make plot_corr() work on mids objects

R/plot_corr.R

@@ -39,45 +39,58 @@ plot_corr <-
     if (is.matrix(data) && ncol(data) > 1) {
       data <- as.data.frame(data)
     }
-    verify_data(data = data, df = TRUE)
+    verify_data(data = data, df = TRUE, imp = TRUE)
     vrb <- rlang::enexpr(vrb)
-    vrb_matched <- match_vrb(vrb, names(data))
+    vrbs_in_data <- if (mice::is.mids(data)) {
+      names(data$imp)
+    } else {
+      names(data)
+    }
+    vrb_matched <- match_vrb(vrb, vrbs_in_data)
     if (length(vrb_matched) < 2) {
       cli::cli_abort("The number of variables should be two or more to compute correlations.")
     }
-    # check if any column is constant
-    constants <- apply(data[, vrb_matched], MARGIN = 2, function(x) {
-      all(is.na(x)) || max(x, na.rm = TRUE) == min(x, na.rm = TRUE)
-    })
-    if (any(constants)) {
-      vrb_matched <- vrb_matched[!constants]
-      cli::cli_inform(
-        c(
-          "No correlations computed for variable(s):",
-          " " = paste(names(constants[which(constants)]), collapse = ", "),
-          "i" = "Correlations are undefined for constants."
+    if (is.data.frame(data)) {
+      # for data: check if any column is constant
+      constants <- apply(data[, vrb_matched], MARGIN = 2, function(x) {
+        all(is.na(x)) || max(x, na.rm = TRUE) == min(x, na.rm = TRUE)
+      })
+      if (any(constants)) {
+        vrb_matched <- vrb_matched[!constants]
+        cli::cli_inform(
+          c(
+            "No correlations computed for variable(s):",
+            " " = paste(names(constants[which(constants)]), collapse = ", "),
+            "i" = "Correlations are undefined for constants."
+          )
         )
-      )
+      }
+      # compute correlations
+      corr <- stats::cov2cor(stats::cov(
+        data.matrix(data[, vrb_matched]),
+        use = "pairwise.complete.obs"
+        ))
+    }
+    if (mice::is.mids(data)) {
+      # check constatnts etc.
+      imps <- mice::complete(data, "all")
+      corrs <- purrr::map(imps, ~ {
+        stats::cor(.x)
+      })
+      corr <- Reduce("+", corrs) / length(corrs)
     }
-    # compute correlations
     p <- length(vrb_matched)
-    corrs <- data.frame(
+    long <- data.frame(
       vrb = rep(vrb_matched, each = p),
       prd = vrb_matched,
-      corr = matrix(
-        round(stats::cov2cor(
-          stats::cov(data.matrix(data[, vrb_matched]), use = "pairwise.complete.obs")
-        ), 2),
-        nrow = p * p,
-        byrow = TRUE
-      )
+      corr = matrix(round(corr, 2), nrow = p * p, byrow = TRUE)
     )
     if (!diagonal) {
-      corrs[corrs$vrb == corrs$prd, "corr"] <- NA
+      long[long$vrb == long$prd, "corr"] <- NA
     }
     # create plot
     gg <-
-      ggplot2::ggplot(corrs,
+      ggplot2::ggplot(long,
                       ggplot2::aes(
                         x = .data$prd,
                         y = .data$vrb,
@@ -95,18 +108,26 @@ plot_corr <-
         limits = c(-1, 1)
       )  +
       theme_minimice()
+    lab_x <- "Imputation model predictor"
+    if (mice::is.mids(data)) {
+      lab_y <- "Column name"
+      lab_note <- "*pooled across imputations"
+    } else {
+      lab_y <- "Column name"
+      lab_note <- "*pairwise complete observations"
+    }
     if (caption) {
       gg <- gg +
         ggplot2::labs(
-          x = "Imputation model predictor",
-          y = "Variable to impute",
+          x = lab_x,
+          y = lab_y,
           fill = "Correlation*
       ",
-          caption = "*pairwise complete observations"
+          caption = lab_note
         )
     } else {
       gg <- gg +
-        ggplot2::labs(x = "Imputation model predictor", y = "Variable to impute", fill = "Correlation")
+        ggplot2::labs(x = lab_x, y = lab_y, fill = "Correlation")
     }
     if (label) {
       gg <-