- Read the documentation for your data
- Read the description of the variables you might use - what they are and what scale they're measured in
- Figure out an interesting insight that you can convey with your data (the "story")
- Is it accurate? Does the data actually support that insight?
- Is it actionable, deeply interesting, or does it help you understand how the world works? Ask "if I learned this insight, would I care?" If the answer is no, try something else. (Tip: "Group X has the highest average value of variable Y" is a snooze of a story for most X and Y. If you're going to do that one, you're gonna have to work to convince the reader to care)
- In order to understand your insight, figure out what kinds of continuity should the reader see (for "the poverty rate has fallen over time" you should be able to see continuous changes in time), and what kinds of contrast they should see (for "the poverty rate has fallen more in country X than country Y" you need it to be easy to contrast changes in X against changes in Y)
- Determine what aesthetics you will use (x, y, color, line type, size, etc.) with each relevant variable. Make sure they are appropriate for making the continuity/contrasts you want
- Determine the geometry (line, bar, point, area, lollipop, etc.) that will best get your story across
- Specific tips: line graphs require an X axis with a natural order. Scatter plots are hard to read with too many points, and need continuous X and Y axes. Most geometries have a hard time graphing lots and lots of separate categories.
- Draw a quick sketch by hand and make sure you like it
- Make an initial rough graph
- Select aesthetic elements - color, shapes, etc. - that are easy to read and are colorblind-friendly. Make sure the font size is large enough to read.
- Label axes, legends, and values using common-language terms, not the variable name in your statistics package (and definitely get rid of any scientific notation)
- Make sure the font size is big enough to read!
- Move information where it's easy to see - labels should go near the data. Legends can often be absorbed into the graph, and so on.
- Add titles, highlighting, and annotation that helps draw the reader's focus to your story
- Remove clutter unrelated to your story
- Ask yourself or a friend "if I didn't know the story I was trying to tell, would I naturally figure it out by looking at the graph?" and if not, make it more clear.