Over the years, Iβve written a lot of one-off functions for formatting numbers in RMarkdown documents. This packages collects them in a single location.
You can install printy from github with:
# install.packages("remotes")
remotes::install_github("tjmahr/printy")
fmt_fix_digits()
prints a number with n digits of precision. R numbers
lose precision when converted to strings. This function converts the
numbers to strings and keeps precision. (Itβs a wrapper for
sprintf()
.)
library(dplyr)
library(printy)
test_cor <- cor(mtcars[, 1:4])
# Typical loss of trailing zeroes
test_cor[1:4, 3] |> round(2) |> as.character()
#> [1] "-0.85" "0.9" "1" "0.79"
test_cor[1:4, 3] |> fmt_fix_digits(2)
#> [1] "-0.85" "0.90" "1.00" "0.79"
fmt_leading_zero()
removes a leading zero on numbers that are bounded
between β1 and 1, such as correlations or p-values.
fmt_leading_zero(c(-0.3, 0.4, 1))
#> [1] "-.3" ".4" "1"
fmt_minus_sign()
formats negative numbers with a minus sign.
fmt_minus_sign(c(1, 2, -3, -0.4, -pi))
#> [1] "1" "2"
#> [3] "−3" "−0.4"
#> [5] "−3.14159265358979"
Putting it all together: Print a correlation matrix with 2 digits, no leading zero and with minus signs.
fmt_correlation <- function(xs, digits = 2) {
xs |> fmt_fix_digits(digits) |> fmt_leading_zero() |> fmt_minus_sign()
}
test_cor |>
as.data.frame() |>
tibble::rownames_to_column(".rowname") |>
tibble::as_tibble() |>
mutate(
across(-.rowname, fmt_correlation)
) |>
rename(` ` = .rowname) |>
knitr::kable(align = "lrrrr")
mpg | cyl | disp | hp | |
---|---|---|---|---|
mpg | 1.00 | β.85 | β.85 | β.78 |
cyl | β.85 | 1.00 | .90 | .83 |
disp | β.85 | .90 | 1.00 | .79 |
hp | β.78 | .83 | .79 | 1.00 |
fmt_p_value()
formats p-values with n digits of precision, with no
leading zero, and with very small values being printed with a <
sign.
p <- c(1, 0.1, 0.01, 0.001, 0.0001)
fmt_p_value(p, digits = 2)
#> [1] "1.00" ".10" ".01" "< .01" "< .01"
fmt_p_value(p, digits = 3)
#> [1] "1.000" ".100" ".010" ".001" "< .001"
fmt_p_value_md()
formats p-values in markdown with nice defaults.
- Use 3 digits of precision for values less than .06
- Otherwise, use 2 digits of precision.
- Include p in markdown
p <- c(1, 0.1, 0.06, 0.059, 0.051, 0.01, 0.001, 0.0001)
fmt_p_value_md(p)
#> [1] "*p* > .99" "*p* = .10" "*p* = .06" "*p* = .059"
#> [5] "*p* = .051" "*p* = .010" "*p* = .001" "*p* < .001"
These render as: p > .99, p = .10, p = .06, p = .059, p = .051, p = .010, p = .001, p < .001.
fmt_effect_md()
is an experimental function for getting model effects
formatted in markdown. You give the function a model, an effect and a
string listing the quantities you want.
model <- lm(breaks ~ wool * tension, warpbreaks)
summary(model)
#>
#> Call:
#> lm(formula = breaks ~ wool * tension, data = warpbreaks)
#>
#> Residuals:
#> Min 1Q Median 3Q Max
#> -19.5556 -6.8889 -0.6667 7.1944 25.4444
#>
#> Coefficients:
#> Estimate Std. Error t value Pr(>|t|)
#> (Intercept) 44.556 3.647 12.218 2.43e-16 ***
#> woolB -16.333 5.157 -3.167 0.002677 **
#> tensionM -20.556 5.157 -3.986 0.000228 ***
#> tensionH -20.000 5.157 -3.878 0.000320 ***
#> woolB:tensionM 21.111 7.294 2.895 0.005698 **
#> woolB:tensionH 10.556 7.294 1.447 0.154327
#> ---
#> Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
#>
#> Residual standard error: 10.94 on 48 degrees of freedom
#> Multiple R-squared: 0.3778, Adjusted R-squared: 0.3129
#> F-statistic: 5.828 on 5 and 48 DF, p-value: 0.0002772
# default to: b (beta), e (error), s (statistic), p (p value)
fmt_effect_md(model, "woolB", "besp")
#> [1] "*b* = −16.33, SE = 5.16, *t* = −3.17, *p* = .003"
b = β16.33, SE = 5.16, t = β3.17, p = .003
# Just a subset of them
fmt_effect_md(model, "woolB", terms = "bp")
#> [1] "*b* = −16.33, *p* = .003"
b = β16.33, p = .003
# B for labeled b
fmt_effect_md(model, "woolB", terms = "Bp", b_lab = "Wool B")
#> [1] "*b*<sub>Wool B</sub> = −16.33, *p* = .003"
bWool B = β16.33, p = .003
# i for interval
fmt_effect_md(model, "woolB", terms = "bi")
#> [1] "*b* = −16.33, 95% CI = [−26.70, −5.96]"
b = β16.33, 95% CI = [β26.70, β5.96]
# S for statistic with df
fmt_effect_md(model, "woolB", terms = "bSp")
#> [1] "*b* = −16.33, *t*(48) = −3.17, *p* = .003"
b = β16.33, t(48) = β3.17, p = .003
# extra digits (except for p-values; those go through `fmt_p_value_md()`)
fmt_effect_md(model, "woolB", terms = "bep", digits = 6)
#> [1] "*b* = −16.333333, SE = 5.157299, *p* = .003"
b = β16.333333, SE = 5.157299, p = .003
These are the currently supported models:
lm()
lme4::lmer()
For lme4 models, Wald confidence intervals are provided. For p-values, the KenwoodβRoger approximation for the degrees of freedom is used by default. We can also choose a method supported by the parameters package.
library(lme4)
data(Machines, package = "nlme")
m <- lmer(score ~ 1 + Machine + (Machine | Worker), data = Machines)
# Default is Kenward
fmt_effect_md(m, "MachineB", terms = "beSp")
#> [1] "*b* = 7.97, SE = 2.42, *t*(5) = 3.29, *p* = .022"
fmt_effect_md(m, "MachineB", terms = "beSp", p_value_method = "kenward")
#> [1] "*b* = 7.97, SE = 2.42, *t*(5) = 3.29, *p* = .022"
# Note residual degrees of freedom for Wald
fmt_effect_md(m, "MachineB", terms = "beSp", p_value_method = "wald")
#> [1] "*b* = 7.97, SE = 2.42, *t*(44) = 3.29, *p* = .002"
# This example doesn't find differences between Satterthwaite and Kenward
fmt_effect_md(m, "MachineB", terms = "beSp", p_value_method = "satterthwaite")
#> [1] "*b* = 7.97, SE = 2.42, *t*(5) = 3.29, *p* = .022"
We can also format effects from glmer()
models. "S"
is not supported
because the model summary uses z statistics, not t statistics.
gm1 <- glmer(
cbind(incidence, size - incidence) ~ period + (1 | herd),
data = cbpp,
family = binomial
)
round(coef(summary(gm1)), 3)
#> Estimate Std. Error z value Pr(>|z|)
#> (Intercept) -1.398 0.231 -6.048 0.000
#> period2 -0.992 0.303 -3.272 0.001
#> period3 -1.128 0.323 -3.495 0.000
#> period4 -1.580 0.422 -3.743 0.000
fmt_effect_md(gm1, "period2", terms = "bespi")
#> [1] "*b* = −0.99, SE = 0.30, *z* = −3.27, *p* = .001, 95% CI = [−1.59, −0.40]"
# Don't use S here
fmt_effect_md(gm1, "period2", terms = "beSp")
#> Error in get_terms.glmerMod(model, effect, terms, ci_width = ci_width, : S is not supported for glmer models
I use fmt_
for formatting functions. The other convention in the
package is skel_
to plug values into a formatting skeleton.
skel_conf_interval_pair()
creates a confidence interval from two
numbers.
skel_conf_interval_pair(c(1, 2))
#> [1] "[1, 2]"
skel_conf_interval()
is the vectorized version. It is suitable for
working on columns of numbers.
model <- lm(breaks ~ wool * tension, warpbreaks)
ci_starts <- confint(model)[, 1] |>
fmt_fix_digits(2) |>
fmt_minus_sign()
ci_ends <- confint(model)[, 2] |>
fmt_fix_digits(2) |>
fmt_minus_sign()
skel_conf_interval(ci_starts, ci_ends)
#> [1] "[37.22, 51.89]" "[−26.70, −5.96]"
#> [3] "[−30.93, −10.19]" "[−30.37, −9.63]"
#> [5] "[6.45, 35.78]" "[−4.11, 25.22]"
skel_stat_n_value_pair()
creates t-test-like or correlation-like
statistic from a vector of two numbers.
skel_stat_n_value_pair(c("20", "2.0"))
#> [1] "t(20) = 2.0"
skel_stat_n_value_pair(c("39", ".98"), stat = "*r*")
#> [1] "*r*(39) = .98"
skel_se()
and skel_ci()
are shorthand functions to help with inline
reporting.
skel_se(c(10, 4))
#> [1] "SE = 10" "SE = 4"
skel_ci(c("[1, 2]"))
#> [1] "95% CI = [1, 2]"
skel_ci(c("[1, 2]"), ci_width = 90)
#> [1] "90% CI = [1, 2]"
One thing Iβve had to do a lot is summarize mixed effects models fit
with lme4. This package provides pretty_lme4_ranefs()
which creates a
dataframe random effect variances and covariances like those printed by
summary()
.
For example, we can fit the model.
library(lme4)
model <- lmer(Reaction ~ Days + (Days | Subject), sleepstudy)
summary(model)
#> Linear mixed model fit by REML ['lmerMod']
#> Formula: Reaction ~ Days + (Days | Subject)
#> Data: sleepstudy
#>
#> REML criterion at convergence: 1743.6
#>
#> Scaled residuals:
#> Min 1Q Median 3Q Max
#> -3.9536 -0.4634 0.0231 0.4634 5.1793
#>
#> Random effects:
#> Groups Name Variance Std.Dev. Corr
#> Subject (Intercept) 612.10 24.741
#> Days 35.07 5.922 0.07
#> Residual 654.94 25.592
#> Number of obs: 180, groups: Subject, 18
#>
#> Fixed effects:
#> Estimate Std. Error t value
#> (Intercept) 251.405 6.825 36.838
#> Days 10.467 1.546 6.771
#>
#> Correlation of Fixed Effects:
#> (Intr)
#> Days -0.138
pretty_lme4_ranefs()
creates the following dataframe.
pretty_lme4_ranefs(model)
#> Group Parameter Variance SD Correlations
#> 1 Subject (Intercept) 612.10 24.74 1.00
#> 2 Days 35.07 5.92 .07 1.00
#> 3 Residual 654.94 25.59
Which in markdown renders as
knitr::kable(
pretty_lme4_ranefs(model),
align = c("l", "l", "r", "r", "r")
)
Group | Parameter | Variance | SD | Correlations | |
---|---|---|---|---|---|
Subject | (Intercept) | 612.10 | 24.74 | 1.00 | |
Days | 35.07 | 5.92 | .07 | 1.00 | |
Residual | 654.94 | 25.59 |
Hereβs a dumb model with a lot going on in the random effects.
model <- lmer(mpg ~ wt * hp + (drat | gear) + (hp * cyl | am), mtcars)
#> boundary (singular) fit: see help('isSingular')
model
#> Linear mixed model fit by REML ['lmerMod']
#> Formula: mpg ~ wt * hp + (drat | gear) + (hp * cyl | am)
#> Data: mtcars
#> REML criterion at convergence: 152.7432
#> Random effects:
#> Groups Name Std.Dev. Corr
#> gear (Intercept) 1.556809
#> drat 0.166292 -1.00
#> am (Intercept) 1.940271
#> hp 0.004055 -0.96
#> cyl 0.456219 -0.98 0.93
#> hp:cyl 0.001508 0.95 -0.94 -0.99
#> Residual 2.113554
#> Number of obs: 32, groups: gear, 3; am, 2
#> Fixed Effects:
#> (Intercept) wt hp wt:hp
#> 48.98745 -7.80904 -0.12118 0.02737
#> optimizer (nloptwrap) convergence code: 0 (OK) ; 0 optimizer warnings; 1 lme4 warnings
knitr::kable(
pretty_lme4_ranefs(model),
align = c("l", "l", "r", "r", "r", "r", "r", "r", "r")
)
Group | Parameter | Variance | SD | Correlations | |||
---|---|---|---|---|---|---|---|
am | (Intercept) | 3.76 | 1.94 | 1.00 | |||
hp | 0.00 | 0.00 | β.96 | 1.00 | |||
cyl | 0.21 | 0.46 | β.98 | .93 | 1.00 | ||
hp:cyl | 0.00 | 0.00 | .95 | β.94 | β.99 | 1.00 | |
gear | (Intercept) | 2.42 | 1.56 | 1.00 | |||
drat | 0.03 | 0.17 | β1.00 | 1.00 | |||
Residual | 4.47 | 2.11 |