nice_lm_contrasts {rempsyc}

Easy planned contrasts using lm models

Description

Easily compute planned contrast analyses (pairwise comparisons similar to t-tests but more powerful when more than 2 groups), and format in publication-ready format. In this particular case, the confidence intervals are bootstraped on chosen effect size (default to Cohen's d).

Usage

nice_lm_contrasts(
  model,
  group,
  data,
  p_adjust = "none",
  effect.type = "cohens.d",
  bootstraps = 2000,
  ...
)

Arguments

model

The model to be formatted.

group

The group for the comparison.

data

The data frame.

p_adjust

Character: adjustment method (e.g., "bonferroni") – added to options

effect.type

What effect size type to use. One of "cohens.d" (default), "akp.robust.d", "unstandardized", "hedges.g", "cohens.d.sigma", or "r".

bootstraps

The number of bootstraps to use for the confidence interval

...

Arguments passed to bootES::bootES.

Details

Statistical power is lower with the standard t test compared than it is with the planned contrast version for two reasons: a) the sample size is smaller with the t test, because only the cases in the two groups are selected; and b) in the planned contrast the error term is smaller than it is with the standard t test because it is based on all the cases (source).

The effect size and confidence interval are calculated via bootES::bootES, and correct for contrasts but not for covariates and other predictors. Because this method uses bootstrapping, it is recommended to set a seed before using for reproducibility reasons (e.g., sed.seet(100)).

Does not for the moment support nested comparisons for marginal means, only a comparison of all groups. For nested comparisons, please use emmeans::contrast() directly, or for the easystats equivalent, modelbased::estimate_contrasts().

When using nice_lm_contrasts(), please use as.factor() outside the lm() formula, or it will lead to an error.

Value

A dataframe, with the selected dependent variable(s), comparisons of interest, degrees of freedom, t-values, p-values, Cohen's d, and the lower and upper 95% confidence intervals of the effect size (i.e., dR).

See Also

nice_contrasts, Tutorial: https://rempsyc.remi-theriault.com/articles/contrasts

Examples

# Make and format model (group need to be a factor)
mtcars2 <- mtcars
mtcars2$cyl <- as.factor(mtcars2$cyl)
model <- lm(mpg ~ cyl + wt * hp, mtcars2)
set.seed(100)
nice_lm_contrasts(model, group = "cyl", data = mtcars, bootstraps = 500)

# Several models at once
mtcars2$gear <- as.factor(mtcars2$gear)
model2 <- lm(qsec ~ cyl, data = mtcars2)
my.models <- list(model, model2)
set.seed(100)
nice_lm_contrasts(my.models, group = "cyl", data = mtcars, bootstraps = 500)

# Now supports more than 3 levels
mtcars2$carb <- as.factor(mtcars2$carb)
model <- lm(mpg ~ carb + wt * hp, mtcars2)
set.seed(100)
nice_lm_contrasts(model, group = "carb", data = mtcars2, bootstraps = 500)