ggcorrmat

Correlation matrix plot with ggcorrmat

For the first example, we will use the gapminder dataset (available in eponymous package on CRAN) provides values for life expectancy, Gross Domestic Product (GDP) per capita, and population, every five years, from 1952 to 2007, for each of 142 countries and was collected by the Gapminder Foundation. Let’s have a look at the data-

library(gapminder)
library(dplyr)

dplyr::glimpse(x = gapminder)
#> Observations: 1,704
#> Variables: 6
#> $ country   <fct> Afghanistan, Afghanistan, Afghanistan, Afghanistan, ...
#> $ continent <fct> Asia, Asia, Asia, Asia, Asia, Asia, Asia, Asia, Asia...
#> $ year      <int> 1952, 1957, 1962, 1967, 1972, 1977, 1982, 1987, 1992...
#> $ lifeExp   <dbl> 28.801, 30.332, 31.997, 34.020, 36.088, 38.438, 39.8...
#> $ pop       <int> 8425333, 9240934, 10267083, 11537966, 13079460, 1488...
#> $ gdpPercap <dbl> 779.4453, 820.8530, 853.1007, 836.1971, 739.9811, 78...

Let’s say we are interested in studying correlation between population of a country, average life expectancy, and GDP per capita across countries only for the year 2007.

The simplest way to get a correlation matrix is to stick to the defaults-

library(ggstatsplot)

# select data only from the year 2007
gapminder_2007 <- dplyr::filter(.data = gapminder::gapminder, year == 2007)

# producing the correlation matrix
ggstatsplot::ggcorrmat(
  data = gapminder_2007,             # data from which variable is to be taken
  cor.vars = lifeExp:gdpPercap       # specifying correlation matrix variables
)

This plot can be further modified with additional arguments-

ggstatsplot::ggcorrmat(
  data = gapminder_2007,                      # data from which variable is to be taken
  cor.vars = lifeExp:gdpPercap,               # specifying correlation matrix variables
  cor.vars.names = c("Life Expectancy", "population", "GDP (per capita)"),
  corr.method = "kendall",                    # which correlation coefficient is to be computed
  lab.col = "red",                            # label color
  ggtheme = ggplot2::theme_light,             # selected ggplot2 theme
  ggstatsplot.theme = FALSE,                  # turn off default ggestatsplot theme overlay
  type = "lower",                             # type of correlation matrix
  colors = c("green", "white", "yellow"),     # selecting color combination
  title = "Gapminder correlation matrix",     # custom title
  subtitle = "Source: Gapminder Foundation"   # custom subtitle
)

As seen from this correlation matrix, although there is no relationship between population and life expectancy worldwide, at least in 2007, there is a strong positive relationship between GDP, a well-established indicator of a country’s economic performance.

Given that there were only three variables, this doesn’t look that impressive. So let’s work with another example from ggplot2 package: the diamonds dataset. This dataset contains the prices and other attributes of almost 54,000 diamonds.

Let’s have a look at the data-

library(ggplot2)

dplyr::glimpse(x = diamonds)
#> Observations: 53,940
#> Variables: 10
#> $ carat   <dbl> 0.23, 0.21, 0.23, 0.29, 0.31, 0.24, 0.24, 0.26, 0.22, ...
#> $ cut     <ord> Ideal, Premium, Good, Premium, Good, Very Good, Very G...
#> $ color   <ord> E, E, E, I, J, J, I, H, E, H, J, J, F, J, E, E, I, J, ...
#> $ clarity <ord> SI2, SI1, VS1, VS2, SI2, VVS2, VVS1, SI1, VS2, VS1, SI...
#> $ depth   <dbl> 61.5, 59.8, 56.9, 62.4, 63.3, 62.8, 62.3, 61.9, 65.1, ...
#> $ table   <dbl> 55, 61, 65, 58, 58, 57, 57, 55, 61, 61, 55, 56, 61, 54...
#> $ price   <int> 326, 326, 327, 334, 335, 336, 336, 337, 337, 338, 339,...
#> $ x       <dbl> 3.95, 3.89, 4.05, 4.20, 4.34, 3.94, 3.95, 4.07, 3.87, ...
#> $ y       <dbl> 3.98, 3.84, 4.07, 4.23, 4.35, 3.96, 3.98, 4.11, 3.78, ...
#> $ z       <dbl> 2.43, 2.31, 2.31, 2.63, 2.75, 2.48, 2.47, 2.53, 2.49, ...

Let’s see the correlation matrix between different attributes of the diamond and the price.

ggstatsplot::ggcorrmat(
  data = diamonds,             
  cor.vars = c(carat, depth:z),        # note how the variables are getting selected
  cor.vars.names = c("carat", "total depth", "table", "price", "length (in mm)", "width (in mm)", "depth (in mm)"),
  hc.order = TRUE                      # use hierarchical clustering
)

We can make a number of changes to this basic correlation matrix. For example, since we were interested in relationship between price and other attributes, let’s make the price column to the the first column. Additionally, since we are running 6 correlations that are of a priori interest to us, we can adjust our threshold of significance to (0.05/6 ~ 0.008). Additionally, let’s use a non-parametric correlation coefficient. Please note that it is important to always make sure that the order in which cor.vars and cor.vars.names are entered is in sync. Otherwise, wrong column labels will be displayed.

ggstatsplot::ggcorrmat(
  data = diamonds,             
  cor.vars = c(price, carat, depth:table, x:z),    # note how the variables are getting selected
  cor.vars.names = c("price", "carat", "total depth", "table",  "length (in mm)", "width (in mm)", "depth (in mm)"),
  corr.method = "spearman",
  sig.level = 0.008,
  type = "lower",
  title = "Relationship between diamond attributes and price",
  subtitle = "Dataset: Diamonds from ggplot2 package",
  colors = c("#0072B2", "#D55E00", "#CC79A7"),
  lab.col = "yellow",
  lab.size = 6,
  pch = 7,
  pch.col = "white",
  pch.cex = 14,
  caption = expression(                            # changing the default caption text for the plot 
    paste(italic("Note"), ": Point shape denotes correlation non-significant at p < 0.008; adjusted for 6 comparisons")
  )
)

As seen here, and unsurprisingly, the strongest predictor of the diamond price is its carat value, which a unit of mass equal to 200 mg. In other words, the heavier the diamond, the more expensive it is going to be.

Correlation statistics matrix with ggcorrmat

Another utility of ggcorrmat is in obtaining matrix of correlation coefficients and their p-values for a quick and dirty exploratory data analysis. For example, for the correlation matrix we just ran, we can get a coefficient matrix and a p-value matrix. We will use robust correlation for this illustration as this correlation is supported only to get correlation coefficients and p-values.

# to get correlations
ggstatsplot::ggcorrmat(
  data = diamonds,             
  cor.vars = c(price, carat, depth:table, x:z),
  output = "correlations",
  corr.method = "robust",
  digits = 3
)
#> # A tibble: 7 x 8
#>   variable  price carat  depth  table      x      y     z
#>   <chr>     <dbl> <dbl>  <dbl>  <dbl>  <dbl>  <dbl> <dbl>
#> 1 price     1     0.936 -0.006  0.145  0.9    0.901 0.896
#> 2 carat     0.936 1      0.025  0.195  0.983  0.982 0.981
#> 3 depth    -0.006 0.025  1     -0.284 -0.02  -0.023 0.085
#> 4 table     0.145 0.195 -0.284  1      0.202  0.197 0.167
#> 5 x         0.9   0.983 -0.02   0.202  1      0.999 0.991
#> 6 y         0.901 0.982 -0.023  0.197  0.999  1     0.991
#> 7 z         0.896 0.981  0.085  0.167  0.991  0.991 1

# to get p-values
ggstatsplot::ggcorrmat(
  data = diamonds,             
  cor.vars = c(price, carat, depth:table, x:z),
  output = "p-values",
  corr.method = "robust",
  digits = 3
)
#> # A tibble: 7 x 8
#>   variable price        carat       depth table         x          y     z
#>   <chr>    <dbl>        <dbl>       <dbl> <dbl>     <dbl>      <dbl> <dbl>
#> 1 price    0     0                1.43e-1     0   0.         0.          0
#> 2 carat    0     0                1.15e-8     0   0.         0.          0
#> 3 depth    0.143 0.0000000115     0.          0   2.45e-6    1.25e-7     0
#> 4 table    0     0                0.          0   0.         0.          0
#> 5 x        0     0                2.45e-6     0   0.         0.          0
#> 6 y        0     0                1.25e-7     0   0.         0.          0
#> 7 z        0     0                0.          0   0.         0.          0

Grouped analysis with grouped_ggcorrmat

What if we want to do the same analysis separately for each type of quality of the diamond cut (Fair, Good, Very Good, Premium, Ideal)? In that case, we will have to either write a for loop or use purrr, none of which seem like an exciting prospect.

ggstatsplot provides a special helper function for such instances: grouped_ggcorrmat. This is merely a wrapper function around ggstatsplot::combine_plots. It applies ggcorrmat across all levels of a specified grouping variable and then combines list of individual plots into a single plot. Note that the grouping variable can be anything: conditions in a given study, groups in a study sample, different studies, etc.

ggstatsplot::grouped_ggcorrmat(
  # arguments relevant for ggstatsplot::ggcorrmat
  data = ggplot2::diamonds,
  grouping.var = cut,
  title.prefix = "Quality of cut",
  cor.vars = c(carat, depth:z),
  cor.vars.names = c(
    "carat",
    "total depth",
    "table",
    "price",
    "length (in mm)",
    "width (in mm)",
    "depth (in mm)"
  ),
  lab.size = 3.5,
  # arguments relevant for ggstatsplot::combine_plots
  title.text = "Relationship between diamond attributes and price across cut",
  title.size = 16,
  title.color = "red",
  caption.text = "Dataset: Diamonds from ggplot2 package",
  caption.size = 14,
  caption.color = "blue",
  labels = c("(a)","(b)","(c)","(d)","(e)"),
  nrow = 3,
  ncol = 2
)

Note that this function also makes it easy to run the same correlation matrix across different levels of a factor/grouping variable. For example, if we wanted to get the same correlation coefficient matrix for color of the diamond, we can do the following-

ggstatsplot::grouped_ggcorrmat(
  data = ggplot2::diamonds,
  grouping.var = cut,
  cor.vars = c(price, carat, depth:table, x:z),
  output = "correlations",
  corr.method = "robust",
  digits = 3
)
#> # A tibble: 35 x 9
#>    Group variable  price  carat  depth  table      x      y      z
#>    <chr> <chr>     <dbl>  <dbl>  <dbl>  <dbl>  <dbl>  <dbl>  <dbl>
#>  1 Fair  price     1      0.898  0.022 -0.032  0.884  0.885  0.845
#>  2 Fair  carat     0.898  1      0.138 -0.077  0.973  0.968  0.965
#>  3 Fair  depth     0.022  0.138  1     -0.587  0.001 -0.009  0.327
#>  4 Fair  table    -0.032 -0.077 -0.587  1     -0.003 -0.013 -0.204
#>  5 Fair  x         0.884  0.973  0.001 -0.003  1      0.993  0.936
#>  6 Fair  y         0.885  0.968 -0.009 -0.013  0.993  1      0.934
#>  7 Fair  z         0.845  0.965  0.327 -0.204  0.936  0.934  1    
#>  8 Good  price     1      0.941 -0.048  0.175  0.916  0.918  0.915
#>  9 Good  carat     0.941  1     -0.054  0.212  0.985  0.985  0.982
#> 10 Good  depth    -0.048 -0.054  1     -0.546 -0.145 -0.147  0.044
#> # ... with 25 more rows

As this example illustrates, there is a minimal coding overhead to explore correlations in your dataset with the grouped_ggcorrmat function.

Grouped analysis with ggcorrmat + purrr

Although grouped_ function is good for quickly exploring the data, it reduces the flexibility with which this function can be used. This is the because the common parameters used are applied to plots corresponding to all levels of the grouping variable and there is no way to adapt them based on the level of the grouping variable. We will see how this can be done using the purrr package from tidyverse.

Note before: Unlike the function call so far, while using purrr::pmap, we will need to quote the arguments.

# splitting the dataframe by cut and creting a list
cut_list <- ggplot2::diamonds %>%
  base::split(x = ., f = .$cut, drop = TRUE)

# this created a list with 5 elements, one for each quality of cut
# you can check the structure of the file for yourself
# str(cut_list)

# checking the length and names of each element
length(cut_list)
#> [1] 5
names(cut_list)
#> [1] "Fair"      "Good"      "Very Good" "Premium"   "Ideal"

# running function on every element of this list note that if you want the same
# value for a given argument across all elements of the list, you need to
# specify it just once
plot_list <- purrr::pmap(
  .l = list(
    data = cut_list,
    cor.vars = list(c("carat", "depth", "table",
                 "price", "x", "y", "z")),
    cor.vars.names = list(c(
      "carat",
      "total depth",
      "table",
      "price",
      "length (in mm)",
      "width (in mm)",
      "depth (in mm)"
    )),
    corr.method = list("pearson", "spearman", "spearman", "pearson", "pearson"),
    sig.level = list(0.05, 0.001, 0.01, 0.05, 0.003),
    lab.size = 3.5,
    colors = list(
      c("#56B4E9", "white", "#999999"),
      c("#0072B2", "white", "#D55E00"),
      c("#CC79A7", "white", "#F0E442"),
      c("#56B4E9", "white", "#D55E00"),
      c("#999999", "white", "#0072B2")
    ),
    ggstatsplot.theme = list(FALSE),
    ggtheme = list(
      ggplot2::theme_grey,
      ggplot2::theme_classic,
      ggplot2::theme_minimal,
      ggplot2::theme_bw,
      ggplot2::theme_dark
    )
  ),
  .f = ggstatsplot::ggcorrmat
)

# combining all individual plots from the list into a single plot using combine_plots function
ggstatsplot::combine_plots(
  plotlist = plot_list,
  title.text = "Relationship between diamond attributes and price across cut",
  title.size = 16,
  title.color = "red",
  caption.text = "Dataset: Diamonds from ggplot2 package",
  caption.size = 14,
  caption.color = "blue",
  labels = c("(a)", "(b)", "(c)", "(d)", "(e)"),
  nrow = 3,
  ncol = 2
)

As can be seen from this example, combination of purrr and ggstatsplot much more flexibility in analyzing the data and preparing a combined plot across studies or conditions.

Suggestions

If you find any bugs or have any suggestions/remarks, please file an issue on GitHub: https://github.com/IndrajeetPatil/ggstatsplot/issues