Plot improvement: Strickler and Lawson 2020

The plot below is from Strickler and Lawson 2020 "Racial conservatism, self-monitoring, and perceptions of police violence":

I thought that the plot might be improved:

---

Key differences between the plots:

1. The original plot has a legend, which requires readers to match colors in a legend to colors of estimates. The revised plot labels the estimates without using a legend.

2. The original plot reports treatment effects on a relative scale. The revised plot reports estimates on an absolute scale, so that readers can directly see the mean percentages that rated the shooting justified, for each group in each condition.

3. The revised plot uses 83% confidence intervals, so that readers can use non-overlaps in the confidence intervals to get a sense of whether the p-value is p<0.05 for a given comparison.

4. The revised plot reverses the axes and stacks the plots vertically, so that, for instance, it's easier to perceive that the percentage of nonWhite respondents in the control that rated the shooting as justified is lower than the percentage of White respondents in the control that rated the shooting as justified, at about p=0.05.

---

The plot below repeats the plot above (left) and adds the same plot but with x-axes for each panel (right):

---

NOTES

1. Thanks to Ryan Strickler for sending me data and code for the article.

2. Code for the paired plot. Data for the plots.

3. Prior discussion of Strickler and Lawson 2020.

4. Other plot improvement posts.