James Gleeson 2022-09-18
The PublicHouse dataset is a compilation of publicly available data on the number of dwellings and people in cities and countries around the world. The README provides details about the creation of the dataset, while this note sets out some basic analysis using the R programming language. It also includes the code used to carry out the analysis, so you can re-create or adapt the charts.
This is the second edition of this analysis note, published in May 2022. The dataset has been expanded and updated, but in a more significant change the analysis uses interpolation to fill in gaps between data points, which improves some of the charts.
Let’s start by loading the R packages we’re going to use.
library(tidyverse) # for various analysis functions
library(khroma) # for the Okabe-Ito colour scale
library(ggbraid) # for the 'braid' geom
library(ggdensity) # for shaded probability areas on scatterplot diagrams
library(geomtextpath) # for annotated line charts
library(lubridate) # for working with dates
library(zoo) # for time series interpolationAnother bit of setup is to set some chart design defaults.
theme_set(theme_minimal(base_size = 14))
theme_update(text = element_text(family = "Calibri"),
plot.title = element_text(size = 12, hjust = 0.5)) Now we load the dataset straight from Github.
d <- read_csv("https://raw.githubusercontent.com/jgleeson/PublicHouse/main/dataset.csv") %>%
mutate(ref_date = dmy(ref_date)) As a first step, let’s see what places and years we have data for. PublicHouse includes data on both cities and countries, so let’s start with countries. Each dot is an observation.
d %>%
filter(area_level != "city-region") %>%
mutate(variable = str_to_title(variable)) %>%
ggplot(aes(x = year, y = variable, colour = variable)) +
geom_point(alpha = 0.5) +
geom_line(alpha = 0.5) +
scale_colour_okabeito() +
labs(x = NULL, y = NULL,
title = "PublicHouse data coverage by country") +
facet_wrap(~area_name) +
theme(legend.position = "none",
strip.background = element_rect(colour = "grey80", fill = "grey80"),
panel.border = element_rect(colour = "grey80", fill = NA),
panel.spacing.x = unit(15, units = "pt"))
And here’s the same thing but for cities. The coverage is not as good for cities as for countries, with fewer places and points in time covered. Sometimes this is because there simply isn’t any sub-national data reported, but often it’s because there is sub-national data but not at a city level. Note, they’re not included in the chart below but I have labelled Singapore and Hong Kong as ‘city states’ and included them in both the country and city charts in the rest of the document.
d %>%
filter(area_level == "city-region") %>%
mutate(variable = str_to_title(variable)) %>%
ggplot(aes(x = year, y = variable, colour = variable)) +
geom_point(alpha = 0.5) +
geom_line(alpha = 0.5) +
scale_colour_okabeito() +
labs(x = NULL, y = NULL,
title = "PublicHouse data coverage by city") +
facet_wrap(~area_name) +
theme(legend.position = "none",
strip.background = element_rect(colour = "grey80", fill = "grey80"),
panel.border = element_rect(colour = "grey80", fill = NA),
panel.spacing.x = unit(15, units = "pt"))
d_interpolated <- d %>%
group_by(area_name, variable) %>%
complete(variable, year = min(year):max(year)) %>%
mutate(value = na.approx(value, na.rm = FALSE)) %>%
fill(full_area_name:grouping, source_org:source_title, frequency:notes) Here’s a simple example of the kind of comparison that can be made with this data, showing the trend in the population and the dwelling stock in the UK and France.
d %>%
filter(area_name %in% c("France", "UK")) %>%
mutate(variable = str_to_title(variable)) %>%
ggplot(aes(x = year, y = value, group = area_name, colour = area_name)) +
geom_line() +
scale_colour_okabeito() +
scale_y_continuous(labels = scales::comma,
limits = c(0, NA)) +
labs(x = NULL, y = NULL, colour = NULL,
title = "Trend in population and number of dwellings in France and UK") +
facet_wrap(~variable, scales = "free")
That comparison broadly works for the UK and France because they have similar population sizes. But given the differences in the populations of all the countries we’re looking at, it doesn’t make much sense to just compare them by their absolute number of dwellings, so instead we calculate a measure of dwellings per person (or dwellings per 1,000 people in the charts below, for presentation purposes). This measures the stock of housing, scaled to the population of the area. The inverse of dwellings per person is persons per dwelling, which is similar to the average household size - but household size is affected by things like vacancy rates, second homes, sharing and the non-household population so is harder to compare across countries.
Is a higher rate of dwellings per person a good thing? Well, housing is a ‘normal’ good in the sense that people generally demand more of it as their incomes increase, so within a given place we would expect to see an increase over time.
The chart below shows the trend in dwellings per 1,000 people over time for countries, colour-coded by a country grouping of my own devising (which is inevitably slightly arbitrary - I’ve put France (and Paris) in the ‘Southern Europe’ category, for example). Note, the x axis starts at 1980 as there are relatively few countries with data before that, and the y axis starts at 200.
d %>%
pivot_wider(id_cols = c(area_name, year, area_level, grouping),
names_from = variable,
values_from = value) %>%
mutate(dpp = dwellings*1000/population) %>%
arrange(area_name, year) %>%
filter(!is.na(dpp)) %>%
filter(area_level != "city-region") %>%
filter(year > 1979) %>%
ggplot(aes(x = year, y = dpp, group = area_name, colour = grouping)) +
geom_line(size = 0.75) +
scale_colour_okabeito() +
scale_x_continuous(breaks = c(1980, 2000, 2020)) +
scale_y_continuous(limits = c(200, 650)) +
labs(x = NULL, y = NULL, title = "Dwellings per 1,000 people by country") +
facet_wrap(~area_name) +
theme(legend.position = "bottom",
legend.title = element_blank(),
strip.background = element_rect(colour = "grey80", fill = "grey80"),
panel.border = element_rect(colour = "grey80", fill = NA),
panel.spacing.x = unit(15, units = "pt"))
A couple of patterns stand out here. First, there is a lot of variation in both the level of dwellings per person and the rate of growth over time. Second, in many countries there is a flattening of the curve in roughly the second half of the period (since 2000), Spain and Ireland being quite extreme examples. Third, East Asian and Southern European countries have generally seen bigger increases over time, while Anglophone countries haven’t seen much increase at all since 2000.
Next is the trend over time for cities. The slope of the lines for cities are more often horizontal or downwards, reflecting the resurgence in population growth in many cities and the failure to increase urban housing stocks fast enough.
d %>%
pivot_wider(id_cols = c(area_name, year, area_level, grouping),
names_from = variable,
values_from = value) %>%
mutate(dpp = dwellings*1000/population) %>%
arrange(area_name, year) %>%
filter(!is.na(dpp)) %>%
filter(area_level != "country") %>%
filter(year > 1979) %>%
ggplot(aes(x = year, y = dpp, group = area_name, colour = grouping)) +
geom_line(size = .75) +
scale_colour_okabeito() +
scale_x_continuous(breaks = c(1980, 2000, 2020)) +
scale_y_continuous(limits = c(200, 650)) +
labs(x = NULL, y = NULL, title = "Dwellings per 1,000 people by city") +
facet_wrap(~area_name) +
theme(legend.position = "bottom",
legend.title = element_blank(),
strip.background = element_rect(colour = "grey80", fill = "grey80"),
panel.border = element_rect(colour = "grey80", fill = NA),
panel.spacing.x = unit(15, units = "pt"))
Here’s another way of looking at the same data - taking the most recent year’s figure for dwellings per 1,000 people (as long as there is one more recently than 2015) and calculating the percentage change from a decade earlier. The chart combines countries (in bold font) and cities, again coloured according to country grouping. The vertical dark line divides places where dwellings per person increased (to the right) from places where it fell (to the left). The horizontal dark line divides places with higher than average dwellings per person (above the line) from places with lower than the average (below the line).
scatter_data <- d_interpolated %>%
pivot_wider(id_cols = c(area_name, year, area_level, grouping),
names_from = variable,
values_from = value) %>%
mutate(dpp = dwellings*1000/population) %>%
arrange(area_name, year) %>%
filter(!is.na(dpp)) %>%
group_by(area_name) %>%
filter(max(year) > 2015) %>%
filter(year == max(year) | (year == max(year)-10)) %>%
mutate(dpp_change = dpp/dpp[1]-1) %>%
filter(dpp_change !=0) %>%
ungroup()
dpp_average <- scatter_data %>% summarise(mean(dpp)) %>% as.double()
scatter_data %>%
mutate(face = case_when(
area_level != "city-region" ~ "bold",
TRUE ~ "italic"
)) %>%
ggplot() +
geom_text(aes(x = dpp_change, y = dpp, colour = grouping,
label = area_name, fontface = face), size = 3) +
geom_hdr(aes(x = dpp_change, y = dpp, fill = grouping),
xlim = c(-0.08,0.20), ylim = c(250, 600),
probs = c(0.75)) +
geom_hline(yintercept = dpp_average) +
geom_vline(xintercept = 0) +
scale_x_continuous(labels = scales::percent_format(accuracy = 1)) +
scale_y_continuous(limits = c(NA, 650)) +
labs(x = "% change in 10 years", y = "Dwellings per person, latest year", title = "Dwellings per 1,000 people and change over 10 years (where available)", caption = "City names in italics, countries in bold") +
scale_colour_okabeito() +
scale_fill_okabeito() +
theme(panel.grid.major.x = element_line(colour = "grey80"),
panel.grid.major.y = element_line(colour = "grey90"),
legend.position = "bottom",
legend.title = element_blank())
There are some quite striking patterns here. First, all of the Anglophone places have fewer dwellings per person than the average, and around half of them have also seen declines over the last decade. Southern European places almost all have more dwellings per person than average and have seen increases (Madrid and Barcelona being the notable exceptions). East Asian places have very different amounts of housing per person, but all saw significant increases over the period (particularly Singapore).
But it’s important to remember that the number of dwellings per person can change due to shifts in population as well as net housing supply. The following two charts separate out the contribution of changes in population and changes in the housing stock.
d_interpolated %>%
filter(area_level != "city-region") %>%
pivot_wider(id_cols = c(area_name, year, area_level, grouping),
names_from = variable,
values_from = value) %>%
mutate(dpp = dwellings*1000/population) %>%
arrange(area_name, year) %>%
filter(!is.na(dpp)) %>%
group_by(area_name) %>%
filter(max(year) > 2015) %>%
filter(year == max(year) | (year == max(year)-10)) %>%
mutate(dpp_change = dpp - dpp[1],
dpp_constant_pop = dwellings*1000/population[1],
"Contribution of dwelling change" = dpp_constant_pop-dpp[1],
dpp_constant_dwellings = dwellings[1]*1000/population,
"Contribution of population change" = dpp_constant_dwellings-dpp[1]) %>%
filter(dpp_change != 0) %>%
select(-c(dwellings, population, dpp, dpp_constant_pop, dpp_constant_dwellings)) %>%
rename("Change in dwellings per 1,000 people" = dpp_change) %>%
pivot_longer(-c(area_name, year, area_level, grouping), names_to = "variable", values_to = "value") %>%
ungroup() %>%
ggplot(aes(x = reorder(area_name, value), y = value, fill = variable)) +
geom_col(data = . %>% filter(variable != "Change in dwellings per 1,000 people")) +
geom_point(data = . %>% filter(variable == "Change in dwellings per 1,000 people"),
shape = 21, size = 2) +
coord_flip() +
scale_fill_okabeito() +
labs(x = NULL, y = NULL, title = "Contribution of population and housing stock changes to change in dwellings per 1,000 people - \nmost recent 10 years") +
facet_grid(grouping ~ ., scales = "free_y", space = "free_y") +
theme(legend.position = "bottom",
legend.title = element_blank(),
legend.key.width = unit(8, "pt"),
strip.text.y = element_text(angle = 0),
strip.background = element_rect(fill = "grey80", colour = "grey80"),
panel.spacing.y = unit(10, "pt")) 
At the country level, we can see that only three countries (Poland, Japan and Portugal) saw a fall in their populations over the decade, contributing to an increase in dwellings per person. Unsurprisingly, no countries saw a decrease in their housing stock. Population growth was highest in the Anglophone countries, which contrbuted to their low or negative growth in dwellings per person. East Asian countries had low population growth but still added a significant number of new homes.
d_interpolated %>%
filter(area_level != "country") %>%
pivot_wider(id_cols = c(area_name, year, area_level, grouping),
names_from = variable,
values_from = value) %>%
mutate(dpp = dwellings*1000/population) %>%
arrange(area_name, year) %>%
filter(!is.na(dpp)) %>%
group_by(area_name) %>%
filter(max(year) > 2015) %>%
filter(year == max(year) | (year == max(year)-10)) %>%
mutate(dpp_change = dpp - dpp[1],
dpp_constant_pop = dwellings*1000/population[1],
"Contribution of dwelling change" = dpp_constant_pop-dpp[1],
dpp_constant_dwellings = dwellings[1]*1000/population,
"Contribution of population change" = dpp_constant_dwellings-dpp[1]) %>%
filter(dpp_change != 0) %>%
select(-c(dwellings, population, dpp, dpp_constant_pop, dpp_constant_dwellings)) %>%
rename("Change in dwellings per 1,000 people" = dpp_change) %>%
pivot_longer(-c(area_name, year, area_level, grouping), names_to = "variable", values_to = "value") %>%
ungroup() %>%
ggplot(aes(x = reorder(area_name, value), y = value, fill = variable)) +
geom_col(data = . %>% filter(variable != "Change in dwellings per 1,000 people")) +
geom_point(data = . %>% filter(variable == "Change in dwellings per 1,000 people"),
shape = 21, size = 2) +
coord_flip() +
scale_fill_okabeito() +
labs(x = NULL, y = NULL, title = "Contribution of population and housing stock changes to change in dwellings per 1,000 people - \nmost recent 10 years") +
facet_grid(grouping ~ ., scales = "free_y", space = "free_y") +
theme(legend.position = "bottom",
legend.title = element_blank(),
legend.key.width = unit(8, "pt"),
strip.text.y = element_text(angle = 0),
strip.background = element_rect(fill = "grey80", colour = "grey80")) 
At the city level, Seoul was the only place with a shrinking population (and it still built a lot of housing). The fastest growth in urban populations was generally in the Central/Northern European cities, and the lowest in Southern European cities (including Paris). Only in East Asia did cities consistently manage to grow their housing stock faster than their population.
The next chart plots all the country-level data on one chart and derives an overall average trend for each country grouping. These average trends are more representative for some groupings than for others: for example, the East Asian group is quite small and very diverse, with huge differences in trends between, say, Singapore and Japan. The other key point to note about these average trend lines is that they aren’t weighted for the population size of each country - so in the Anglophone grouping, for example, the data for the USA has no more effect on the trend than the data for Ireland. I think that’s okay in this case as the focus of this analysis is on comparing places.
With those caveats, this chart does reveal some striking general patterns. In particular, there has on average been very little increase in the amount of housing per person in the Anglophone countries since the start of the 1980s, while there has been a steady increase in Southern Europe, and increases over time in Central/Northern Europe and East Asia.
d_interpolated %>%
pivot_wider(id_cols = c(area_name, year, area_level, grouping),
names_from = variable,
values_from = value) %>%
mutate(dpp = dwellings*1000/population) %>%
arrange(area_name, year) %>%
filter(!is.na(dpp)) %>%
filter(area_level != "city-region") %>%
filter(year > 1979) %>%
ggplot(aes(x = year, y = dpp, colour = grouping)) +
geom_point(alpha = 0.1, size = 3) +
# geom_smooth(se = F, span = 0.3) +
geom_textsmooth(aes(x = year, y = dpp, colour = grouping, label = grouping),
se = F, span = 0.3, method = "loess", size = 4, linewidth = 1) +
scale_colour_okabeito() +
scale_x_continuous(breaks = c(1980, 2000, 2020)) +
coord_cartesian(ylim = c(200, 650)) +
labs(x = NULL, y = NULL, title = "Dwellings per 1,000 people by country grouping") +
theme(legend.position = "bottom",
legend.title = element_blank(),
panel.grid.major.y = element_line(colour = "grey80"))
Finally, this chart shows an overall average trend for every country except England, and compares that to the England trend. Broadly speaking, England started out with more housing per person than the international average, but has stagnated since 2000 and has ended up with less.
d_interpolated %>%
pivot_wider(id_cols = c(area_name, year, area_level, grouping),
names_from = variable,
values_from = value) %>%
mutate(dpp = dwellings*1000/population) %>%
arrange(area_name, year) %>%
filter(!is.na(dpp)) %>%
filter(area_level != "city-region") %>%
filter(year > 1959) %>%
ggplot() +
geom_smooth(data = . %>% filter(area_name != "England"), aes(x = year, y = dpp, colour = "International average trend"), span = 0.4) +
geom_line(data = . %>% filter(area_name == "England"),
aes(x = year, y = dpp, colour = "England")) +
scale_colour_okabeito() +
scale_x_continuous(breaks = c(1960, 1980, 2000, 2020)) +
labs(x = NULL, y = NULL, title = "Dwellings per 1,000 people - England and aggregate international trends") +
theme(legend.position = "bottom",
legend.title = element_blank())
The previous section disaggregated changes in dwellings per capita by comparing growth in the population and housing stock over a period of time. We can get a finer-grained understanding by tracking the proportional change in housing and population from a particular baseline. The chart below does this for countries from the year 2000 up to the latest available year (mostly 2020).
d_interpolated %>%
mutate(variable = str_to_title(variable)) %>%
group_by(variable, area_name) %>%
filter(year > 1999) %>%
mutate(index = (value/value[1])*100) %>%
filter(area_level != "city-region") %>%
filter(area_name %in% c("Australia", "England", "Finland", "France", "Germany",
"Ireland", "Japan", "Netherlands", "New Zealand",
"Poland", "Portugal", "Spain",
"Sweden", "Switzerland", "Taiwan", "USA")) %>%
ggplot(aes(x = year, y = index, colour = variable,
group = variable)) +
geom_line() +
scale_x_continuous(breaks = c(2000, 2010, 2020)) +
scale_y_continuous(breaks = c(100, 120, 140)) +
scale_colour_okabeito() +
labs(x = NULL, y = NULL,
title = "Index of change in dwelling stock and population, selected countries \n(2000 = 100)") +
facet_wrap(~area_name) +
theme(legend.title = element_blank(),
panel.grid.minor = element_blank())
The next chart does the same thing for cities. You’ll note that there are more cities than countries where growth in the housing stock failed to keep up with or only barely kept up with growth in the population.
d_interpolated %>%
mutate(variable = str_to_title(variable)) %>%
group_by(variable, area_name) %>%
filter(year > 1999) %>%
mutate(index = (value/value[1])*100) %>%
filter(area_level != "country") %>%
filter(!area_name %in% c("Seoul", "Amsterdam", "Lisbon", "Copenhagen", "Oslo")) %>%
ggplot(aes(x = year, y = index, colour = variable,
group = variable)) +
geom_line() +
scale_x_continuous(breaks = c(2000, 2010, 2020)) +
scale_y_continuous(breaks = c(100, 120, 140)) +
scale_colour_okabeito() +
labs(x = NULL, y = NULL,
title = "Index of change in dwelling stock and population, selected cities \n(2000 = 100)") +
facet_wrap(~area_name) +
theme(legend.title = element_blank(),
panel.grid.minor = element_blank())
We’ve compared housing growth with population growth in previous sections because population data is easily available and it serves as an obvious denominator for measuring housing growth against. But population by itself is not a good measure of the demand for housing, since factors such as the housing preferences, incomes and access to credit of the population change over time.
There is probably no perfect single measure of demand for housing, but
the next chart compares change in the housing stock of a country with
change in its Gross National Income, as an admittedly imperfect proxy
for the buying/renting power of the population. To get this data and
match it against our existing data we use the
[WDI](https://github.com/vincentarelbundock/WDI) and
[countrycode](https://github.com/vincentarelbundock/countrycode)
packages by Vincent Arel-Bundock.
library(countrycode) # for matching country names to the World Bank data
library(WDI) # for World Bank data
# get GNI data by country and year
gni <- WDI(indicator = "NY.GNP.MKTP.PP.KD", start = 2000, end = 2020)
# assign country codes
d_codes <- d_interpolated %>%
filter(area_level != "city-region") %>%
mutate(code = countrycode(area_name, origin = "country.name", destination = "iso2c")) %>%
mutate(code = case_when(area_name == "England" ~ "GB", TRUE ~ code)) %>% # We use UK GDP for England (the trend growth will be similar even if the level isn't)
left_join(gni, by = c("code" = "iso2c", "year" = "year"))
# filter to the data we're interested in and calculate index trends
d_codes <- d_codes %>%
filter(variable == "dwellings") %>%
filter(year > 1999) %>%
group_by(area_name) %>%
mutate(dwellings = (value/value[1])*100) %>%
mutate(GNI = (NY.GNP.MKTP.PP.KD/NY.GNP.MKTP.PP.KD[1])*100) %>%
filter(area_name %in% c("Austria", "England", "Finland", "France", "Germany",
"Ireland", "Japan", "Netherlands", "New Zealand",
"Poland", "Portugal", "Spain", "Belgium",
"Sweden", "Switzerland", "USA")) %>%
select(area_name, year, "Dwellings" = dwellings, GNI)
# create a 'long' version for plotting purposes
d_codes_long <- d_codes %>%
pivot_longer(cols = c(Dwellings, GNI), names_to = "variable", values_to = "value")
# plot it
ggplot() +
geom_line(data = d_codes_long, aes(x = year, y = value, colour = variable, group = variable)) +
geom_braid(data = d_codes, aes(x = year, ymin = Dwellings, ymax = GNI, fill = Dwellings < GNI),
alpha = 0.4) +
scale_x_continuous(breaks = c(2000, 2010, 2020)) +
scale_colour_okabeito() +
scale_fill_okabeito() +
labs(x = NULL, y = NULL, title = "Index of change in number of dwellings and Gross National Income, selected countries \n (2000 = 100)") +
guides(fill = "none") +
facet_wrap(~area_name, scales = "free") +
theme(legend.title = element_blank(),
panel.grid.minor = element_blank())
There are a number of other measures of housing stock or supply you could use, but this section focuses on the net annualised change in dwellings per 1,000 population. Whereas dwellings per capita is a measure of the stock of housing, this is a measure of the flow (albeit in net rather than gross terms), and it is not affected by either the previous size of the stock or by the change in population (at least not very much). So countries that are building a lot of homes relative to their population size can ‘look good’ by this measure even if that new supply isn’t actually enough to keep up with population growth.
The first chart in this section shows this measure for countries. The trend in most countries actually looks fairly flat by this measure, largely because of some extremely high values in a few countries (notably Ireland and South Korea)
d %>%
pivot_wider(id_cols = c(area_name, year, area_level, grouping),
names_from = variable,
values_from = value) %>%
group_by(area_name) %>%
mutate(net_change = (dwellings - lag(dwellings, 1))/(year - lag(year, 1)),
net_change_per_1k = net_change/((population + lag(population, 1))/2)*1000) %>%
filter(!is.na(net_change_per_1k)) %>%
filter(area_level != "city-region") %>%
filter(year > 1979) %>%
ggplot(aes(x = year, y = net_change_per_1k, group = area_name, colour = grouping)) +
geom_line(size = 0.75) +
scale_colour_okabeito() +
scale_x_continuous(breaks = c(1980, 2000, 2020)) +
scale_y_continuous(limits = c(0,NA)) +
labs(x = NULL, y = NULL, title = "Annualised net additional dwellings per 1,000 people by country") +
facet_wrap(~area_name) +
theme(legend.position = "bottom",
legend.title = element_blank(),
strip.background = element_rect(colour = "grey80", fill = "grey80"),
panel.border = element_rect(colour = "grey80", fill = NA),
panel.spacing.x = unit(15, units = "pt"))
Here’s the same chart for cities. Again the extreme outliers (hello Seoul!) tend to flatten the trend for everywhere else. The breaks in the trends for Berlin and Brussels are due to the net dwelling change briefly going negative - I suspect due to data collection reasons rather than an actual shrinkage of the housing stock.
d %>%
pivot_wider(id_cols = c(area_name, year, area_level, grouping),
names_from = variable,
values_from = value) %>%
group_by(area_name) %>%
mutate(net_change = (dwellings - lag(dwellings, 1))/(year - lag(year, 1)),
net_change_per_1k = net_change/((population + lag(population, 1))/2)*1000) %>%
filter(!is.na(net_change_per_1k)) %>%
filter(area_level != "country") %>%
# filter(area_name %in% c("London", "Paris", "New York", "Tokyo", "Berlin", "Singapore")) %>%
filter(year > 1979) %>%
ggplot(aes(x = year, y = net_change_per_1k, group = area_name, colour = grouping)) +
geom_line(size = 0.75) +
# geom_point(size = 2.5) +
scale_colour_okabeito() +
scale_y_continuous(limits = c(0, NA)) +
scale_x_continuous(breaks = c(1980, 2000, 2020)) +
labs(x = NULL, y = NULL, title = "Annualised net additional dwellings per 1,000 people by city") +
facet_wrap(~area_name) +
theme(legend.position = "bottom",
legend.title = element_blank(),
strip.background = element_rect(colour = "grey80", fill = "grey80"),
panel.border = element_rect(colour = "grey80", fill = NA),
panel.spacing.x = unit(15, units = "pt"))
Can we extract any aggregate trends? The chart below shows smoothed averages for country groupings, but as the quite widely dispersed dots around each trend should suggest, these are fairly indicative trends at best. Broadly speaking it looks like new supply peaked in East Asian around 2000 (possibly as a result of the late 1990s financial crisis in that region), while there was a peak in supply around the time of the global financial crisis in Anglophone countries and (a much stronger effect) in Southern European countries. The European and (in particular) the Anglophone countries have increased supply in recent years while in East Asia supply has fallen.
d %>%
pivot_wider(id_cols = c(area_name, year, area_level, grouping),
names_from = variable,
values_from = value) %>%
group_by(area_name) %>%
mutate(net_change = (dwellings - lag(dwellings, 1))/(year - lag(year, 1)),
net_change_per_1k = net_change/((population + lag(population, 1))/2)*1000) %>%
filter(!is.na(net_change_per_1k)) %>%
filter(area_level != "city-region") %>%
filter(year > 1979) %>%
ggplot(aes(x = year, y = net_change_per_1k, colour = grouping)) +
geom_point(alpha = 0.1, size = 3) +
geom_smooth(se = FALSE, span = 0.4) +
scale_colour_okabeito() +
coord_cartesian(ylim = c(0, 10)) + # use this rather than scale_x_continuous to 'zoom' into a section of the chart without losing any data
scale_x_continuous(breaks = c(1980, 2000, 2020)) +
labs(x = NULL, y = NULL, title = "Annualised net additional dwellings per 1,000 people by country") +
theme(legend.position = "bottom",
legend.title = element_blank(),
panel.grid.major.y = element_line(colour = "grey80"))
And finally, here’s the trend for England compared to the aggregate trend for all other countries. At the international level, new supply seemed to peak in the late 60s / early 70s, then was relatively steady from the early 80s through to the mid 2000s, but took a serious hit as a result of the global financial crisis. England has had lower net housing supply per person than the international trend throughout this period, and the long-term trend has been generally downwards.
d_interpolated %>%
pivot_wider(id_cols = c(area_name, year, area_level, grouping),
names_from = variable,
values_from = value) %>%
group_by(area_name) %>%
mutate(net_change = (dwellings - lag(dwellings, 1))/(year - lag(year, 1)),
net_change_per_1k = net_change/((population + lag(population, 1))/2)*1000) %>%
filter(!is.na(net_change_per_1k)) %>%
filter(area_level != "city-region") %>%
filter(year > 1959) %>%
ggplot() +
geom_smooth(data = . %>% filter(area_name != "England"),
aes(x = year, y = net_change_per_1k,
colour = "International average trend"), span = 0.4) +
geom_line(data = . %>% filter(area_name == "England"),
aes(x = year, y = net_change_per_1k, colour = "England")) +
scale_colour_okabeito() +
scale_x_continuous(breaks = c(1960, 1980, 2000, 2020)) +
labs(x = NULL, y = NULL, title = "Annualised net additional dwellings per 1,000 people - \nEngland and aggregate international trends") +
theme(legend.position = "bottom",
legend.title = element_blank(),
panel.grid.major.y = element_line(colour = "grey80"))