dlookr (version 0.6.3)

correlate: Compute the correlation coefficient between two variable

Description

The correlate() compute the correlation coefficient for numerical or categorical data.

Usage

correlate(.data, ...)
# S3 method for data.frame
correlate(
  .data,
  ...,
  method = c("pearson", "kendall", "spearman", "cramer", "theil")
)
# S3 method for grouped_df
correlate(
  .data,
  ...,
  method = c("pearson", "kendall", "spearman", "cramer", "theil")
)
# S3 method for tbl_dbi
correlate(
  .data,
  ...,
  method = c("pearson", "kendall", "spearman", "cramer", "theil"),
  in_database = FALSE,
  collect_size = Inf
)

Value

An object of correlate class.

Arguments

.data

a data.frame or a grouped_df or a tbl_dbi.

...

one or more unquoted expressions separated by commas. You can treat variable names like they are positions. Positive values select variables; negative values to drop variables. If the first expression is negative, correlate() will automatically start with all variables. These arguments are automatically quoted and evaluated in a context where column names represent column positions. They support unquoting and splicing.

See vignette("EDA") for an introduction to these concepts.

method

a character string indicating which correlation coefficient (or covariance) is to be computed. One of "pearson" (default), "kendall", or "spearman": can be abbreviated. For numerical variables, one of "pearson" (default), "kendall", or "spearman": can be used as an abbreviation. For categorical variables, "cramer" and "theil" can be used. "cramer" computes Cramer's V statistic, "theil" computes Theil's U statistic.

in_database

Specifies whether to perform in-database operations. If TRUE, most operations are performed in the DBMS. if FALSE, table data is taken in R and operated in-memory. Not yet supported in_database = TRUE.

collect_size

a integer. The number of data samples from the DBMS to R. Applies only if in_database = FALSE.

correlate class

The correlate class inherits the tibble class and has the following variables.:

• var1 : names of numerical variable

• var2 : name of the corresponding numeric variable

• coef_corr : Correlation coefficient

When method = "cramer", data.frame with the following variables is returned.

• var1 : names of numerical variable

• var2 : name of the corresponding numeric variable

• chisq : the value the chi-squared test statistic

• df : the degrees of freedom of the approximate chi-squared distribution of the test statistic

• pval : the p-value for the test

• coef_corr : theil's U correlation coefficient (Uncertainty Coefficient).

Details

This function is useful when used with the group_by() function of the dplyr package. If you want to compute by level of the categorical data you are interested in, rather than the whole observation, you can use grouped_df as the group_by() function. This function is computed stats::cor() function by use = "pairwise.complete.obs" option for numerical variable. And support categorical variable with theil's U correlation coefficient and Cramer's V correlation coefficient.

See Also

cor, summary.correlate, plot.correlate.

Examples

# \donttest{
# Correlation coefficients of all numerical variables
tab_corr <- correlate(heartfailure)
tab_corr

# Select the variable to compute
correlate(heartfailure, "creatinine", "sodium")

# Non-parametric correlation coefficient by kendall method
correlate(heartfailure, creatinine, method = "kendall")

# theil's U correlation coefficient (Uncertainty Coefficient)
tab_corr <- correlate(heartfailure, anaemia, hblood_pressure, method = "theil")
tab_corr
   
# Using dplyr::grouped_dt
library(dplyr)

gdata <- group_by(heartfailure, smoking, death_event)
correlate(gdata)

# Using pipes ---------------------------------
# Correlation coefficients of all numerical variables
heartfailure %>%
  correlate()
  
# Non-parametric correlation coefficient by spearman method
heartfailure %>%
  correlate(creatinine, sodium, method = "spearman")
 
# ---------------------------------------------
# Correlation coefficient
# that eliminates redundant combination of variables
heartfailure %>%
  correlate() %>%
  filter(as.integer(var1) > as.integer(var2))

# Using pipes & dplyr -------------------------
# Compute the correlation coefficient of 'creatinine' variable by 'smoking'
# and 'death_event' variables. And extract only those with absolute
# value of correlation coefficient is greater than 0.2
heartfailure %>%
  group_by(smoking, death_event) %>%
  correlate(creatinine) %>%
  filter(abs(coef_corr) >= 0.2)

# extract only those with 'smoking' variable level is "Yes",
# and compute the correlation coefficient of 'Sales' variable
# by 'hblood_pressure' and 'death_event' variables.
# And the correlation coefficient is negative and smaller than 0.5
heartfailure %>%
  filter(smoking == "Yes") %>%
  group_by(hblood_pressure, death_event) %>%
  correlate(creatinine) %>%
  filter(coef_corr < 0) %>%
  filter(abs(coef_corr) > 0.5)
# }

# If you have the 'DBI' and 'RSQLite' packages installed, perform the code block:
if (FALSE) {
library(dplyr)
# connect DBMS
con_sqlite <- DBI::dbConnect(RSQLite::SQLite(), ":memory:")

# copy heartfailure to the DBMS with a table named TB_HEARTFAILURE
copy_to(con_sqlite, heartfailure, name = "TB_HEARTFAILURE", overwrite = TRUE)

# Using pipes ---------------------------------
# Correlation coefficients of all numerical variables
con_sqlite %>% 
  tbl("TB_HEARTFAILURE") %>% 
  correlate()

# Using pipes & dplyr -------------------------
# Compute the correlation coefficient of creatinine variable by 'hblood_pressure'
# and 'death_event' variables.
con_sqlite %>% 
  tbl("TB_HEARTFAILURE") %>% 
  group_by(hblood_pressure, death_event) %>%
  correlate(creatinine) 

# Disconnect DBMS   
DBI::dbDisconnect(con_sqlite)
}