R | Robin Lovelace

Data Science for Transport Planning: 2 day course

Thu, 18 Sep 2025 10:00:00 +0100

Note: tickets are now on sale! Get yours at store.leeds.ac.uk.

Based on demand, I’m running a 2-day course with colleagues teaching modern data science skills for transport planning, focussed on transport planning practitioners and transport planning researchers looking to develop their skills for real-world applications.

Any questions welcome, looking forward to seeing you at the course!

Inaugural Lecture: Data Science for Future-Proof Transport Planning

Thu, 08 May 2025 16:30:00 +0100

Mapping, classifying, and integrating diverse street network datasets

Thu, 24 Apr 2025 14:30:00 +0100

See a video from the talk below:

Street Networks in R

Fri, 29 May 2020 00:00:00 +0000

CycleStreets.net in the Propensity to Cycle Tool

Wed, 21 Dec 2016 00:00:00 +0000

After 2 years in the making, the paper describing the Propensity to Cycle Tool (PCT), and the thinking behind it, has finally been published (Lovelace et al. 2016). The PCT is an online tool for helping to decide where to prioritise cycling policies such as new cycle paths.

The PCT would not have been possible without CycleStreets.net, so as well as describing the PCT and it’s use of their routing services, this article serves as a big Thank You from PCT to CycleStreets.net.

What is the Propensity to Cycle Tool?

For those new to the PCT, it’s an online tool for helping to decide where to prioritise cycling policies such as new cycle paths. It lives at http://www.pct.bike - check it out!

The context of its development is explained in the accompanying video on that page. This article reports how the tool itself works and how it uses data from CycleStreets.net.

The PCT is best understood by using it to explore current cycling levels, at regional, area, desire line, route and route network levels. We will take a look at how the PCT works at each of these levels, after a brief look at the scenario results at the regional level (the senarios are described in more detail in the paper).

Under the 2011 Census scenario, the PCT represents levels of cycling to work based on the Census. This is a reasonable proxy for levels of utility cycling overall. We used origin-destination (OD) data from the Census as the basis of the PCT as this is best publicly available dataset on English travel patterns. The input data is described in the paper and can be freely downloaded from the official http://wicid.ukdataservice.ac.uk/ website.

The regional picture and scenarios

The first thing the user sees on the front page is a map of England, broken into 44 regions:

We used deliberately large regions because successful cycling plans should be strategic and joined up, covering both large areas and large spans of time. This discourages the stop-start investment plans that have typified funding for active travel.

By hovering over different regions, the user can see what the current level of cycling to work is. We can discover that West Yorkshire has a low current level of cycling to work, 1.3% in the 2011 census, and that Cambridgeshire has a relatively high (but low by Dutch standards) level of cycling of 9.7%.

An exciting feature of the PCT is its ability to allow the user to imagine ‘cycling futures’. This can be seen on the front page map by clicking on the different scenarios (set to Census 2011 by default). We can see, for example, that under the Government Target to double cycling levels by 2025, West Yorkshire’s level would rise to 3.3% (more than a doubling) whereas Cambridgeshire would see cycling levels grow to 13.7% (a larger rise in absolute terms):

Under the Go Dutch scenarios, these regions would see 23.1 and 13.5% of people cycling to work, respectively. This represents a huge leveling-out of cycling levels across the country, but still highlights the fact that some regions have higher cycling potentials than others, due to average trip distances and levels of hilliness.

Cycling levels at the area level

To launch the PCT for a region, click on it. Try clicking on West Yorkshire. You should be presented with the following image, which shows the area-based level of cycling to work from the 2011 Census. (When using the PCT, it is worth remembering that the visualisations work for every scenario.)

This shows that West Yorkshire has very low levels of cycling to work, hovering around 1% to 2% in most places. This suggests strongly that the region has low levels of utility cycling overall (despite the successes of the region’s sport cyclists). There is a cluster of zones with a higher level of cycling to the north of Leeds city centre (around Headingly) but even there the percentage of people cycling as their main mode of travel to work does not exceed 5%.

Cycling potential at the desire line level

This is all useful information, especially when we look at how the cycling potential could shift in the future. However, it provides little information about where current and future cyclists actually go. This is where the desire line level can be useful. This can be selected by clicking on the Straight Lines option from the Cycling Flows dropdown menu. The results (zoomed in for Leeds) are shown in the figures below (see Figure 3 in the paper).

What the above figures show is that as the level of cycling increases in a city, the spatial distribution of cycling can be expected to change. Under current conditions (be they related to socio-demographics or infrastructure or other factors), cycling in Leeds is dominated by the travel corridor to the north of the city centre. Yet there are clearly many short trips taking place from the south into the centre, as illustrated by the high cycling potential south of the city under the Go Dutch scenario.

Allocating cycling potential to the route network

This is where CycleStreets.net comes into play.

We know how many people go from A to B by cycling from Census data. But we have very little idea of how they are likely to travel. This is where the routing algorithm of CycleStreets.net comes in handy. We used their routing API to estimate the ‘fastest’ route for all short (well, up to 20 km in Euclidean distance) desire lines in England.

Not only does CycleStreets.net allow us to find all the fastest routes, a good indication of where new infrastructure should be built as people (especially women and elderly) have a strong preference for cycling along the most direct routes.

The results of all this routing work is illustrated in the future below, which shows the fastest and quietest routes associated with the top cycled routes in Leeds.

Interestingly, the big fat line up to the north-west is Otley Road, well-known to have very high level of cycling. This also shows up in Strava data as having high current levels of cycling:

This is not formal validation but it is a good sign that the PCT and other data sources line-up for the current level of cycling. The big question is whether the PCT’s estimates of cycling levels under various cycling futures, including Go Dutch.

Here is not the place to answer such a question. Only the passage of time, and commitment from people (perhaps informed by models such as the PCT) to sustainable travel will help answer that one.

There is much more to say about the use of CycleStreets.net in the PCT but it gets rather technical very quickly. Suffice to say at this stage that it involved writing lots of code in R, a language for statistical programming, and that this has now resulted in the publication of stplanr, an R package for sustainable transport. (For more on how to install R and (for bells and whistles) RStudio, which this blog post was written in, please see the relevant sections of the book Efficent R Programming (Gillespie and Lovelace, 2016).) With R installed, stplanr can be installed with:

install.packages("stplanr")

With this package installed, you can start using the CycleStreets.net routing algorithm with the following function:

library(stplanr)
route = route_cyclestreet(from = "Leeds", to = "Cambridge")

which results in spatial data, which can be visualised as follows:

library(leaflet)
leaflet() %>% addTiles() %>% addPolylines(data = route)

There is much more I could say about the technical side of things but at the request of the editors I will leave it there for now. For more info please see the stplanr vignette.

I plan to follow this overview article up with a more technical blog post in the New Year. Watch this space!

Reference

Lovelace, R., Goodman, A., Aldred, R., Berkoff, N., Abbas, A., & Woodcock, J. (2016). The Propensity to Cycle Tool: An open source online system for sustainable transport planning. Journal Of Transport And Land Use, 0. doi:http://dx.doi.org/10.5198/jtlu.2016.862

Coxcomb plots and 'spiecharts' in R

Fri, 27 Dec 2013 00:00:00 +0000

After switching to a new site I decided to revive some old posts. I found this one that was written back 7 years ago (in January 2021 when this update was written), back in December 2013. The results of the book are now published in the book Low Impact Living: A Field Guide to Ecological, Affordable Community Building (Chatterton, 2015, for more info on the Lilac project in particular and cohousing in general see lilac.coop/resources/). I was amazed to find that, with some tweaks, the ggplot2 code still ran.

I was contacted recently by a housing organisation who wanted an attractive visualisation of their finances, arranged in a circular form. Because there were two 4 continuous variables to include, all of which were proportions of each other, the client suggested a plot similar to a pie chart, but with each segment extending out a different radius from the segment. I realised later that what I had been asked to make was a modified coxcomb plot, invented by Florence Nightingale to represent statistics on cause of death during the Crimean War. In fact, I had been asked to make a “spie chart.” This post demonstrates, for the first time to my knowledge, how it can be done using ggplot2. A reproducible example of this, including sample data input, can be found on the project’s github repository: https://github.com/Robinlovelace/lilacPlot . Please fork and attribute as appropriate!

Reading and looking at the data

This is the original dataset I was given:

u <- "https://github.com/Robinlovelace/lilacPlot/raw/master/F2.csv"
f <- read.csv(u)
knitr::kable(f[1:3, ])

H	Value	Value.P	Allocation	Deposit	Captial	Debt	Cap	Contribution	Repayments
q	163827	0.065	0.979	16382	147445	0	2457.405	1287.24	0.00
a	165994	0.066	1.022	16599	5488	138847	2489.910	208.02	208.02
z	159425	0.063	0.933	15943	76632	63601	2391.375	995.46	995.46

Without worrying too much about the details, the basics of the dataset are as follows:

One observation per row, these will later be bars on the box plot
Two components of data - captital and revenue
Different orders of magnitude: some data is in absolute monetary terms, some in percentages

Base on the above points, a prerequisite was to create preliminary plots and manipulate the data so it would better fit in a coxcomb plot.

The first stage, however, is to demonstrate how the addition of coord_polar to a barchart can conver it into a pie chart:

library(ggplot2)
(p <- ggplot(f, aes(x = H, y = Allocation)) + geom_bar(color = "black", stat = "identity",
width = 1))

plot of chunk unnamed-chunk-2

p + coord_polar()

The above example works well, but notice that all the bars are of equal widths. What we want is to be proportional to a value (variable “Value”) of each observation. To do this we use the age-old function cumsum, as described in an answer to a stackexchange question.

w <- f$Value
pos <- 0.5 * (cumsum(w) + cumsum(c(0, w[-length(w)])))
(p <- ggplot(f, aes(x = pos)) + geom_bar(aes(y = Allocation), width = w, color = "black",
stat = "identity"))

p + coord_polar(theta = "x") + scale_x_continuous(labels = f$H, breaks = pos)

Finally a spie chart has been created. After that revelation, it was essentially about adding the ‘bells and whistles’, including a 10% line to represent how much more or less than their share each observation was paying.

Adding the 10 %

f$Deposit/f$Value

## [1] 0.09999573 0.09999759 0.10000314 0.09999837 0.10000120 0.10000000
## [7] 0.10000311 0.10000085 0.10000511 0.10000356 0.09999676 0.09999700
## [13] 0.09999812 0.10000511 0.10000085 0.10000240 0.10000000 0.10000694
## [19] 0.09999901 0.09999883

# add 10% in there
p <- ggplot(f)
p + geom_bar(aes(x = pos, y = Allocation), width = w, color = "black", stat = "identity") +
geom_bar(aes(x = pos, y = 0.1), width = w, color = "black", stat = "identity",
fill = "green") + coord_polar()

# make proportional to area
f$Allo <- sqrt(f$Allocation)
p <- ggplot(f)
p + geom_bar(aes(x = pos, y = Allo, width = w), color = "black", stat = "identity") +
geom_bar(aes(x = pos, y = sqrt(0.1), width = w), color = "black", stat = "identity",
fill = "green") + coord_polar()

## Warning: Ignoring unknown aesthetics: width
## Warning: Ignoring unknown aesthetics: width

# add capital
capital <- (f$Captial + f$Deposit)/(f$Value) * f$Allocation
capital <- sqrt(capital)
p + geom_bar(aes(x = pos, y = Allo, width = w), color = "black", stat = "identity") +
geom_bar(aes(x = pos, y = capital, width = w), color = "black", stat = "identity",
fill = "red") + geom_bar(aes(x = pos, y = sqrt(0.1), width = w), color = "black",
stat = "identity", fill = "green") + coord_polar() + scale_x_continuous(labels = f$H,
breaks = pos)

## Warning: Ignoring unknown aesthetics: width
## Warning: Ignoring unknown aesthetics: width
## Warning: Ignoring unknown aesthetics: width

# add ablines
p + geom_bar(aes(x = pos, y = Allo, width = w), color = "grey40", stat = "identity",
fill = "lightgrey") + geom_bar(aes(x = pos, y = capital, width = w), color = "grey40",
stat = "identity", fill = "red") + geom_bar(aes(x = pos, y = sqrt(0.1),
width = w), color = "grey40", stat = "identity", fill = "green") + geom_abline(intercept = 1,
slope = 0, linetype = 2) + geom_abline(intercept = sqrt(1.1), slope = 0,
linetype = 3) + geom_abline(intercept = sqrt(0.9), slope = 0, linetype = 3)

## Warning: Ignoring unknown aesthetics: width
## Warning: Ignoring unknown aesthetics: width
## Warning: Ignoring unknown aesthetics: width

plot of chunk unnamed-chunk-4

# calculate vertical ablines of divisions
v1 <- 0.51 * f$Value[1]
v2 <- cumsum(f$Value)[17] + f$Value[18] * 0.31
v3 <- cumsum(f$Value)[17] + f$Value[18] * 0.64
p + geom_bar(aes(x = pos, y = Allo, width = w), color = "grey40", stat = "identity",
fill = "lightgrey") +
geom_vline(x = v1, linetype = 5, xintercept = 0) +
geom_vline(x = v2, linetype = 5, xintercept = 0) +
geom_vline(x = v3, linetype = 5, xintercept = 0) +
coord_polar()

## Warning: Ignoring unknown aesthetics: width

## Warning: Ignoring unknown parameters: x
## Warning: Ignoring unknown parameters: x
## Warning: Ignoring unknown parameters: x

# putting it all together
p <- ggplot(f)
p + geom_bar(aes(x = pos, y = Allo, width = w), color = "grey40", stat = "identity",
fill = "lightgrey") + geom_bar(aes(x = pos, y = capital, width = w), color = "grey40",
stat = "identity", fill = "red") + geom_bar(aes(x = pos, y = sqrt(0.1),
width = w), color = "grey40", stat = "identity", fill = "green") + geom_abline(intercept = 1,
slope = 0, linetype = 2) + geom_abline(intercept = sqrt(1.1), slope = 0,
linetype = 3) + geom_abline(intercept = sqrt(0.9), slope = 0, linetype = 3) +
geom_vline(x = v1, linetype = 5, xintercept = 0) +
geom_vline(x = v2, linetype = 5, xintercept = 0) +
geom_vline(x = v3, linetype = 5, xintercept = 0) +
coord_polar() + scale_x_continuous(labels = f$H, breaks = pos) +
theme_classic()

## Warning: Ignoring unknown aesthetics: width
## Warning: Ignoring unknown aesthetics: width
## Warning: Ignoring unknown aesthetics: width

## Warning: Ignoring unknown parameters: x
## Warning: Ignoring unknown parameters: x
## Warning: Ignoring unknown parameters: x

The above looks great, but ideally, for an ‘infographic’ feel, it would have no annoying axes clogging up the visuals. This was done by creating an entirely new ggpot theme.

Create theme with no axes

theme_infog <- theme_classic() + theme(axis.line = element_blank(), axis.title = element_blank(),
axis.ticks = element_blank(), axis.text.y = element_blank())
last_plot() + theme_infog

Creating a ring

To add the revenue element to the graph is not a task to be taken likely. This was how I tackled the problem, by creating a tall, variable-width bar chart first, and later adding the original spie chart after:

f$Cap.r <- f$Cap/mean(f$Cap) * 0.1 + 1.2
f$Cont.r <- f$Contribution/mean(f$Cap) * 0.1 + 1.2
f$Rep.r <- f$Cont.r + f$Repayments/mean(f$Cap) * 0.1
f$H <- c("a", "b", "c", "d", "e", "f", "g", "h", "i", "j", "k", "l", "m", "n",
"o", "p", "q", "r", "s", "t")
p <- ggplot(f)
p + geom_bar(aes(x = pos, y = Allo, width = w), color = "grey40", stat = "identity",
fill = "lightgrey")

## Warning: Ignoring unknown aesthetics: width

# we need the axes to be bigger for starters - try 1.3 to 1.5
p + geom_bar(aes(x = pos, y = Cap.r, width = w), color = "grey40", stat = "identity",
fill = "white") + geom_bar(aes(x = pos, y = Rep.r, width = w), color = "grey40",
stat = "identity", fill = "grey80") + geom_bar(aes(x = pos, y = Cont.r,
width = w), color = "grey40", stat = "identity", fill = "grey30") + geom_bar(aes(x = pos,
y = 1.196, width = w), color = "white", stat = "identity", fill = "white")

## Warning: Ignoring unknown aesthetics: width
## Warning: Ignoring unknown aesthetics: width
## Warning: Ignoring unknown aesthetics: width
## Warning: Ignoring unknown aesthetics: width

last_plot() + geom_bar(aes(x = pos, y = Allo, width = w), color = "grey40",
stat = "identity", fill = "grey80") + geom_bar(aes(x = pos, y = capital,
width = w), color = "grey40", stat = "identity", fill = "grey30") + geom_bar(aes(x = pos,
y = sqrt(0.1), width = w), color = "grey40", stat = "identity", fill = "black") +
geom_abline(intercept = 1, slope = 0, linetype = 5) + geom_abline(intercept = sqrt(1.1),
slope = 0, linetype = 3) + geom_abline(intercept = sqrt(0.9), slope = 0,
linetype = 3) + coord_polar() + scale_x_continuous(labels = f$H, breaks = pos) +
theme_infog

## Warning: Ignoring unknown aesthetics: width
## Warning: Ignoring unknown aesthetics: width
## Warning: Ignoring unknown aesthetics: width

Just inner

After all that it was decided it looked nicer with only the inner ring anyway. Here is the finished product:

p <- ggplot(f)
p + geom_bar(aes(x = pos, y = Allo, width = w), color = "grey40", stat = "identity",
fill = "grey80") + geom_bar(aes(x = pos, y = capital, width = w), color = "grey40",
stat = "identity", fill = "grey30") + geom_bar(aes(x = pos, y = sqrt(0.1),
width = w), color = "grey40", stat = "identity", fill = "black") + geom_abline(intercept = 1,
slope = 0, linetype = 5) + geom_abline(intercept = sqrt(1.1), slope = 0,
linetype = 3) + geom_abline(intercept = sqrt(0.9), slope = 0, linetype = 3) +
coord_polar() + scale_x_continuous(labels = f$H, breaks = pos) + theme_infog

## Warning: Ignoring unknown aesthetics: width
## Warning: Ignoring unknown aesthetics: width
## Warning: Ignoring unknown aesthetics: width

ggsave("just-inner.png", width = 7, height = 7, dpi = 800)

Moving house by bicycle trailer

Sun, 13 Oct 2013 00:00:00 +0000

Back in 2009 I moved from York to Sheffield to start a PhD on the Energy Futures 4 year Doctoral Training Centre. Young and naive, some friends and I decided to do it by bicycle, with one bicycle trailer filled to the brim and 4 large panniers carrying most of my stuff.

4 years later and I’ve moved again, this time to Leeds. I’ve got a new a job at Leeds University, working on the Geotalisman project in the department of geography. Older and hopefully wiser, I decided to move house by bicycle again: it’s shorter than the trip to York (40 miles, not 60), I was more organised with planning (providing a month’s notice for friends and family to attend and after 4 years in Sheffield I knew many people who would be up for the ride. To make it more exciting, this time we brought music, in the form of a 12v sound system. See the map, words, images and video below for the full low-down on this epic ride.

Introduction: preparation and kit

The first stage was to recruit people. With a month’s notice, I managed to get around 15 people to sign-up to the challenge. One issue for some people was lack of a suitable bicycle, so additional ones were found and leant out. The best of these was a large tandem leant to me by a friend. Trailers would be a major limiting factor for most people, but, as a member of EWB Sheffield who had helped on the bike trailer project, this was less of an issue. We borrowed one trailer from EWB, and the other one was my own, purchased from Ebay for around £60.

The plan was to set-off at 11 am. But it took almost an hour for to say final goodbyes, for everyone to sort their stuff out and to decide who would be riding which bikes. So it was essentially a midday start.

Stage one: navigating the urban jungle

Getting out of any major city is usually a challenge by bicycle, and Sheffield is no exception, with the giant M1 dominating the norther route. Fortunately we were ably guided for the first 10 miles or so by Sustrans Ranger Simon Geller, who took us safely out via Northern General hospital and out onto the car-free Trans Pennine Trail to Leeds.

It was a major relief to get off the road, despite the hassle of hauling the big bike trailers over the barriers - see video below. After that the route was quite tranquil until, realising we were behind schedule, we decided to just get onto the A61 for the final 5 miles to Barnesley, where we had agreed to meet other people and from where people could go home early.

Stage two: Barnesley to Wakefield

We were all quite tired in Barnesley, so a good rest was had by all in the lush sunshine. Our only puncture of the trip was fixed, we ate and waited for Nokolai - a new recuit from Leeds, who would help out greatly with lugging the trailers. By the time we set off again it was almost 5. Conscious of the need to go faster to arrive before sunset, we changed our route plan completely and decided just to get our heads down and pedal along the A61 all the way.

The difference in speed was noticeable. With fresh legs towing the bicycle trailers and some of the less experienced cyclists leaving us in Barnesley it felt like we were flying. The road is wide and flat, so it did not feel especially dangerous. Cycling in convoy is probably safer anyway because people notice a flock of cyclists more than a single one or two. There was a hard climb up to Wakefield where we had a number of small stops. The overall feeling was “let’s just get there” so we pressed on, after another couple of people headed to the train station. The proximity of rail stations to the route was one of its major advantages, allowing people to leave when they wanted to, and meaning that setting off was not such a major commitment. It was all downhill from Wakefield, or so we thought.

The final straight: into Leeds

As dusk descended, the group re-united. With various distractions, we had become dispersed on the fast downhill stretch directly north of Wakefield. As the traffic volume increased again near central Leeds, and as our level of visibility decreased it made sense to stick together. At this stage we turned the volume up on the bicycle trailer soundsystem. This helped a great deal to raise spirits that were flagging after a longer than expected ride.

We arrived in Leeds city centre at around 8pm. Knackered, we went directly to the Wharf Chambers for a pint and some food. Before I knew it we were watching live music, most of my Sheffield friends had gone back after heartfelt goodbyes, and it was approaching midnight. But we still had two bicycle trailers to get back to my final destination, which was still another mile away, uphill. This final stretch was a real slog, but, thanks to Nikolai’s indominatable spirit we made it there before 12. Finally I was home!

Why move house by bike trailer?

With all the extra hassle involved, you may wonder why I went to all these lengths. A van could have done the job in under 2 hours, with a lot less sweat and in fact I used this option a week later to move residual stuff that I could not fit into the trailers. So why bother?

The main reason for doing this was social: instead of having a big ‘piss-up’ to see me off I much preferred people stayed in the night before. That way I got to have a sober conversation with some of my best friends in Sheffield whilst doing something fun. For a few people on the ride, it was the furthest they had ever travelled by bicycle, and this alone was worth-it for me.

Another reason was that bicycle trailer moves are cheap. To hire a van can easily cost £100 for the day, and the price of breakfast for the ride (and buying many people drinks in Leeds) was considerably less.

Finally, bicycling in general is healthy. It’s good for you because you get your heart pumping and some fresh air. It’s healthy for the environment too, as you get to leave the car at home.

Conclusion

Moving house by bicycle trailer was one of the the most enjoyable experiences of 2013 for me. Being surrounded by friends, music and sunshine was infinitely preferable to the van-based alternative. The even also brought people together in a way that simply going out and drinking alcohol cannot. It was fun, free (in every sense of the word) but definitely not fast, so if you have a tight schedule to move bicycle trailers may not be the best way forward. There are also obvious physical limits in how far you can actually move bulky items in a single day and 40 miles was already pushing the limit for us.

I would strongly recommend anyone who’s moving house to a not too distant new location to consider bicycle trailers. They are so much more fun and sociable than just jumping in a van, you just can’t go wrong. If this little article on the matter has not persuaded you, check the video below, sounds and all.

I will be forever grateful to all the people who helped out with this move.