Skip to contents

Automatically select statistical tests for quantitative variables

Source: R/find.test.R
find.test.Rd

This function evaluates quantitative variables across stratification groups and automatically determines the appropriate parametric or non-parametric hypothesis test. See Details for an exhaustive explanation of the criteria followed. Designed to work seamlessly with tbl_summary, this function generates the input lists for arguments test and test.args of the function add_p, specifying the recommended test function and any additional arguments different from the defaults.

Usage

find.test(data, variables, strata, alpha = 0.05)

Arguments

data

name of the dataset, of class data.frame, containing at least two columns (quantitative variable, strata).

variables

column name(s) of the quantitative variable(s) to analyse. Can be specified as a character string (with quotes) or as an unquoted symbol.

strata

strata indicator column name. Can be specified as a character string (with quotes) or as an unquoted symbol. Must be a factor with at least 2 levels.

alpha

significance level threshold used for normality (shapiro.test) and homoscedasticity (var.test or bartlett.test) testing.
The default value is 0.05.

Value

A list of length 2 intended to be passed directly to arguments test and test.args of the function add_p when using tbl_summary:

test

a list of formulas matching each variable name to its selected test string (e.g., age ~ "t.test").

test.args

a list of formulas specifying additional test parameters (e.g., age ~ list(var.equal = TRUE)). Variables requiring no additional parameters are omitted from this second list.

Details

If any level of strata has less than five non-missing observations, no statistical test is performed for that variable, and a warning is issued. Otherwise, the appropriate test is selected based on the following criteria:

Two Groups (strata with 2 levels):

  • t-test (t.test): selected if sample sizes in both groups are \(\ge 30\) and Shapiro-Wilk tests indicate normality in both groups. Variance equality is tested using an F-test.

  • Mann-Whitney test (wilcox.test): selected if sample sizes are small, this is \(< 30\) in some group, or if normality is rejected in either group.

More than Two Groups (strata with > 2 levels):

  • ANOVA (oneway.test): selected if ANOVA model residuals adhere to a normal distribution according to the Shapiro-Wilk test. Homoscedasticity is tested via Bartlett's test.

  • Kruskal-Wallis test (kruskal.test): selected if ANOVA residuals deviate from normality.

Examples

df <- as.data.frame(gtsummary::trial)
df$trt <- factor(df$trt)
df$grade <- factor(df$grade)

# Two levels variable (trt: Drug A, Drug B)
tests2 <- find.test(df, variables = c(age, marker), strata = trt)
tests2

# Use outputs directly with tbl_summary()
gtsummary::tbl_summary(df, include = c(age, marker), by = trt) |>
  gtsummary::add_p(test = tests2$test, test.args = tests2$test.args)

# Three levels variable (grade: I, II, III)
tests3 <- find.test(df, variables = c(age, marker), strata = grade)
tests3

# Use outputs directly with tbl_summary()
gtsummary::tbl_summary(df, include = c(age, marker), by = grade) |>
  gtsummary::add_p(test = tests3$test, test.args = tests3$test.args)
  
# Using function show.p() from BiMaUmisc package
gtsummary::tbl_summary(df, include = c(age, marker), by = grade) |>
  gtsummary::add_p(test = tests3$test, test.args = tests3$test.args,
                   pvalue_fun = function(x) show.p(x))