Jiakang Zhang
Housing & Urban Data Analyst · MA Sociology, University of Guelph (2026)
Kitchener, Ontario · jiakangz.0907@gmail.com · 647-740-9412 · zhangjiakang.pro · linkedin.com/in/jz0907
Six independent, reproducible studies on public data, plus an award-winning MA thesis.
Each study is three committed Python scripts (fetch, analyze, chart); every number below
traces to saved model output. Full plain-language write-ups, technical notes, estimate
tables, and academic references for each study are at zhangjiakang.pro/projects.html.
Limits are stated up front throughout, and each study documents a real data problem
caught and fixed before shipping.
01 · Is high rent bad for your health?
Health & housing · survey-weighted logistic regression · Canadian Housing Survey 2022 (14,110 renter households)
Fair-or-poor mental health climbs from 23% to 37% as the share of income going to shelter rises. Adjusted with Statistics Canada's 1,000 bootstrap replicate weights, the ratio itself stops predicting mental health; what survives is hardship.
Missed or delayed a housing payment: 2.3× the odds of poor mental health, after adjustment for income and other measured characteristics (95% CI 1.7 to 3.1).
Cross-sectional associations, stated as such. Exposure variable switched after the questionnaire showed the first candidate measured general money trouble.
02 · Did the 2023 refugee crisis change Toronto's shelter numbers?
Program evaluation · controlled interrupted time series · Toronto Shelter System Flow + IRCC asylum claims, 54 months
Refugee counts rose before the crisis and peaked in 2024. The August 2023 contrast is +21% relative to non-refugees (revised 95% CI about +4% to +41%). This is a difference at the break; later trend terms change it. The refugees-only level estimate is near zero, making the comparison an important assumption.
Linear post-trend rejected for visible underfit; published model uses a curve, R² = 0.97. Fixed snapshot through June 2026; a refresh workflow is prepared.
Newey-West errors; identification limits stated on the page.
03 · Are apartments worse in poorer neighbourhoods?
Housing quality & program evaluation · descriptive + OLS · RentSafeTO audits (3,452 buildings) + census incomes
The expected income gradient is nearly flat: audit scores average about 89 in every income third (a doubling of income is worth +2.4 points, adjusted). The clearest findings are citywide heat risk and program effects over time.
84 in 100 audited buildings provide no air conditioning, in rich and poor neighbourhoods alike. Under the pre-2023 scheme, citywide averages climbed 65.5 to 80.9 in six years; 93% of re-inspected buildings improved, with gains largest exactly where enforcement pressed (regression to the mean stated).
A punctuation mismatch initially dropped 239 buildings from the income join; normalized keys matched every one.
04 · What does Airbnb's 28-day loophole rent for?
Short-term rentals · hedonic regression, fixed effects · Inside Airbnb (9,109 listings) + CMHC benchmarks
58.7% of entire-home listings sit past the 28-night line where Toronto's by-law stops. Like-for-like, they price 53% below short stays (95% CI 51 to 55) in every distance ring, which still implies far more than ordinary rent.
Median loophole one-bedroom: $3,671 a month, versus $2,365 for an average rental condo (CMHC, Oct 2025). Extends the thesis finding that unregulated listings carried 2.5× the rent impact.
Listed prices, never transactions; benchmark extracted programmatically from CMHC's own data tables.
05 · How long do Toronto building applications wait?
Urban planning · current-status survival + logistic regression · 25,944 development applications
The tracker records no decision dates, so the design follows what the data honestly supports: each year's share still open. The open queue holds 5,065 files, median age 4.1 years; half of everything submitted since 2022 is still open. Rezonings have lower adjusted odds of being open than site plans at the same age, but 54.6% of open rezonings sit at the appeals tribunal (vs 8.7% of site plans).
Best bug story on the site: hidden trailing whitespace silently dropped 3,799 records and flipped a key odds ratio; a 98% figure failed the smell test and exposed it.
"Closed" bundles approvals, refusals, and withdrawals; no approval rate is claimed.
06 · What did the pandemic do to opioid deaths?
Public health · single-series interrupted time series · PHAC / Health Infobase, 40 quarters
Deaths jumped when the pandemic hit (immediate shift bracketed +54% to +85% across functional forms, both p < .001), peaked at 2,082 in late 2021, and stand a third below peak in late 2025. Ontario's rate has halved from its peak; B.C.'s remains about four times as high.
A nationwide shock has no comparison group; the page says so and contrasts the design with study 02. The post-peak trend (−2.6%/quarter, p = .063) is reported as not quite clearing significance.
Quadratic-form sensitivity published on the page; recent quarters flagged as revisable.
MA thesis · Does Airbnb make rent more expensive?
Econometrics · instrumental variables (2SLS) · original 10-source dataset, 6,165 rental listings · University of Guelph
Citywide, entire-home Airbnb activity is linked to rents about 5.6% higher than they would otherwise be; listings that sidestep the by-law carry about 2.5× the overall per-listing estimate. Awarded for "original dataset creation, complex data analysis methods, and clear communication of findings" (Methodological Excellence Award, 2026).
Full story in eight chapters at zhangjiakang.pro/research.html; one-page brief available.