right in saying that this is in part covered by Tufte and the like, but this bears repeating anytime we can. The most well-executed plot means nothing if a scatterplot is used in place of a boxplot or anything else. I wonder how feasible a flowchart would be, that link different properties of the dataset to the recomended type of visualization?
On an unrelated note, kudos on putting the code for the figures online.
