Gallery

58 worked examples. Every figure was rendered by ggnext when this site was built, and the code under each one is exactly what produced it.

Essentials

The everyday layers. Each takes the same aesthetics you would expect, and several can be stacked in one plot - layers draw in the order they are added.

Scatter plot

The default view of a relationship between two continuous variables.

Aesthetics: x, y, color, size, alpha

ggnext(cars, aes(speed, dist)) + geom_point() + labs(title = "Stopping distance rises with speed", 
    x = "Speed (mph)", y = "Distance (ft)")

Jittered points

Points nudged by a reproducible random offset so overlapping observations stay countable; pair with a boxplot or violin.

Aesthetics: x, y, color

ggnext(iris, aes(Species, Sepal.Width, color = Species)) + geom_jitter(alpha = 0.7) + 
    theme(legend_position = "none")

Trend line with confidence band

A loess or linear fit with a confidence ribbon; use method = "lm" for a straight line.

Aesthetics: x, y, color

ggnext(cars, aes(speed, dist)) + geom_point(alpha = 0.6) + geom_smooth(method = "lm") + 
    theme_minimal()

Line chart

One polyline per group, ordered by x - the standard time-series view.

Aesthetics: x, y, color, group, linewidth, dash

ggnext(data.frame(t = rep(1:12, 2), v = c(cumsum(rnorm(12, 2)), 
    cumsum(rnorm(12, 1))), g = rep(c("A", "B"), each = 12)), aes(t, 
    v, color = g)) + geom_line() + theme_minimal()

Step chart

Holds each value until the next observation - right for quantities that change discretely, like a policy rate.

Aesthetics: x, y, color, group

ggnext(data.frame(t = 1:8, v = c(2, 2, 3, 3, 5, 4, 4, 6)), aes(t, 
    v)) + geom_step() + geom_point() + theme_minimal()

Area chart

A line closed to a zero baseline; reads as magnitude over time rather than rate of change.

Aesthetics: x, y, color, group, alpha

ggnext(data.frame(t = 1:10, v = c(2, 4, 3, 6, 8, 7, 9, 8, 11, 13)), 
    aes(t, v)) + geom_area(alpha = 0.5) + theme_minimal()

Ribbon

A band between two series - forecast intervals, min/max envelopes, uncertainty around a fit.

Aesthetics: x, ymin, ymax, color, alpha

ggnext(data.frame(t = 1:12, lo = (1:12) * 0.8, hi = (1:12) * 1.4), 
    aes(t, ymin = lo, ymax = hi)) + geom_ribbon(alpha = 0.45) + 
    theme_minimal()

Segments

Straight lines between explicit endpoints; the building block for arrows, connectors, and slope charts.

Aesthetics: x, y, xend, yend, color

ggnext(data.frame(x = 1:4, y = c(2, 4, 3, 5), xe = 1:4 + 0.7, ye = c(4, 
    6, 2, 7)), aes(x, y, xend = xe, yend = ye)) + geom_segment(linewidth = 2) + 
    theme_minimal()

Reference lines

geom_hline() and geom_vline() span the panel and ignore the plot's aes(), so they never disturb the data mapping.

Aesthetics: (literal intercepts)

ggnext(cars, aes(speed, dist)) + geom_point(alpha = 0.6) + geom_hline(mean(cars$dist), 
    dash = "5,4", color = "#C1462F") + geom_vline(mean(cars$speed), 
    dash = "5,4", color = "#C1462F") + theme_minimal()

Text labels

Draws the label column at each position; use for annotating a handful of points, not hundreds.

Aesthetics: x, y, label, color, size

ggnext(data.frame(x = c(1, 2, 3), y = c(3, 1, 2), l = c("alpha", 
    "beta", "gamma")), aes(x, y, label = l)) + geom_point(size = 5, 
    alpha = 0.3) + geom_text() + theme_minimal()

Tile heatmap

A grid of cells shaded by a continuous value - correlation matrices, calendars, any two-way table.

Aesthetics: x, y, color

ggnext(local({
    g <- expand.grid(x = 1:8, y = 1:6)
    g$z <- as.vector(outer(1:8, 1:6, function(a, b) sin(a/2) + 
        cos(b/2)))
    g
}), aes(x, y, color = z)) + geom_tile() + theme_minimal()

Distributions

Summaries of one variable, or of one variable split by a category. Where a stat is involved, plot_data() will show you exactly what was computed.

Bar chart

geom_bar() counts rows per category; geom_col() takes the height from y directly.

Aesthetics: x, color, position

ggnext(data.frame(g = c("alpha", "beta", "gamma", "delta"), v = c(12, 
    27, 19, 8)), aes(g, v, color = g)) + geom_col() + theme(legend_position = "none")

Stacked and dodged bars

position = "stack" (default) shows totals, "dodge" compares groups side by side, "fill" shows proportions.

Aesthetics: x, color, position

ggnext(data.frame(g = rep(c("Q1", "Q2", "Q3"), each = 3), grp = rep(c("a", 
    "b", "c"), 3), v = c(4, 6, 3, 7, 4, 5, 5, 8, 2)), aes(g, v, 
    color = grp)) + geom_col(position = "dodge") + theme_minimal()

Histogram

Bins a continuous variable and counts each bin; bins = n gives exactly n bins.

Aesthetics: x, bins or binwidth

ggnext(cars, aes(speed)) + geom_histogram(bins = 8) + theme_minimal()

Density

A smoothed distribution estimate - easier to overlay across groups than histograms.

Aesthetics: x, color, adjust

ggnext(iris, aes(Sepal.Length, color = Species)) + geom_density(alpha = 0.5) + 
    theme_minimal()

Box plot

Tukey's five-number summary with outliers beyond 1.5 IQR drawn individually.

Aesthetics: x, y, color

ggnext(iris, aes(Species, Sepal.Length, color = Species)) + geom_boxplot() + 
    theme(legend_position = "none")

Violin

A mirrored density per category - shows bimodality that a box plot hides.

Aesthetics: x, y, color

ggnext(iris, aes(Species, Sepal.Width, color = Species)) + geom_violin() + 
    theme(legend_position = "none")

Layered distribution view

Violin for shape, box for summary, jitter for the raw data - layers draw in the order added.

Aesthetics: x, y, color

ggnext(iris, aes(Species, Sepal.Length, color = Species)) + geom_violin(alpha = 0.25) + 
    geom_boxplot() + geom_jitter(alpha = 0.4) + theme(legend_position = "none")

Ridgeline (joyplot)

One density per group, offset vertically; compares many distributions in little vertical space.

Aesthetics: x, y (the group)

ggnext(iris, aes(Sepal.Length, Species)) + geom_ridgeline() + theme_minimal()

Error bars and point ranges

An interval per observation; geom_pointrange() adds the estimate marker.

Aesthetics: x, y, ymin, ymax

ggnext(data.frame(g = c("A", "B", "C", "D"), m = c(5, 7, 4, 8), 
    lo = c(4, 6.2, 3.1, 7.1), hi = c(6, 7.8, 4.9, 8.9)), aes(g, 
    m, ymin = lo, ymax = hi)) + geom_pointrange() + theme_minimal()

Dumbbell

Two endpoints joined by a connector - before/after comparison across categories.

Aesthetics: x, xend, y

ggnext(data.frame(g = c("North", "South", "East", "West"), before = c(12, 
    18, 9, 14), after = c(19, 21, 15, 13)), aes(before, xend = after, 
    y = g)) + geom_dumbbell() + theme_minimal()

Waterfall

Running-total bars showing how each contribution moves a starting value to an ending one.

Aesthetics: x, y

ggnext(data.frame(step = factor(c("Start", "Sales", "Costs", "Tax", 
    "End"), levels = c("Start", "Sales", "Costs", "Tax", "End")), 
    v = c(100, 45, -30, -12, 0)), aes(step, v)) + geom_waterfall() + 
    theme_minimal()

Layout

Geoms whose positions come from a layout algorithm rather than straight from the data. Each algorithm is implemented directly - squarified treemaps, force-directed graphs, Sankey node stacking.

Facets

One panel per subset. Axes are shared by default, which is what makes panels comparable.

Aesthetics: facet_wrap(var), facet_grid(rows, cols)

ggnext(iris, aes(Sepal.Length, Sepal.Width, color = Species)) + 
    geom_point() + facet_wrap(Species) + theme(legend_position = "none")

Facets with free scales

Each panel scales to its own data - right when panels differ in magnitude and shape matters more than comparison.

Aesthetics: scales = "free"

ggnext(iris, aes(Sepal.Length, Sepal.Width, color = Species)) + 
    geom_point() + facet_wrap(Species, scales = "free") + theme(legend_position = "none")

Radar / spider

A closed profile per series across categorical axes. The radial axis starts at zero so areas stay honest.

Aesthetics: x (axis), y (value), color (series)

ggnext(data.frame(axis = rep(c("Speed", "Power", "Range", "Cost", 
    "Safety"), 2), value = c(8, 6, 7, 4, 9, 5, 9, 4, 8, 6), model = rep(c("A", 
    "B"), each = 5)), aes(axis, value, color = model)) + geom_radar() + 
    coord_polar() + theme_minimal()

Circular bar chart

Any cartesian geom bends into polar coordinates; bars become wedges.

Aesthetics: coord_polar()

ggnext(data.frame(g = c("Mon", "Tue", "Wed", "Thu", "Fri", "Sat", 
    "Sun"), v = c(4, 7, 6, 9, 12, 15, 11)), aes(g, v, color = g)) + 
    geom_col() + coord_polar() + theme(legend_position = "none")

Treemap

Area-proportional tiles laid out with the squarified algorithm, so tiles stay near-square and comparable.

Aesthetics: size (area), label, color

ggnext(data.frame(region = c("North", "South", "East", "West", 
    "Central"), revenue = c(52, 38, 27, 19, 11)), aes(size = revenue, 
    label = region, color = region)) + geom_treemap() + theme(legend_position = "none")

Sankey / alluvial

Flows between stages. One value-to-height scale is shared across stages, so a flow keeps its thickness end to end.

Aesthetics: x (source), xend (target), y (value)

ggnext(data.frame(from = c("Visited", "Visited", "Signed up", "Signed up"), 
    to = c("Signed up", "Left", "Purchased", "Churned"), n = c(400, 
        600, 150, 250)), aes(x = from, xend = to, y = n)) + geom_sankey()

Network

A Fruchterman-Reingold force layout: repulsion between all nodes, attraction along edges, with a cooling schedule.

Aesthetics: x (source), xend (target)

ggnext(data.frame(from = c("A", "A", "B", "C", "D", "E", "B"), 
    to = c("B", "C", "C", "D", "E", "A", "E")), aes(x = from, xend = to)) + 
    geom_network()

Chord

Entities on a circle joined by ribbons whose ends are arcs proportional to the flow.

Aesthetics: x (source), xend (target), y (value)

ggnext(data.frame(from = c("A", "A", "B", "C"), to = c("B", "C", 
    "C", "A"), n = c(5, 3, 7, 2)), aes(x = from, xend = to, y = n)) + 
    geom_chord()

Streamgraph

Stacked areas on a wiggle baseline rather than zero, so each band's thickness stays readable.

Aesthetics: x, y, color (series)

ggnext(data.frame(t = rep(1:8, 3), v = c(2, 4, 6, 5, 3, 2, 4, 5, 
    1, 3, 5, 8, 6, 4, 3, 2, 5, 4, 3, 4, 6, 7, 5, 4), grp = rep(c("a", 
    "b", "c"), each = 8)), aes(t, v, color = grp)) + geom_stream() + 
    theme_minimal()

Bump chart

Rank trajectories with sigmoid interpolation, so crossings read cleanly instead of as zigzags.

Aesthetics: x (time), y (rank), color (series)

ggnext(data.frame(year = rep(2020:2024, 4), rank = c(1, 2, 3, 3, 
    2, 2, 1, 1, 2, 1, 3, 3, 2, 1, 3, 4, 4, 4, 4, 4), team = rep(c("A", 
    "B", "C", "D"), each = 5)), aes(year, rank, color = team)) + 
    geom_bump() + scale_y_reverse() + theme_minimal()

Funnel

Centred bars tapering through the stages of a conversion or triage process.

Aesthetics: x (stage), y (count)

ggnext(data.frame(stage = factor(c("Visits", "Signups", "Trials", 
    "Paid"), levels = c("Visits", "Signups", "Trials", "Paid")), 
    n = c(10000, 3200, 1100, 420)), aes(stage, n, color = stage)) + 
    geom_funnel() + theme(legend_position = "none")

Parallel coordinates

One line per observation across independently rescaled axes - eyeball structure in high-dimensional data.

Aesthetics: x (variable), y (value), group

ggnext(local({
    d <- iris[c(1, 20, 60, 80, 110, 140), ]
    data.frame(id = rep(rownames(d), 4), var = rep(c("SL", "SW", 
        "PL", "PW"), each = nrow(d)), val = c(d$Sepal.Length, d$Sepal.Width, 
        d$Petal.Length, d$Petal.Width), sp = rep(as.character(d$Species), 
        4))
}), aes(var, val, group = id, color = sp)) + geom_parallel() + 
    theme_minimal()

UpSet (set intersections)

Intersection sizes with a membership matrix - readable where a 4-way Venn diagram is not.

Aesthetics: label (membership, e.g. "A&B")

ggnext(data.frame(sets = c("A", "A&B", "B", "A&B&C", "C", "A&B", 
    "A", "B&C", "A&C", "A&B")), aes(label = sets)) + geom_upset()

Machine learning

Model diagnostics as first-class layers. Each takes a tidy data frame rather than a fitted model object, so any framework that can produce the columns will work.

SHAP beeswarm

Every observation's contribution per feature, nudged vertically where values collide; colour shows the direction of effect.

Aesthetics: x (SHAP value), y (feature), color (feature value)

ggnext(local({
    rng <- local_rng(1)
    data.frame(feature = rep(c("age", "income", "tenure"), each = 40), 
        shap = c(rng$norm(40, 0.3, 0.2), rng$norm(40, -0.1, 0.3), 
            rng$norm(40, 0, 0.15)), value = rng$unif(120))
}), aes(shap, feature, color = value)) + geom_shap() + geom_vline(0, 
    dash = "3,3") + labs(x = "SHAP value", y = NULL) + theme_minimal()

Partial dependence + ICE

Thin per-observation ICE curves under a bold average, so heterogeneous effects are visible.

Aesthetics: x (feature), y (prediction), group

ggnext(local({
    rng <- local_rng(9)
    data.frame(x = rep(1:10, 8), id = rep(1:8, each = 10), pred = as.vector(sapply(1:8, 
        function(i) {
            (1:10) * 0.1 * i + rng$norm(10, 0, 0.15)
        })))
}), aes(x, pred, group = id)) + geom_partial_dependence() + theme_minimal()

ROC curve

The true-positive against false-positive staircase over every score threshold.

Aesthetics: score, truth

ggnext(local({
    rng <- local_rng(2)
    s <- rng$unif(300)
    data.frame(score = s, truth = rng$bernoulli(300, s))
}), aes(score = score, truth = truth)) + geom_roc() + theme_minimal()

Calibration curve

Binned predictions against observed rates; distance from the diagonal is over- or under-confidence.

Aesthetics: x (predicted prob), y (outcome)

ggnext(local({
    rng <- local_rng(3)
    p <- rng$unif(400)
    data.frame(pred = p, obs = rng$bernoulli(400, p^1.3))
}), aes(pred, obs)) + geom_calibration() + theme_minimal()

Cumulative gain

Share of positives captured against share of population targeted - how to size a cutoff.

Aesthetics: score, truth

ggnext(local({
    rng <- local_rng(11)
    s <- rng$unif(250)
    data.frame(score = s, y = rng$bernoulli(250, s))
}), aes(score = score, truth = y)) + geom_lift_gain() + theme_minimal()

Confusion matrix

Row-normalised shading with raw counts annotated, so class imbalance cannot hide errors.

Aesthetics: x (predicted), y (actual), size (count)

ggnext(data.frame(predicted = c(rep("cat", 14), rep("dog", 9), 
    rep("bird", 6)), actual = c(rep("cat", 11), rep("dog", 3), 
    rep("dog", 7), rep("cat", 2), rep("bird", 5), "cat")), aes(predicted, 
    actual)) + geom_confusion_matrix() + theme(legend_position = "none")

Residual diagnostics

Residuals against fitted values with a zero line and a loess trend - the first check on a linear model.

Aesthetics: x (fitted), y (residual)

ggnext(local({
    m <- lm(dist ~ speed, cars)
    data.frame(fitted = fitted(m), resid = resid(m))
}), aes(fitted, resid)) + geom_residual() + theme_minimal()

Learning curve

Train and validation score against training-set size - shows whether a model is data-limited or over-fitting.

Aesthetics: x (size/epoch), y (score), color (split)

ggnext(data.frame(n = rep(c(50, 100, 200, 400, 800), 2), score = c(0.72, 
    0.8, 0.85, 0.88, 0.9, 0.66, 0.75, 0.81, 0.85, 0.88), split = rep(c("train", 
    "validation"), each = 5)), aes(n, score, color = split)) + 
    geom_learning_curve() + theme_minimal()

Embedding with hulls

A t-SNE/UMAP/PCA scatter with convex hulls, so cluster shape is visible rather than inferred from colour.

Aesthetics: x, y, color (cluster)

ggnext(local({
    rng <- local_rng(4)
    data.frame(d1 = c(rng$norm(40), rng$norm(40, 4)), d2 = c(rng$norm(40), 
        rng$norm(40, 3)), cluster = rep(c("a", "b"), each = 40))
}), aes(d1, d2, color = cluster)) + geom_embedding() + theme_minimal()

Silhouette

Sorted silhouette widths per cluster - the standard visual check on cluster separation.

Aesthetics: x (width), y (cluster)

ggnext(local({
    rng <- local_rng(12)
    data.frame(cluster = rep(c("1", "2", "3"), each = 25), width = c(rng$unif(25, 
        0.3, 0.9), rng$unif(25, 0.1, 0.7), rng$unif(25, -0.1, 0.6)))
}), aes(width, cluster, color = cluster)) + geom_silhouette() + 
    theme(legend_position = "none")

Decision boundary

A shaded prediction grid; overlay geom_point() for the training data.

Aesthetics: x, y, color (predicted class)

ggnext(local({
    g <- expand.grid(x = seq(0, 1, 0.04), y = seq(0, 1, 0.04))
    g$cls <- ifelse(g$x + g$y > 1, "a", "b")
    g
}), aes(x, y, color = cls)) + geom_decision_boundary() + theme_minimal()

Forecast with interval

History solid, forecast dashed, interval as a ribbon on the forecast rows only.

Aesthetics: x, y, ymin, ymax, group

ggnext(data.frame(t = 1:12, v = c(3, 4, 4, 5, 6, 6, 7, 8, 9, 10, 
    11, 12), lo = c(rep(NA, 7), 6.5, 7, 7.5, 8, 8.5), hi = c(rep(NA, 
    7), 9.5, 11, 12.5, 14, 15.5), part = rep(c("actual", "forecast"), 
    c(7, 5))), aes(t, v, ymin = lo, ymax = hi, group = part)) + 
    geom_forecast_band() + theme_minimal()

Clinical

Figures that clinical reporting needs constantly and that otherwise take dozens of lines of manual layering. Estimators such as Kaplan-Meier and Aalen-Johansen are computed in-package.

Kaplan-Meier

Product-limit survival curves with censoring ticks, per treatment arm.

Aesthetics: time, status, color (arm)

ggnext(local({
    rng <- local_rng(5)
    data.frame(t = c(rng$exp(60, 0.08), rng$exp(60, 0.14)), ev = rng$bernoulli(120, 
        0.75), arm = rep(c("Treatment", "Control"), each = 60))
}), aes(time = t, status = ev, color = arm)) + geom_km() + theme_minimal()

Cumulative incidence

Aalen-Johansen curves per event type - the correct estimator when competing risks make 1 - KM biased upward.

Aesthetics: time, status (0 = censored, 1..k = event types)

ggnext(local({
    rng <- local_rng(8)
    data.frame(t = rng$exp(150, 0.1), ev = rng$choice(0:2, 150, 
        c(0.4, 0.35, 0.25)))
}), aes(time = t, status = ev)) + geom_cuminc() + theme_minimal()

Forest plot

Estimates with confidence intervals and a no-effect reference; marker area encodes study weight.

Aesthetics: x (estimate), y (study), ymin, ymax, size (weight)

ggnext(data.frame(study = c("Trial A", "Trial B", "Trial C", "Trial D", 
    "Pooled"), hr = c(0.82, 0.71, 0.95, 0.88, 0.83), lo = c(0.65, 
    0.52, 0.78, 0.7, 0.74), hi = c(1.03, 0.97, 1.16, 1.1, 0.93), 
    weight = c(30, 22, 28, 20, 100)), aes(hr, study, ymin = lo, 
    ymax = hi, size = weight)) + geom_forest() + labs(x = "Hazard ratio (95% CI)", 
    y = NULL) + theme_minimal()

Swimmer plot

Per-subject time on treatment, with arrowheads for subjects still ongoing at data cutoff.

Aesthetics: x (duration), y (subject), color, label (ongoing)

ggnext(data.frame(subject = paste0("S", 1:8), months = c(4, 9, 
    14, 6, 20, 11, 17, 7), response = c("PR", "CR", "CR", "SD", 
    "PR", "SD", "CR", "PD"), ongoing = c(FALSE, FALSE, TRUE, FALSE, 
    TRUE, FALSE, TRUE, FALSE)), aes(months, subject, color = response, 
    label = ongoing)) + geom_swimmer() + labs(x = "Months", y = NULL) + 
    theme_minimal()

Oncology spider plot

Per-subject change from baseline over time, with the RECIST +20% / -30% thresholds marked.

Aesthetics: x (time), y (% change), color (subject)

ggnext(data.frame(month = rep(c(0, 2, 4, 6, 8), 4), pct = c(0, 
    -20, -35, -40, -42, 0, 10, 25, 40, 55, 0, -5, -10, -8, -12, 
    0, -30, -45, -50, -48), subject = rep(c("S1", "S2", "S3", "S4"), 
    each = 5)), aes(month, pct, color = subject)) + geom_spider_response() + 
    labs(x = "Month", y = "% change from baseline") + theme_minimal()

RECIST waterfall

Best response per subject, ordered worst to best and shaded by RECIST category.

Aesthetics: x (subject), y (% change)

ggnext(local({
    rng <- local_rng(6)
    data.frame(subject = paste0("S", 1:24), pct = sort(rng$unif(24, 
        -78, 48), decreasing = TRUE))
}), aes(subject, pct)) + geom_waterfall_response() + labs(y = "% change from baseline", 
    x = NULL) + theme(axis_text_x = FALSE)

Spaghetti trajectories

Individual longitudinal paths with a bold group mean - shows change without hiding spread.

Aesthetics: x (time), y (measure), group (subject)

ggnext(local({
    rng <- local_rng(7)
    data.frame(week = rep(0:5, 10), id = rep(1:10, each = 6), score = as.vector(sapply(1:10, 
        function(i) {
            50 + i + (0:5) * 2 + rng$norm(6, 0, 3)
        })))
}), aes(week, score, group = id)) + geom_spaghetti() + theme_minimal()

Bland-Altman

Difference against mean with bias and 95% limits of agreement - the standard method-comparison plot.

Aesthetics: x, y (the two methods)

ggnext(local({
    rng <- local_rng(13)
    a <- rng$norm(80, 100, 12)
    data.frame(method_a = a, method_b = a + rng$norm(80, 2, 5))
}), aes(method_a, method_b)) + geom_bland_altman() + theme_minimal()

Dose-response

A four-parameter log-logistic fit with the EC50 marked; pair with scale_x_log10().

Aesthetics: x (dose), y (response)

ggnext(data.frame(dose = rep(c(0.1, 1, 10, 100, 1000), each = 3), 
    resp = c(5, 7, 6, 18, 22, 20, 52, 48, 55, 82, 79, 85, 95, 97, 
        93)), aes(dose, resp)) + geom_dose_response() + scale_x_log10() + 
    theme_minimal()

Adverse-event heatmap

Incidence by preferred term and treatment arm, shaded by rate and annotated with values.

Aesthetics: x (arm), y (term), size (incidence)

ggnext(local({
    d <- expand.grid(arm = c("Placebo", "Low", "High"), ae = c("Nausea", 
        "Fatigue", "Headache", "Rash"))
    d$pct <- c(5, 12, 22, 8, 15, 26, 3, 6, 11, 2, 9, 17)
    d
}), aes(arm, ae, size = pct)) + geom_ae_heatmap() + theme(legend_position = "none")

CONSORT flow

Participant flow from screening to analysis, laid out automatically from a stage/count table.

Aesthetics: label (stage), size (count)

ggnext(data.frame(stage = c("Assessed for eligibility", "Randomised", 
    "Received allocation", "Completed follow-up", "Analysed"), 
    n = c(420, 300, 291, 276, 271)), aes(label = stage, size = n)) + 
    geom_consort()