Accessible Area, Error Bar, Point and Lollipop Plots in matplotlib and seaborn with py-maidr

Make matplotlib area charts, error bar plots, seaborn point plots and stem (lollipop) charts accessible with py-maidr, intervals included.

These four charts share one idea: a value with something drawn around it. An area chart is read like a line, but a stacked area announces each band’s own value beside the running total. Error bar and point plots are read as a grid, so Left and Right walk the samples and Up and Down at one sample walk the lower bound, the estimate and the upper bound. A lollipop (ax.stem()) is read as a bar chart whose bars have been thinned to lines: one stop per stem, carrying its value.

AREA, STACKED_AREA, ERRORBAR and LOLLIPOP are experimental and may change without a deprecation period; a point plot is read through the error bar layer, so it inherits that status. See Plot Type Stability.

Setup

Every example on this page needs only the import below. The plot cells repeat it so each one can be copied on its own.

import matplotlib.pyplot as plt
import seaborn as sns

# Just import maidr package: plt.show() now renders accessible output 
import maidr  

# The point plot below reuses this dataset.
penguins = sns.load_dataset("penguins")

Area Plot

Prototype

This is one of the experimental plot types. It has not been through a user study, and it may change without a deprecation period. See Plot type stability.

ax.stackplot() reads as an area chart rather than a line one, and the distinction matters: a stacked area draws two numbers at each point that a line would conflate. The band’s height is the series’ own value; its top edge is the running total. maidr announces the value and reports the total beside it, so a reader always knows which they heard.

A single band is a plain area — nothing is stacked on it, so a running total equal to its own value would be noise. baseline="wiggle" (a streamgraph) reads the same way as any stacked area: floating the stack moves where the bands sit without changing what any band measures.

import matplotlib.pyplot as plt

import maidr 

years = [2019, 2020, 2021, 2022]
subscriptions = [10, 20, 25, 30]
services = [5, 8, 12, 14]

fig, ax = plt.subplots(figsize=(8, 5))
ax.stackplot( 
    years, subscriptions, services, labels=["Subscriptions", "Services"]
)

ax.set_title("Revenue by Source")
ax.set_xlabel("Year")
ax.set_ylabel("Revenue (millions)")
ax.legend(loc="upper left")

plt.show()

Pass labels= and each band is announced by name. Without it the bands are still read, but they can only be told apart by their order in the stack.

ax.fill_between() reads as an area too, when it draws one: fill_between(x, y) fills from zero up to a curve, which is the same chart a single stackplot band draws. ax.fill_betweenx() is the same picture turned over and reads the same way.

import matplotlib.pyplot as plt

import maidr 

months = list(range(1, 13))
rainfall = [78, 62, 71, 55, 48, 39, 31, 35, 52, 74, 89, 84]

fig, ax = plt.subplots(figsize=(8, 4))
ax.fill_between(months, rainfall, alpha=0.6, label="Rainfall") 

ax.set_title("Monthly Rainfall")
ax.set_xlabel("Month")
ax.set_ylabel("Rainfall (mm)")

plt.show()

A band drawn between two curves — fill_between(x, lower, upper), the usual way to shade a confidence interval — is a different chart, and maidr does not read it as an area. Its content is the gap between the edges rather than the height of either, so announcing it as an area would report the upper edge as a magnitude and drop the lower one. Those charts render as a static image for now; see issue #339.

Error Bar Plot

Prototype

This is one of the experimental plot types. It has not been through a user study, and it may change without a deprecation period. See Plot type stability.

Uncertainty is usually the finding, not the decoration: whether two group means differ is answered by whether their intervals overlap. ax.errorbar() draws that, and maidr reads the estimate and its interval, so the comparison the chart was drawn to support can be made by ear.

Navigation is a grid. Left and right walk the samples; up and down at one sample walk the lower bound, the estimate and the upper bound — so a reader traces one interval rather than rebuilding it from three passes over the chart.

The bounds are read off the drawn bars rather than from the yerr you passed, so asymmetric errors, one-sided errors, uplims/lolims and fmt="none" all read correctly without you having to present them any particular way.

import matplotlib.pyplot as plt

import maidr 

groups = ["control", "low dose", "high dose"]
means = [4.2, 5.1, 7.3]

# Asymmetric intervals, as a real confidence interval usually is: the lower
# and upper distances from the mean are given separately.
lower = [0.4, 1.1, 0.2]
upper = [0.4, 1.5, 0.1]

fig, ax = plt.subplots(figsize=(8, 5))
ax.errorbar(groups, means, yerr=[lower, upper], fmt="o", capsize=5)

ax.set_title("Mean Response by Dose, with 95% Confidence Intervals")
ax.set_xlabel("Group")
ax.set_ylabel("Response")

plt.show()

Point Plot

sns.pointplot() is the same reading from the other direction: it estimates a group mean from raw observations and draws the interval around it for you. It reads as an error bar layer, so navigation is the same grid — left and right walk the groups, up and down walk the lower bound, the estimate and the upper bound.

capsize, orient and errorbar= change how the chart is drawn, not how it is read; the bounds come from the drawn interval either way. A hue splits the chart into groups, and each group keeps its own estimate and interval on the error bar layer, so up and down at a category walk that group’s bounds. A point plot drawn with errorbar=None, or one whose groups each hold a single observation, has no intervals to carry and is read as a line chart instead.

import matplotlib.pyplot as plt
import seaborn as sns

import maidr 

fig, ax = plt.subplots(figsize=(8, 5))
sns.pointplot( 
    penguins, x="species", y="body_mass_g", capsize=0.1, ax=ax
)

ax.set_title("Mean Body Mass by Species, with 95% Confidence Intervals")
ax.set_xlabel("Species")
ax.set_ylabel("Body Mass (g)")

plt.show()

Lollipop Plot

Prototype

This is one of the experimental plot types. It has not been through a user study, and it may change without a deprecation period. See Plot type stability.

ax.stem() draws a stem down to each value with a marker at the top — a bar chart with the bars thinned to lines, which is the usual answer when there are too many categories for bars to stay readable. maidr reads it as a lollipop: one term per stem, carrying its value.

import matplotlib.pyplot as plt
import numpy as np

import maidr 

positions = np.arange(1, 9)
values = [2.1, 5.4, 3.3, 9.2, 4.0, 7.1, 6.4, 8.3]

fig, ax = plt.subplots(figsize=(8, 5))
ax.stem(positions, values) 
ax.set_xlabel("Sample")
ax.set_ylabel("Concentration (mg/L)")
ax.set_title("Concentration by sample")

plt.show()