If you want to draw meaningful conclusions from a dataset, you should visualize it as it might otherwise mislead you (Anscombe, 1973; Matejka & Fitzmaurice, 2017). Raincloud plots are great for this because they provide a maximum of statistical information while allowing for ‘inference at a glance’ from key summary statistics (Allen et al., 2021). Figure 1 visualizes the body mass of penguins (Horst et al., 2020) with a raincloud plot: It consists of raw data points (the rain) and half a violin which shows their probability density (the cloud). Further, key summary statistics are visualized by a box plot. Instead of a boxplot, it is also possible to show the mean with a confidence interval (see following examples). From Figure 1 we learn that most of the penguins weigh between 3000 and 5000g and that heavier penguins also tend to have longer flippers (as shown by darker points).
Figure 1. In this Sample, Most Penguins Weigh 3000 – 5000g and Heavier Ones have Longer Flippers.
Raincloud plots paint a comprehensive, transparent picture of the data and have several advantages which become apparent when comparing Figures 2 and 3 below. Both figures show the bill length for penguins on three different islands. Figure 2 is the “standard” figure, which is relatively impoverished (see also “barbarplots“):
Figure 2. Mean (and 95% Confidence Interval) of Bill Length for Penguins on Three Different Islands.
In contrast, the raincloud plot below shows all of the data, together with the means and associated 95% confidence intervals:
Figure 3. The Distribution of Bill Length in Penguins on Three Different Islands. Intervals around means show 95% confidence intervals.
It is evident that by just plotting means (Figure 2), a lot of information remains hidden. In contrast, the raincloud plot (Figure 3) reveals three crucial aspects of the data. First, the observations stretch much further than the confidence intervals. The raw data is more than its central tendency. Second, each of the three groups has some outliers. Third, the distribution of bill length on Dream island is bimodal (as shown by the violin and the clustering of the raw data points. This suggests that the data should be inspected further. In fact, adding a second factor, species, reveals that there live two different species on Dream island (see Figure 4). This is something important that one ought to notice, but is impossible to see when visualizing only means (Figure 2).
Figure 4. Different Penguin Species Live on Biscoe and Dream Island. Intervals around means show 95% confidence intervals.
From the points and violins in raincloud plots it is possible to check distributions (e.g., bimodal vs. normal distribution); something that is highly relevant for statistical tests with distributional assumptions (e.g., t-tests of normal distributions). But further, the degree to which violins of groups overlap also informs us about differences between groups (see Figure 4). For example, while green Adelie penguins do not notably vary in their bill length across islands, brown Gentoo penguins and grey Chinstrap penguins have remarkably longer bills compared to green Adelie penguins.
From version 0.19 onward, JASP offers a flexible, generic, and user-friendly way to create raincloud plots. The raincloud functionality in JASP is based on the R package ggrain (Judd, van Langen, Allen, & Kievit, 2023) and includes the following features:
- Intuitive graphical user interface – no programming required (see Figure 5)
- Add up to two factors (e.g., 4 x 3 design) and a covariate
- Show means and confidence intervals, connect means with lines
- Fine-tune elements (e.g., color, position, or size)
- Connect individual observations over time to inspect trends and outliers
- Inspect key statistics in a table (e.g., N per group, median, confidence intervals)
To create raincloud plots for your own data, download the latest version of JASP, select “Raincloud Plots” from the Descriptives Module (see Figure 5), and you are ready to roll.
Figure 5. The Intuitive Graphical User Interface for Raincloud Plots in JASP.
About The Author
Vincent Ott
Vincent Ott is a Research Master student at the University of Amsterdam.
References
Anscombe, F. J. (1973). Graphs in statistical analysis. The American Statistician, 27(1), 17-21.
Allen, M., Poggiali, D., Whitaker, K., Marshall, T. R., van Langen, J., & Kievit, R. A. (2021). Raincloud plots: a multi-platform tool for robust data visualization [version 2; peer review: 2 approved]. Wellcome open research, 4:63. https://doi.org/10.12688/wellcomeopenres.15191.2
Horst, A. M., Hill, A.P., & Gorman, K.B. (2020). palmerpenguins: Palmer Archipelago (Antarctica) penguin data. R package version 0.1.0. https://allisonhorst.github.io/palmerpenguins/. https://doi.org/10.5281/zenodo.3960218
Judd, N., van Langen, J., Allen, M., & Kievit, R. A. (2023). ggrain: A Rainclouds Geom for ‘ggplot2’. R package version 0.0.3. CRAN 2023, https://CRAN.R-project.org/package=ggrain.
Matejka, J., & Fitzmaurice, G. (2017). Same stats, different graphs: Generating datasets with varied appearance and identical statistics through simulated annealing. Proceedings of the 2017 CHI Conference on Human Factors in Computing Systems, 1290–1294. https://doi.org/10.1145/3025453.3025912
About the author
