Merge pull request #17 from malcolmbarrett/smd

Use smd and gtsummary

Author: malcolmbarrett

exercises/01-whole-game-exercises.Rmd

@@ -9,8 +9,6 @@ library(broom)
 library(rsample)
 library(ggdag)
 library(causaldata)
-library(survey)
-library(tableone)
 ```
 
 ## Causal Modeling: Whole Game
@@ -194,45 +192,39 @@ The main goal here is to *break* the non-causal associations between quitting sm
 
 Standardized mean differences (SMD) are a simple measurement of differences that work across variable types. In general, the closer to 0 we are, the better job we have done eliminating the non-causal relationships we drew in our DAG. Note that low SMDs for everything we adjust for does *not* mean that there is not something else that might confound our study. Unmeasured confounders or misspecified DAGs can still distort our effects, even if our SMDs look great!
 
-We'll use the {survey} and {tableone} package to calculate the SMDs, then visualize them.
+We'll use the {smd} package to calculate the SMDs, then visualize them.
 
 ```{r}
-svy_des <- svydesign(
-  ids = ~ 1,
-  data = nhefs_complete_uc,
-  weights = ~ wts)
-
-smd_table_unweighted <- CreateTableOne(
-  vars = c("sex", "race", "age", "education", "smokeintensity", "smokeyrs", 
-           "exercise", "active", "wt71"),
-  strata = "qsmk",
-  data = nhefs_complete_uc,
-  test = FALSE)
-
-smd_table <- svyCreateTableOne(
-  vars = c("sex", "race", "age", "education", "smokeintensity", "smokeyrs", 
-           "exercise", "active", "wt71"),
-  strata = "qsmk",
-  data = svy_des,
-  test = FALSE)
-
-
-plot_df <- data.frame(
-  var = rownames(ExtractSmd(smd_table)),                        
-  Unadjusted = as.numeric(ExtractSmd(smd_table_unweighted)),                      
-  Weighted = as.numeric(ExtractSmd(smd_table))) %>%
-  pivot_longer(-var, names_to = "Method", values_to = "SMD")
+vars <- c("sex", "race", "age", "education", "smokeintensity", "smokeyrs", 
+          "exercise", "active", "wt71")
+
+plot_df <- nhefs_complete_uc %>% 
+  summarise(
+    across(
+      all_of(vars), 
+      list(
+        unweighted = ~smd(.x, qsmk)$estimate, 
+        weighted = ~smd(.x, qsmk, wts)$estimate
+      )
+    )
+  ) %>% 
+  pivot_longer(
+    everything(), 
+    values_to = "SMD", 
+    names_to = c("variable", "Method"),
+    names_sep = "_"
+  )
 
 ggplot(
   data = plot_df,
-  mapping = aes(x = var, y = SMD, group = Method, color = Method)
+  aes(x = abs(SMD), y = variable, group = Method, color = Method)
 ) +
-  geom_line() +
+  geom_line(orientation = "y") +
   geom_point() + 
-  geom_hline(yintercept = 0.1, color = "black", size = 0.1) +  
-  coord_flip() +
-  theme_minimal()
-
+  geom_vline(xintercept = 0.1, color = "black", size = 0.1) +
+  theme_minimal() +
+  scale_color_manual(values = c("grey85", "#00BFC4")) +
+  xlim(0, .3)
 ```
 
 These look pretty good! Some variables are better than others, but weighting appears to have done a much better job eliminating these differences than an unadjusted analysis.

exercises/05-pscores-diagnostics-exercises.Rmd

@@ -6,9 +6,8 @@ output: html_document
 
 ```{r}
 library(tidyverse)
-library(survey)
-library(tableone)
 library(broom)
+library(smd)
 library(causaldata)
 ```
 
@@ -32,55 +31,46 @@ df <- propensity_model %>%
 
 _After updating the code chunks below, change `eval = TRUE` before knitting._
 
-Create the survey design object to incorporate the weights.
+Calculate the standardized mean differences with and without weights
 
 ```{r, eval = FALSE}
-svy_des <- ____(
-  ids = ~ 1,
-  data = ___,
-  weights = ___
-)
-```
-
-Create the **unweighted** standardized mean differences data frame
-
-```{r, eval = FALSE}
-smd_table_unweighted <- ____(
-  vars = _____,
-  strata = _____,
-  data = ____,
-  test = FALSE)
+smds <- df %>% 
+  summarise(
+    across(
+      # variables to calculate SMD for
+      ________, 
+      list(
+        unweighted = ~_____(.x, qsmk)$estimate, 
+        weighted = ~_____(.x, qsmk, w = ___)$estimate
+      )
+    )
+  )
 ```
 
-Create the **weighted** standardized mean differences data frame
+Pivot `smds` so that it is in tidy format with the columns "SMD", "variable", and "Method".
 
 ```{r, eval = FALSE}
-smd_table <- ____(
-  vars = _____,
-  strata = _____,
-  data = ____,
-  test = FALSE)
-```
-
-Create a data frame that merges `smd_table_unweighted` and `smd_table` and pivots it to prepare for plotting
-
-```{r, eval = FALSE}
-plot_df <- data.frame(
-  var = rownames(____),                        
-  Unadjusted = _____,                      
-  Weighted = _____) %>%
-  pivot_longer(-var, names_to = "Method", values_to = "SMD")
+plot_df <- smds %>% 
+  pivot_longer(
+    _______, 
+    values_to = "___", 
+    names_to = c("___", "___"),
+    
+    # don't change this one. We really mean _!
+    names_sep = "_"
+  )
 ```
 
 Create the Love Plot using ggplot
 
 ```{r, eval = FALSE}
-ggplot(data = _____, 
-       mapping = aes(x = ____, y = ____, group = ____, color = ____)) +
-  geom_line() +
+ggplot(
+  data = ____,
+  aes(x = abs(____), y = ____, group = ____, color = ____)
+) +
+  geom_line(orientation = "y") +
   geom_point() + 
-  geom_hline(yintercept = 0.1, color = "black", size = 0.1) +  
-  coord_flip()
+  geom_vline(xintercept = 0.1, color = "black", size = 0.1)
 ```
 
 

exercises/06-outcome-model-exercises.Rmd

@@ -6,8 +6,6 @@ output: html_document
 
 ```{r}
 library(tidyverse)
-library(survey)
-library(tableone)
 library(broom)
 library(causaldata)
 library(rsample)