Henderson’s Guide to Pandemic History

What will happen when the Pandemic ends?

Will pre-Pandemic patterns, like people moving to Vancouver, go back to normal? Or will small towns, far-flung suburbs, and rural areas see a boost at the expense of cities, reflecting perhaps a new aversion to density and/or embrace of the rise in telecommuting acceptability? (we’ve seen such speculation in certain corners of City Hall).

Or indeed, might we see the opposite? Will people flock to cities like Vancouver as we return to mobility (including newly amped up immigration along with outreach to Hong Kong) and enjoyment of all the urban pleasures we’ve given up during the pandemic?

It’s all speculation at this point. But it’s got me curious about the past. What happened after the 1918-1919 influenza pandemic? And here I struggle with two things: 1) there was a LOT going on during and prior to the 1918-1919 flu pandemic, making it hard to isolate any response, and 2) the census data skips right around the two key years, with timing gaps too large for zooming in.

I can’t fully fix the overlapping events (WWI, and prior to that a big speculative economic crash), but I can kind of get around some of the data limitations of the Census by playing with some historical data sources I’ve been meaning to give more attention, in particular, the brilliant collection of BC City Directories archived by the VPL, including especially Henderson’s City and Greater Vancouver Directories and Wrigley’s BC Directories.

First, a couple of quick notes about the 1918-1919 Pandemic, brought to you by Margaret Andrews (1977) enlightening research in “Epidemic and Public Health: Influenza in Vancouver, 1918-1919” open access in BC Studies vol. 34. According to Andrews, the Pandemic hit Vancouver especially hard relative to other cities in Canada and the USA. It was also very different from today’s Pandemic in targeting mostly young and middle-aged adults.

At the same time, it was similar to today’s Pandemic in arriving across multiple waves, though the first (in 1918) took the greatest toll.

So what can we add by looking at City Guides? Well, we can compare them to Census results to get a more fine-grained sense of how the City responded to and potentially bounced back from the Pandemic of 1918-1919. The guides include, especially, the Henderson’s City of Vancouver Directories and related Wrigley’s Guides (which swallowed up Henderson’s in 1924), all providing listings of businesses (and households) across Greater Vancouver. I estimate the number of listings for each year, folding businesses and households together. While this isn’t a perfect match for population, or even households, it provides a relatively consistent method for a fine-grained look at how Greater Vancouver businesses and households together experienced the concentrated events piling up between census years (more details below!)

What’s our fine-grained examination of directory listings in combination with census data tell us? It appears we really do miss a lot with census data alone, especially between 1911 and 1921, where we saw a gigantic speculative bubble crash in 1913, followed by the Dominion’s entrance into WWI in 1914, and the Influenza Pandemic itself in 1918.

Where Census data from 1901, 1911, 1921, and 1931 make Vancouver’s growth look relatively steady and nearly linear, directory data demonstrate the enormous upset and losses of 1913-1915 in Vancouver, followed by a bottoming out and start at recovery during WWI (when many otherwise unemployed men went to fight in the war), finally interrupted by effective stasis during the Pandemic of 1918-1919. Then boom! Vancouver was off to the races again, climbing rapidly in listings from 1919-1923 and again (jumping different guides & methods) from 1924 seemingly only slowing a bit in 1926. From there, the trajectory of growth seemingly carried right through the beginnings of the Great Depression to 1931, when the next census was carried out.

Is past prelude? If so, Vancouver looks set to recover quite spectacularly from the Pandemic once it ends, as people flock back to the joys of the city. Maybe we’ll get our own Roaring 2020s!

But of course, for now we’re still here in the middle of the damn thing. So I’m still singing “Come On Vaccine.”

You know the tune…

APPENDIX

A couple quick methods notes for my beloved nerds. Historical census data was taken from Norbert MacDonald’s “Population Growth and Change in Seattle and Vancouver, 1880-1960” from Pacific Historical Review 39(3): 297-321 (unfortunately paywalled). MacDonald combines South Vancouver and Point Grey into the City of Vancouver boundaries for 1921, but I believe he considers the populations of these municipalities effectively too low to matter in earlier years. Henderson’s Directories were released on a yearly basis with a pretty standard, two column format, from 1905-1923, and seemingly covered all of Greater Vancouver during this time, with listings showing up in North Vancouver, New Westminster, and Burnaby, for instance (though North Vancouver was sometimes also reported separately). Ads were placed somewhat randomly within the text, rather than as full pages. In 1924, the Henderson directories were absorbed by Wrigley’s directories, using a new three column format (and smaller type) with interspersed full page ads. I attempted to estimate the listings for each year of these two different sources by gathering page numbers for alphabetized listings (of resident households and businesses) and multiplying by an estimate of the number of listings per page, excluding full page ads where possible. I estimated ~95 listings per page for Henderson’s and ~184 listings per page for Wrigley’s, based upon a quick count on what seemed representative pages (the second A listings), but this estimate could certainly use further checking.

Return to the Airport!

A couple of months ago I took the blog for a visit to the airport to check out historical passenger data and see what’s happened since COVID. Today I want to return, both to provide an update and to pull YVR Passenger data (enplaned & deplaned pdf) together with BC CDC Flight Exposure data (full pdf), providing a check on air travel’s contributions to spreading COVID.

First the update!

We can see that through August (last month of data available as of today), flights are still gradually rising toward a return to 2019 levels, but they’ve still got a loooong way to go. Mostly the rise has been led by domestic air travel within Canada. We can zoom in, looking at monthly passenger totals for 2020 as a percentage of passenger totals for 2019.

Sure enough, by the end of August we’re back up to over a quarter of the Domestic air travel from the same month in 2019. International flights still remain far below 2019 levels, with the biggest drop in Transborder trips between Vancouver and cities in the USA. Miscellaneous International trips that mostly cover Latin America and the Caribbean have seen a recent decline from slightly higher numbers in June and July. Passengers to and from Asia Pacific destinations never dropped as much as other international passengers and have bounced back a little, and passengers to Europe appeared to rise through July and August.

So how are we doing containing COVID exposures on these flights? The BC CDC lists exposures by flight number, origin and destination, and affected rows, and as of today includes exposures through September 30, though given lags in reporting it’s possible the September listings aren’t yet complete (none have yet been listed for October). Here I separate inbound and outbound flight exposures for Vancouver by Origin/Destination Stream roughly matching YVR categories (I remain less certain exactly how flights to and from Mexico fit in, and have included them here as Misc. Intl).

Overall, it’s clear that COVID exposures on flights have declined and then risen again with flights overall between March and August, with the pattern likely continuing into September (again, we don’t yet know if September data is complete and we don’t have YVR passenger data for September yet). Domestic exposures dominate flight exposures overall, especially the rise in August and September.

Finally, we can combine the two sources of data to provide a rough estimate of the inbound and outbound specific risks associated with exposures. How many exposures do we see per 100,000 passengers for different streams of travel? Here I’ve given outbound exposures negative values, and inbound exposures positive values, which tells us something about the direction COVID is traveling relative to YVR during exposure events on flights. I’ve proxied September passenger data with August passenger data to match with September exposure data, and I’ve dropped International Miscellaneous flights, which mostly involve flights to and from Mexico and harder for me to confidently link to passenger data.

A few takeaways:

  1. We get the sense that risks of exposures per 100,000 boardings are real, but generally pretty low, at least as discovered and reported by the BC CDC (where are there have been occasional transparency issues).
  2. We can also see that while most YVR related COVID exposures are happening on Domestic flights between Vancouver and other Canadian cities, the risks of exposure on these flights tend to be lower than the risks of exposure on inbound international flights.
  3. We get a peek at the gateway pattern by which international exposures tend to arrive at YVR from elsewhere, while YVR has tended, in recent months, to send more exposures to the rest of Canada than it receives from Domestic flights.
  4. Finally, while all inbound international travel remains risky relative to domestic travel, European and Transborder (USA) flights generally alternate the lead for most risk, with Asia Pacific flights trailing. That said scanning the international exposure data reveals that European and Transborder risks are generally diverse across cities, while most recent Asia Pacific exposures seem to relate specifically to flights to and from Delhi.

Big takeaway: the tentative and on-going return of air travel will likely continue to contribute to the on-going return of COVID infections, both Domestic and International. Air travel provides a key link between the rise in cases elsewhere and what happens here, potentially turning visitors into vectors. Definitely something to keep an eye on as we continue into Fall!

Why People Move in Canada & the USA: Comparing CHS, AHS, & CPS results

Why do people move? I’ve taken up this question in a series of recent posts (some co-authored), and though the available data to address the question remains sparse, it’s getting richer all the time. Today I want to compare three different sources of information, highlighting how much it matters just how we ask people about their reasons for moving.

The Canadian Housing Survey (CHS) is the newest source of information on reason for move. Its format borrows heavily from the American Housing Survey (AHS). But the Current Population Survey (CPS) also provides information on reason for move in the USA. Each survey asks about reason for move in slightly different ways.

In the USA, the CPS and the AHS ask about reason for move in different ways that might at first seem subtle, but have a big impact on results. The CPS tracks individuals, and asks where they lived one year ago. If they lived somewhere different from their current residence, they’re asked “what was your main reason for moving to this house?” This directs them to choose only one reason as their main reason, with options to specify reasons not on the list. The AHS, but contrast, tracks households, and asks only the reference person for the household if they moved in the last two years.  If so, they’re directed to a “recent movers” section, providing a little preamble and asking them repeated yes or no questions about their move, each of which might constitute one of multiple reasons to characterize their last move.

Reason4Move-A

There are a few major differences in these questions which I’ll detail in a moment, but one is worth talking about insofar as it’s especially subtle given its possible impact. Researchers often think of two separable but related processes as involved in moving. There are the “push” reasons you might leave a home and the “pull” reasons that might draw you to a new one. Reading the different questions carefully, the CPS clearly cues for “pull” reasons in specifying “reason for moving to this house.” The implicit comparison is “as compared to some other house” you might’ve moved to, rather than “why did you leave your old house.” The AHS more neutrally refers to moves overall, letting respondents sort through push or pull factors relevant to each option. I’ll come back to why this might be importantly in a moment. First let’s jump over to the Canadian Housing Survey question, which asks the responding member of each household about their previous residence and the move to their current residence, no matter how long ago it occurred.

Reason4Move-B

The set up is then quite similar to the AHS, except the CHS appears to provide all of the options at once instead of one at a time (people can still choose more than one). There is significant overlap (one might say “copying”) in the language of each option, though the CHS also provides a few extra options unavailable in the AHS, concerning moves for school, personal health, and to become a homeowner (all closely related to options available in the CPS).

Let’s quickly summarize major points of difference:

  1. individual (CPS) v. household (AHS, CHS)
  2. one-year (CPS) v. last move within two years (AHS) v. last move (CHS)
  3. different option lists (CPS, AHS, CHS)
  4. choose only “main” option (CPS) v. all relevant explanations (AHS, CHS)
  5. cued for place moving to (CPS) v. cued for many reasons for moving (AHS, CHS)

All of these differences create real problems for comparing results, but its also clear that the CHS and AHS are closest (rather than the CHS and CPS, which I’ve compared before). So let’s compare CHS (StatCan 46-10-0036-01) and AHS (Interactive Table) first. Here I’ll compare countries overall and also the four biggest metro areas within each country to get at some of the variation.

Reason4Move-C

The Canadian data is themed in the “cool” colors of blue, purple, and green, while the USA is in “hot” shades of red, orange, and yellow (“hot zone” references entirely unintentional, but perhaps apt). Here we see only the categories where AHS and CHS options map – almost identically – onto one another. For many options, the percentage of movers indicating the option at least partially explains their last move matches pretty closely. In particular, the “forced move;” “new job;” “change in household size;” and maybe “upgrade to bigger dwelling” all look like the AHS and CHS could plausibly be drawing upon the same distributions. But there are some big differences with the other options, with Americans reporting greater likelihood a move relates to “form own household;” “be closer to family;” “reduce commuting time;” “reduce housing cost;” and move to a “more desirable neighbourhood.” Are these real differences between countries or artifacts of the different surveys themselves?

Let’s zoom in on a few areas and add in the CPS comparison (here accessed via IPUMS for contemporary metro data) to provide more information. First up: forced moves!

Reason4Move-D

I’ve written about “forced moves” before, with special attention to those relating to landlords, banks and other financial institutions, and government actions in Canada and evictions and foreclosures in the USA. I puzzled over the differences between Canadian (CHS) and American (CPS) data. But looking across all surveys, we can see that the CHS and AHS data actually look very similar. It’s the CPS that seems to report an unusually low percentage of evictions and foreclosures rather than forced moves. So what’s happening? If one were reporting only the main “reason for move,” it would seem like being forced out of one’s previous residence would rise to the top, so it’s probably not just a matter of choosing a single “main” reason vs. multiple reasons. BUT let’s remember that the CPS also conditions peoples’ choices toward “pull” factors relating to the “main reason for moving to this house.” So CPS respondents are likely drawn toward considering why they ended up in their current residence, as opposed to other possible places they could’ve moved, rather than reporting on why they left their old place. Like I said, it’s a subtle difference in question wording, but here it probably has a big impact.

Returning to the AHS and CHS comparison, it looks like forced moves have been a little bit more common in the USA than in Canada, which matches with my rough expectations given differences in tenant protections, mortgage finance regimes, and economic turmoil. (If anything, I suspect these differences may become more stark, with more Americans experiencing forced moves as pandemic restrictions loosen). There remains big variation within each country, with Metro Vancouver topping forced moves in Canada and Chicago topping forced moves (and exceeding Metro Vancouver’s rate) in the USA. Of note, the CPS data is probably less reliable at distinguishing between Metros, but it’s notable that Chicago still stands out.

Let’s try moving for work!

Reason4Move-E

We can consider two different work-related options explaining moves: moving for a new job and moving to reduce commuting time. Interestingly, new jobs or job transfers account for more moves than reducing commutes in Toronto, Calgary, Vancouver, and Dallas. This is likely related to the high in-migration to these metro areas. Reducing commutes accounts for more moves in generally slower-growing metros (Montreal, NYC, LA, and Chicago). A notably smaller proportion of respondents in the CPS chose job transfer or reducing commute as the MAIN reason for moving to their current house, indicating lots of people considered a job-related move as likely just one of multiple reasons for moving – and possibly less related to why they chose a particular residence from multiple possibilities.

Let’s take a look at a suite of other, more housing-oriented reasons people might choose to move.

Reason4Move-I

“Form own household” as a reason to move is commonly thought of as capturing people like young adults (and/or divorcees) splitting off from existing households to start their own. This is a pretty regular demographic process, so it’s somewhat surprising that it seems to be related to so many more moves in the AHS than the CHS. Is this a Canadian-USA difference? Maybe, maybe not. Here the CHS and the CPS actually look more similar. What’s going on? One likely possibility is related to the fact that the AHS doesn’t have an option for people to choose “to become a homeowner” unlike both the CHS and the CPS. The closest SOUNDING option is “to form own home.” It seems entirely possible that this ambiguity in the meaning of “own home” – whether it means to become a homeowner or to separate from a previous household – explains much of the difference between the AHS results relative to both the CPS and the CHS.

Let’s compare moving for a larger dwelling with moving because of new household members.

Reason4Move-G

Change in household or family size and upgrading to a larger dwelling might be understood as related options. Again, very basic demographic processes – having children, partnering, etc. – often motivates a move to a larger home. Other demographic processes can result in smaller households, of course, but it’s less often people move in direct response. If a change in household size typically operates as a “push” (e.g. “this place is too small for us now”) then moving to a bigger dwelling operates as a “pull” (“this place is just right!”). What’s interesting here is that the CPS is predisposed to capture the “pull” part of this kind of move, and has no option at all for the “push” part. Perhaps as a result, here the CPS seems to “overperform” with “new or better home” as the MAIN reason for move almost reaching the prevalence of “upgrade to a larger of better dwelling” as one of many reasons for a move in the AHS.

Finally, let’s consider neighbourhood desirability and reduced housing costs

Reason4Move-H

Comparing the CHS and the AHS alone would make it appear that neighbourhood desirability is much more important as a reason for move in the USA than in Canada. We could spin all kinds of possible reasons for this (e.g. greater neighbourhood segregation and inequality in the USA). But adding information from the CPS reveals that moving for a better neighbourhood is very seldom the MAIN reason for a move. People mostly don’t move in search of better neighbourhoods, it’s just a kind of side feature. So maybe it doesn’t actually tell us much that Americans mention this feature more often as describing their reason for moving (when presented with it as a “yes/no” option) than Canadians (provided as one of many options). By contrast, the CPS results more closely track both the AHS results (which still run higher) and the CHS results for moving to cheaper housing as a reason for moving.

LONG STORY SHORT: every move is a story in itself. We only partially capture this story with survey questions about why people move, and how we structure those survey questions really matters for the results we get. Compare with caution!

COVID deaths in context by weeks

co-authored with Jens von Bergmann & cross-posted over at MountainMath

In our previous post on COVID mortality in context, we tried to place COVID deaths, as recorded so far this year, in the context of expected deaths from previous years. There have been a lot more developments since that post. And unfortunately a lot more deaths too.

Here we’re providing an update to our previous post, but also expanding on that post by talking a bit more about new mortality analyses and the progression of outbreaks in terms of expected deaths on a weekly basis. First, an update! We previously placed COVID deaths in the context of expected deaths at the national level, starting after the 20th death was recorded. What does that look like now?

COVID-2-a

COVID-2-b

As visible in the mortality data, Belgium has moved to the forefront of the COVID outbreak in Europe in terms of COVID deaths relative to expected deaths from years prior. Ireland, the UK, and the US appear to continue to climb. By contrast, Spain and Italy, early centres of the outbreak in Europe, have largely leveled off. Though the USA “leads” in deaths from COVID-19, this doesn’t (yet) show up in the relationship between COVID deaths and expected deaths because the USA is enormous, with a lot of expected deaths every year, and the outbreaks of COVID deaths have been heavily concentrated in a few locales so far.

Overall, and as mentioned previously, there’s still a lot we don’t know in these comparisons. For instance, we don’t know if we’re actually counting all of the deaths due to COVID. Lots of people don’t get tested, and cause of death is always tricky to determine in the best of times, let alone with an overloaded medical system and coroners’ offices. As a result, revisions to the data can add dramatically to the death toll, as happened recently in New York City. In addition to good COVID death data, we’d also like updated data on mortality overall. We’ve seen recent – and very preliminary – data out of NYC and scattered other locales suggesting that all-cause mortality has risen dramatically in places with severe COVID outbreaks.

Where we have it, we can put updated all-cause mortality in conjunction with COVID mortality and expected mortality all together. Putting this on a weekly basis really provides a sense of the progression of outbreaks and how overloaded they leave medical systems in terms of the normal deaths they have to deal with. Given some of the data from NYC, here’s roughly what that looks like.

COVID-2-c

 

We notice a downturn in deaths as recorded by the CDC FluView for the last week they report data (the week ending on 2020-04-12). This is not a REAL downturn. Rather it illustrates the reporting lag for data on deaths. It can take several weeks for the numbers to fill in and stabilize. We added the reported Covid-19 related deaths as assembled by the JHU for reference. JHU data was aggregated up the week ending 2020-04-19, so it’s nominally a week ahead of the FluView data. However, these deaths are coded by date reported, unlike the CDC data that is coded by date of death, which causes the JHU data to lead a bit. Even accounting for a possible time shift in JHU data, it appears that JHU data does not account for the full increase in all-case mortality, hinting at likely under-reporting of Covid-19 deaths in the JHU data.

Unfortunately we still don’t have updated all-cause mortality on the country level. As suggested by the lag in NYC data, it takes awhile to compile in the best of times (here’s a look at efforts to gather some of the European data). So here we’ll provide a replication of our previous analysis, but breaking out COVID deaths against expected deaths on a weekly basis for countries instead of across the entire length of the outbreak.

COVID-2-d

COVID-2-e

 

Overall, weekly COVID deaths as a percentage of expected deaths looks broadly similar to our earlier figure, which charted the rise in COVID deaths as a percentage of expected deaths since outbreak deaths began. But there are a few significant differences. The weekly chart better highlights the evolving overload on hospitals and health systems, as well as coroners’ offices, and this is reflected in the y-axis, demonstrating that COVID deaths in Belgium have more than doubled the expected deaths in the last week for which we have data. The weekly chart also more quickly identifies declines in the relative impact of COVID deaths in places where the worst of the outbreak has passed, like Spain, Italy, and France. It will take a long time for the expected death toll to diminish the impact of the overall death toll of COVID in our figures at the top of the post. But on a weekly basis, we can already see the toll of COVID receding in many places.

As we’ve noted previously, it will still take a long time to sort out the overall effects of COVID on mortality. Why? Well, we’re still nowhere near done with the outbreak, and we can expect deaths to continue until we have a vaccine and have reached some level of “herd immunity.” But we’ll also be sorting through the mortality data for years to come. Also important: the toll at national levels, while helpful in assessing cross-national differences, masks the impact at local levels where outbreaks often occur. So it is that the estimate from Belgium, where most recent weekly COVID deaths appear to have more than doubled expected mortality, is dwarfed by the estimate from New York City, where the most recent weekly COVID deaths appear to be more than six times the expected (pre-COVID) mortality.

As usual, the code for the post is available on GitHub in case anyone wants to refine or adapt it for their own purposes.

Context for COVID-19 Mortality so far

co-authored with Jens von Bergmann & cross-posted over at MountainMath

Unfortunately, more and more people are dying due to COVID-19. We won’t know the full toll from COVID-19 for quite some time. But we can at least start to get a sense of its impact. One useful way of assessing the impact, of course, is just to plot deaths attributed to COVID-19. This highlights the real loss of human lives associated with outbreaks. But as any demographer can tell you, deaths are a normal part of life. Within a given population, we can reliably expect a certain number of deaths to occur over any given time period. So another way of visualizing COVID-19 deaths is also useful: How many deaths attribute to COVID-19 are occurring as compared to the deaths we would normally expect to occur?

Below we follow the rise in deaths attributed to COVID-19 through time relative to the expected number of deaths that likely would have occurred without COVID-19 during the same time. [UPDATE April 20: our newest post plots this on a weekly basis]

 

COVID-mortality1

 

This visualization places deaths reported from COVID-19 in the context of expected deaths overall. This helps establish where we know the mortality toll has already been enormous. As of March 31, the end-point of the animation, Italy leads the overall count in deaths attributed to COVID-19. Here we can also report that in just over a month, Italy’s deaths so far attributed to COVID-19 already add more than 20% to its expected deaths. But Spain’s toll relative to its expected number of deaths is ever higher. In just over three weeks time, we can see that COVID-19 already accounts for more than a 30% rise over the deaths that would’ve been expected without COVID-19.

 

COVID-mortality2

 

Unfortunately, most curves are still rising. So far. Initially curves grow exponentially, until aggressive containment or mitigation strategies flatten them. Curves that stabilize and flatten, or even begin to turn downward, reflect countries where deaths attributed to COVID-19 are being overtaken by deaths that might’ve been expected to occur anyway. Hopefully this reflects an outbreak coming increasingly under control – GOOD NEWS – rather than a data gap.

But the possibility for data gaps is very real. It will be quite awhile before we can properly estimate the overall toll from COVID-19. We already have preliminary data on deaths attributed to COVID-19 rolling in. But this data will be messy, excluding cases where COVID-19 was missed as a cause, despite being present, and possibly over-including cases where the cause was actually not COVID-19 (e.g. instead common influenza), or COVID-19 was present but the death should be attributed primarily to a different underlying condition claiming the life. Cause of death data is never clean to begin with. As COVID-19 overwhelms medical systems and coroners’ offices, we should fully expect that data quality will suffer further.

More concretely, COVID-19 deaths will show up in the mortality databases with code U07.1 or U07.2 in the current ICD-10 classification system (or RA01.0 and RA01.1 once ICD-11 comes into effect). But many will likely also get classifed as J11, J18 or J22. When the dust settles, we will have to check how these cases have evolved over time and estimate how many cases in 2020 (or late 2019 in the case of China) are likely misclassified COVID-19 cases.

 

COVID-mortality3

 

We will also eventually get data about overall mortality. We will likely see deaths increase beyond those attributed directly to COVID-19. Deaths will rise both in response to complications introduced by COVID-19 in those with pre-existing conditions, and in response to people dying due to failure of overloaded medical systems to be able to respond to non COVID-19 cases they way they normally would. At the same time, some other non-COVID deaths may go down. This can happen when COVID-19 claims lives that otherwise might’ve been claimed by something else (e.g. an underlying condition). But it can also relate to deaths that don’t occur due to lockdown and the measures related to dealing with COVID-19. For instance, the regular toll of influenza may diminish in response to the lockdown targeted at Coronavirus (making it unclear what the “expected” baseline case count for 2020 should be). Similarly, fewer cars on the road will likely result in fewer deaths from car accidents. For references, see the most common causes of death in Canada in normal years here. A similar discussion of the eventual breakdown we’ll need in mortality data can be found in this demographer thread attempting to summarize some of this complexity via twitter feed.

The mortality data coming in bears watching, both in terms of COVID-19 attributed deaths and deaths overall. Some analysts (e.g. in Italy and Spain) as well as some China skeptics, are already drawing upon anecdotal mortality data to suggest that the toll from COVID-19 is far greater than revealed in the official data so far. These kinds of analyses are especially potent when applied to cities and regions as opposed to countries. But ultimately it will take years for demographers to sort this all out. In the meantime, we can at least get a rolling sense of COVID-19’s toll by looking at deaths attributed to Coronavirus relative to deaths otherwise expected based on past data from the same rough period of time.

As usual, the code for the post is available on GitHub in case anyone wants to refine or adapt it for their own purposes.

 

Update (2020-04-06)

It’s been a week since we posted this, and things are changing fast with covid-19 related deaths increasing exponentially and background mortality estimates only increasing linearly with time. The traces in the animated GIF already highlight this, but here is a quick update of what the graph looks like using data from a week later.

covid_mortality2

And for completeness, here is the static graph with the latest available numbers.

covid-19-mortality-final2-1

 

Overnight Visitors and Crude Travel Vectors

co-authored with Jens von Bergmann & cross-posted over at MountainMath

The spread of Coronavirus is reminding us of just how often people travel around, especially as various locations become quarantined and international travel corridors get shut down. So let’s take a look at some basic data on travel patterns here of relevance to us here in Vancouver. Then we’ll put them back in the context of Coronavirus.

TLDR: travel data is really interesting, don’t be frightened of travelers, and there’s still a lot we don’t know about coronavirus.

We’ve looked at the movement of people before in terms of migration, immigration and commuting patterns. But these are movements that are either regularized, everyday, and routine (e.g. commuting) or shuffle people between one settled set of routines and another (e.g. migration). Travel data gives us something different, representing something more like the unsettled movement of people. People travel for work, to visit family, and of course, for tourism. The Tourism Industry is interested enough in travel data that they ask Statistics Canada to compile data for them. Stats Canada combines Canadian travel surveys and border crossing administrative data to get us a decent look at overnight stays. So it is that we get overnight stayer data for Vancouver!

Let’s look at where people are visiting Metro Vancouver from. The Tourism Vancouver data has an interesting selection of countries available, with special breakdowns for Canada and the USA. More than a quarter of all overnight stays in Metro Vancouver are trips from elsewhere in British Columbia. Another quarter plus of trips arrive from elsewhere in Canada, with Ontario and Alberta leading the way. The USA accounts for just under a quarter of overnight visits. Altogether, Canada and the USA account for over 8 million of the roughly 10 million visits. Most American visitors to Metro Vancouver arrive from nearby neighbours down the Pacific Coast (WA, OR, CA), which together account for over half of travel from the USA. About as many people visit from all of Mexico as from nearby Oregon (140k).

Overnight1

Of the slightly less than two million international visitors from beyond NAFTA borders, a little over half arrive from Asian/Pacific countries, with most of the remainder from Europe. China, the UK, and Australia, Japan, India, and Germany each accounted for more than 100k visitors in 2019, South Korea, Hong Kong, and Taiwan not far behind. Let’s put all these flows together on a map (click for interactive access).

Overnight1a

Of some concern, lots of the places identified above have had recent outbreaks of Coronavirus. We’re still in early days of tracking the virus. And we know it’s already having major effects on travel. But can we look at current prevalence estimates and recent travel patterns to give some insights into crude vector risks for Metro Vancouver? Maybe. It’s worth keeping in mind that everything is still pretty much up in the air in terms of what we know!

First let’s look at up-to-date active confirmed Coronavirus cases drawing on data collected at Johns Hopkins.

Overnight2

Wuhan, of course, appears as the centre of the outbreak, and Hubei Province in China contains most of the active confirmed cases to date (as of March 03, 2020!) The number of cases is important to track, obviously, and the starting point for healthcare workers and epidemiologists alike. But focusing on these numbers can provide a misleading impression of how widespread the Coronavirus has become. So let’s come up with a crude estimate of prevalence instead of case numbers. Here we’re going to use active confirmed cases as our starting point. Another option is to track all confirmed cases, including those who have recovered (no longer testing positive) or died from coronavirus. But active confirmed cases might arguably give us a better sense of current spread.

We can plot the evolving nature of active confirmed cases in terms of prevalence estimates across places, effectively dividing total number of active confirmed cases by population for our data reported so far. Setting this to motion, we can track outbreaks by prevalence across time. Even just looking at active confirmed cases, we get a sense that recorded prevalence has recently stopped climbing for Hubei province. Meanwhile, outbreaks in South Korea, Iran, Hong Kong, and the nearby state of Washington continue to grow. Also worth noting, some countries (e.g. South Korea) seem to have a better handle on testing the virus, providing better confidence in their numbers. The numbers coming out of other locales (Iran and the USA) seem far less reliable, either because of inconsistent testing, untrustworthy reporting by officials, or both. This sets a real limit on what we can know so far.

overnight3

Overall it needs to be stressed that – given the numbers we have so far – the prevalence of coronavirus is still very low. Even in Hubei province, the centre of the outbreak, not much more than a single active confirmed case per thousand people has been confirmed. Comparing locations of cases to surrounding populations, most places around with the world with outbreaks still see only about one active confirmed case per hundred thousand people. Even setting aside the hyper-cautious mood around the world and its effects on travel, if you met a visitor from one of these places in Metro Vancouver, fairly unlikely that they would be a carrier. There’s little reason to be scared of individual travellers!

But what about travel patterns writ large? Surely even if any individual presents a very low risk as a vector, by sheer number, the masses of people travelling through Vancouver from places with coronavirus outbreaks represent a risk. Indeed, that’s how the coronavirus has spread so far. We can very crudely estimate this risk by setting a base likelihood that each individual traveller from a given outbreak location is coronavirus-free (1 – cases / population). In other words, we might use currently active confirmed cases as our measure of prevalence, estimating we can be 99.99975% certain that a given traveller from Washington State will not be a carrier for coronavirus. But what if a LOT of people travel from Washington? Then we exponentiate 99.99975% by the number of visitors (126,493 for the first three months of 2019 as a proxy) to come up with an estimate that none of these travellers carry the virus (we really should be drawing without replacement here, but this is a good approximation), with the complement giving a rough estimate of at least one visitor being a carrier. This comes out at 27% using our current estimates. This only considers Washington residents travelling to Vancouver and still neglects Vancouver residents travelling to Washington and getting infected there. And it relies on current active confirmed cases, it does not include active but not yet confirmed cases. And it assumes travel patterns similar to a year ago. Still, it provides us with a measure of vector risk to Metro Vancouver that combines risk of coronavirus with travel volumes.

Let’s run with this for recent coronavirus outbreak data based on travel volumes similar to past years – EXCEPT excluding cases from Hubei province in China after January 23rd (when the quarantine went in place). What does our crude evolving overnight travel vector risk look like?

overnight4

Here we can see rapidly changing vector possibilities. Conditions are changing fast! Still, it’s hard to know how much to trust these numbers. Given what we understand about testing at the moment, it’s likely we’re still overstating the risk from high quality testing locales (South Korea), as well as understating the risk from places where testing has been poor (Washington) and places where we don’t have any visitor data at all (Iran). We’re also missing current data on how travel is changing as well as data on where people from Metro Vancouver are traveling, which is a big deal given that most of our cases so far represent returned travelers from abroad.

Here is a still of the most recent snapshot as of the writing of this.

Overnight5

Upshot

So here are the big takeaways from our exercise: 1) Visitor data to Metro Vancouver is actually really interesting, even for those outside of the tourism business. 2) Don’t shun travelers from abroad! The likelihood of anyone you meet, even coming from an outbreak centre, being a carrier of coronavirus is very, very low. 3) The combination of travel patterns plus coronavirus prevalence gives us some interesting ways to model evolving vector risks in Metro Vancouver. 4) But it’s not clear how much we should trust our data. Travel patterns have surely altered, and we need better coronavirus testing fast, especially in places like Washington State.

Overall, integrating travel data with coronavirus data may, if nothing else, help people and agencies prepare and plan better. Practically any planning is better than some of the ad hoc decisions being made out there, as when American Airlines suspended its flights to Milan only after pilots refused to fly there. For most people, the important thing is to listen to local health agencies, like the BC Centre for Disease Control, wash your hands, and be kind to those around you, wherever they come from.

As usual, the code for the post is available on GitHub in case anyone wants to refine or adapt it for their own purposes.

 

UPDATE: For a look at how the professionals are joining international travel data to coronavirus data, see Gardner (et al) here (now unfortunately outdated!)

Wealth vs. Income

co-authored with Jens von Bergmann & cross-posted over at MountainMath

Wealth and income are different things. Wealth is measured in terms of assets minus debts at any given point in time. It can accumulate or deplete over a lifetime and across generations. By contrast, income represents some variation of how much money one makes over a given time period (usually a year). Most people get this on some level. But since both income and wealth deal with people and their money, the terms are also often used interchangeably. So it was that the CBC yesterday reported that “B.C. budget 2020 promises new tax on wealthy to help ensure future surpluses” despite the actual new tax being a tax on high-income individuals.

Here the difference matters for two reasons:

  1. it matters because wealthy people aren’t always high income, and high income people aren’t always wealthy, and
  2. it matters because a wealth tax is quite distinct from an income tax, and in this headline the two are blurred together (fortunately the article clarified).

With wealth taxes in the news (and in multiple Democrats’ platforms in the US), it’s important to separate out wealth taxes from income taxes. Here in Vancouver, as we’ve noted before, our property taxes actually do a pretty good job of taxing wealth.1

In this post we’ll focus on our first point: just how well do wealth and income line up together? Underneath this is also the question of how to measure wealth and what to include as income, we will just go with the standard definitions from StatCan’s Survey of Financial Security to answer this question for family net wealth and family income. The data allows us to divide up the Canadian population into equally sized quintiles (fifths) by net wealth and by income. What overlap do we see? The data also allows us to break out sub-areas of Canada, including the Atlantic provinces, Quebec, Ontario, the Prairie provinces, and British Columbia. So let’s run those too!

First let’s look at how income quintiles break down by wealth quintiles, as assessed all across Canada. How many families in the lowest income bracket fit into each wealth bracket? Are they all the lowest wealth bracket? Nope.

wealth-v-inc-1

 

We can see a clear relationship between wealth and income. But only about half of lowest income families in Canada fit into the lowest wealth category. The same is true on the other side of the distribution. Only about half of the highest income families fit into the wealthiest category. Moreover, there are wealthy (highest quintile) and poor (lowest quintile) households in each and every income quintile. Counter-intuitive as it may seem, there are clearly poor high income folks and wealthy low income folks. Not very many, but at any given point in time they definitely exist.

Let’s look at some of the provincial differences, remembering that we’re using Canada-wide quintiles. Looking at raw numbers, it’s quickly evident that some provinces (Quebec and Atlantic Canada) are disproportionately lower-income, while others (the Prairie provinces) tend toward higher income. Ontario and BC are more inbetween. Looking at what percentage of each income quintile fit in each wealth quintile by province, the general pattern of a correlation between wealth and income is evident in all provinces. But looking more carefully, a few differences jump out, especially between BC and the Prairies. In BC, each income quintile has a higher proportion of families in the top wealth quintile than one might expect – including the lowest income quintile: wealthy low income folks. In the Prairies, by contrast, each income quintile looks less wealthy than one might expect. In each case, despite the correlation between wealth and income, there are also people showing up in each category.

Flipping the chart around, we can look at how many families in the highest wealth bracket fit into each income bracket. Only about half of the wealthiest families in Canada are in the highest income quintile. There’s even greater diversity in BC, where only about 40% of the wealthiest are in the highest income quintile.

wealth-v-inc-2

Let’s pull out BC from the rest of Canada and run the numbers matrix style. If there were a perfect correlation between income quintile and wealth quintile, then we’d see a bright diagonal line filled with 20% of families in each of the five diagonal cells, surrounded by twenty cells with 0% of families. If there were NO relationship between income quintile and wealth quintile, we’d see each of our twenty-five cells filled with roughly 4% of families. What we see is somewhere inbetween. For Canada as a whole, we see strong evidence of correlation at the margins (for highest and lowest quintiles), but the middle looks very mushy. For BC, we see a strong relationship between being in the top income quintile and the top wealth quintile. But everything else looks mushier than expected. In effect, BC stands out for its generally limited correspondence between wealth and income.

wealth-v-inc-3

What throws off the relationship? Many peoples’ wealth represents savings over one or more lifetimes. So age matters, as does inheritance. Immigration can also affect patterns, with different results evidenced by program (e.g. investor), time in Canada, and wealth accrued in country of origin (Vancouver’s far from the only place where rapid escalation in prices have made millionaires of home owners). Asset inflation also matters, and BC’s rapid appreciation in real estate wealth surely plays a role in its weirdness. As a reminder, capital gains accruing to primary residence don’t show up in income statistics, but they definitely represent wealth. We could cap current exemptions on this enormous tax break for home owners, taxing these capital gains more like income. But we could also just levy an overall wealth tax. Returning to a theme, taxing wealth is distinct from taxing income.

All of which is to say: wealth and income are not the same thing. And it matters. Especially in BC!

As usual, the code for the analysis is available on GitHub.

 


  1. And our property taxes are still too low! [return]

Why Do People Move? New Data, Mysteries, and Agendas

How often do people move, and why? Canada has ok data on the first question, and as of yesterday (!) also some ok data on the second. The USA just released its most recent data, with even better answers for both questions. The big finding out of the USA data, attracting significant media coverage, is that Americans just aren’t moving as much as they used to… which is pretty interesting.

Let’s start by comparing the USA to Canada in broad terms. Here I’m looking only at moves over the course of a year (the one-year mover rate), and I’ll just pull from the USA data on movers for recent Canadian census years (2001, 2006, 2011, 2016), and add the most recent year available (for 2018-2019). I’ll also break the numbers down into their component types of moves: short-distance mobility (within county in the USA, within municipality in Canada), longer-distance migration (between counties and states, or within and across provincial lines), and immigration (from another country).*

Mobility1

Overall mobility for both Canadians and Americans dropped between 2006 and 2011, with the intervening Great Recession likely a big explanation for the decline (as well as its greater severity in the USA). But Canadian mobility rebounded, while the Americans continued to… well… stay at home. Just under 10% of Americans moved in the last year, compared to just over 11% in 2015-2016, when a comparable 13% of Canadians moved.

What’s apparent for both countries is that short-distance moves (within the same county or municipality) dominate moves overall, and correspondingly tend to drive broader trends in mobility and migration. Even though geographies of moving can be funky (and US counties are especially weird in this regard), this is a pretty stable pattern. Given the different geographies, it’s hard to read too much into the differences in longer-distance moves between Canada and the USA, but more long-distance moves cross state lines in the USA than provincial lines in Canada. And finally, while still small overall, immigrants (crossing international lines in the last year) make up a bigger proportion of movers in Canada than the United States, actually exceeding the proportion of movers crossing provincial lines.

But why do people move? The USA has good data on that! (Tables 17-18). Here we’ve got the main reason for a given move (often there are more than one), divided into a set of common categories. Let’s break it down by distance moved to show off some general patterns and how short-distance moves are different than longer distance and international moves.

Mobility2

Pretty neat! Short-distance moves (within counties) are dominated by those moving for housing reasons. Longer-distance moves (between counties) are much more heavily focused on work reasons, chief among these moving for a new job. International movers respond primarily to other concerns, with education being a big one! (Housing reasons drop away almost entirely). Strikingly, moves for family reasons are pretty constant across all distances. Thinking about immigration, the categories we get, including: Work, Family, Education, and Other (including refugees) map onto a variety of federal immigration programs, both in the USA and Canada.

Let’s also talk a little bit about the actual reasons given, starting with the work-related categories (in green), including moves because of a new job, moves because of looking for work or recently losing a job, moves to be closer to work (reducing a commute), moves because of retirement, and other job moves. Most work-related moves are for a new job or to be closer to work. Next come family-related categories (in yellow), including moves because of changes in marital status (e.g., moving in, getting a divorce), starting a new household (e.g. moving out of the parental home), and other family (e.g. moving to be closer to a parent, needing room for more kids, etc.). After that I’ve placed a variety of miscellaneous reasons for moving in shades of brown and red. The largest of these, separated out from a generalized “other,” are moving for school (e.g. university), moving for health reasons (e.g. closer to care), and change of climate (e.g. moving to Florida). But natural disasters also motivate a significant number of moves, especially for international movers, and in a world of climate change that’s definitely a category to keep an eye on. Finally let’s turn to housing-related categories (in blue). Here we see people moving because they wanted to own a home (usually after renting), because they wanted a better home, to live in a better neighbourhood, to live in cheaper housing, or because they were evicted or foreclosed upon, with a residual of other housing-related reasons bringing up the rear.

Let’s look at historical variation in reasons for move with handy data from the past twenty years.

Mobility3

Work-related and Family-related reasons for moving seem to have declined only slightly over time. The big decline in American mobility is strikingly concentrated in the decline in moving for Housing-related reasons. We might think of this as reflecting a real decline in housing opportunities, leaving younger people, in particular, “stuck in place,” as per this Brookings report. “Other” reasons for moving may have gone up slightly in recent years, though it’s difficult to fully compare given a variety of changes to survey instruments and coding (e.g., an instrument error may explain truncation in the 2012-2015 era, and new coding procedures for write-in reasons were adopted in 2016).

What’s the new Canadian data on reason for move look like? Unfortunately, it’s different and slightly less useful for some questions than the US data. But it’s something! (Hat tip to Jens, who told me it was out & already wrote up a blog post about it). What the Canadian Housing Survey has done is ask people about whether they’ve moved in the last FIVE years (rather than the last one year). If they’ve moved, the survey asked the reasons for their last move. Canadians could report more than one, which reflects the complexity behind peoples’ actual moves, but unfortunately also makes it difficult to distinguish and compare the main reason for peoples’ move. But let’s look at reasons overall. We don’t have quite the same set of reasons codified in Canada as in the USA, but there is significant overlap, and broad categories can be grouped in more or less similar fashion. Here (for selfish reasons) I also provide a cut-out for my province of British Columbia (BC).

Mobility5

In broad terms, we can see that the categories and their relative importance match up pretty well with what we get in the USA. Housing factors dominate reasons for move, and the largest reason people in Canada give for their moves is that they moved “to upgrade to a larger dwelling or better quality dwelling,” an explanation involved in over a quarter of all moves. Moving for family-related reasons comes next, followed by moving for work-related reasons, as in the data for the USA. Leftover “other reasons” in Canada is a little more inclusive in Canada than in the USA, but we can see that it’s still a residual category, without as much overall explanatory power as the others.

Looking at specific reasons, where they match up to reasons in the USA data, they tend to carry the same general explanatory power. Most moves are about finding housing, matching it to one’s family or household, and matching up to a job. But there’s one reason for move that really jumps out in the Canadian data, despite playing a much smaller role in the American data. So let’s talk more about evictions and foreclosures!

Being “forced to move by a landlord, a bank or other financial institution or the government” is a factor in over 6% of Canadian moves, jumping up to a staggering 10% of moves in British Columbia. One-in-ten moves involves a shove out the door! Those are big numbers. I’ve got ninety-nine reasons for why we might expect BC to see a higher proportion of moves involving these kinds of interactions than Canada as a whole (e.g., we don’t have enough homes, we’re dominated by Metro Vancouver‘s super-tight housing market, and we rely much too heavily on unstable secondary suites and condo rentals that can be reclaimed for use by their owners). But assuming this is mostly about eviction and foreclosure, I really don’t have any good explanation for why they would be playing such an outsized role in explaining moves in Canada relative to the USA. It’s a mystery!

To get a sense of how big of a difference we’re talking about, let’s go back to the data from the USA. In the most recent year, less than 1% of moves (an estimated 216,000 in total) were mainly the result of an eviction or foreclosure. We can go back further. The USA only began providing and recording evictions and foreclosures as a standard option in 2012, but they include a coding of write-in answers in 2011. Good timing, with respect to the aftermath of the Great Recession, as foreclosures piled up, weighing heavily on peoples’ lives as well as the post-Recession recovery more broadly. In the peak year of 2011-2012, an astonishing 792,000 Americans reported moving due to eviction or foreclosure. And yet… that number still represented just over 2% of all movers, with over 35 million moving in that year.

Mobility4

By contrast with the USA, Canada has low mortgage arrears and foreclosure rates and tends toward relatively strong tenant protections. It might simply come down to the survey options available for people to choose. “Forced to move” may be read as more inclusive than “eviction or foreclosure” in such a way that people more readily recognize their circumstances in the former (language of everyday life) than in the latter (legal language). Canadians may also be expanding the range of reasons they were forced to move to encapsulate more ambiguous situations like “my landlord kept trying to sell the place, with showings every week, so we had to get out of there.” So maybe the US and Canadian data just aren’t fully comparable here.

Returning to my ninety-nine reasons for BC’s high rate of forced moves relative to Canada as a whole, it’s worth noting that we do actually have some data on evictions, thanks to Nick Blomley’s team at SFU. Eviction proceedings mostly follow missed rent checks, just as foreclosures almost entirely follow borrowers missing their mortgage payments. Overall, even in Metro Vancouver, the proportion of evictions related to landlords reclaiming dwellings for their own use appears to be pretty small, involving less than 4% of tenant-landlord disputes between 2006-2017 (compared to nearly 40% involving missed rental payments, p. 9 & 12). That said, landlords reclaiming dwellings for their own use seems to be on the rise (p. 10). But overall, the informal ways people feel forced to move by their landlords, banks, or governments, may play a significantly larger role than formal eviction or foreclosures, perhaps even pointing to some shortcomings of the US data for missing a more expansive understanding of forced moves. Can you guess what I’m going to say next? We need more research on this topic!

Forced moves attract attention because they’re the kinds of outcomes we should be working hard to prevent, and it’s important to provide strong protections enabling and supporting people to stay put in their housing where possible. There are good reasons to support an anti-displacement agenda, especially providing for tenant protections. But bearing this in mind, it’s also important to recognize and normalize moving.

Most moves represent positive experiences for people: leaving home, getting married, making room for a child, getting a new job, moving closer to work, moving to better housing or a better neighbourhood. Sometimes such moves are vital, as when people need to escape from a bad family situation. The right to move is protected in some form or another in both the USA and Canada (Charter of Rights!). But it’s largely meaningless without the right to housing. We should be protecting the right to move, together with the right to housing in places people want to move.

To put the matter differently, an anti-displacement agenda is important to protect peoples’ existing housing arrangements, focused on those currently lacking legal standing to remain in place (i.e. most tenants). But anti-displacement efforts must be coupled with a broad pro-mobility, pro-housing agenda in order to fully enact, protect and expand peoples’ right to move and right to housing. Fortunately, evictions and foreclosures seem to be declining in the USA, but moving overall has also declined. Evidence suggests that the decline in moving in America may be most strongly related to a decline in housing opportunities (e.g. Glaeser & Gyourko). We know moving overall has rebounded in Canada, though we don’t yet know if people are increasingly feeling forced to move. The numbers out of BC are certainly disturbing. Pushing for an expansive right to housing means continuing to work toward strong protections for existing tenants, but also – and crucially – working to make sure people can move pretty close to the places they want and need to go.

Let me end by proposing a simple motto for our governments to work toward: Freedom to move and freedom to stay, we’ll get you housing either way.

 

*- I use the data with the most recent base in the US dataset (e.g., 2010 census for 2010-2011 year in USA), and for the 2001 Census year in Canada I extrapolate the finer categories here from cruder categories available using the corresponding proportions in the 2006 Census year.  Check original files for a variety of other cautions with the data.

Metrics and Bird Memes

 

Working with Jens von Bermann, I gave a talk yesterday at #HousingCentral on housing metrics! Specifically, we talked through and expanded upon our earlier joint blog post on the same topic. Click the image below to visit our full slides.

Image-Talk-HousingCentral

Included in the slides are a variety of graphics, mostly from past posts of mine and Jens’. In case you’re curious, follow the links below to find out more about them:

Rent correlation with vacancy rates

Price correlation with inventory (borrowed from YVR Housing Analyst)

Crowding measures

Urban Density

Homeless Counts

Empty Homes

Core Housing Need

and Job Vacancies

As for the conference, Housing Central is an annual shindig put on by the BC Non-Profit Housing Association (BCNPHA), including a special set of panels on research from the fine folks at the Pacific Housing Research Network (PHRN). Check the PHRN Symposium website for calls if you’re interested in presenting!

Last but not least, I took some bird pictures down along the southern edge of the Fraser River delta, and I really, REALLY want to turn them into as many housing memes as I can. So here’s me summarizing our Housing Central talk with a bird-based housing meme.

Birds-per-Post-2

Enjoy!

Simple Metrics for Deciding if you have enough Housing

(co-authored with Jens von Bergmann & cross-posted over at mountainmath)

What are the best metrics for understanding if a given place has enough housing, just the right amount, or too much? Whether you’re a potential renter or buyer or an analyst or policymaker, the answer really depends on what you’re looking for.

For potential renters and buyers, if you can’t find what you’re looking for and/or it’s not in your price range, then there’s not enough housing. If you can find it, then there’s just the right amount. When is there too much housing? Mostly if you’re already comfortably housed, but concerned about changes to your neighbourhood and/or you’re looking to maximize the price you can get for selling your housing. So we can root a set of foundational answers to questions about housing supply in peoples’ direct experiences interacting with the housing market. We can also extend this to non-market housing. If there are people on the waitlist, there’s not enough non-market housing (note: there are ALWAYS people on the waitlist and we definitely need more non-market housing).

But decisions about whether we have enough housing aren’t actually left to people interacting directly with housing markets. Most people can’t add much to the supply of housing by themselves. Housing has become exceptionally technical, and a vast slew of regulations now prevent most self-building except in informal sectors (in Vancouver most notably the subdivision of existing dwellings into suites, only a minority of which comply with building codes and have a permit). Instead most decisions about how much housing we have are produced via a combination of developers working through their financial models in conjunction with planners, regulators, and politicians working with tight existing constraints on what can be built where. Interestingly, both the comfortably housed and those looking to maximize their prices for selling housing DO get a voice. Why? They tend to be the ones electing (and speaking directly to) local politicians. This group notably includes local developers, who are both actively engaged in maximizing the prices they can get for selling housing and actively engaged in local politics (if you think market developers are unambiguously pro-supply, think again).

So how do we know if we have enough housing in a given place? Or, since the answer always depends upon the perspective, how do we hear from potential residents (including renters and buyers) about whether THEY have enough housing? Their voices are the ones that tend to get left out of debates. Usually, to the extent their voices are heard at all, it’s through some set of metrics informing decision-makers. So let’s return to metrics, because different metrics tell us different things!

Ideally decision-makers consider metrics with specific goals in mind: do we have enough housing in a given place for what purpose? Are we interested in enough housing to meet demand, preserve affordability, or address need? Enough to promote the right kind of growth? Enough to support transit, reduce greenhouse gas emissions, promote urban vitality? Or perhaps we’re worried about too much housing to support our preferred sales price, keep out the wrong kind of people, preserve our favourite aesthetic, maintain green space, or just generally keep our neighbourhood the way we like it? Being clear about these goals is helpful, insofar as they set the criteria for which metrics can provide meaningful answers. If we can decide on our criteria, then we still have to figure out the right metric. Let’s start by looking at the four common elements that make up most metrics:

  • Dwellings
  • Money
  • People
  • Land

These are the things we tend to track with our metrics for whether or not we have enough housing, just the right amount, or too much. Dwellings are housing. If we want to figure out if we have enough, then we definitely need to keep track of dwellings. Of note, dwellings can also be differentiated by square footage, number of bedrooms, and related characteristics. Money is an expression of desire, weighted by wealth and/or income (and hence also inherently unequal). People are bodies, variously disposed to live together and share space. Both money and people move around, unlike most dwellings, which are fixed in place. Land is how we fix dwellings in place, and can support various numbers of dwellings. By virtue of fixing dwellings in place, land also defines various kinds of places we might be concerned about: e.g. neighbourhoods, cities, and metropolitan areas. Places are connected to one another: what happens Downtown has an impact on nearby neighbourhoods (e.g. Kitsilano), just as what happens in the City of Vancouver has effects on what happens in the City of Surrey. As a result, metrics should pay careful attention both to place of interest and interconnection between places. In the background, fitting these elements together, we also want to keep in mind that time matters to how we construct metrics.

The key metrics we tend to track often involve just two of the elements above, measured at varying scales of aggregation, places, and times. We can provide a quick and dirty guide to the different questions answered by the key metrics we use to measure if we have enough housing or too much as well as the underlying logistical mechanism guiding our inquiries.

Class of Metric Q. Do we have enough / too much housing to… Logistical Mechanism
Money per Dwelling Preserve Affordability Market Allocation / Inequality
People per Dwelling Fit people into dwelling units Rationing / Sharing Rules
Dwelling per Land Support Urbanism / Reduce Env. Impact Rationing / Sharing Rules (via zoning)

How we define elements matters to how the metrics work, as does how we incorporate time and the level of aggregation (individuals, households, census tracts, cities, metro areas). We’ll keep coming back to these throughout, often with reference to examples from Vancouver, the metro area we know best, but it’s helpful to start by keeping things simple.

Money per Dwelling (a.k.a. price)

Perhaps the most obvious way to bring these elements together is by asking how much dwellings cost. Given the persistence of market allocation for housing, there will always be enough housing to meet demand… at some price. That’s because the price mechanism sets prices at where demand curves and supply curves meet. Put differently, the demand for $1 dwellings is practically limitless. The demand for $100 million dwellings is practically zero (so far). In between, there’s a demand curve specifying how many dwellings would sell at what price. On the supply side, self-interested owners would rarely sell dwellings if they could only sell them for $1. But they’d probably sell as many as they could get away with if they could sell them for $100 million. In between there’s a supply curve specifying how many dwellings will be sold at what price. The market pricing mechanism moves prices toward equilibrium where demand and supply curves meet. This is the stuff of basic economic analysis (brought to you by a mathematician and a sociologist).

How about if you don’t just want to meet demand, but you want to meet it at a particular price? Maybe you want the market to meet a certain affordability threshold for a certain kind of dwelling? Let’s define this better: do we have enough housing if we want the average two bedroom dwelling priced at $250,000? In some places (e.g. Edmonton), this isn’t far off the mark. There’s enough housing there relative to demand that two bedroom dwellings sell for about $250,000. In other places (e.g. Vancouver), there’s not enough two bedroom dwellings to go around to everyone who might want them at that price, so they’re bid up to a far higher price. It would take the addition of a lot more dwellings to bring prices down to $250,000. So if that’s where you want prices to go, then there is definitely not enough housing.

The same general dynamics apply to the market pricing mechanism for apartment rents. Landlords respond to their understanding of local supply and demand when setting their asking rents. The longer their apartments stay on the market without being rented, the more likely they are to lower their asking rents accordingly. Vacancy rates measure the supply of apartments for rent. Correspondingly, the negative correlation between vacancy rates and rent change is very strong. As vacancy rates go up, rents come down. Here’s a comparison by metropolitan area in Canada.

 

image1

Say you want to ensure average rents for two bedroom apartments are affordable, at about $1,200/mo (again, around the rent level of Edmonton, vacancy rate around 5%). The takeaway from the above would appear to be that if you want to lower rents to this level in a market like Metro Vancouver (average rent @ $1,650, asking rents much higher, vacancy rate around 1%), then you need to ensure that a lot more two bedroom apartments come on the market to rent. In short, you don’t have enough housing.

Exactly how many two bedroom apartments would you need to add to bring average two bedroom rents down to $1,200/mo in Vancouver? This is a tricky (and worthy) question to answer. It would require knowing the shape of the demand curve (made up by knowing how many apartments would be rented at each rent from, say $1/mo to $1 million/mo). It would be difficult to figure this out, even if we could ask everyone in Vancouver what rent they’d be willing to pay for a two bedroom apartment. Why? Two reasons: 1) at lower rent points, some people might be willing to pay for multiple two bedroom apartments (rich people do all kinds of odd things, and when we use price as our metric, the whims of the wealthy matter more than the needs of the poor); 2) we should almost certainly assume that there are a lot of people living outside of Vancouver (including former residents) who would love to move here if they could find a two bedroom apartment for $1,200/mo. They only get a vote in how much housing gets built through their influence on the demand curve. Otherwise they don’t get heard at all. So it’s difficult to tell just how many two bedroom apartments we would need to add to bring Metro Vancouver rents down to $1,200/mo.

Another way to set a metric is to set an ideal vacancy rate instead of a specific rent. Vacancy rate targeting was explicitly mentioned by several candidates in the last City of Vancouver civic election. Inflation-adjusted rents tend to fall when vacancy rates rise above 3%. Setting a vacancy rate target of 4% or 5% will work to deflate rents.

In general, if your goal in asking if a place has enough housing is to preserve the affordability of market housing, then prices (or rental vacancy rates) should be your metric. If prices are higher (or lower) then you want them to be, then you should work to add to (or reduce) the supply of housing accordingly.

But how do we add to the supply of housing? Generally the most important way to add supply is to build more housing. It’s what builders do. But it’s worth noting that if they want to build more housing, builders get stuck in the middle of even more demand and supply curves. Labour, materials, and (most variably) land all influence the costs of constructing new housing. Just like buyers and sellers in the housing market, builders also watch price signals, and they tend to build when they think they can sell the housing they construct for a significantly higher price than they pay to purchase labour, materials, and land, with the difference equal to profit. The Minimum Profitable Production Cost (MPPC), or the minimum cost to bring a new unit to market, sets a hard cap on when builders have any incentive at all to try and add housing. As a result, it also provides a lower bound on the price of new market housing. And this minimum cost rises as density increases and construction becomes more involved and expensive (the minimum profitable production cost of new rental housing in Vancouver is currently too high for market developers to offer new two bedroom apartments at $1,200 market rents). Not surprisingly, holding other characteristics constant, new housing always tends to be more expensive than old housing. As a result, when you compare new housing to old housing, it might seem like new housing is doing nothing at all to bring down prices. But when you consider that building new housing is the primary way of adding more dwellings to the market overall then you get how new housing might “soak up” some of the demand in a given market, thereby lowering the prices of older housing from where they’d otherwise be and bringing down prices overall. Of course, building new housing only adds to the total housing market to the extent that you build more new housing than you demolish, a point to which we’ll return below.

Aside from demolitions, how would one reduce the supply of housing? Generally speaking, we seldom see demolitions exceed new construction, so this doesn’t happen much. But there are a few examples we can talk through, perhaps most prominently AirBnB. In response to new profit-making incentives of AirBnB, many property owners have removed dwellings from the long-term rental market into the short-term, hotel-style market (these markets once weren’t so distinct, but they have become so over time with the passage of laws like BC’s Residential Tenancy Act). As dwellings get removed from the long-term rental market, it drives down vacancy rates and correspondingly drives up asking rents for those units remaining.

What else matters? Location, location, location. Additions and subtractions from the supply of dwellings for sale or rent don’t just have local effects. Their effects spill over into places near and far, tied together by their fixture to land and to transportation networks. For instance, the effects of building and renting out a bunch of new housing in Downtown Vancouver may be felt in asking rents in suburban Surrey. The degree to which additions of housing in one place affect rents in another is heavily dependent upon how long it takes and how much it costs to travel between them as well as to job centres and amenities. That said, some observers suggest that hyper-local “induced demand” may come in to play, meaning that new construction in Downtown Vancouver could potentially drop asking rents in suburban Surrey more than asking rents Downtown. The evidence gathered to date suggests this likely doesn’t happen much, but certainly the scale of the metric matters when thinking about how the addition of new supply affects prices and rents.

So far we’re also talking strictly about dwelling characteristics like bedrooms and size, but not about the structural type of dwellings. We can’t add more single family homes in the inner municipalities in Vancouver, so market mechanisms are constrained in terms of reducing the rent or price when we restrict ourselves to single family homes in the inner municipalities. Being very picky on location can have similar effects. Adding condos or rental properties in the downtown peninsula is more expensive than adding them in e.g. Dunbar. Adding housing in downtown requires concrete high-rise, which is substantially more expensive than 4 or 6 storey low rise which can still add significant housing in Dunbar. Providing amenities like public spaces and libraries for a growing population is also more expensive in areas that are already denser. Given demand and various constraints, it’s quite possible that the market won’t ever be able to supply rental housing at a cost that can push rents down into the $1,200/month range (or push the sale price into the $250,000 range) for a 2 bedroom apartment in Downtown Vancouver. But Surrey seems possible. Regardless, if we want to try we have clear price signals that we’d need to add a lot more 2 bedroom apartments than we have now.

Considered as a class, metrics for Money per Dwelling, including prices (per dwelling, per sq ft, etc.), rents (per BR, etc.), rental vacancy rates, and sales listings, represent transactional data reflecting market pricing mechanisms. Inequality is built into these measures as a reflection of how market allocation weighs the whims of the wealthy of greater importance than the desperate desires of the poor. Correspondingly, reductions in inequality make for more egalitarian housing outcomes. Given market allocation of housing, this is the class of metrics people should turn to if they’re interested in achieving or preserving affordability. They provide the clearest path for identifying if there’s enough (or too much) housing when affordability is the criteria of interest. Of course, these metrics don’t resolve the debate between those who want prices and rents to rise (home sellers and landlords) and those who want them to come down (home buyers and renters), but at least they provide a common empirical grounding.

People per Dwelling (a.k.a. residential crowding)

People per dwelling provides a different class of metrics for thinking about whether there’s enough housing, focused on residential crowding. Fundamentally these metrics ask if there are there enough dwellings to “fit” the number of people we have in a given place. Of course, this is only a potential measure of fit when houses are mostly distributed by the market. Wealthy people probably take up way more room (and rooms) than they need, while poor people more often end up stuffed together. There are two solutions to this situation: one is to ration housing, so that extra rooms are shared around. We see this only for the small proportion of our housing stock that’s non-market housing. Market housing isn’t at all rationed according to need, but instead doled out by wealth-weighted desire (money). The other solution, far more common across North America, is to outlaw too much residential crowding via maximum occupancy codes and sharing rules. This is very common, and in the absence of rationing housing according to need this tends to lead to the exclusion of poor people altogether.

Across most of Canada residential crowding remains low. This is especially true of those places with strong municipal regulations against crowding (e.g. fire codes and occupancy standards) and market distribution of housing. Non-urban, non-market housing, especially on First Nations reserves and in Nunavut, where rationing is more common, tends to be where we see the greatest number of people per dwelling. Here we see a real failure of investment in non-market housing to match occupancy standards observed elsewhere, though differences in family sizes and cultural openness to different rules for living together also play a role.

image2

While crude aggregate crowding metrics can help reveal the lack of housing across reservations and Northern territories, they don’t tell us much about differences between metropolitan areas, which stick together in a relatively narrow range between two to three people per dwelling. The narrow range reflects how crowding is both generally outlawed and also discouraged by market mechanisms distributing the vast majority of housing (above). We also know residential crowding is on the decline in most places, resulting from long-term declines in childbearing, family size, and tolerance for living together combined with the general rise of affluence, occupancy standards and enforcement. Correspondingly, crude aggregate crowding metrics should probably not be used to answer questions about whether metros or municipalities have enough housing. They don’t tell us much.

image-jpg

 

Despite their problematic nature, people per dwelling metrics are commonly used to answer questions for which they’re not suited. Several municipal planners and even a couple of academics have used new persons (or new households) per new dwelling as a metric for whether a place is adding enough housing. Given constraints on crowding and market mechanisms, this is equivalent to asking whether housing supply is meeting demand (as above). Of course it is! By definition, local housing is ALWAYS meeting demand (at some price). Similarly, by definition if you count all of the housed people added and all of the new housing added in a given location, there will always appear to be enough housing added to house everyone (at some level of crowding). After all, only housed people are counted, meaning only the net “winners” able to out-compete others for the dwellings being offered by the market. Net “losers” not provided housing by the market don’t get counted at all! Put differently, if price metrics weigh the whims of the wealthy too high relative to the needs of the poor (a valid critique), then crowding metrics ignore everyone without local housing entirely: all the people who want to live in a place but are prevented from finding housing there don’t get a vote.

Contributing to this fundamental problem, net housing additions are also often poorly counted, either because of changing census methods or failure to combine completions data with demolitions data. This has proven a particular problem for analyses that take for granted how people distribute themselves into households and simply compare new households to new dwellings, taking the leftover number of new dwellings as “empty” excess (in this case, the number of net new housed households can never exceed the number of net new dwellings except in cases where there were previous “empty” dwellings). Given the myriad of problems involved, crude aggregate measures of new persons or new household per new dwelling are especially poor metrics for determining if metro areas or municipalities are building enough. The answer they provide, by default, is practically always “yes.” For similar reasons, reinterpreting past census counts into population projections as the basis for how much housing development to allow is backwards. In high demand places, the availability of housing limits population growth rather than the other way around. Planners and academics should stop using metrics that count only local winners as answers to whether we’re building enough housing.

What about more refined measurements of crowding at different levels of analysis? These are often worthwhile to consider. Given a few strong assumptions about the privacy needs of people while they sleep (practically the least interesting activity they undertake), residential crowding can be measured in terms of bedrooms rather than simply dwellings. Measured at the household level, we can get a sense of how many households are living in dwellings that force more than two people to share a bedroom. We can come up with even more elaborate rules, as in the Canadian National Occupancy Standard, where we assume people need one bedroom per sleeper, but we allow couples to share with each other, and kids to share with other kids (below age 6) and other kids of the same gender (below age 18). Applying these rules more clearly demonstrates the residential crowding on First Nations and in Nunavut. But once again, the metric tells us little about most municipal and metropolitan variation.

We can also refine measures to explore residential sharing at particular ages. When do children leave home? It might be that adult children remaining living with their parents is a sign of need for more dwellings. This is tenuous as an indicator (some children want to stay home, others do not), but interesting!

image3

We can also count individuals without dwellings. This is a form of mismatch. Given the current distribution of housing, how many people are going without? Homeless counts offer an important signal about whether there’s enough housing: if we can count people who are homeless, then there is not enough housing. But this is a broader problem with inequality. Bringing more housing to market may not solve the problem, especially since the demand for housing isn’t just local, and the whims of the wealthy will continue to outweigh the needs of the homeless. Homeless counts are an especially good signal of the need for more non-market housing. Of course, another good signal of the need for non-market housing are the waitlists for cooperativesubsidized and supportive housing. Effectively, both homeless counts and non-market housing waitlists register urgent local needs not being met by the market distribution of housing. That said, homeless counts and waitlists suffer some of the same problems as other crowding metrics insofar as they only tend to record housing need that’s already in a given locale. But people fall in and out of need and they also move. The dire needs of refugees in tent camps tens or thousands of miles away do not get considered, even if those refugees might eventually show up in a municipality. As a result, there remain difficulties in determining just how much need to meet: there are probably no ethically satisfactory stopping points. And even if there were, under rationing systems of all sorts, housing waitlists can grow to enormous lengths. As with attempts to preserve market affordability, we can know we need to build a lot more non-market housing without necessarily knowing when (or if) we should stop.

Finally, returning to the notion of “excess” dwellings, we can also count dwellings without people in them. This is ultimately a bad measure of whether there’s enough housing without a) greater knowledge of the reasons why units appear to be empty and without b) a corresponding will to expropriate “bad” empty units and ration them out according to need. Speaking to the first point, if dwellings register as “vacant” and available to the market (e.g. rental vacancies or unoccupied sales listings), then these dwellings will help reduce prices (see above). If they’re not on the market, they may reflect development processes (pre-demolition or recently constructed dwellings) working toward adding more housing. A variety of other procedural transitions (deaths, inheritances, etc.) may also account for dwellings without people in them before we get to second “vacation” residences (whims of the wealthy, etc.), and alternative uses (AirBnBs, etc.). To the extent these kinds of unoccupied dwellings are rising, they may result in reductions to the market supply of housing, pushing up prices for dwellings that remain. Finally, keeping housing empty and off the market may result from attempts to reduce transaction costs and/or speculatively manipulate market pricing. This is of greatest concern from the standpoint of maintaining market stability and affordability. The diversity of reasons that dwellings might show up as unoccupied means that, by itself, keeping track of unoccupied or empty dwellings is probably a bad measure of whether the market is building enough housing. After all, empty units may be adding to supply or detracting from supply, with varying affects on affordability, depending upon whether they’re on the market. That said, like homeless counts, “empty home” counts can be useful as an indicator of how the market is working to match people to dwellings (given underlying and unmeasured inequality). Moreover, empty homes can be bad in their own right, potentially deadening neighbourhoods. A Lincoln Institute report defines thresholds at which vacancy becomes a problem, with “low” vacancy (a problem for facilitating moves) below 4%, “reasonable” vacancy between 4%-8%, and high vacancies at 8%-20%. “Hypervacancy” (20% or more) poses special problems, especially in the case of declining cities. All major Canadian metro areas fit in the “reasonable range.”

image4

But in high demand cities, lots of empty homes can point toward the desirability of higher property taxes, potentially including Empty Homes Taxes, which can distinguish between types of vacancies and induce owners of empty units and second homes to more quickly return them to market, boosting supply and lowering prices. This will reduce the profitability of any speculative market manipulation. But of course another response to that kind of manipulation is to add more dwellings and credibly promise to keep adding dwellings, placing pressure on prices and rents to lower over time and make speculation unprofitable.

Dwellings per Land (a.k.a. dwelling density)

Dwellings per unit of land as a class of metrics measures dwelling density, constituting yet a different aspect of whether there’s enough (or too much) housing in a given place. This class of metrics has important implications for urban dynamism and environmental impact. It also has potential effects on parking, noise, and the preferred aesthetics of many neighbourhood organizers. Dwellings per unit of land is often measured as dwellings per acre or hectare. Beyond definitional issues, there are tricky aspects to measuring this, insofar as both the areal unit (lot, block, neighbourhood, municipality, metro area) and what gets counted as potential land for dwellings (in the denominator) really matters. If we’re interested in housing density, should one count only land allowing dwellings? What about streets? Or other land uses, like industrial parks? What about recreational parks? Schools? Subtracting out streetscapes makes a big difference, and when other features fall within small areal units, like blocks, they can really affect measures of housing density, making a block with a park look much less than dense than the block next door, even if both are made up of entirely the same kind of housing. Counting only land allowing dwellings constitutes “net housing density” while counting all land and uses constitutes “gross housing density.”

Overall it’s worth noting that this class of metrics is also a bit of a dodge, since often what we’re really interested is people per unit of land, better known as population density. After all more people in a given place constitute more potential interactants in public spaces, more likely transit riders, more shares of infrastructure, and more possible “eyes on the street.” More people also constitute more potential competition for parking and services. People sharing space are also often understood to be poor and potentially dangerous, bringing down property values. So debates over housing density as a class of metrics are often really about how many people should be encouraged or tolerated in a given place. But the regulatory powers of cities are stronger over buildings than bodies, so the focus often ends up being on dwelling density rather than population density. Aside from population density, dwellings per unit of land can have independent effects on the aesthetic “character” of neighbourhoods, as expressed by many peoples’ aversions to high-rises. As noted above, we can, more or less, substitute between population density and housing density just by dividing population density by average household size. This doesn’t always work, insofar as denser housing tends to hold smaller households, but it still gives us a rough translation. We can even figure in unoccupied dwellings if we want, which would give us an overall standard of about 2.34 people per dwelling in Metro Vancover. Alternatively, instead of measuring dwellings per acre, we could measure bedrooms per acre. Bedrooms relate more closely to population than dwellings, and are often similarly regulated by cities.
image5

In terms of impact, housing density (or dwellings per acre) has been linked to urban vitality. Jane Jacobs famously set a few thresholds for what she considered suburban (six or fewer dwellings per acre) and truly urban (one hundred or more dwellings per acre). She considered “in-between densities” as less conducive to the “lively diversity and public life” of the city. Needless to say, the vast majority of the landscape of North American cities fall in Jacobs’ “in-between” ranges, “fit, generally, for nothing but trouble.” Outside of Downtown and a few other scattered census tracts, the same is also true of Metro Vancouver. Where the best threshold for urban vitality might be located remains a matter for debate.

image6

Similar thresholds have been suggested for what kind of densities can support urban transit. Commonly cited thresholds suggest about 12 dwellings per acre around a large central business district is enough to support a decent urban transit system. Guerra & Cervero provide more careful updates on this estimate, exploring capital costs in conjunction with what can be supported by population and jobs located near stations. Using their estimates, a project like Vancouver’s forthcoming skytrain extension along Broadway, at a capital cost of nearly $500/km2 CAD (nearly $600/sq mile USD), would require over 120 people per acre gross population density to support, or more than 50 dwellings per acre near skytrain stations.

Generally speaking, higher dwelling densities enable more transit viability, encourage people to get out of their cars (when coupled with jobs and commercial destinations), promote lower energy useage and support transitions to more sustainable cities. But higher dwelling densities also challenge some peoples’ conceptions of what they want their neighbourhoods to look like and how many people they want to compete with for parking. Moreover, higher dwelling densities tend to be forbidden on the vast majority of North America’s urban land base. Why? Zoning.

Most residential land, including in the City of Vancouver and surrounding suburbs, is zoned to support single-family residential character. At its strictest, single-family zoning insures only one dwelling can be built per lot, and in some cases minimum lot sizes can be enormous. Dwellings are often rationed out according to quite draconian land use rules. Even on the relatively modest 33’ x 122’ standard residential lots that make up a large part of Vancouver’s urban landscape, a single dwelling per lot standard nets only about 10 dwellings per residential acre. Initiatives to add and legalize secondary suites, laneway houses, and most recently duplexes (with secondary suites) means that the actual range of legal dwellings per lot on most single-family zoned land in the City of Vancouver can get all the way up to 40 dwellings per acre. Not bad, but nowhere near the densities supportive of urban vitality or skytrains.

Old-Apartment-Comparison-2018-B

 

On the other hand, a 33’ x 122’ lot located within a commercial zone in Vancouver is allowed greater dwelling density and the ability to build out to lot lines. Even under the same broad height restrictions applied to single-family zoning, twelve dwellings can easily be fit into a given lot while retaining a central courtyard, achieving a dwelling density of about 120 dwellings per residential acre, like this low-rise apartment building in a C-2 (where one of the co-authors of this post lived when he first moved to Vancouver). This moves solidly into Jane Jacobs & heavy transit supportive territory, though the difference between net density and gross density suggests we’re still not quite there yet.

Setting Rules to Metrics

A lot of the metrics we describe above are set into rules (e.g. by-laws, policies, etc.) for regulating cities. In particular: zoning by-laws often set hard limits to dwelling density (dwellings per land) and maximum square footage (Floor Space Ratios) for given lots. The metrics embedded in our zoning effectively mean that we’re rationing out how many dwellings we allow per land parcel. Through the sharing rules embedded in our occupancy standards, we’re also disallowing most residential crowding. But after we apply these rationing and sharing rules to structure housing production and occupancy, we switch to the market in terms of how we develop and distribute most housing. In high demand locations, the net result of these general policies is construction for rich people and the gradual exclusion of poor people. Their dire needs in the market weigh as less important than the whims of the wealthy. Since poor people are also prevented from sharing existing dwellings in high concentrations, they can’t even get a foot in the door, and don’t show up in crowding metrics at all.

While some rules set to metrics are built to be responsive and flexible, automatically adjusting to conditions (e.g. setting rent control to inflation, and setting below-market rates at a set discount from market rates), others require lengthy hearings and political debates to change (changing zoning). As presently configured, debates about dwelling density largely exclude everyone not currently living in our cities. Indeed, this is one reason legislators in places like California and Oregon have moved to erode the power of municipalities to exclude development near transit hubs. They want to give potential renters and buyers a bigger say in whether we have enough housing by allowing them to speak through the demand curve, encouraging developers to build more housing in these places. To date the political process hasn’t let them get away with much, which ironically insures that developers profit hansomely from the scarcity of new housing being added to the market. In a high demand place like Vancouver, this means that in the long term, rents and prices tend to just keep going higher (though as we’re learning, in the short term prices can still swing up and down in line with speculative booms and busts, just like anywhere else!)

If we’re concerned about the exclusionary effects of high prices, we could reform our zoning regulations to be responsive, automatically adjusting to both transit development and market conditions (just like with rent control or the setting of below-market rents). There seems to be a lot of potential in considering this possibility. One example would be to set affordability thresholds. We could, for instance, automatically enable a rise in the number of dwellings permitted on a lot equal to one for every $250,000 in its assessed value. Once a lot hits three million in value, we could automatically enable up to twelve dwellings, looking something like the building above. Thresholds for non-market housing could be set even lower, enabling non-market developers (including the City) a competitive advantage in securing lots. Cities could also take over the production of non-market dwellings themselves, purchasing low-density lots and using their power over zoning to upzone and redevelop for the higher densities needed to support a more economically diverse population.

Conclusion (and Preview)

Overall, there’s still lots to think through when asking if we have enough housing! But metrics can establish crucial common ground for providing answers. Stripping down our metrics to their basics helps demonstrate their utility in terms of what answers they can provide and who they give voice. Overall, price (and rent) metrics provide the best indicators of whether we have enough housing to preserve or achieve market affordability. Non-market waitlists and homeless counts provide the best indicators of local non-market housing need (though they still exclude need from elsewhere). By contrast, residential crowding metrics (people per dwelling) don’t generally tell us much in urbanized Canada, and tend to privilege the voices of those already living in a place (e.g. the “winners” in finding housing). Dwelling per land metrics point toward the limits often imposed upon getting to enough housing in a place, and potentially spell out the rewards for getting there in terms of sustainability and urban vitality.

In terms of underlying logics, the market distribution of housing tracked by price metrics is problematic insofar as the whims of the wealthy far outweigh the dire needs of the poor. But when we simply wave away price metrics, and pretend we’re rationing out housing by need instead (by only tracking persons per dwelling), then we’re saying we don’t care who wins for the limited amount of housing we’re willing to offer when we ration out dwellings to land. Really addressing housing need is a monumental and important task, and requires a much greater investment in non-market housing. But questions quickly arise as to how non-market housing should be rationed, and advocates should pay more attention to providing answers that don’t assume that no one ever moves.

In future posts on housing metrics, we’ll compare across specific measurements within the same class and dig further into more complicated metrics that combine multiple classes (e.g. price to income multiples, core housing needs, shelter & transportation cost to income rations, etc.) So consider this a preview. Rest assured we’ll keep playing around with metrics!

..