Skip to contents

Ordered effect sizes with bootstrap intervals for one clinical variable

Source: R/clinical.R
fig_clinical_forest.Rd

WHAT IS ON IT. One row per population (or marker), ordered by effect: the point is Spearman's rho or Cliff's delta, the bar is its 95% percentile bootstrap interval, and the dashed line is no effect. Rows whose interval clears the line are the ones a follow-up cohort should be powered on.

Usage

fig_clinical_forest(
  assoc,
  key_col,
  var_name,
  outfile,
  dpi = 200,
  colors = fcs_colors()
)

Arguments

assoc

data frame from stats_clinical_association().

key_col

"population" or "marker".

var_name

which clinical variable to draw.

outfile

path.

dpi

resolution.

colors

palette.

Details

A three-level variable is drawn on epsilon-squared instead, which is unsigned, so its axis starts at zero and no interval is drawn: there is no direction to put an interval around.

WHY IT IS ORDERED BY EFFECT AND NOT BY p. At these sample sizes the p-value ordering is close to the effect ordering but not identical, and where they disagree the effect is the one worth reading: p carries how many samples there were as much as how large the difference is.