Between-group differences for every population in one figure

WHAT IT IS. One point per population: the effect on the x-axis, the evidence on the y-axis. group_comparison.png gives each population a panel and shows every sample, which is the right figure for reading one population carefully and the wrong one for finding which population to read – a dozen panels have to be compared by eye, and the effect sizes never appear on it at all. This is the shortlist: everything far from the centre line moved, everything high up moved consistently.

Usage

fig_group_volcano(
  stats,
  outfile,
  effect = c("cliff", "log2fc"),
  p_source = c("raw", "BH"),
  label_n = 8L,
  title_noun = "Population abundance",
  dpi = 200,
  colors = fcs_colors()
)

Arguments

stats

output of stats_group_comparison().

outfile

path.

effect

"cliff" (default) or "log2fc".

p_source

"raw" or "BH" – which p-value on the y-axis.

label_n

how many of the strongest populations to name on the figure.

title_noun

the subject of the title, e.g. "Population abundance".

dpi

resolution.

colors

palette.

Details

WHY CLIFF'S DELTA AND NOT A FOLD CHANGE by default. A fold change of medians is unbounded and, at these group sizes, dominated by whichever sample sits at the median; a population whose reference median is near zero produces a fold change of 40 that means nothing. Cliff's delta is bounded [-1, 1], is the effect size the rank test actually corresponds to, and is directly readable: 0.5 means that in three quarters of the cross-sample pairs the comparison group was higher. effect = "log2fc" gives the conventional axis for anyone who wants it.

WHAT THE TWO HORIZONTAL LINES MEAN. The upper is p = 0.05. The lower is the smallest p this design can produce at all (see rank_sum_p_floor()); when it sits BELOW the 0.05 line the design can reach significance, and when it sits above, no population can, and an empty upper region says nothing about biology.