The UK's Political Housing Crisis

Can we build it? No we can't...

This Substack post continues a series of deep-dive data analyses of British politics. This time we’re looking at building new houses and whether that offers a (politically) viable way of solving Britain’s housing crisis. In coming weeks, I’ll look at how people feel about inheritance taxation and other wealth taxes, whether Europeans are different to Brits, and more questions on fairness and social mobility.

But there’s more! On March 30th, my book Why Politics Fails, comes out in the UK (North Americans, you have to wait until May but you get an exciting different cover). I’ll be covering some of the themes of the book over the next few months. So please sign up to this Substack and encourage friends and colleagues to do so. And authors love pre-orders of their books, so… you know what to do!

---

Why are British houses so expensive? In part, because there aren’t enough of them, at least not in the right places. But, given this deficit, why then have we slipped from building over 300,000 houses a year from the 1950s the 1970s to under 200,000 in the 1990s to fewer than 150,000 between 2010 and 2014? Even today we’re struggling to get over 200,000 new dwellings a year. The slowdown in housebuilding during the 1980s and 1990s might have been understandable - annual population growth collapsed over the 1970s. But population growth in the first decade of the twenty-first century hit levels higher than the baby boom. More people; not enough houses.

Generations of commentators, politicians, and policymakers - actually, generations of Tweeters at this point - have cast blame on an endemic NIMBYism in the UK. Planning laws empower anyone who would do anything for love but they won’t… tolerate seeing another house from the edge of their garden if they squint.

Of course, the government could muscle in, in its imitable way, and ignore local concerns, building reams of public housing to accommodate all these new citizens… Ah, forget it. In the 1950s and 1960s, local authorities averaged around 175,000 new dwellings a year. By the 1990s, that figure was around 1,750. An impressive drop of two orders of magnitude.

Now of course, private builders and housing associations did pick up some of the slack - but even in the case of the former, private constructors are today building only around three quarters of the 200,000 houses they built annually in the 1960s. And housing associations at their peak in the early 1990s were building under 40,000 new houses a year. A mid-sized Premier League stadium number of new houses.

Britain is not the only country with a housing crisis. But we do lag our peers. For the last decade, France, which has a very similar population, has constructed between 400,000 and 500,000 houses a year. Over the long-run Britain has been much slower in constructing new dwellings than most other OECD countries. Between 1990 and 2015 the annual growth rate in housing stock in the UK was under 0.8%. This compares to over 1.25% in Australia, Belgium, New Zealand, Canada, Spain, Switzerland, and Japan. Some of these countries also had high rates of immigration (e.g. Canada and Switzerland) - unlike the UK they built houses.

The collapse of house-building in the UK is stark. But we know precious little about how people feel about it. After all, it’s not obvious that people should care - or to be more precise - that the swing voter, on who elections depend, should care. Around two thirds of Brits own their property - and that number is likely substantially higher among people who actually vote. They already have theirs, Jack. But perhaps I’m wrong - SPOILER, I’m not wrong - and people are far-sighted and other-centred and want to build new houses for those who can’t get on the property ladder?

In the absence of survey data, it’s hard to know. We have to rely on anecdote or personal sentiment and grievances - which when it comes to housing in Britain are pretty much our national go-to topics of conversation.

To get at this question more rigorously, my WEALTHPOL team and I ran a series of surveys in 2021 and 2022. If you’re already a subscriber to this Substack (you're not? go and click the button), you will have seen some results from these surveys in my earlier posts. Our main interest in these surveys was in how people feel about wealth, particularly from the property market, so asking about house-building was a core question for us. What we find very clearly is there is no majority supporting building new houses locally in the UK. We can’t build because we won’t build.

In two surveys we conducted with YouGov, one in June 2021, one in October 2022, we asked people a fairly straightforward question “Thinking about new housing in your local area. How much would you support or oppose more homes being built in your local area?” People answered on a five point scale: strongly oppose, somewhat oppose, neither support nor oppose, somewhat support, strongly support. Combining these surveys we have 6624 responses. That’s a pretty big survey size. And that means we can do something rather interesting…

The Magical MRP

Three initials have struck fear into politicians and innumerate political reporters over the past few years: M.R.P. Let me spell that out: Multilevel Regression with Post-stratification. You’re probably not feeling any better, are you?

I’m not going to lie and say MRP is simple. It’s not. In fact to produce the results I am going to show you, I had to run a model on my pretty high-powered laptop overnight. And it still hadn’t finished when I had coffee the next morning. It’s a complex algorithm that requires producing tons of simulated data to get final results.¹

But the idea of MRP, if not the process itself, is simple-ish. We often want to know about public opinion in small areas. Such as parliamentary constituencies. But few of us - Michael Ashcroft apart - are wealthy enough to run statistically viable samples in individual constituencies. To even have a margin of error of four percent you would need around 500 respondents. Multiply times 650 constituencies and… that’s a big (and expensive) poll.

But what about all those headlines in the Times before every election, showing predicted results for each constituency? How do those happen, then? Here’s where the MRP magic comes in. If we have a fairly large national survey with lots of demographic information and at least a few observations in each of the units we care about (constituencies, here) AND we have demographic data for each unit, then we can combine the two.

We use the info we have from the whole survey about how demographics such as age, gender, education, etc, connect to the thing we care about - perhaps vote choice - along with info about average support at the local level from the few cases we have in each constituency, and we combine this with what we know about local demographics.

For example, we use the survey as a whole to see the relationship between age and vote choice; then for each constituency we look at the known age structure of that constituency. Then do the same for education. And homeownership. And gender. Etc. We can combine all this to simulate the vote choice for the constituency.²

In other words, you take relationships from the data as a whole and combine with local demographics from the census (or similar) and you build up a picture of vote choice for each constituency. Obviously this is not the same (nor probably as accurate) as a poll in each constituency. But it’s a lot cheaper and has been broadly effective and predicting political outcomes in the UK. Remember 2017, where the YouGov MRP poll was the only one to predict a hung parliament.

And that’s what I am going to do here, but for building houses rather than vote choice. Drumroll please… This is the first set of MRP estimates by UK constituency for support for building houses. Let’s say you are an MP - as I am sure a vast majority of my subscribers are - and you want to know how many people in your constituency support building houses in the local area. I am comfortable in saying that this Substack post has the best (also, only) estimates in existence for that.

Now our five point scale is great and all but it’s a little hard to interpret. So what I am going to do is to compact it into a single binary indicator - do you answer either “somewhat support” or “strongly support” as opposed to the other options? If you do, I will call you a supporter. And now for each constituency we can run an MRP analysis to estimate the proportion of people who support building new houses locally.

OK, OK, I get it

Enough build-up. What’s the take-home? The average level of support across constituencies, according to the MRP is 36.8%. For people who think the country desperately needs to build new houses, that’s not… great. If it makes you feel any better, the average level of support in the combined polls (i.e. of the 6624 respondents) is 38%. The difference comes from compacting things down into constituency averages. But look, either way, this is bad news for building houses.

Now we could umm and er a little here - I have thrown in people who neither support nor oppose with the opposers. I do this because ultimately I think relying on allegedly indifferent people to get houses built is foolhardy. But you probably won’t feel hugely relieved to discover that 39% of people oppose any local house-building. We may have a housing supply crisis in the UK but more people oppose building houses than support it.

Anyway, whatever we think of the precise level of support for house-building, the really interesting stuff is how it varies across the country. If you want to see how your constituency fares an interactive table is available here. Or you can wait for the pretty graphs.

Let’s begin by looking at the top and bottom ten constituencies in terms of support for building new houses. The top group read like a Rough Guide to Hipster Haunts of London. Otherwise known as the home constituencies of the Corbyn-era Shadow Cabinet. We have over sixty percent estimated support in both Hackney constituencies, in JC’s own Islington North and in David Lammy’s Tottenham constituency. Keir Starmer is bringing up the rear with 58.3% support.

How about the bottom ten? Here we have a series of the most Conservative parts of the country - the Lonely Planet Special ERG edition. They are more geographically varied - Dorset, Lancashire, Essex, Derbyshire, Lincolnshire. But they are largely rural, right-leaning, older locations.

Now in a way none of this is surprising. The MRP model uses information on local indicators including past voting as well as demographics. And the MRP also uses age, education, subjective social class, gender, and homeownership from the YouGov surveys. Because many of the the latter are strong predictors of support for building new houses, places with young, non-owning, high-education, high SES people - like Hipster London and Bristol - will be estimated to have high support for building new houses. And places with older, lower-education, lower-SES homeowners, will be estimated to have low support. That’s how the model works. We don’t know what Wyre and Preston North’s true support for house-building is. But this is our best estimate given what we do know about (a) what kind of people support building houses, and (b) what Wyre and Preston North looks like demographically.

You can see these relationships in the next two figures. These show the estimates and their confidence intervals (the 5th to 95th percentile from the simulations), arrayed by first Conservative support in the General Election of 2019 and second by the 2011 Census estimate of homeownership rates. Estimates are coloured green if the entire confidence interval is above the national mean (i.e. we are really sure these places are big supporters of building houses) and red if the entire confidence interval is below the national mean (these places definitely dislike building houses). There are just under one hundred constituencies in each group. Everyone else is the messy middle.

What we see is what we’d now expect. Conservative places are ‘red’ - these constituencies have well below national average support for building houses: often only around 20% of people supporting building. And non-Conservative places - either Labour or SNP seats mostly - are strong supporters - often well above fifty percent support.

We also see that the constituency’s level of home-ownership is very closely related to estimated support for building new houses. Where lots of people own houses, they don’t want any more, thank you very much.

Mapping Support for Housing

I like a scatterplot with confidence intervals as much as the next academic but perhaps you’d prefer something a little more tangible. Lucky you. Here are some maps.

The first map shows each constituency in Westminster (save Northern Ireland - sorry guys!), and the estimate of support for building new houses. Sadly, Substack does not yet offer interactive mapping facilities, so if you want to see how your own constituency fares you’ll need to go this dandy site I created, which is interactive.

The pretty obvious thing from this figure is that support for building new houses is concentrated in four areas. 1. London. 2. Scotland. 3. Liberal cities (Bristol, Oxford, Cambridge, Manchester), and 4. the Red Wall!

What of course is in common is all of these places are, or were, major centres of support for the Labour Party. This is very apparent in the next map, which I rather like.³ This is a cartogram where the size of the constituency depends on their estimated support for building new houses and the colour is who won it in 2019.

It’s easy to see from this just how different the constituencies of each party are with respect to house-building. Is it really surprising that the Conservative Party has largely failed to create a sea-change in building houses, given many - perhaps most - of their constituencies hate it? Those places that are big and blue are either in London or in the Red Wall.

Conversely, the implication is that if and when Labour take power, their base will be much more supportive of house-building. But of course nothing is constant. Marginal seats that Labour needs for a majority are less likely to be supportive of house-building. So electoral politics may override any other pro-construction factors.

Who Likes Building Houses?

Let’s leave the MRP behind and turn back to the YouGov polls. Here I’ll focus on the most recent one (October 2022) because we also did something interesting here - we asked people to provide a one-line justification for their answer. We’ll explore that text data in a bit but let’s start with some simple graphs for the 3600 or so respondents in that survey. At the individual level, what correlates with support for building new houses? Just to give you a sense of total support - for the 2022 survey we have 39% support for building new houses locally.

Let’s begin with politics. We have information both on how people voted in 2019 and in their vote intention in October 2022. (I’ll let you in on a secret, the former was good for the Conservatives, the latter, not so much). Starting with voting in 2019 we can see a pretty clear partisan divide. Around thirty percent of Conservative voters supported building new houses locally, whereas around fifty percent of Labour voters did. What about other parties? Brexit Party voters looked just like Conservatives. Green voters and SNP voters just like Labour voters and Lib Dems - God bless them - were right in the middle. Neither ‘winning here’, nor ‘building here’ apparently.

What about current vote intention? Pretty much the same thing. A near twenty gap between Conservatives and Labour. Other parties behaving similarly except the Reform Party, whose supporters hate building new houses. Perhaps they think they are for illegal immigrants…

One interesting thing to think about is non-voters (or current Don’t Knows) who are down as NA. This group looks a lot like Conservative voters. And this helps to explain why Conservatives don’t feel in any great rush to build houses - there isn’t some pool of potential voters for them out there who want new houses. If anything, the incentive for disenchanted non-voters is… build fewer houses.

Let’s shift gear and look at other factors predicting support. You will not, I am sure, be surprised to find that non-homeowners have higher levels of support for building new houses locally than homeowners. Frankly, I’m a bit surprised it’s only a ten point difference. Interestingly, if you split it out - homeowners who own outright and mortgage-holders look pretty much the same, as do private renters and people in social housing. People living in housing associations are somewhere in the middle.

What about house prices? Amusingly, the group of homeowners with most support for building houses are those who think their house is worth more than £750,000. Other than them there’s a mild decline in support as house prices rise. This clearly speaks to something about urban homeowners and their partisan preferences (clue: not Conservative).

As we’ll shortly see, when you ask people to explain their views on housing, the Green Belt regularly comes up. I am able to match people to their local authority and then in turn to the proportion of land in the local authority that is either Green Belt or an Area of Outstanding National Beauty. And, well there’s something to this. It definitely looks like people living in these ‘prettier’ areas don’t want new housing.

Finally, let’s quickly look at the other trifecta of demographics that predict voting: income, education and age. What we see is more support for home-building among high income households (over £100k), among postgraduates (and maybe graduates) and among younger people. In other words - yuppies. A group who seem to have drifted a long long way from the Conservative Party and their housing policies.

Give Me a Reason

One great thing about online surveys is you can ask people to explain their answers. Obviously, this is time-consuming and you’d want to limit its scope otherwise it might feel like you were punishing survey respondents. But what I’ve learned from doing this a number of times is that people take it really seriously and write quite considered and thoughtful responses. Which from a civics perspective I find pretty encouraging.

Less encouraging for supporters of house-building is what people give as their justifications in this survey. This becomes apparent the moment we look at some word clouds using the 3600 odd responses we have. For single words we get some pretty obvious things (I excluded the word ‘house’, which would otherwise dominate everything!). Words such as ‘need’ ‘build’ ‘new’ and ‘area’ largely describe the wording used in the prompt. But we have some others that are more meaningful such as ‘infrastructure’, ‘already’, ‘enough’, ‘green’ and ‘school’, as well as ‘affordable’

When we switch to pairs of words we get more meaningful things - ‘green space’ is the biggie along with ‘green belt’, ‘new build’, ‘people need’, ‘need affordable’ and ‘school doctor’. You can see overall it’s splitting in two directions - a large number of people raising concerns about what building new houses might mean for existing green space or infrastructure and another group of responses mentioning the need for affordable housing.

With ‘trigrams’ (three word groups for noobs) we see a very similar pattern including lots of things mentioning the ‘green belt’ and the holy trinity of public goods infrastructure ‘school, doctor, dentist’.

Word clouds are fun and all but they don’t show us the different arguments people make. And that’s where we can see Britain’s political divide over housing more clearly. The way we look at this is to do what’s called ‘keyness’ analysis - this identifies words that are statistically more associated with one group of respondents than another (it’s a simple chi-squared test). I will do that for two groupings - homeowners versus non-homeowners and people who voted Conservative in 2019 versus those who voted Labour.

Let’s start with the former pairing. The two graphs below show how homeowners and non-homeowners speak differently using (a) single words and (b) pairs of words. The little dots on the vertical line in the middle correspond to the words to the left (or right) and measure how ‘distinct’ that word is across the groups. So for non-homeowners the most distinct word is ‘homeless’ and others include ‘affordable’, ‘desperately’, ‘crisis’, ‘ever’, ‘need’, ‘opportunity’… it’s kind of heartbreaking.

For homeowners the most distinct words are - by a mile - ‘infrastructure’, followed by ‘road’, ‘school’, ‘doctor’ and things like ‘appointment’ and ‘village’. A very different set of priorities.

With pairs of words, for non-homeowners we get things like ‘afford buy’, ‘desperately need’ and ‘need social’, and for homeowners ‘local infrastructure’, ‘infrastructure school’ and, amusingly, ‘young people’ (I cannot confirm whether fists were shaken when the last pair of words was written).

What about differences by party politics. They look, unsurprisingly, rather like the non-homeowner, homeowner split. Labour voters use ‘affordable’, ‘social’, ‘homeless’, ‘landlord’ and ‘desperately need’ or ‘afford buy’. Conservative voters use ‘doctor’, ‘cope’, ‘village’, ‘school’, ‘school doctor’, ‘local infrastructure’ and our favourite ‘young people’ (grrr).

The final way I want to look at this to use some magic. OK machine learning. Here I run what’s called a Structural Topic Model. The idea is that I ask the computer to look for groups of words that seem to cluster together and to sort them into ten distinct topics. We can then sort each answer people gave by how much it fits into each ‘topic’. Here are the ten topics that my magical machine learning mechanism provided.

Topics 10 and 4 are the most common, appearing in 15 to 20% of people’s responses. We can then ask the computer to pick out some especially choice answers that fit into the topics. Topic 10 as you can see has a bunch of responses that really emphasise infrastructure. I want to emphasise these are real and meaningful concerns - that building more houses means local infrastructure can’t logistically or physically (in the case of flooding and sewage) cope.

Topic 4 picks up on another set of real and meaningful concerns - but quite distinct ones: no-one can afford houses, especially young people; homelessness is being created. I am grateful that respondents to the survey made such thoughtful comments. I think it’s easy for people to mock NIMBYs, or indeed the proverbial avocado-eating millennials, and therefore ignore their concerns or claim they aren’t real. But these are completely coherent answers. They just stress different problems in British life - fears of overcrowding and system under strain; fears of never being able to get on the property ladder at all.

And it won’t surprise you to learn that people’s politics do shape their answers. The bottom figure shows how Conservative versus Labour 2019 voters focus on different problems. The frequency of ‘topic 10’ style answers about pressure on infrastructure is about 7.5 percent points higher for Conservatives than Labour voters. By contrast, the frequency of Topic 4 - affordability and homelessness - is about four percent points lower.

Britons understand that Britain has a housing crisis. But they define it differently. For some people, it’s about scarce resources and nature already under massive pressure. For other people, it’s about not enough houses being built and the unaffordability of property. To poorly paraphrase Tolstoy, we are all unhappy voters, just unhappy in our own ways.

Is this Anything New?

Britain was having a housing crisis when I left for graduate school in the States in 2000. When I came back in 2013 it was worse. I can’t say things have improved measurably during the time I’ve been back either. Is this it? Are we destined to be in a housing perm-crisis?

Think back to the start of this, admittedly lengthy, post. We did use to build houses. It’s funny, in a way, that Boomers on Facebook absolutely love the ‘remember when’ style posts about the glories of 1960s and 1970s biscuits or TV shows. And yet, there was something else we did back then too: construct dwellings for people to live in. “‘Memba affordable houses, eh? What were that all about?”

In my last post on generational differences I looked at each generations’s average support for the Conservative Party, relative to that election’s average, going back to the 1970s. Like John Burn Murdoch at the FT, I found that Millenials appear to be diverging away from the Conservatives, rather than becoming more Conservative as they aged.

Below I show what happens if you split each generation by whether they are homeowners or not. So we are looking at each generation’s average Conservative support among non-homeowners and homeowners respectively.

Doing that we see something quite surprising. In the past non-owners (mostly renters, public or private) were pretty systematically less supportive of the Conservatives by ten to twenty points. Differences among generations were also pretty small (see 2001 for example). But then in the last two elections, there has been a huge divergence. Boomer and Silent generation renters moved en masse to the Conservative Party. Millenials and Gen Xers kept away.

But with homeowners, the opposite happened. Owners used to be systematically more Conservative, albeit with a slight age gradient. But… Gen X and Millennial homeowners have actually shifted away from the Conservatives, especially Millennials. Even getting on the housing ladder hasn't made them (vote) more conservative.

So it feels to me like the coalition of homeowners has now broken. If millennial and Gen X homeowners look more like younger renters than older citizens - if age rather than ownership divides us - then ‘generational turnover’ might break the logjam. For those in the back, that was a euphemism for older homeowners dying off. If younger homeowners vote for a Labour Party that, given all the evidence above, seems to be more pro house-building, then there might finally be a governing political coalition large enough to make real progress in house-building.

But the Labour Party won’t find it easy-going. People opposing building new houses can be called NIMBYs all you want - but the reasons they give for opposing new developments are consistent, coherent, and strongly held. Perhaps a PM Keir Starmer won’t want to burn up his political capital fighting against guerrilla armies of anti-development rural homeowners. He could assuage their fears perhaps by building the mythical infrastructure that everyone demands and yet never appears. But that then becomes a national mission to build houses and doctors’ surgeries and schools and roads and…

What about the current Prime Minister? Rishi Sunak is of course a Millennial homeowner, so on the face of it, it’s surprising he’s a Conservative… I kid, but the problem of how to get ‘geriatric millennials’ to vote Conservative has been much in the news. Centre-right think tanks from Onward to CPS, and commentators from Robert Colville to Sam Bowman have argued the Tories need to ‘build build build’. But politically, given the views of their base, that would be very… bold. You can’t ‘Build Back Better’ if your voters like the ‘back’ much more than the ‘build’.

And so, for now I think we’re stuck. No politician has been willing to face down the critics of mass housing construction. It’s no surprise, given what we saw above about the general national reluctance to build houses and the strong arguments people give in their opposition. Britain, no longer a nation of shopkeepers, has become a nation of gatekeepers.

1. I owe Chris Hanretty big time for his MRP code and his help with some debugging issues. Most of quantitative British political science would fall apart without Chris.

2. As ever, things are a little more complicated. We have combinations of age, gender, homeownership, education etc for each constituency. So we find out that 1.5% of the constituency’s population is men over 65 with a degree who own their house. We use our statistical estimates from the poll to estimate their support for the thing we care about - let’s say building houses. We then do the same for the next group - 30-40 year-old women with degrees who don’t own their house and so forth. We weight each category by their proportion of population in the constituency and add it all up to get our final estimate, which may be adjusted by the average level of local support for house-building from our survey. We then run the model lots of times (several hundred) to get a range of estimates, which provides our measure of uncertainty. That’s why it took all night.

3. Many thanks to Evan O’Dell for his wonderful parlitools R package which makes creating these cartograms simple.