This document summarizes graphics best practices for data visualization. It discusses establishing hierarchy, minimizing non-data ink known as "chartjunk", finding the smallest effective difference between data groups, and accounting for multivariate data. Specific chart types like line charts, bar charts, and pie charts are examined as well as displaying numbers, scales, and error bars. Guidelines are provided for choosing colors, formatting, and addressing color blindness.
1 of 58
Download to read offline
More Related Content
Graphics Best Practices
1. Lunch and Learn
Graphics Best Practices
Anna Jursik, Lester Shen, and Jenny Edwards, March 13, 2013
41. Use legends if space is tight
Make sure legend order reads in the same order as data, and use
color to assist
Page 41
42. General Color and Format Rules:
Dont overuse: maintain smallest effective difference
Use color and formatting to group similar types, to
show scale or trends
Reserve different formatting dimensions to highlight
what you care about
Page 42
#3: graphics are a communication tool, not ornamental organize and communicate key informationIf you use them effectively, youll tell the story you want to tell to your audience If you use them ineffectively, you could distract actually distract from the data
#4: Cant assume they have the same level of knowledge, interest, or commitment as youGraphics should make it easier for the audience to understand and take interest in the data or information
#5: eyes are drawn to big and bold and brightnon-linear: bounce around, can't assumethe viewer will look through everything in orderProvide handholds but do include all the data for curious/nit picky audience members
#6: stick w/established symbols and conventionsrepetition: look for patterns we've learned maps: symbols consistent; capitol cities star, cities bold, lakes and rivers blue, interstates red
#9: Huge amounts of data, limited amount of spacePaper is two-dimensional, so
#10: after this, therefore because of thisCorrelation not causationFalse causationCoincidental causation
#12: Mean arithmetic mean, standard average the sum of all measurements divided by the number of observations in the datasetMedian middle value, half the population is above and half is belowMode most likely value
#32: More acceptable even with a lot of categories. Because color is used well, labeled. They start with the largest value and go around clockwise. But the major take away message will be that TV is big.
#34: Too hard to interpret, label all the values. (Radial as treemap)
#36: They make your reader do extra work, and they cant focus on the trends when eyes dart back and forth
#38: Dont use legend if you are plotting the same values
#41: Here, there are multiple labels, so a simple legend is good.
#45: Many Errors here:Dont use 3-dUse ascending colorsDont need graphicDont need decimal