add_p() and add_difference()Below is a listing of tests available internally within gtsummary.
Tests listed with ... may have additional arguments
passed to them using add_p(test.args=). For example, to
calculate a p-value from t.test() assuming equal variance, use
tbl_summary(trial, by = trt) %>% add_p(age ~ "t.test", test.args = age ~ list(var.equal = TRUE))
| alias | description | pseudo-code |
"t.test" |
t-test | t.test(variable ~ as.factor(by), data = data, conf.level = 0.95, ...) |
"aov" |
One-way ANOVA | aov(variable ~ as.factor(by), data = data) %>% summary() |
"kruskal.test" |
Kruskal-Wallis test | kruskal.test(data[[variable]], as.factor(data[[by]])) |
"wilcox.test" |
Wilcoxon rank-sum test | wilcox.test(variable ~ as.factor(by), data = data, ...) |
"chisq.test" |
chi-square test of independence | chisq.test(x = data[[variable]], y = as.factor(data[[by]]), ...) |
"chisq.test.no.correct" |
chi-square test of independence | chisq.test(x = data[[variable]], y = as.factor(data[[by]]), correct = FALSE) |
"fisher.test" |
Fisher's exact test | fisher.test(data[[variable]], as.factor(data[[by]]), conf.level = 0.95, ...) |
"mcnemar.test" |
McNemar's test | mcnemar.test(data[[variable]], data[[by]], conf.level = 0.95, ...) |
"lme4" |
random intercept logistic regression | lme4::glmer(by ~ (1 \UFF5C group), data, family = binomial) %>% anova(lme4::glmer(by ~ variable + (1 \UFF5C group), data, family = binomial)) |
"paired.t.test" |
Paired t-test | tidyr::pivot_wider(id_cols = group, ...); t.test(by_1, by_2, paired = TRUE, conf.level = 0.95, ...) |
"paired.wilcox.test" |
Paired Wilcoxon rank-sum test | tidyr::pivot_wider(id_cols = group, ...); wilcox.test(by_1, by_2, paired = TRUE, conf.int = TRUE, conf.level = 0.95, ...) |
"prop.test" |
Test for equality of proportions | prop.test(x, n, conf.level = 0.95, ...) |
"ancova" |
ANCOVA | lm(variable ~ by + adj.vars) |
| alias | description | pseudo-code |
"svy.t.test" |
t-test adapted to complex survey samples | survey::svyttest(~variable + by, data) |
"svy.wilcox.test" |
Wilcoxon rank-sum test for complex survey samples | survey::svyranktest(~variable + by, data, test = 'wilcoxon') |
"svy.kruskal.test" |
Kruskal-Wallis rank-sum test for complex survey samples | survey::svyranktest(~variable + by, data, test = 'KruskalWallis') |
"svy.vanderwaerden.test" |
van der Waerden's normal-scores test for complex survey samples | survey::svyranktest(~variable + by, data, test = 'vanderWaerden') |
"svy.median.test" |
Mood's test for the median for complex survey samples | survey::svyranktest(~variable + by, data, test = 'median') |
"svy.chisq.test" |
chi-squared test with Rao & Scott's second-order correction | survey::svychisq(~variable + by, data, statistic = 'F') |
"svy.adj.chisq.test" |
chi-squared test adjusted by a design effect estimate | survey::svychisq(~variable + by, data, statistic = 'Chisq') |
"svy.wald.test" |
Wald test of independence for complex survey samples | survey::svychisq(~variable + by, data, statistic = 'Wald') |
"svy.adj.wald.test" |
adjusted Wald test of independence for complex survey samples | survey::svychisq(~variable + by, data, statistic = 'adjWald') |
"svy.lincom.test" |
test of independence using the exact asymptotic distribution for complex survey samples | survey::svychisq(~variable + by, data, statistic = 'lincom') |
"svy.saddlepoint.test" |
test of independence using a saddlepoint approximation for complex survey samples | survey::svychisq(~variable + by, data, statistic = 'saddlepoint') |
| alias | description | pseudo-code |
"logrank" |
Log-rank test | survival::survdiff(Surv(.) ~ variable, data, rho = 0) |
"petopeto_gehanwilcoxon" |
Peto & Peto modification of Gehan-Wilcoxon test | survival::survdiff(Surv(.) ~ variable, data, rho = 1) |
"survdiff" |
G-rho family test | survival::survdiff(Surv(.) ~ variable, data, ...) |
"coxph_lrt" |
Cox regression (LRT) | survival::coxph(Surv(.) ~ variable, data, ...) |
"coxph_wald" |
Cox regression (Wald) | survival::coxph(Surv(.) ~ variable, data, ...) |
"coxph_score" |
Cox regression (Score) | survival::coxph(Surv(.) ~ variable, data, ...) |
| alias | description | difference statistic | pseudo-code |
"t.test" |
t-test | mean difference | t.test(variable ~ as.factor(by), data = data, conf.level = 0.95, ...) |
"paired.t.test" |
Paired t-test | mean difference | tidyr::pivot_wider(id_cols = group, ...); t.test(by_1, by_2, paired = TRUE, conf.level = 0.95, ...) |
"paired.wilcox.test" |
Paired Wilcoxon rank-sum test | rate difference | tidyr::pivot_wider(id_cols = group, ...); wilcox.test(by_1, by_2, paired = TRUE, conf.int = TRUE, conf.level = 0.95, ...) |
"prop.test" |
Test for equality of proportions | rate difference | prop.test(x, n, conf.level = 0.95, ...) |
"ancova" |
ANCOVA | mean difference | lm(variable ~ by + adj.vars) |
"ancova_lme4" |
ANCOVA with random intercept | mean difference | lme4::lmer(variable ~ by + adj.vars + (1 \UFF5C group), data) |
"cohens_d" |
Cohen's D | standardized mean difference | effectsize::cohens_d(variable ~ by, data, ci = conf.level, ...) |
To report a p-value (or difference) for a test not available in gtsummary, you can create a
custom function. The output is a data frame that is one line long. The
structure is similar to the output of broom::tidy() of a typical
statistical test. The add_p() and add_comparison() functions will look for columns called
"p.value", "estimate", "conf.low", "conf.high", and "method" for the
p-value, difference, confidence interval, and the test name used in the footnote.
Example calculating a p-value from a t-test assuming a common variance between groups.
ttest_common_variance <- function(data, variable, by, ...) {
data <- data[c(variable, by)] %>% dplyr::filter(complete.cases(.))
t.test(data[[variable]] ~ factor(data[[by]]), var.equal = TRUE) %>%
broom::tidy()
}trial[c("age", "trt")] %>%
tbl_summary(by = trt) %>%
add_p(test = age ~ "ttest_common_variance")
A custom add_difference() is similar, and accepts arguments conf.level=
and adj.vars= as well.
ttest_common_variance <- function(data, variable, by, conf.level, ...) {
data <- data[c(variable, by)] %>% dplyr::filter(complete.cases(.))
t.test(data[[variable]] ~ factor(data[[by]]), conf.level = conf.level, var.equal = TRUE) %>%
broom::tidy()
}
For tbl_summary() objects, the custom function will be passed the
following arguments: custom_pvalue_fun(data=, variable=, by=, group=, type=, conf.level=, adj.vars=).
While your function may not utilize each of these arguments, these arguments
are passed and the function must accept them. We recommend including ...
to future-proof against updates where additional arguments are added.
The following table describes the argument inputs for each gtsummary table type.
| argument | tbl_summary | tbl_svysummary | tbl_survfit |
data= |
A data frame | A survey object | A survfit() object |
variable= |
String variable name | String variable name | NA |
by= |
String variable name | String variable name | NA |
group= |
String variable name | NA |
NA |
type= |
Summary type | Summary type | NA |
conf.level= |
Confidence interval level | NA |
NA |
adj.vars= |
Character vector of adjustment variable names (e.g. used in ANCOVA) | NA |
NA |