170102

util/util.R

@@ -1114,7 +1114,7 @@ check_moderator_levels <- function(df, experiment, outcome) {
   return(x)  # Return the formatted string
 }
 
-run_sse_SMD <- function(df, rho_value = 0.5) {
+run_sse_SMD_L <- function(df, rho_value = 0.5) {
 
   #  df<-filter_experiment_outcome_type(df, experiment, outcome)
 
@@ -1149,7 +1149,42 @@ run_sse_SMD <- function(df, rho_value = 0.5) {
   return(SMD_sse)
 }
 
-run_sse_plot_SMD <- function(df, rho_value = 0.5) {
+run_sse_SMD_C <- function(df, rho_value = 0.5) {
+
+  #  df<-filter_experiment_outcome_type(df, experiment, outcome)
+
+  df<-df %>% 
+    filter(!is.na(SMDv)) %>%
+    filter(outcome_type == "Cognition") %>%
+    filter(SortLabel == "TvC")
+
+  df <- df %>% mutate(effect_id = row_number()) # add effect_id column
+  df$SMDN <- 1/sqrt(as.numeric(df$NumberOfAnimals_C) + as.numeric(df$NumberOfAnimals_I))
+
+  #calculate variance-covariance matrix of the sampling errors for dependent effect sizes
+
+
+  VCVM_SMD <- vcalc(vi = SMDv,
+                    cluster = StudyId, 
+                    subgroup= ExperimentID_I,
+                    obs=effect_id,
+                    data = df, 
+                    rho = rho_value)
+
+  SMD_sse <- rma.mv(yi = SMD,
+                    V = VCVM_SMD,
+                    random = ~1 | Strain / StudyId / ExperimentID_I, # nested levels
+                    mods = ~ SMDN, # sampling error (squart root of N);
+                    test = "t", # use t- and F-tests for making inferences
+                    data = df,
+                    dfs="contain",
+                    control=list(optimizer="nlm")
+  )
+
+  return(SMD_sse)
+}
+
+run_sse_plot_SMD_L <- function(df, rho_value = 0.5) {
 
   #  df<-filter_experiment_outcome_type(df, experiment, outcome)
 
@@ -1185,6 +1220,41 @@ run_sse_plot_SMD <- function(df, rho_value = 0.5) {
   return(plot)
 }
 
+run_sse_plot_SMD_C <- function(df, rho_value = 0.5) {
+
+  #  df<-filter_experiment_outcome_type(df, experiment, outcome)
+
+  df<-df %>% 
+    filter(!is.na(SMDv)) %>%
+    filter(outcome_type == "Cognition") %>%
+    filter(SortLabel == "TvC")
+
+  df <- df %>% mutate(effect_id = row_number()) # add effect_id column
+  df$SMDN <- 1/sqrt(as.numeric(df$NumberOfAnimals_C) + as.numeric(df$NumberOfAnimals_I))
+
+  #calculate variance-covariance matrix of the sampling errors for dependent effect sizes
+
+
+  VCVM_SMD <- vcalc(vi = SMDv,
+                    cluster = StudyId, 
+                    subgroup= ExperimentID_I,
+                    obs=effect_id,
+                    data = df, 
+                    rho = rho_value)
+
+  SMD_sse <- rma.mv(yi = SMD,
+                    V = VCVM_SMD,
+                    random = ~1 | Strain / StudyId / ExperimentID_I, # nested levels
+                    mods = ~ SMDN, # sampling error (squart root of N);
+                    test = "t", # use t- and F-tests for making inferences
+                    data = df,
+                    dfs="contain",
+                    control=list(optimizer="nlm")
+  )
+
+  plot <- bubble_plot(SMD_sse, mod = "SMDN", group = "StudyId", xlab = "1/SQRT(N) associated with SMD estimate", ylab = "SMD estimate", legend.pos = "none")
+  return(plot)
+}
 subgroup_SMD <- function(df, experiment_type, outcome, moderator, rho_value) {
   # with intercept, to allow calculation of effect of moderators - returns intercept 
   # as beta-coefficient for first category, and beta coefficients for other categories