Merge pull request #269 from r-causal/style2

Run styler

R/setup.R

@@ -62,20 +62,20 @@ est_ci <- function(.df, rsample = FALSE) {
 # based on https://github.com/hadley/r-pkgs/blob/main/common.R
 status <- function(type) {
   status <- switch(type,
-                   unstarted = "is unstarted, but don't worry, it's on our roadmap",
-                   polishing = "has its foundations written but is still undergoing changes",
-                   wip = "is actively undergoing work and may be restructured or changed. It may also be incomplete",
-                   complete = "is mostly complete, but we might make small tweaks or copyedits",
-                   stop("Invalid `type`", call. = FALSE)
+    unstarted = "is unstarted, but don't worry, it's on our roadmap",
+    polishing = "has its foundations written but is still undergoing changes",
+    wip = "is actively undergoing work and may be restructured or changed. It may also be incomplete",
+    complete = "is mostly complete, but we might make small tweaks or copyedits",
+    stop("Invalid `type`", call. = FALSE)
   )
-  
+
   class <- switch(type,
-                  complete = ,                 
-                  polishing = "callout-note",
-                  wip = "callout-warning",
-                  unstarted = "callout-warning"
+    complete = ,
+    polishing = "callout-note",
+    wip = "callout-warning",
+    unstarted = "callout-warning"
   )
-  
+
   knitr::asis_output(paste0(
     "::: ", class, "\n",
     "## Work-in-progress 🚧\n",

appendices/A-bootstrap.qmd

@@ -81,7 +81,7 @@ That means that some of the included rows are there more than once; each row is
 This spread is close to what we expect on average: about two-thirds of the original dataset ends up in each bootstrapped dataset.
 
 ```{r}
-boot_resample <- bootstrapped_resamples$splits[[1]] |> 
+boot_resample <- bootstrapped_resamples$splits[[1]] |>
   as.data.frame()
 
 boot_resample
@@ -111,7 +111,7 @@ Rather than doing this one by one, we'll use iteration to run the regression on
 We'll take advantage of the existing structure in `bootstrapped_resamples` to store the results as another list-column.
 
 ```{r}
-bootstrapped_resamples <- bootstrapped_resamples |> 
+bootstrapped_resamples <- bootstrapped_resamples |>
   mutate(lm_results = map(splits, fit_lm))
 ```
 
@@ -127,11 +127,11 @@ Now, we have ten estimates of each of the three coefficients in the model (the i
 
 ```{r}
 library(broom)
-bootstrapped_resamples <- bootstrapped_resamples |> 
+bootstrapped_resamples <- bootstrapped_resamples |>
   mutate(tidy_results = map(lm_results, tidy))
 
-unnested_results <- bootstrapped_resamples |> 
-  select(id, tidy_results) |> 
+unnested_results <- bootstrapped_resamples |>
+  select(id, tidy_results) |>
   unnest(tidy_results)
 
 unnested_results
@@ -140,10 +140,10 @@ unnested_results
 ```{r}
 #| label: fig-r-10
 #| fig-cap: "The bootstrapped distributions of the coefficients from the model `lm(y ~ x + z, data = .df)`. The distributions were calculated with 10 bootstrapped resamples."
-unnested_results |> 
-  ggplot(aes(estimate)) + 
+unnested_results |>
+  ggplot(aes(estimate)) +
   geom_density(fill = "steelblue", color = NA) +
-  facet_wrap(~ term, scales = "free")
+  facet_wrap(~term, scales = "free")
 ```
 
 The more times we resample, the smoother the distribution of estimates.
@@ -153,20 +153,20 @@ Here's 1000 times (@fig-r-1000).
 #| label: fig-r-1000
 #| fig-cap: "The bootstrapped distributions of the coefficients from the model `lm(y ~ x + z, data = .df)`. The distributions were calculated with 1000 bootstrapped resamples."
 bootstrapped_resamples_1k <- bootstraps(
-  sampled_data, 
+  sampled_data,
   times = 1000
-) |> 
+) |>
   mutate(
     lm_results = map(splits, fit_lm),
     tidy_results = map(lm_results, tidy)
   )
 
-bootstrapped_resamples_1k |> 
-  select(id, tidy_results) |> 
-  unnest(tidy_results) |> 
-  ggplot(aes(estimate)) + 
+bootstrapped_resamples_1k |>
+  select(id, tidy_results) |>
+  unnest(tidy_results) |>
+  ggplot(aes(estimate)) +
   geom_density(fill = "steelblue", color = NA) +
-  facet_wrap(~ term, scales = "free")
+  facet_wrap(~term, scales = "free")
 ```
 
 We can calculate information about the spread of these coefficients using the confidence interval functions in rsample, which follow the pattern `int_*(nested_results, list_column_name)`.
@@ -207,8 +207,8 @@ A given observation may end up in more than one study.
 Suppose 20 such studies have been done, all from the same population.
 
 ```{r}
-samples <- map(1:20, ~ population[sample(n, size = 200), ]) |> 
-  bind_rows(.id = "sample") |> 
+samples <- map(1:20, ~ population[sample(n, size = 200), ]) |>
+  bind_rows(.id = "sample") |>
   mutate(sample = as.numeric(sample))
 ```
 
@@ -219,19 +219,19 @@ Each of these sample estimates hovers around the population estimate (@fig-pop-s
 #| label: fig-pop-samples
 #| fig-cap: "The distributions of the sample means of `x` for twenty samples. Each sample is sampled from `population` and has a sample size of 200."
 #| warning: false
-sample_means <- samples |> 
-  group_by(sample) |> 
+sample_means <- samples |>
+  group_by(sample) |>
   summarize(across(everything(), mean))
 
-samples |> 
+samples |>
   ggplot(aes(x = x)) +
   geom_histogram() +
   geom_vline(
-    data = sample_means, 
-    aes(xintercept = x), 
+    data = sample_means,
+    aes(xintercept = x),
     color = "firebrick"
-  ) + 
-  facet_wrap(~ sample)
+  ) +
+  facet_wrap(~sample)
 ```
 
 You may notice that sampling from the population bears a similarity to the sampling we do in the bootstrap.
@@ -241,10 +241,10 @@ A key difference from sampling the population is that the bootstrap will determi
 Let's look at sample 8 a bit closer.
 
 ```{r}
-sample_8 <- samples |> 
+sample_8 <- samples |>
   filter(sample == "8")
 
-sample_8 |> 
+sample_8 |>
   summarize(across(everything(), mean))
 ```
 
@@ -259,24 +259,24 @@ The distribution of bootstrapped estimates in @fig-boot-x falls symmetrically ar
 calculate_mean <- function(.split, what = "x", ...) {
   .df <- as.data.frame(.split)
   t <- t.test(.df[[what]])
-  
+
   tibble(
-    term = paste("mean of", what), 
+    term = paste("mean of", what),
     estimate = as.numeric(t$estimate),
     std.error = t$stderr
   )
 }
 
 s8_boots <- bootstraps(sample_8, times = 1000, apparent = TRUE)
-s8_boots <- s8_boots |> 
+s8_boots <- s8_boots |>
   mutate(boot_mean_x = map(splits, calculate_mean))
 
-s8_boots |> 
-  mutate(boot_mean_x = map_dbl(boot_mean_x, \(.df) .df$estimate)) |> 
+s8_boots |>
+  mutate(boot_mean_x = map_dbl(boot_mean_x, \(.df) .df$estimate)) |>
   ggplot(aes(x = boot_mean_x)) +
   geom_histogram() +
   geom_vline(
-    data = sample_means |> filter(sample == "8"), 
+    data = sample_means |> filter(sample == "8"),
     aes(xintercept = x),
     color = "firebrick"
   )
@@ -294,19 +294,19 @@ We'll also increase the number of resamples to approximate the coverage better.
 ```{r}
 n_samples <- 1000
 
-samples <- map(seq_len(n_samples), ~ population[sample(n, size = 200), ]) |> 
-  bind_rows(.id = "sample") |> 
+samples <- map(seq_len(n_samples), ~ population[sample(n, size = 200), ]) |>
+  bind_rows(.id = "sample") |>
   mutate(sample = as.numeric(sample))
 
-cis <- samples |> 
-  group_by(sample) |> 
+cis <- samples |>
+  group_by(sample) |>
   group_modify(~ t.test(.x$x) |> tidy())
 
 between(
-  rep(mean(population$x), n_samples), 
-  cis$conf.low, 
+  rep(mean(population$x), n_samples),
+  cis$conf.low,
   cis$conf.high
-) |> 
+) |>
   mean()
 ```
 
@@ -317,22 +317,22 @@ We won't run this since it requires `r_bootstraps * n_samples` calculations, but
 #| eval: false
 bootstrap_ci <- function(.sample_df, ...) {
   sample_boots <- bootstraps(.sample_df, times = 1000)
-  sample_boots <- sample_boots |> 
+  sample_boots <- sample_boots |>
     mutate(boot_mean_x = future_map(splits, calculate_mean))
-  
-  sample_boots |> 
+
+  sample_boots |>
     int_pctl(boot_mean_x)
 }
 
-boot_cis <- samples |> 
-  group_by(sample) |> 
+boot_cis <- samples |>
+  group_by(sample) |>
   group_modify(bootstrap_ci)
 
 coverage <- between(
-  rep(mean(population$x), n_samples), 
-  boot_cis$.lower, 
+  rep(mean(population$x), n_samples),
+  boot_cis$.lower,
   boot_cis$.upper
-) |> 
+) |>
   mean()
 ```
 
@@ -372,7 +372,7 @@ library(furrr)
 n_cores <- availableCores() - 1
 plan(multisession, workers = n_cores)
 
-s8_boots <- s8_boots |> 
+s8_boots <- s8_boots |>
   mutate(boot_mean_x = future_map(splits, calculate_mean))
 ```
 :::
@@ -410,7 +410,7 @@ We see that, for instance, with confidence intervals from the traditional t-test
 
 ```{r}
 c(
-  mean(mean(population$x) < cis$conf.low), 
+  mean(mean(population$x) < cis$conf.low),
   mean(mean(population$x) > cis$conf.high)
 )
 ```
@@ -450,18 +450,18 @@ For example, bootstrapping the minimum of `x` results in a strange distribution.
 #| warning: false
 calculate_min <- function(.split, what = "x", ...) {
   .df <- as.data.frame(.split)
-  
+
   tibble(
-    term = paste("min of", what), 
+    term = paste("min of", what),
     estimate = min(.df[[what]])
   )
 }
 
-s8_boots <- s8_boots |> 
+s8_boots <- s8_boots |>
   mutate(boot_min_x = map(splits, calculate_min))
 
-s8_boots |> 
-  mutate(boot_min_x = map_dbl(boot_min_x, \(.df) .df$estimate)) |> 
+s8_boots |>
+  mutate(boot_min_x = map_dbl(boot_min_x, \(.df) .df$estimate)) |>
   ggplot(aes(x = boot_min_x)) +
   geom_histogram()
 ```

chapters/03-counterfactuals.qmd

@@ -356,8 +356,10 @@ data <- data.frame(
 )
 
 # partner's happiness increases by 2 when they get a different flavor
-data <- data |> mutate(y_chocolate_vanilla = y_chocolate_chocolate + 2,
-                       y_vanilla_chocolate = y_vanilla_vanilla + 2)
+data <- data |> mutate(
+  y_chocolate_vanilla = y_chocolate_chocolate + 2,
+  y_vanilla_chocolate = y_vanilla_vanilla + 2
+)
 
 set.seed(11)
 data_observed <- data |>

chapters/04-target-trials-std-methods.qmd

@@ -441,9 +441,9 @@ Let's examine three models: (1) an unadjusted model (@tbl-panel-1), (2) a linear
 #| label: tbl-panel
 #| layout-ncol: 2
 #| tbl-cap: Three ways to estimate a causal effect.
-#| tbl-subcap: 
-#|   - Unadjusted regression 
-#|   - Adjusted regression 
+#| tbl-subcap:
+#|   - Unadjusted regression
+#|   - Adjusted regression
 #|   - Propensity score weighted regression
 #| code-fold: true
 #| message: false
@@ -455,7 +455,7 @@ lm(y ~ treatment, d) |>
 
 lm(y ~ treatment + age + weight, d) |>
   tbl_regression() |>
-  modify_column_unhide(column = std.error) 
+  modify_column_unhide(column = std.error)
 
 d |>
   mutate(
@@ -543,13 +543,13 @@ dagify(
 #| warning: false
 #| layout-ncol: 2
 #| tbl-cap: Three ways to estimate a causal effect in a non-randomized setting
-#| tbl-subcap: 
-#|   - Unadjusted regression 
-#|   - Adjusted regression 
+#| tbl-subcap:
+#|   - Unadjusted regression
+#|   - Adjusted regression
 #|   - Propensity score weighted regression
 lm(y ~ treatment, d) |>
   tbl_regression() |>
-  modify_column_unhide(column = std.error) 
+  modify_column_unhide(column = std.error)
 
 lm(y ~ treatment + age + weight, d) |>
   tbl_regression() |>
@@ -576,7 +576,7 @@ tibble(
   SE = round(x$std.wt, 3),
   `95% CI` = glue::glue("{round(x$est.wt - 1.96 * x$std.wt, 1)}, {round(x$est.wt + 1.96 * x$std.wt, 1)}"),
   `p-value` = "<0.001"
-) |> 
+) |>
   knitr::kable()
 ```
 

chapters/11-estimands.qmd

@@ -870,7 +870,7 @@ The default is for the ATE, but we can also target other estimands.
 ```{r}
 library(lmw)
 implied_weights <- lmw(
-  ~ park_extra_magic_morning + park_ticket_season + park_close + park_temperature_high, 
+  ~ park_extra_magic_morning + park_ticket_season + park_close + park_temperature_high,
   data = seven_dwarfs_with_ps,
   treat = "park_extra_magic_morning"
 )
@@ -881,21 +881,21 @@ These implied weights have some nice properties.
 They are perfectly balanced on the mean and have the lowest variance of such weights.
 
 ```{r}
-seven_dwarfs_with_ps |> 
-  mutate(park_close = as.numeric(park_close)) |> 
+seven_dwarfs_with_ps |>
+  mutate(park_close = as.numeric(park_close)) |>
   tidy_smd(
     .vars = c(park_ticket_season, park_close, park_temperature_high),
     .group = park_extra_magic_morning,
     .wts = iw
-  ) |> 
-  ggplot(aes(abs(smd), variable, color = method, group = method)) + 
+  ) |>
+  ggplot(aes(abs(smd), variable, color = method, group = method)) +
   geom_love()
 ```
 
 They also have a mean of 1, meaning they sum to the original size of the dataset.
 
 ```{r}
-seven_dwarfs_with_ps |> 
+seven_dwarfs_with_ps |>
   summarize(mean = mean(iw), sum = sum(iw), n = n())
 ```
 
@@ -909,7 +909,7 @@ So, even though we've targeted the ATE, we may not have done so very precisely.
 
 ```{r}
 implied_weights_int <- lmw(
-  ~ park_extra_magic_morning * (park_ticket_season + park_close + park_temperature_high), 
+  ~ park_extra_magic_morning * (park_ticket_season + park_close + park_temperature_high),
   data = seven_dwarfs_with_ps,
   treat = "park_extra_magic_morning"
 )