Skip to main content

Table of the effect sizes of binary variables

Description

Formats the output of cor_binary_effects(): groups and sizes, means (SD), difference in means with its CI and t-test, rank-biserial correlation r_rb with its CI and Mann-Whitney test, and the observed r_s next to its maximum for this split, so that the attenuation of the correlation is visible.

Usage

tbl_cor_binary(eff, labels = NULL, digits = 2, title = NULL, subtitle = NULL)

Arguments

eff output of cor_binary_effects()
labels named vector variable -> label
digits decimals
title, subtitle override the automatic texts

Value

a gt_tbl

Examples

library("corrkit")

d <- simulate_cor_data(150)
tbl_cor_binary(cor_binary_effects(d, c("BIN_CRP", "BIN_IL6"), c("depression", "fatigue")),
               labels = attr(d, "labels"))
Binary variables: effect sizes to report
Group 1 = upper level of each binary variable | N = 142–143 | raw p-values (no adjustment)
Binary variable Groups 0 / 1: n (% group 1) Mean (SD), group 0 Mean (SD), group 1
Mean difference
Rank-biserial correlation
Observed rₛ / max |rₛ|
M₁ − M₀ [95% CI] p (t-test) r_rb [95% CI] p (Mann-Whitney)
Depression score
CRP high low / high: 74 / 69 (48%) 11.50 (3.30) 12.82 (3.22) 1.32 [0.24, 2.40] 0.017 0.22 [0.04, 0.39] 0.021 0.19 / 0.87
IL-6 high low / high: 115 / 27 (19%) 11.77 (3.31) 13.78 (2.80) 2.01 [0.65, 3.37] 0.004 0.34 [0.11, 0.53] 0.007 0.23 / 0.68
Fatigue score
CRP high low / high: 73 / 69 (49%) 8.02 (3.51) 8.61 (3.69) 0.59 [-0.60, 1.79] 0.327 0.09 [-0.10, 0.27] 0.367 0.08 / 0.87
IL-6 high low / high: 115 / 28 (20%) 7.61 (3.42) 10.29 (3.07) 2.69 [1.29, 4.08] < 0.001 0.40 [0.18, 0.58] 0.001 0.27 / 0.69
M₁ − M₀: difference in means, Student's t-test with equal variances (= test of the point-biserial r); 95% CI from stats::t.test() {stats}.
r_rb = 2U / (n₀ n₁) − 1 = P(Y₁ > Y₀) − P(Y₁ < Y₀) = 2 AUC − 1, U = Mann-Whitney statistic of stats::wilcox.test() {stats} (Kerby 2014, doi:10.2466/11.IT.3.1); 95% CI from effectsize::rank_biserial() {effectsize}.
Observed rₛ / max |rₛ|: Spearman correlation with the 0/1 variable and its maximum √(3pq) for this split (no ties): the correlation is bounded by the split and must not be compared with that of a continuous variable (Babchishin & Helmus 2016, doi:10.3758/s13428-015-0627-7).
Back to top