What sort of visuals do you find most useful?

Hey DD community,

For one of the designs I was working on with a friend, we were able to create a really nice visual where we compared sample size to power:

change_vote_ate %>% 
  get_diagnosands() %>% 
  mutate(N_per_country = N_per_country * 2) %>% 
  mutate(N = factor(N_per_country)) %>% 
  ggplot(aes(N, power, fill = estimator_label, group = estimator_label)) +
  geom_bar(position = "dodge", stat = "identity") +
  geom_hline(yintercept = 0.8)

In your designs, what do you think is useful to visualize? What variables do you end up graphing a lot?

For power curves, I tend to use geom_line. Also since you get bootstrapped se’s you could experiment with geom_ribbon.

Another plot that I think is quite nice is variance over bias - the tradeoff is usually taught theoretically but rarely visualized.


I like your plot!

One thing I think about when visualizing dd output is whether I want to be looking at the simulations or the diagnosands – or both!

I could imagine a faceted graph with facet_grid(N~estimator_label), where each facet is a geom_histogram of the estimates, possibly colored by whether the estimate is statistically significant or not. That graph helps to show how the sampling distributions tighten up with N, possibly differently by estimator!

1 Like