In j3ypi/inductive: A pipe-friendly and consistent framework for frequentist statistics in R

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

Example of a t-test

Before we get started lets load the three packages we need.

library(inductive)
library(magrittr)
library(purrr)

Within the inductive package there is a data set called big_five, which can be loaded with the data() function without further arguments when the library() call was executed before.

data(big_five)

big_five %>%  
  head() %>%  
  knitr::kable() 

There are three different ways to conduct Student’s t-Test in the inductive package. Note that all the information provided for the t-test can be applied to any other function from the package. Lets say we want to investigate whether there is a difference between the average degree of the big five personality trait extraversion compared to neuroticism. The recommend way of passing those columns to the function is the formula syntax, which is well-known from base R functions within the stats package like lm() or aov(). Up until now you could only use the formula syntax for a t-test when you pass a grouping variable like gender as a second argument. Now it can be used with two metrical columns as well. All arguments from stats::t.test() can be passed to sta_t(). To conduct a t-test and not a Welch test one has to set the var.equal argument to TRUE.

big_five %>% 
  sta_t(Extraversion ~ Neuroticism, var.equal = TRUE)

The option to use the formula syntax with a grouping variable still exists. The function is more strict though, so that the grouping variable must be a factor and not just a string.

big_five %>% 
  sta_t(Extraversion ~ Gender)

For those who prefer the known way of passing the columns as x and y separately, this option is included as well.

big_five %>% 
  sta_t(x = Extraversion, y = Neuroticism) 

Even the way of passing the arguments without a data argument as a vector to the function (e.g. with the dollar-operator) works exactly like it does for the corresponding base R functions from the stats package.

sta_t(x = big_five$Extraversion, y = big_five$Neuroticism)

Note that the formula syntax will always be preferred. If the data is provided alongside the formula argument, x and y are ignored even when they are passed to function additionally. Other combinations which are not shown above will result in an error.

The obj argument

The only thing you see when you conduct a statistical test in the inductive package is a clean result table. But behind the scenes all functions return a list with two elements by default. To illustrate that lets save the results from above to the variable res.

res <- big_five %>% 
  sta_t(Extraversion ~ Neuroticism, var.equal = TRUE)

Now we can inspect those two elements with purrr's function pluck() where one could either pass numbers (element 1 or 2) or the name of the list elements (result, obj).

res %>% 
  pluck("result")

Of course, one could also use the dollar-operator or [[ to access those list elements. The first element result contains the data.frame and the second element obj the object returned by the function from the stats package the function is wrapped around (e.g. stats::t.test()).

res %>% 
  pluck("obj")

There are certain scenarios where the default output of the obj element comes in handy. For example to conduct post-hoc tests after the calculation of an ANOVA or when you want to return additional fit indices like information criteria from you linear model with broom::glance(). To avoid this behavior and to save only the result table, one would need to set the obj argument to FALSE, which is practical e.g. when working with tidyr::nest() and purrr::map, if the objects are extremely large or if you want export the result table to LaTeX with xtable::xtable() or to Word with rMarkdown.

big_five %>% 
  sta_t(Extraversion ~ Neuroticism, obj = FALSE) %>% 
  class()

All functions return the result with an additional stats class, which is only necessary for the adapted generic print method and can be ignored otherwise.

j3ypi/inductive documentation built on Feb. 7, 2020, 12:37 p.m.