1. Perceived area of a circle = (actual area)^X, where X = .8 ± .3
2. Don't use more dimensions than variables (e.g. area to represent a linear change). Preferably, show each variable in a different dimension (or other feature, if >3)
3. Don't use the same dimension to represent different things (or different scales or units) in the same graphic
1. includes using multiscalar color - but color as one variable and darkness as another should be OK
4. Use adjusted standardized measures - e.g. show budget in inflation-adjusted spending per capita, not just nominal cost
5. Show context - both before/after of the same measure, and other similar measures for comparison
6. Don't add false 3d; it drastically distorts the perception of change (e.g. a x=y bar graph w/ 3d vanishing point towards center looks like slow-stable-FAST growth)
7. Graphics are best used to show large amounts of data densely
8. Erase whatever doesn't add information
9. Graphs can be revised to be denser & clearer (quasi-minimalist in that function>>form):
1. Quartile plot instead of box-and-whisker plot (turned horizontal and ASCIIfied):
   

   -----  *    ---

   Lower line is from minimum to 25%ile; * = median; white space in center = 25%-75% (the "box"); etc.
1. his "preferred version" with an offset middle 50%ile and gap median is horrible though IMO
2. ... but yes the box-and-whisker plot is 80% clutter - especially if it's combined with a shaded bar graph (which I've seen in numerous research papers... ugh what a bad design; you can't even see the lower tick because it's obscured by the shading)
1. don't shade using hashing or Moiré patterns, or better, don't shade at all
3. another alternate:  ---==*======--
2. don't show top & right borders; they're superfluous
3. cut off border ruler lines at the min & max rather than at origin & rounded-max
1. even better: replace the simple border line with a quartile plot of the marginal probability, leaving just the tick marks and scale numbers (or here with his "preferred" style quartile plot & frequency graphs)
2. improvement: replace cut-off parts with light gray lines, so the visual reference is still there
4. for low-density graphs use regular spacing but make the numbering individual for each datum (e.g. instead of 0 5 10 15 show 3.2  9.3 etc [if those are the data])
5. go very very easy on graph lines, and if using any at all make sure they're very thin and gray
1. or use lines to add data - e.g. on a graph of years, show events with graph lines
2. for a bar graph, use "white space" graph lines and eliminate ticks, frame, and graph lines
6. don't use alphabetic ordering for labels that appear on the chart if some other order (e.g. date) would add useful info

I like all of these (except his offset quartile graph - it's very hard to read). I think though that the whole book can be summarized like I did above with perhaps a halfdozen annotated examples to illustrate before-and-after.