One of the five core guidelines in my data visualization work is what I call “Integrating the Graphics and Text.” In Better Data Visualizations, I quote Amanda Cox—recently on the podcast—who once said, “The annotation layer is the most important thing we do…otherwise it’s a case of ‘here it is, you go figure it out.’”
So I agree: the second option of your example is crisp and complete, the third misses necessary redundancy. In the Washington Post example the bars are close to the colored text in the Legend, plus the bars already having intrinsic meaning through the negative and positive signs.
Another relative argument against colored text in the legend: lighter colors don't really work well with letters, so then you additionally have to darken the tone. I also do that for colored labels inside my graphs that match to light lines or symbols, but it's extra hassle.
Oops, hit the return button.
So I agree: the second option of your example is crisp and complete, the third misses necessary redundancy. In the Washington Post example the bars are close to the colored text in the Legend, plus the bars already having intrinsic meaning through the negative and positive signs.
Another relative argument against colored text in the legend: lighter colors don't really work well with letters, so then you additionally have to darken the tone. I also do that for colored labels inside my graphs that match to light lines or symbols, but it's extra hassle.