Left in the Cold

Toronto Shelter System — people unplaced at 4 a.m. on freezing nights, and how we get to zero

On 95.9% of nights below −5 °C across six Toronto winters, people were left unplaced at 4 a.m. The overnight shelter queue tripled, then fell — proof that zero is a budget choice the city makes each summer.

People are being left in the cold — and we can count them

Toronto’s shelter system keeps a number every night: how many people asked for a safe place to sleep and were still without one at 4 a.m., when Central Intake closes its overnight count. On warm nights, that number is a logistics problem. On freezing nights, it is people outside in weather the city itself has declared dangerous.

The City of Toronto’s own policy recognizes temperatures at −5 °C as cold enough to trigger emergency action — a threshold that is effectively a matter of life and death. Yet our data reveals a stark, repeated failure: on 95.9% of nights hitting this limit over the last six winters, people were still left unplaced at 4 a.m. This is not a weather event; it is a choice to expose residents to severe harm. We know this is fixable: last winter proved that when the system has room to breathe, we can reach zero. This winter, it is a choice to do it again.

The chart below shows every one of those nights.

seasonRug = FileAttachment("../data/season_rug.csv").csv().then(rows =>
  rows.map(d => ({
    min_temp:      +d.min_temp,
    winter_season: d.winter_season
  }))
)
binnedSC = FileAttachment("../data/cold_response_binned_by_winter_sc.csv").csv()
  .then(rows => rows.map(d => ({
    min_temp_bin:  +d.min_temp_bin,
    winter_season: d.winter_season,
    mean_unplaced: +d.mean_unplaced,
    n:             +d.n
  })))

// ── Theme-aware palette ──────────────────────────────────────────────────────
cs      = getComputedStyle(document.body)
refLine = cs.getPropertyValue("--bq-chart-stroke").trim()

// ── Season color map ─────────────────────────────────────────────────────────
// Explicit domain+range so the rug and dots share identical season → colour
// bindings across any chart added to this page.
seasonColorMap = new Map([
  ["2020-21", "#4DF884"],
  ["2021-22", "#2F9DF5"],
  ["2022-23", "#DEDD32"],
  ["2023-24", "#E63946"],
  ["2024-25", "#F65F18"],
  ["2025-26", "#900C00"]
])

viewof selectedWinters = {
  const allSeasons = [
    "2020-21","2021-22","2022-23","2023-24","2024-25","2025-26"
  ];
  const selected = new Set(allSeasons);

  // Inject custom checkbox styles once — guarded so OJS re-runs don't duplicate
  if (!document.getElementById("bq-season-cb-style")) {
    const s = document.createElement("style");
    s.id = "bq-season-cb-style";
    s.textContent = `
      .bq-season-cb {
        appearance: none;
        -webkit-appearance: none;
        width: 14px;
        height: 14px;
        border-radius: 2px;
        margin: 0;
        border: 1.5px solid #888;
        cursor: pointer;
        display: inline-flex;
        align-items: center;
        justify-content: center;
        flex-shrink: 0;
        background: transparent;
      }
      .bq-season-cb:checked {
        background-color: var(--season-color);
        border-color: var(--season-color);
      }
      .bq-season-cb:checked::after {
        content: "✓";
        color: #fff;
        font-size: 10px;
        font-weight: 700;
        line-height: 1;
      }
      .bq-season-cb:hover {
        transform: scale(1.15);
        box-shadow: 0 0 0 2px rgba(0,0,0,0.15);
      }
      .bq-season-lbl {
        display: inline-flex;
        align-items: center;
        gap: 8px;
        cursor: pointer;
        user-select: none;
      }
    `;
    document.head.appendChild(s);
  }

  const root = document.createElement("div");
  root.style.cssText = "font-family:var(--sans-serif,sans-serif);font-size:13.5px;";

  const labelEl = document.createElement("div");
  labelEl.textContent = "Seasons shown";
  labelEl.style.cssText = "font-weight:600;margin-bottom:3px;";

  const hint = document.createElement("div");
  hint.textContent = "Click a season to show or hide it";
  hint.style.cssText =
    "font-size:11.5px;font-weight:400;color:var(--bq-fg-faint);margin-bottom:6px;";

  const row = document.createElement("div");
  row.style.cssText =
    "display:flex;flex-wrap:wrap;gap:6px 20px;align-items:center;";

  for (const season of allSeasons) {
    const color = seasonColorMap.get(season);

    const cb = document.createElement("input");
    cb.type = "checkbox"; cb.value = season; cb.checked = true;
    cb.className = "bq-season-cb";
    cb.style.setProperty("--season-color", color);

    const txt = document.createElement("span");
    txt.textContent = season;

    const lbl = document.createElement("label");
    lbl.className = "bq-season-lbl";
    lbl.append(cb, txt);
    row.append(lbl);

    cb.addEventListener("change", () => {
      if (cb.checked) selected.add(season); else selected.delete(season);
      root.value = [...selected];
      root.dispatchEvent(new Event("input"));
    });
  }

  root.append(labelEl, hint, row);
  root.value = allSeasons;
  return root;
}

scSeasons = selectedWinters

Note

Reading the chart. The chart shows data from up to six cold seasons (November 15 to April 15), colour-coded by winter year. The vertical lines across the background act as a temperature rug — each line marks one night’s recorded overnight minimum temperature, coloured to match its season. Where these lines cluster together, data is plentiful; where they thin out (below about −15 °C), individual nights carry more weight. Use the checkboxes above to show or hide individual seasons.

{
  const visibleBinned = binnedSC.filter(d => scSeasons.includes(d.winter_season))
  const visibleRug    = seasonRug.filter(d => scSeasons.includes(d.winter_season))

  return Plot.plot({
    width:        720,
    height:       400,
    marginLeft:   70,
    marginRight:  20,
    marginTop:    20,
    marginBottom: 50,
    color: {
      type:   "ordinal",
      legend: true,
      label:  "Winter season",
      domain: [...seasonColorMap.keys()],
      range:  [...seasonColorMap.values()]
    },
    r: { range: [2, 8] },
    x: {
      label: "Overnight minimum temperature (°C)",
      labelAnchor: "center",
      reverse: true
    },
    y: {
      label: "Mean people unplaced (singles/couples, observed)",
      labelAnchor: "top",
      grid: true
    },
    marks: [
      Plot.tickX(visibleRug, {
        x: "min_temp",
        stroke: "winter_season",
        frameAnchor: "bottom",
        strokeOpacity: 0.35,
        strokeWidth: 1
      }),
      Plot.ruleX([0, -5, -10, -15], {
        stroke: refLine,
        strokeWidth: 1,
        strokeDasharray: "4,3"
      }),
      Plot.dot(visibleBinned, {
        x: "min_temp_bin",
        y: "mean_unplaced",
        fill: "winter_season",
        r: "n"
      }),
      Plot.tip(visibleBinned, Plot.pointer({
        x: "min_temp_bin",
        y: "mean_unplaced",
        title: d =>
          `${d.winter_season}\n` +
          `${d.min_temp_bin} °C bin  ·  ${d.n} nights\n` +
          `Observed mean: ${d.mean_unplaced.toFixed(1)} unplaced`
      }))
    ]
  })
}

Figure 1: Observed mean unplaced people (singles and couples) at 4 a.m. by overnight minimum temperature and cold season, binned in 2 °C intervals. Bins are labelled by their midpoint (e.g., the −3 °C dot covers −4 °C up to but not including −2 °C). Each dot represents one temperature bin for one winter; dot size scales with the number of nights in that bin (larger = more nights). The 2025–26 season (bottom-right cluster) sits visibly lower than earlier winters at comparable temperatures.

This is a choice, and the choice is being made right now

The number is not fixed. As recently as the winters of 2020–21 and 2021–22, the overnight queue averaged about 35 people a night. By 2022–23 it had tripled, to over a hundred, and stayed there for three winters. Then, this past winter, it fell back to roughly 34.

Table 1. Average number of people (singles and couples) left unplaced at 4 a.m., December through February, Toronto shelter system (2020–21 through 2025–26).

Cold season	Mean unplaced per night
2020–21	37.6
2021–22	34.7
2022–23	100.3
2023–24	155.5
2024–25	101.4
2025–26	34.2

Note. Values are nightly means of people (singles and couples) still unmatched to a bed at the 4 a.m. Central Intake closeout, averaged across December, January, and February. The increase between 2021–22 and 2022–23 reflects a structural change in the series; the two periods should not be averaged together. Source: Toronto Central Intake (SMIS), BonQuery analysis.

So what was different in the winters when the number was low? Not the cold — those winters were no milder. And not emergency warming centres, which were at their largest ever this past winter. The difference was that the regular shelter system had room to breathe: it ran near 97% full, instead of jammed against 99% or more as in the worst years. When the everyday system isn’t already at capacity, a cold snap doesn’t have to end with people on the street.

That is the whole lever. And it is pulled in the summer — now — when the fall budget decides how many beds will exist on the first freezing night of November. By the time the cold arrives, the decision has already been made.

Zero is the only acceptable number of people left outside below freezing in a city as wealthy as Toronto. Reaching it next winter does not require a new program or a breakthrough. It requires deciding, in this summer’s budget, to fund enough standing shelter capacity that the system carries the same breathing room it had last winter — before the first freezing night, not after. The city has already shown it can get the number down. The only question left is whether it will choose to.

---

Data: Toronto Central Intake call records (SMIS), six cold seasons November 15–April 15, 2020–21 through 2025–26. Analysis by BonQuery. Weather: Environment and Climate Change Canada hourly records, Toronto station, retrieved via the weathercan R package.

Contains information licensed under the Open Government Licence – Toronto.