inst/tinytest/test-rounding.R

source("helpers.R")
requiet("tinysnapshot")
using("tinysnapshot")

# rounding cleans up NaN inside \\num
dat <- mtcars
dat$cyl <- factor(dat$cyl)
expect_snapshot_print(
    datasummary(cyl + mpg ~ SD + N, data = dat, output = "latex"),
    "rounding-datasummary_latex")

# named list
mod <- lm(mpg ~ hp + factor(cyl), data = mtcars)
tab <- modelsummary(
    mod,
    statistic = c("SE: {std.error}", "conf.int"),
    fmt = fmt_statistic(estimate = 0, std.error = 4),
    output = "data.frame")
expect_equivalent(
    tab[["(1)"]][1:3],
    c("29", "SE: 1.5878", "[25.398, 31.903]"))

# glue function
m <- glm(am ~ mpg, data = mtcars, family = binomial)
tab <- modelsummary(
    m,
    output = "data.frame",
    fmt = NULL,
    estimate = "{round(exp(estimate), 5)}",
    statistic = "{round(exp(estimate) * std.error, 3)}")
known <- c("0.00136", "0.003", "1.35938", "0.156")
expect_equivalent(tab[["(1)"]][1:4], known)

# bugfix: format() is not vectorized (in fact, it is vectorized and vincent just didn't understand the function)
mod <- lm(mpg ~ I(hp * 1000) + drat, data = mtcars)
f <- function(x) format(x, digits = 3, nsmall = 2, scientific = FALSE)
tab <- modelsummary(mod, fmt = f, statistic = NULL, gof_map = NA, output = "data.frame")
expect_equivalent(tab[["(1)"]], c("10.7898612", "-0.0000518", "4.6981578"))

# very small numbers
mod <- lm(mpg ~ I(hp * 1000) + drat, data = mtcars)
tab <- modelsummary(mod, fmt = 2, output = "dataframe")
expect_true("10.79" %in% tab[["(1)"]])
tab <- modelsummary(mod, fmt = fmt_significant(2), output = "dataframe")
expect_true("-5.2e-05" %in% tab[["(1)"]])

# per term rounding
mod <- lm(mpg ~ qsec + factor(cyl), data = mtcars)
tab <- modelsummary(
    mod,
    output = "dataframe",
    statistic = NULL,
    gof_map = NA,
    fmt = fmt_term("(Intercept)" = 0, "qsec" = 2)
)
expect_equivalent(c("35", "-0.44", "-7.431", "-12.603"), tab[["(1)"]])
vincentarelbundock/gtsummary documentation built on Nov. 6, 2024, 11:07 p.m.