corrcat facilitates to explore correlations between categorical and ordinal variables in a tidy way using data.table as back-end.
The development version from GitHub:
# install.packages("pak")
pak::pak("AngelFelizR/corrcat")To use the functions of this library you need to input a data.table, tibble or data.frame, then the functions will work to calculate all possible values of the metric returning as a result other data frame with all the metrics.
Validating correlations is an important step for every EDA, if all the columns are numeric is really easy to use the pearson correlation to make our estimation, there are also alternatives for categorical variables that can simplify a lot our analysis.
Data
set.seed(2024)
df <- data.frame(
gender = sample(c("male", "female"), 100, replace = TRUE) |> factor(),
race = sample(c("white", "black", "asian", "other"), 100, replace = TRUE),
education = sample(c("high school", "college", "graduate"), 100, replace = TRUE),
income = sample(c("low", "medium", "high"), 100, replace = TRUE)
)
set.seed(NULL)
head(df)
# gender race education income
# 1 female other high school high
# 2 male black high school medium
# 3 male white graduate low
# 4 female other graduate high
# 5 male white high school high
# 6 male other graduate medium
Cramer correlation for each pair of columns
cramerV_df(df)
# V1 V2 Cramer.V
# 1 gender race 0.15370
# 2 gender education 0.07143
# 3 gender income 0.10350
# 4 race education 0.18480
# 5 race income 0.18920
# 6 education income 0.19900
cohenW correlation for each pair of columns
cohenW_df(df)
# V1 V2 Cohen.w
# 1 gender race 0.15370
# 2 gender education 0.07143
# 3 gender income 0.10350
# 4 race education 0.26130
# 5 race income 0.26750
# 6 education income 0.28140