Skip to main content

ECDF plot

An ECDF plot draws F(x) = P(X ≤ x) — the fraction of observations at or below each value — as a right-continuous step function. It is one of the most informative single-distribution diagnostics: no binning, no bandwidth to choose, and the whole distribution is visible at once.

SELECT kuva_render(to_json({
'title': 'ECDF',
'x_axis': {'name': 'expression'},
'y_axis': {'name': 'F(x)'},
'series': [{
'type': 'ecdf',
'groups': [{'label': 'Control', 'values': vals}],
'color': 'steelblue'
}]
})) AS chart
FROM (
SELECT list(expression) AS vals
FROM read_csv_auto('{{DFK_BASE_URL}}data/samples.tsv')
WHERE "group" = 'Control'
);

Multi-group comparison​

One entry per group in groups overlays the curves, and a group's own color overrides the shared one. Several ECDFs on one pair of axes is the honest way to ask "are these distributions different?" — the curves never cross if the groups are ordered, and they cross exactly where one group overtakes another.

SELECT kuva_render(to_json({
'title': 'Treatment vs control',
'x_axis': {'name': 'expression'},
'y_axis': {'name': 'F(x)'},
'legend': {'position': 'outside_right_top'},
'series': [{
'type': 'ecdf',
'groups': groups
}]
})) AS chart
FROM (
SELECT list({'label': g, 'values': vals} ORDER BY g) AS groups
FROM (
SELECT "group" AS g, list(expression) AS vals
FROM read_csv_auto('{{DFK_BASE_URL}}data/samples.tsv')
GROUP BY "group"
)
);

Complementary CDF​

complementary flips the curve to 1 − F(x) — the survival function, or exceedance probability. That is the standard view for read lengths (what fraction of reads are at least N bp?), coverage (what fraction of positions have at least N× depth?), and any heavy tail where the bulk is uninteresting. It pairs naturally with a log x axis.

SELECT kuva_render(to_json({
'title': 'Read length — CCDF',
'x_axis': {'name': 'read length (bp)', 'log': true, 'tick_format': 'integer'},
'y_axis': {'name': 'fraction ≥ length'},
'series': [{
'type': 'ecdf',
'groups': [{'label': 'reads', 'values': lens}],
'complementary': true,
'rug': true,
'color': 'steelblue'
}]
})) AS chart
FROM (
SELECT list("end" - "start") AS lens
FROM read_csv_auto('{{DFK_BASE_URL}}data/reads.tsv')
);

Confidence bands​

confidence_band shades a DKW 95 % band around each curve, with the half-width ε = √(ln 40 / 2n) — wide for small samples, tight for large ones. This is the diagnostic that turns "these curves look different" into "they are further apart than the sampling noise", and band_alpha controls the fill.

SELECT kuva_render(to_json({
'title': 'DKW confidence bands',
'x_axis': {'name': 'expression'},
'y_axis': {'name': 'F(x)'},
'legend': {'position': 'outside_right_top'},
'series': [{
'type': 'ecdf',
'groups': groups,
'confidence_band': true,
'band_alpha': 0.15
}]
})) AS chart
FROM (
SELECT list({'label': g, 'values': vals} ORDER BY g) AS groups
FROM (
SELECT "group" AS g, list(expression) AS vals
FROM read_csv_auto('{{DFK_BASE_URL}}data/samples.tsv')
WHERE "group" IN ('Control', 'Drug_A')
GROUP BY "group"
)
);

Rug plot​

rug draws a short tick at the bottom of the plot area for every observation, at its own x position. It shows where the raw samples actually are — clusters, gaps and outliers that the step function alone smoothes over. With several groups the ticks are offset slightly so they do not hide each other.

SELECT kuva_render(to_json({
'title': 'With a rug',
'x_axis': {'name': 'expression'},
'y_axis': {'name': 'F(x)'},
'series': [{
'type': 'ecdf',
'groups': [{'label': 'Control', 'values': vals}],
'color': 'steelblue',
'rug': true,
'rug_height': 8
}]
})) AS chart
FROM (
SELECT list(expression) AS vals
FROM read_csv_auto('{{DFK_BASE_URL}}data/samples.tsv')
WHERE "group" = 'Control'
);

Percentile reference lines​

percentile_lines draws dashed horizontal lines at given F levels and labels them at the right edge. The values are F levels in 0–1, so [0.25, 0.5, 0.75] marks the quartiles and the median — read where each line meets a curve to get that group's quartile.

SELECT kuva_render(to_json({
'title': 'Quartile reference lines',
'x_axis': {'name': 'expression'},
'y_axis': {'name': 'F(x)'},
'legend': {'position': 'outside_right_top'},
'series': [{
'type': 'ecdf',
'groups': groups,
'percentile_lines': [0.25, 0.5, 0.75]
}]
})) AS chart
FROM (
SELECT list({'label': g, 'values': vals} ORDER BY g) AS groups
FROM (
SELECT "group" AS g, list(expression) AS vals
FROM read_csv_auto('{{DFK_BASE_URL}}data/samples.tsv')
GROUP BY "group"
)
);

Step markers​

markers puts a circle at every step endpoint, which makes the discrete nature of the ECDF explicit. It is worth it for small samples (say under ~30 points); past that the dots merge into the line.

SELECT kuva_render(to_json({
'x_axis': {'name': 'value', 'tick_format': 'integer'},
'y_axis': {'name': 'F(x)'},
'series': [{
'type': 'ecdf',
'groups': [{'label': 'n=8', 'values': [1.2, 2.4, 2.9, 3.5, 4.1, 5.0, 5.8, 7.2]}],
'color': 'steelblue',
'markers': true,
'marker_size': 4
}]
})) AS chart;

Smooth CDF​

smooth replaces the staircase with a KDE-integrated smooth CDF (bandwidth from Silverman's rule), stroke_width sets the line, and line_dash dashes it.

SELECT kuva_render(to_json({
'title': 'Smooth CDF',
'x_axis': {'name': 'expression'},
'y_axis': {'name': 'F(x)'},
'legend': {'position': 'outside_right_top'},
'series': [{
'type': 'ecdf',
'groups': groups,
'smooth': true,
'stroke_width': 2
}]
})) AS chart
FROM (
SELECT list({'label': g, 'values': vals} ORDER BY g) AS groups
FROM (
SELECT "group" AS g, list(expression) AS vals
FROM read_csv_auto('{{DFK_BASE_URL}}data/samples.tsv')
GROUP BY "group"
)
);

Fields​

FieldTypeWhat it sets
groupsgroup[]Required. One curve per group, each {label, values, color?}.
complementarybooleanDraw the complementary CDF (1 − F, starts at 1 and falls).
confidence_bandbooleanDraw a DKW 95 % confidence band around each curve.
band_alphanumberThe band's opacity (default 0.15).
rugbooleanDraw a short tick on the axis for every observation.
rug_heightnumberHeight of those ticks, in pixels (default 6).
percentile_linesnumber[]Horizontal reference lines at these F levels (0–1), e.g. [0.25, 0.5, 0.75].
markersbooleanDraw a circle at every step endpoint.
marker_sizenumberMarker radius (default 3).
smoothbooleanReplace the steps with a KDE-integrated smooth CDF.
smooth_samplesintegerGrid points for the smooth CDF (default 200).
stroke_widthnumberCurve width (default 1.5).
line_dashstringA dash pattern (e.g. "4 2").

color and legend come from series & shared fields; a group's own color overrides it.

Notes​

  • Every group needs at least one value; an empty groups list is an error.
  • percentile_lines takes fractions, not percentages — [0.25, 0.5, 0.75], not [25, 50, 75].
  • On a dense sample, rug plus percentile_lines is usually clearer than markers.
  • complementary pairs with x_axis.log for tail questions.

See also​

  • kuva — ECDF plot — the plotting library's own reference for this chart.
  • Q-Q plot — the same data compared against a theoretical distribution.
  • Density plot — a smoothed curve instead of the exact steps.