Developing an R package

The guide also contains technical instructions for authors on how to edit the settings of their visualisation software to better style their graphics. The guide covers examples of doing so in R and Python (with examples added by the community for other software), with R being used often (both base R and {ggplot2}). To help authors with styling their charts to fit in with the guidance in this document, we have developed an R package, {RSSthemes}.

This R package:

• Stores and load font files
• Defines colour palettes with RSS colours
• Creates theme_*() and scale_* functions for {ggplot2}
• Sets (and unsets) different base R defaults

Let’s see an example in action! Let’s say you have simple data set that you want to use to make a bar chart with {ggplot2}

library(ggplot2)
plot_df <- data.frame(x = LETTERS[1:4], y = 1:4)
ggplot(data = plot_df,
       mapping = aes(x = x, y = y, fill = x)) +
  geom_col()

You can restyle it, using the {RSSthemes} package and two lines of code:

library(RSSthemes)
ggplot(data = plot_df,
       mapping = aes(x = x, y = y, fill = x)) +
  geom_col() +
  scale_fill_rss_d(palette = "signif_qual") +
  theme_significance()

As of March 2024, the guide has been viewed by over 14,000 users, in 152 countries.

TidyTuesday

TidyTuesday is a weekly data challenge aimed at the R community. Every week a new dataset is posted alongside a chart or article related to that dataset, and ask participants explore the data. You can access the data and find out more here.

Below is a selection of my contributions to the TidyTuesday challenge. All contributions and the code used to generate them can be found on GitHub. You can also browse through all contributions using this dashboard.

Plotnine contest

Plotnine is a visualisation library that brings the Grammar of Graphics to Python, and the 2024 Plotnine Contest aims to bring the community together to create and share with others great plotnine examples. You can read the blog post about how I created this annotated area chart for my entry into the Plotnine contest. This visualisation was a runner up in the Plotnine contest.

#30DayChartChallenge

The #30DayChartChallenge is a data visualisation challenge organised by Cédric Scherer and Dominic Royé. Participants make one chart each day of the challenge, inspired by the daily prompt. The prompts are also split across 5 different categories. Check out the challenge at twitter.com/30DayChartChall. I used a range of tools to create my charts, including R, Python, Tableau, rawgraphs, Inkscape, and Figma.

#30DayMapChallenge

The #30DayMapChallenge is a data visualisation challenge organised by Topi Tjukanov. Participants make one map each day of the challenge, inspired by the daily prompt. Check out the challenge on GitHub, or see contributions on the official website.

Tableau Public

You can find examples of my visualisations created using Tableau on my Tableau Public profile.

Other challenges

Viz For Social Good (VFSG) volunteers create informative and impactful data visualizations for mission-driven organizations across the globe. Check out the organisation here.

Papers and articles

PhD Thesis

Detecting demand outliers in transport systems.
N. Rennie. PhD Thesis. Lancaster University. 2021. doi: doi.org/10.17635/lancaster/thesis/1448.
PhD Thesis

Blog posts

Besides the blog posts you’ll find on this website, I’ve also contributed blog posts to a variety of other organisations:

Software

ggauto: Automatically Create and Style ‘ggplot2’ Charts.
N. Rennie. CRAN. (First released 25 March 2026). doi: doi.org/10.32614/CRAN.package.ggauto.
Software

messy: Create Messy Data from Clean Data Frames.
N. Rennie. CRAN. (First released 3 December 2024). doi: doi.org/10.32614/CRAN.package.messy.
Software

RSSthemes: Royal Statistical Society Palettes and Themes.
N. Rennie. CRAN. (First released 29 August 2023). doi: doi.org/10.32614/CRAN.package.RSSthemes.
Software

ggflowchart: Flowcharts with ‘ggplot2’.
N. Rennie. CRAN. (First released 11 May 2023). doi: doi.org/10.32614/CRAN.package.ggflowchart.
Software

PrettyCols: Pretty Colour Palettes.
N. Rennie. CRAN. (First released 2 September 2022). doi: doi.org/10.32614/CRAN.package.PrettyCols.
Software

See also nrennie.rbind.io/projects for a list of software packages I have developed and published.

ggflowchart

messy

PrettyCols

RSSthemes

aRt

aRtData

colorTools

funcnetout

ggalttext

ggtextcircle

LondonMarathon

national.highways

simnetdemand

usefunc

WBLifeExpectancy

Royal Institution Masterclass

The Royal Institution run Ri Masterclasses - a series of hands-on workshops where students can explore mathematics and computer science beyond the scope of classroom curriculum.

Florence Nightingale Day

In January 2024, I delivered a talk about my career path and data science projects I’ve worked on as part of Lancaster University’s Florence Nightingale Day celebrations, to an audience of around 200 students from local high schools.

STEM Ambassador

I am registered as a STEM Ambassador with STEM Learning - an organisation who deliver teacher CPD (continuing professional development) in STEM subjects, bring STEM role models into schools as part of the STEM Ambassador Programme and providing bespoke, long-term support for groups of schools in collaboration with companies.

You can request a STEM Ambassador here.

Gallery

More examples of interactive generative art made with D3, can be found at nrennie.rbind.io/drawing-with-d3.

Preferences for face-to-face contraceptive services

Data from the Reproductive Health Survey for England was used to look at how people prefer to access contraceptive services. Using a large online survey of women and people assigned female at birth, we found that about one in four still want only in-person appointments rather than phone, online, or hybrid options. People who preferred face-to-face care were more likely to be younger or older people, less educated, struggling financially, living with a disability, and not in a relationship, suggesting that completely digital services might not meet everyone’s needs. We argue that maintaining a mix of in-person and remote contraceptive services is important for fair and person-centred healthcare.

Publication: Preference for face-to-face contraceptive service delivery post-COVID-19 pandemic: a cross-sectional study

Why some areas struggle to recruit and keep doctors

This study explored why some parts of the UK struggle to recruit and keep doctors working there. Through interviews with doctors already working in these under-doctored areas (often rural or deprived communities), we found that a mix of personal life choices, career opportunities, and the characteristics of the place itself influence whether doctors move to and stay in these areas. The research highlights that understanding these factors could help improve doctor recruitment and reduce healthcare inequalities in underserved communities.

Publication: Medical training pathways and underdoctored areas: a qualitative study of doctors working in areas that struggle to recruit and retain

PrettyTypst

RSSpythemes

Motivation and requirements

ScotRail is a publicly-owned company operating rail services in Scotland, owned by the Scottish Government. ScotRail claim to “publish daily figures on our punctuality so customers can see how we performed over the previous 24 hours”. However, these figures are currently (as of April 2026) published as inaccessible, and poorly structured PDF documents. This means the information can be hard to access, and even harder to compare over time. A solution is required that (i) collects existing data and provides it in a more accessible format; and (ii) displays the data to enable comparison.

Features

The dashboard has several features that make it more user friendly and easier to maintain:

• It pre-filters posts based on a set of keywords to ensure posts are relevant to R, Python, and data science.
• The data updates automatically on a schedule, meaning that there is no manual process involved in maintaining the dashboard’s relevance. It can also be triggered manually if required.
• The table is searchable meaning that a user can look for links posted about a specific topic, or by a specific user, or even from a specific website.
• It extracts and displays both the link to the original tweet and the link to the resource, meaning users can browser resources without visiting twitter but also see who and when it was posted.

Tools

The dashboard is built using freely available, open source tools.

• Data: pdf, rvest, gh actions.

• Dashboard: the dashboard is built using Quarto. interactive elements (data downloads, tabsets, tooltips, modals with notes)

• Deployment: the github pages (static site) meaning no server is required in backend, using GitHub Actions to automatically re-deploy the dashboard whenever the data or content changes.

Data

The data is scraped from (poorly structured) PDF documents, so no guarantee is given for the quality of the information provided.

Daily data

Source: www.scotrail.co.uk/performance-and-reliability/previous-daily-statistical-summaries

Data is displayed for January 2026 onwards.

Monthly data

Source: www.scotrail.co.uk/about-scotrail/performance-and-reliability

There is no contents page showing historic data, and each PDF is published under a random, non-structured URL. Data is included from March 2026 onwards, with historic data included if I’ve been able to find it.

Links

The live dashboard can be found online, and the source code is available on GitHub.

Interactive data gaps explorer

Explore the tool at royal-statistical-society.github.io/poverty-data-gaps.

You can read more about the data gaps explorer on the RSS website. The source code can be viewed on GitHub.

Motivation and requirements

When browsing through social media, especially when endlessly scrolling on mobile, we regularly see interesting links about things related to data science. However, those links are either impossible to find again later on desktop or mixed in with lots of non-data science posts if using Twitter’s bookmark feature. We wanted a way to gather together all of the relevant links to interesting data science things without having to do any manual work.

Features

The dashboard has several features that make it more user friendly and easier to maintain:

• It pre-filters posts based on a set of keywords to ensure posts are relevant to R, Python, and data science.
• The data updates automatically on a schedule, meaning that there is no manual process involved in maintaining the dashboard’s relevance. It can also be triggered manually if required.
• The table is searchable meaning that a user can look for links posted about a specific topic, or by a specific user, or even from a specific website.
• It extracts and displays both the link to the original tweet and the link to the resource, meaning users can browser resources without visiting twitter but also see who and when it was posted.

Tools

The dashboard is built using freely available, open source tools.

• Data: the data is extracted through the Twitter API via R; with further processing in R. GitHub Actions is used to automatically run the extraction and processing scripts on a weekly basis.
• Dashboard: the dashboard is built using Shiny in R and makes use of the reactable package.
• Deployment: the dashboard is deployed through shinyapps.io, using GitHub Actions to automatically re-deploy the dashboard whenever the data or content changes. Docker is used to ensure the versions of R and any packages used are consistent, and package updates do not trigger failed deployments.

Read my blog post about creating, deploying, and updating this Shiny app for more information.

Teaching statistics communities

Project description

An outlier is generally considered an abnormal event that doesn’t fit with the pattern of events normally observed. In the transport setting, demand outliers might mean that there are a lot more (or less) passengers than normal, or that they book earlier than normal, or that passengers are willing to pay more than they normally would for the same ticket. There are many things that can cause these changes in demand such as, football matches, weather, or public holidays. Some of these events e.g. Christmas, are known about in advance and can be accounted for in the planning process. Others are not known about. And it’s these changes in demand that we want to identify. Ideally, they’d be identified as early as possible, in order for the transport company to take action.

It’s important to detect changes in demand for multiple reasons. Transport companies set prices and ticket availability based on forecasts of demand. If the demand no longer matches the forecasts, then the prices that have been set are no longer optimal. Then transport companies lose revenue. It could also result in transport services being busier than expected. If you’ve ever gotten on a packed train and been unable to get a seat, you know that incorrectly managed demand isn’t desirable for passengers either.

At the moment, many transport companies employ analysts to monitor the number of bookings. However, as human beings, we’re not very good at spotting things by eye, and we tend to see patterns that don’t exist. This research develops a statistical method to identify demand outliers and send alerts to analysts.

Developing outlier detection methodology

Most transport providers store information about bookings as a time series. For example, for train ABC departing on January 31 at 10:00, there is a series of observations of how many passengers had booked tickets by 3 months per departure, 1 month before departure, 1 week before departure … You get the idea.

Note: we’ll mainly be talking about trains here, but it’s all generalisable to other types of transport systems.

So for a set of departures, we would have a collection of time series that might look a bit like this:

To detect outliers within this set of time series, we use functional analysis. Functional analysis treats each time series of bookings as an observation of a continuous function. We calculate the functional depth of the time series of bookings for each train. I won’t go into all of the details about the equations for calculating the functional depth because they’re not very nice. But generally, depth measures provide us with an ordering of the time series – where the time series closest to the centre i.e., the median, has the highest depth, and that in the tails of the distribution i.e. outliers, have low depth. This allows to consider both changes in magnitude (e.g. a big increase in bookings), and shape of booking patterns (e.g. passengers booking earlier than normal). Outliers are detected by setting a threshold for the functional depth. Any time series with a functional depth below that threshold is classified as an outlier.

Of course, we don’t just want to identify demand outliers in historic data (although this can still be beneficial on its own). We want to identify these outliers as they are happening, so that an analyst can make an adjustment. We found that forecasting the bookings still to come in, and performing the outlier detection on the forecasted bookings, helped us to identify the outliers earlier.

Developing methodology for networks

The vast majority of, if not all, transport systems do not simply consist of a single journey from A to B. There are often many places where passengers can start and end their journey, and multiple combinations to get between the two. It’s quite unlikely that any demand changes will only affect one single part of the transport network e.g. a single leg of a train journey. It’s also quite unlikely that the entire network will be affected by demand changes in the same way at the same time. Before even thinking about demand outliers, we developed a method split up the network into clusters that experience demand in similar ways. Our outlier detection method could then be applied jointly to legs in the same cluster which are likely to share common outliers.

Building a proof-of-concept dashboard

The initial motivation for this research was to support analysts who monitor the number of bookings, sending them alerts when an intervention is required. To communicate the alerts to the analysts, I developed a proof of concept dashboard that uses the results of the statistical methods to generate a list of alerts that analysts can explore in more detail.