Raincloud Plots Use Split Densities and Raw Jitter to Expose Bimodal Clusters Hidden by Boxplots

The standard boxplot summarizes a distribution using just five numbers—the minimum, first quartile, median, third quartile, and maximum. The single respect in which a raincloud plot differs is that it refuses to discard the underlying data, displaying the raw points alongside those exact same summary statistics.[1]

Analysts have relied on boxplots to compare groups and track variables since the 1970s. But a boxplot can easily hide a bimodal distribution, which occurs when data clusters heavily at two distinct extremes rather than grouping neatly in the middle.[4]

If a medical trial produces excellent results for half the patients and terrible results for the other half, a boxplot might show a perfectly healthy median. The summary statistics look identical to a uniform distribution, completely masking the polarized reality of the outcomes.[4]

The symmetric violin plot attempted to solve this problem by drawing the continuous probability density of the data. However, it mirrors the exact same density curve on both sides of the axis, wasting horizontal space while still obscuring the individual observations.

Enter the raincloud plot, formalized in 2021 by Micah Allen and colleagues. It combines a split-half density curve, a narrow boxplot, and raw jittered points into a single, space-efficient graphic that leaves no room for hidden clusters.[1]

The Anatomy of a Raincloud

The visualization gets its name from its distinct, layered visual components. The "cloud" is a half-violin plot, which relies on a kernel density estimate to show the continuous probability distribution of the data across the measurement axis.

By slicing the traditional violin plot in half, the raincloud plot reclaims 50 percent of the horizontal space. This allows the chart to remain compact while making room for the actual data points that generated the curve.[4]

Below the cloud sits the "rain," which consists of the raw, individual data points. These points are jittered, meaning a small amount of random noise is added to their position to prevent them from overlapping into a solid, unreadable line.

Sandwiched between the cloud and the rain is a standard, flattened boxplot. This preserves the scannable summary statistics—specifically the median and the interquartile range—that researchers and journal editors expect to see when comparing multiple groups.[1]

Together, these three elements provide a complete and transparent picture. A reader can instantly spot the median, trace the overall shape of the distribution, and verify the sample size and outliers through the raw points.[2]

Exposing Hidden Bimodal Clusters

The primary advantage of the raincloud plot is its ability to expose multimodal distributions. When data clumps into two or more distinct groups, the "cloud" density curve develops multiple distinct peaks that are impossible to miss.[4]

A standard boxplot completely fails in this scenario. A dataset with 500 points at a value of 10 and 500 points at a value of 90 will produce the exact same boxplot as 1,000 points evenly spread between 10 and 90.[4]

"Previously common approaches... have been criticized for distorting effect size and hiding underlying patterns in the raw data," noted the researchers who developed the raincloud framework in their 2021 Wellcome Open Research paper. They argued that transparent plotting tools are essential for robust science.[1]

By showing the raw "rain," the plot also reveals severe sample size disparities. If one group contains 50 observations and another contains 5,000, the density curves might look similar, but the raw points immediately expose the imbalance.[2]

This transparency is crucial in fields like neuroscience and psychology, where small sample sizes are common. A smooth density curve can imply a robust finding, but the sparse rain underneath might reveal that the curve is based on just twelve participants.[1]

The Mechanics of Split Densities

Creating the "cloud" requires calculating a kernel density estimate, which smooths discrete data points into a continuous probability curve. The algorithm uses a bandwidth parameter to determine how closely the curve hugs the raw data.

If the bandwidth is too narrow, the cloud becomes jagged and over-interprets random noise. If it is too wide, the cloud over-smooths the data, potentially erasing the very bimodal clusters the plot is designed to reveal.[4]

The split-density approach solves the inherent redundancy of the traditional violin plot. Because a standard violin plot is perfectly symmetrical, the left side conveys the exact same mathematical information as the right side.

Slicing the violin in half allows the raincloud plot to stack multiple distributions vertically or horizontally without visual clutter. This makes it possible to compare five or six different groups in the same chart area that two standard violin plots would require.[4]

The technique has gained rapid adoption in open-source statistical software. Packages like the dedicated raincloud repository on GitHub have democratized the format, providing ready-to-use code for researchers across disciplines who want to upgrade their charts.[3]

Jittering and Visual Transparency

The "rain" component relies heavily on a technique called jittering. Without jittering, 100 data points with the exact same value would stack directly on top of each other, appearing as a single, misleading dot on the page.

Jittering adds a tiny, random offset to each point along the non-measure axis. This separates the points visually, allowing the reader to perceive the true density of the cluster without altering the actual data values being measured.

However, jittering introduces its own visual challenges. If the random offset is too large, the points begin to bleed into adjacent categories, confusing the reader about which specific group a given observation actually belongs to.[4]

To manage this, modern raincloud implementations often use structured jittering or beeswarm algorithms. These methods arrange the points deterministically, ensuring they pack tightly together without overlapping or drifting too far from their baseline.

The combination of the smoothed cloud and the discrete rain provides a built-in reality check. If the density curve suggests a massive peak, but the raw points show only a few outliers driving the estimate, the reader can immediately discount the artifact.[2]

Overcoming the Boxplot Legacy

Despite their flaws, boxplots remain deeply entrenched in scientific publishing. They were invented by John Tukey in the 1970s, an era when charts had to be drawn by hand and calculating a continuous density curve was computationally impossible.[4]

"The raincloud plot combines a cloud of points with a box plot and a one-sided violin plot," explains the documentation for the JASP statistical software. "This combination allows the viewer to obtain a wealth of information about a variable from a single plot."

This layered approach means analysts do not have to generate three separate charts to convey the same information. A single graphic handles the heavy lifting, reducing cognitive load for the reader while maximizing the data-to-ink ratio on the page.[4]

Transitioning away from boxplots requires overcoming decades of inertia. Many peer reviewers still demand traditional bar charts with standard error bars, a format derisively known as dynamite plots because they hide almost all underlying data variance.[1]

Raincloud plots offer a compromise. By retaining the standard boxplot in the center of the visualization, they provide the familiar summary statistics that traditionalists expect, while adding the transparency that modern data science demands.[2]

Implementation Across Statistical Software

The rapid spread of raincloud plots is largely due to their open-source availability. The original 2021 Wellcome Open Research paper included ready-to-use code for R, Python, and Matlab, removing the technical barrier to entry.[1]

In the R ecosystem, analysts can build raincloud plots by layering geometries. Users combine split-violin functions for the cloud, standard boxplot functions for the summary, and jitter functions for the rain to create a cohesive final graphic.

Python users typically rely on specialized libraries that offer wrapper functions to generate the complex layered plot with a single line of code. This accessibility has driven adoption across corporate data science teams.[3]

For researchers who do not write code, graphical interfaces have integrated the tool. JASP, an open-source statistical program, added native raincloud plot support in 2021, allowing users to generate the visualization with a single checkbox.

The shift toward raincloud plots represents a broader movement toward scientific transparency. By refusing to hide the raw data behind summary statistics, analysts are forcing their visualizations to be as honest as their underlying research.[4]