TITLE: Pretty correlation matrices with ggplot
DATE: 2020-07-05
AUTHOR: John L. Godlee
====================================================================


I needed to make a correlation matrix plot to show the 
relationships between all variables in a structural equation model 
I was writing. I created a function which takes a dataframe of 
numeric variables and returns a ggplot2 object.

    #' Create a pretty correlation matrix plot
    #'
    #' @param x dataframe of numeric variables to correlate
    #'
    #' @return ggplot object with correlogram
    #' 
    #' @examples
    #' data(iris)
    #' corrplot(iris[,1:4])
    #' 
    #' @import ggplot2 
    #' @importFrom psych corr.test
    #'
    corrplot <- function(x, col = c("red", "white", "blue"), ...) {
      corr <- psych::corr.test(x, ...) 
      corr_ci <- data.frame(raw.lower = corr$ci$lower, raw.r = 
corr$ci$r, 
        raw.upper = corr$ci$upper, raw.p = corr$ci$p, 
        adj.lower = corr$ci.adj$lower.adj, adj.upper = 
corr$ci.adj$upper.adj)
      corr_ci$vars <- row.names(corr_ci)
      corr_ci$conf_x <- unlist(sapply(1:(length(x)-1), function(i){
          c(1:(length(x)-1))[i:(length(x)-1)]
        })) + 1
      rev_mat <- (length(x)-1):1
      corr_ci$conf_y <- unlist(sapply(1:(length(x)-1), function(i){
          rep(i, times = rev_mat[i])
        }))
      n_seq <- 2:length(x)
      corr_ci$y_var <- unlist(sapply(1:(length(x)-1), function(i){
          rep(row.names(corr[[1]])[i], rev_mat[i])
        }))
      corr_ci$x_var <- unlist(sapply(1:(length(x)-1), function(i){
          row.names(corr[[1]])[n_seq[i]:length(x)]
        }))
      corr_ci$x_var <- factor(corr_ci$x_var, levels = 
unique(corr_ci$x_var))
      corr_ci$y_var <- factor(corr_ci$y_var, levels = 
unique(corr_ci$y_var))
      corr_ci$conf <- (corr_ci$raw.lower > 0) == (corr_ci$raw.upper 
> 0)
      corr_ci$raw.r <- round(corr_ci$raw.r, 2)

      ggplot2::ggplot() + 
        ggplot2::geom_tile(data = corr_ci, 
          ggplot2::aes(x = x_var, y = y_var, 
            fill = raw.r), colour = "black") + 
        ggplot2::geom_text(data = corr_ci, 
          ggplot2::aes(x = x_var, y = y_var, label = raw.r),
          size = 3) + 
        ggplot2::geom_point(data = corr_ci[corr_ci$conf == FALSE,], 
          ggplot2::aes(x = x_var, y = y_var), 
          fill = NA, colour = "black", shape = 21, size = 11) + 
        ggplot2::scale_fill_gradient2(name = "r", 
          low = col[1], mid = col[2], high = col[3]) + 
        ggplot2::theme_classic() + 
        ggplot2::labs(x = "", y = "") + 
        ggplot2::coord_equal() +
        ggplot2::theme(legend.position = "none")
    }

    corrplot(iris[,1:4])

  ![Correlation matrix 
plot](https://johngodlee.xyz/img_full/corrplot/corrplot.png)

Here is the actual plot I made for the publication, which 
additionally colours the axis text to group variables into latent 
constructs:

  ![Correlation matrix plot for 
publication](https://johngodlee.xyz/img_full/corrplot/corrplot_pub.j
pg)