Control color and alpha conditionally at the same time

Question

I'm trying to tinker with visualizing more detail about the precision of results based on some color-based variation, but running into trouble when I try to simultaneously create conditions based on both alpha and color at the same time.

How would I do something like:

-If the pvalue is >.10, all colors are gray with alpha =0.5

-If the pvalue is between .05 and .10, the colors remain as specified below in scale_color_manual(), but have alpha =0.5

-If the pvalue is <.05, the colors remain as specified below in scale_color_manual() with "full" alpha

Some simulated data and code:

set.seed(123)
terms <- c("covariate1", "covariate2", "covariate3", "covariate4")
outcomes <- c("Outcome1", "Outcome2", "Outcome3")
treatments <- c("Treatment A", "Treatment B")
results <- expand.grid(term = terms, outcome = outcomes, treatment = treatments) %>%
  mutate(
    term = factor(term, levels = terms),
    estimate = rnorm(n(), 0, 0.2),   
    se = runif(n(), 0.05, 0.2),     
    conf.low = estimate - 1.96 * se, 
    conf.high = estimate + 1.96 * se
  )

ggplot(results, aes(x = estimate, y = term, color = outcome)) +
  geom_vline(xintercept = 0, color = "black", linetype = "dashed") +
  geom_point(size = 2, position = position_dodge(width = 0.2)) +
  geom_errorbarh(aes(xmin = conf.low, xmax = conf.high), height = 0, position = position_dodge(width = 0.2)) +
  facet_wrap(~ treatment) +
  scale_color_manual(values = c("Outcome1" = "gold3", 
                                "Outcome2" = "green3", 
                                "Outcome3" = "salmon")) +
  theme_minimal()

Answer 1

One possible solution:

First, calculate your p-values. Here's an example with a z-test, but use whatever's appropriate for your data:

results <- results %>%
  mutate(
    p.value = 2 * (1 - pnorm(abs(estimate / se)))
  )

Then, map the color aesthetic to a new factor vector that is equal to outcome unless p >= 0.10:

color = factor(
  if_else(p.value < 0.10, outcome, "n.s."),
  levels = c(levels(outcome), "n.s.")
)

In order to keep outcomes in the legend even if none are significant, we need to add drop = FALSE to scale_color_manual() as well as show.legend = TRUE to both geom_point() and geom_errorbarh(). We also need to add the aesthetic group = outcome so that position_dodge() still works as before.

Map alpha to a new factor vector that is one level if less than 0.05, and another otherwise:

alpha = cut(p.value, c(0, 0.05, 1))

And use scale_alpha_manual() to define the alpha values:

scale_alpha_manual(
  values = c(1, 0.5),
  guide = "none"
)

Putting it all together (note I used a different seed):

set.seed(789)
terms <- c("covariate1", "covariate2", "covariate3", "covariate4")
outcomes <- c("Outcome1", "Outcome2", "Outcome3")
treatments <- c("Treatment A", "Treatment B")
results <- expand.grid(term = terms, outcome = outcomes, treatment = treatments) %>%
  mutate(
    term = factor(term, levels = terms),
    estimate = rnorm(n(), 0, 0.2),   
    se = runif(n(), 0.05, 0.2),     
    conf.low = estimate - 1.96 * se, 
    conf.high = estimate + 1.96 * se
  ) %>%
  mutate(
    p.value = 2 * (1 - pnorm(abs(estimate / se)))
  )

ggplot(results, 
    aes(
      x = estimate, 
      y = term, 
      group = outcome,
      color = factor(
        if_else(p.value < 0.10, outcome, "n.s."),
        levels = c(levels(outcome), "n.s.")
      ),
      alpha = cut(p.value, c(0, 0.05, 1))
    )
  ) +
  geom_vline(xintercept = 0, color = "black", linetype = "dashed") +
  geom_point(size = 2, position = position_dodge(width = 0.2), show.legend = TRUE) +
  geom_errorbarh(aes(xmin = conf.low, xmax = conf.high), height = 0, position = position_dodge(width = 0.2), show.legend = TRUE) +
  facet_wrap(~ treatment) +
  scale_color_manual(
    values = c(
      "Outcome1" = "gold3", 
      "Outcome2" = "green3", 
      "Outcome3" = "salmon",
      "n.s." = "grey"
    ),
    drop = FALSE,
    name = "Outcome"
  ) +
  scale_alpha_manual(
    values = c(1, 0.5),
    guide = "none"
  ) +
  theme_minimal()