Toronto Renovations: A data story
I developed a lot of respect for the work of data scientists last week, when I did a little project involving a data set of building permits for the city of Toronto. Like any good research project, it raised more questions than it answered, but I will record some of what I did learn here.
In broad strokes, I was interested in answering the following questions:
1. How has the rate and price of home renovation in Toronto changed over time?
2. How has the incidence of renovation by location in the city changed over time?
3. Has there been a change in the types of renovations being done?
The source of data for the investigation was primarily from the City of Toronto’s open data repository. In particular, they have data on all closed building permits over the last 20 years. I paired this with geographic and census data from Statistics Canada, and with some data about housing prices from the Canadian Mortgage and Housing Corporation.
The building permit data set has been nicely explored in a “Data Stories” project by the Open Data staff. I used their work on compiling and cleaning the data to create a data set to use as a starting point for my research (thank you!). On to the research questions!
Exploring the data
The first step was to spend some time exploring the data. Building permit data includes all types of building permits, from large commercial developments to individual homeowners who just need a permit to re-do the plumbing in their bathroom. To narrow the focus to residential projects, I selected only permits with the following permit types: ‘Small Residential Projects’, ‘Building Additions/Alterations’ and ‘Residential Building Permit’. Note that the ‘Plumbing’ and ‘Mechanical’ permits are typically submitted alongside another more general permit, so I omitted those from this study.
Plumbing(PS) 0.239137
Mechanical(MS) 0.197430
Small Residential Projects 0.156142
Building Additions/Alterations 0.112459
Drain and Site Service 0.091954
Residential Building Permit 0.044277
New Houses 0.040462
Demolition Folder (DM) 0.027335
Fire/Security Upgrade 0.021593
Designated Structures 0.014837
Building Historical data - Converted 0.011777
Non-Residential Building Permit 0.011045
Conditional Permit 0.008198
Temporary Structures 0.007961
Partial Permit 0.006045
Site Inspection(Scarborough) 0.002871
New Building 0.002802
AS Alternative Solution 0.001800
Multiple Use Permit 0.001036
Portable Classrooms 0.000830
Change of Use Permit 0.000010The next observation is that the description of the types of work included in the ‘WORK’ field are not super helpful. Some are highly specific (e.g. “Green Roof”, but the most highly represented categories — “Interior Alterations” and “Multiple Projects” — could capture almost anything.
Interior Alterations 0.431744
Multiple Projects 0.216087
Addition(s) 0.068051
Garage 0.044192
Pool Fence Enclosure 0.041837
Other(SR) 0.030293
Deck 0.026671
Other(BA) 0.024364
Porch 0.021057
Underpinning 0.017863
Second Suite (New) 0.009361
Accessory Building(s) 0.009351
Garage Repair/Reconstruction 0.008703
Party Wall Admin Permits 0.007959
Finishing Basements 0.007254
Balcony/Guard Repairs 0.005948
Walk-Out Stair 0.005872
Carport 0.004661
Canopy 0.004280
MGO Remediation 0.003689
Re-Roofing/Re-Cladding 0.003603
Change of Use 0.001506
Canopy w/o Enclosure 0.001172
Solar Domestic Hot Water (Res) 0.001058
New Building 0.000705
Shoring 0.000562
Interior Demolition 0.000419
Window Replacement 0.000381
Fire Damage 0.000286
Fireplace/Wood Stoves 0.000229
Green Roof 0.000181
Unknown 0.000181
Demolition 0.000143
Addition 0.000095
Addition to Existing Building 0.000057
MGO 565 Remediation 0.000038
Garage/Porch/Balcony 0.000029
Alteration to Existing Building 0.000029
Install/Alter Plumbing - only 0.000019
Install/Alter HVAC - only 0.000019
Other(PS) 0.000010
Other Proposal 0.000010
Addition/Alteration to Existing Building 0.000010
Other(DS) 0.000010
Partial Permit - Foundation 0.000010I chose not to further filter this data, although excluding some of the categories might be a topic of further analyses.
1. How has the rate and price of home renovation in Toronto changed over time?
One of the fields of interest is the ‘Estimated Construction Cost’. Of course this is only an estimate, but my hope is that it will give a rough proxy for the cost of renovations, and how that has changed over time. Here is where it gets a little interesting, because we have a choice for how we measure the date of each permit. We can choose to look at when the permit application was received, or we can look at the date the permit was closed. This choice will paint two very different pictures.
In words, if we look at the date that the permits are closed, it would suggest an increasing pattern over time. In contrast, the number of building permit applications seems to have remained relatively stable over time. What is the more faithful indicator of the rates of renovation? And why is there this discrepancy?
What about the cost? The following shows the comparison of the cost of renovation (adjusted for inflation using the easymoney package) compared to the cost of a new home (from the CMHC housing data).
Here again, we see a difference depending on how we measure the permits (application vs. closure date). Mapping by the permit closure date we observe an upward trend in the estimated construction cost. Measuring by application date suggests that the estimated construction cost by project has remained level. In both cases, the price of renovation has remained relatively stable, while the cost of a new home in Toronto has increased dramatically to over a million dollars.
Conclusion: This does provide evidence for the observation that many homeowners in Toronto would prefer to renovate their current home to better suit their needs, rather than more to a new home. Simply put, it is often more economical to renovate a home than to consider purchasing a new home, given the fixed costs associated with high value real estate transactions.
2. How has the incidence of renovation by location in the city changed over time?
Attempting to answer this question really put my data science (aka googling for the right Medium article) skills to the test. Each building permit has an address associated with it. However, visualizing these on an actual map is not so easy. The process of converting a written address into latitude and longitude is known as geocoding, and in short, it isn’t cheap. It is possible to do a limited amount of geocoding through Open Street Maps for instance (via the geopy project), but for the tens of thousands of addresses I was hoping to map this wasn’t an option (at least not an easy, quick or cheap option).
Because I wanted to group the renovations by regions of the city, I opted to use the Forward Sortation Area (FSA) field of the permit data. That’s a fancy way of saying the first three digits of the postal code. The good thing about this approach is that this information was already available in the data. Unfortunately, the forward sortation areas do not necessarily correspond to neighbourhoods, or even split the city according to standard landmarks. They are also larger than what we typically think of as neighbourhoods (plus I identify a lot more with “Little Portugal” than I do with “M6J”!). Still, with a little help from matplotlib and this article I was able to generate this choropleth map of permits over time:
But what does this tell us, really? There are a lot of limitations to this analysis. The actual numbers get much smaller (between roughly 30 and 100 renovations per year per FSA), and therefore I get suspicious about their ability to reliably tell a story. In addition, the number of dwellings in a particular area may have changed over time, which potentially could make the number of renovations per area misleading. Trying to normalize by FSA population seems challenging because the census data is only available in 5 year intervals, so I stuck with absolute counts. The map above uses dates that the permits were cleared, which indicates a nice increasing trend. But see below for the same map sorted by the date of permit application. It doesn’t seem to tell the same story.
Conclusion: By focusing on a particular FSA in either map it is possible to consider a trend over time. But I don’t see a cohesive story emerging about renovations in Toronto.
Further Questions, Data and Data Science
I didn’t get around to trying to answer my final research question about the type of renovations being done, and what patterns can be observed. Some initial analysis of this type was done by the Toronto Open Data staff (for example they note an increase in drain permits after a summer of heavy rain in 2013). However, as mentioned, the permit data set lacks detailed descriptions of what work is done, which may limit this inquiry.
To conclude, it’s possible that I learned more about doing data science than I did about the actual research questions I set out to answer. I also learned that there are a lot of choices that are made by data scientists about how to select, prepare, analyze and present data. There is a lot of expertise involved and required for making these decisions wisely. In the right hands these are powerful tools, but it also seems like a fragile workflow in which it is easy to make poor decisions accidentally, or even with the intention of misleading.
