Describe regression model (GLM, GLMer, lm, lm.circular, ...)

Usage

# S3 method for class 'glm'
apa(
  obj,
  term = NULL,
  dtype = 1,
  b.digits = 2,
  t.digits = 2,
  test.df = FALSE,
  p.as.number = FALSE,
  term.pattern = NULL,
  eff.size = FALSE,
  adj.digits = FALSE,
  ...
)

# S3 method for class 'lm'
apa(
  obj,
  term = NULL,
  dtype = 1,
  b.digits = 2,
  t.digits = 2,
  test.df = FALSE,
  p.as.number = FALSE,
  term.pattern = NULL,
  eff.size = FALSE,
  adj.digits = FALSE,
  ...
)

# S3 method for class 'glmerMod'
apa(
  obj,
  term = NULL,
  dtype = 1,
  b.digits = 2,
  t.digits = 2,
  test.df = FALSE,
  p.as.number = FALSE,
  term.pattern = NULL,
  eff.size = FALSE,
  adj.digits = FALSE,
  ...
)

# S3 method for class 'lmerModLmerTest'
apa(
  obj,
  term = NULL,
  dtype = 1,
  b.digits = 2,
  t.digits = 2,
  test.df = FALSE,
  p.as.number = FALSE,
  term.pattern = NULL,
  eff.size = FALSE,
  adj.digits = FALSE,
  ...
)

Arguments

obj: model object from stats::glm, stats::lm, lme4::lmer, lme4::glmer, lmerTest::lmer, etc.
term: model term to describe (a string with the term name or its sequential number); if NULL, returns a formatted summary for all coefficients
dtype: description type (1: t, p; 2: B(SE), p; 3: B, SE, t, p; or other: B (SE), t)
b.digits: how many digits to use for B and SE
t.digits: how many digits to use for t
test.df: should we include degrees of freedom in description?
p.as.number: should the p-values be transformed to numbers (TRUE) or shown as strings (FALSE)?
term.pattern: return only the model terms matching the regex pattern (grepl is used)
eff.size: should we include effect size (currently implemented only for simple regression)?
adj.digits: automatically adjusts digits so that B or SE would not show up as "0.00"
...: other parameters passed to format_results

Value

A character result for a selected term. Otherwise, a data.frame with columns B (coefficient estimate), SE (standard error), Stat (test statistic), p (p-value), eff (term name), and str (complete APA-formatted result). The summary columns are character vectors except that p is numeric when p.as.number = TRUE. term.pattern filters the rows of this data frame.

Examples

# Using the Animals dataset from MASS package
Animals <- MASS::Animals
Animals$body <- log(Animals$body)
Animals$brain <- log(Animals$brain)

# Fit a linear model
fit <- lm(brain ~ body, Animals)

# Format results in different styles
apa(fit, "body")
#> [1] "_t_ = 6.35, _p_ < .001"
apa(fit, "(Intercept)")
#> [1] "_t_ = 6.18, _p_ < .001"
apa(fit, "body", 2)   # B(SE), p format
#> [1] "_B_ = 0.50 (0.08), _p_ < .001"
apa(fit, "body", 3)   # B, SE, t, p format
#> [1] "_B_ = 0.50, _SE_ = 0.08, _t_ = 6.35, _p_ < .001"
apa(fit, "body", 4)   # B(SE), t format
#> [1] "_B_ = 0.50 (0.08), _t_ = 6.35"
apa(fit, "body", 3, test.df = TRUE)  # Include df in the output
#> [1] "_B_ = 0.50, _SE_ = 0.08, _t_(26) = 6.35, _p_ < .001"

# Full model summary for all coefficients
apa(fit)
#>                B   SE Stat      p         eff                    str
#> (Intercept) 2.55 0.41 6.18 < .001 (Intercept) _t_ = 6.18, _p_ < .001
#> body        0.50 0.08 6.35 < .001        body _t_ = 6.35, _p_ < .001

# With effect size (requires rockchalk package)
if (requireNamespace("rockchalk", quietly = TRUE)) {
  print(apa(fit, "body", 4, eff.size = TRUE))
}
#> The deltaR-square values: the change in the R-square
#>       observed when a single term is removed.
#> Same as the square of the 'semi-partial correlation coefficient'
#>      deltaRsquare
#> body    0.6076101
#> [1] "_B_ = 0.50 (0.08), _t_ = 6.35, _R__{part}^2= .61"

if (requireNamespace("lmerTest", quietly = TRUE)) {
  fm <- lmerTest::lmer(Reaction ~ Days + (Days | Subject), lme4::sleepstudy)
  print(apa(fm, "Days", test.df = TRUE))
}
#> [1] "_t_(17.00) =  6.77, _p_ < .001"

if (requireNamespace("lme4", quietly = TRUE)) {
  gm <- lme4::glmer(
    cbind(incidence, size - incidence) ~ period + (1 | herd),
    data = lme4::cbpp,
    family = binomial
  )
  print(apa(gm, "period2"))
}
#> [1] "_Z_ = -3.27, _p_ = .001"