Every fall, Maricopa County publishes its heat surveillance report, and the local news covers it. 645 deaths in 2023, a record. 608 in 2024, a crisis that wouldn't quit. 430 in 2025, which got read as progress or as a cooler summer depending on who was talking.
Each number got its news cycle. Each news cycle treated it as the property of one particular summer, explained by that year's weather, that year's outreach money, that year's politics. Then the story ran its course and everybody waited for next year's figure.
So I read the three reports together. The totals moved. Not much else did.
What the sky did
Temperature figures here come from NWS Phoenix annual reviews for 2023, 2024, and 2025, measured at Sky Harbor Airport.
| Measure | 2023 | 2024 | 2025 |
|---|---|---|---|
| Days at or above 110°F | 55 | 70 | 37 |
| Nights at or above 90°F | 35 | 39 | 23 |
| Longest consecutive run above 110°F | 31 days | 12 days | 14 days |
| Heat-associated deaths | 645 | 608 | 430 |
Weather and deaths don't line up in a straight line. 2024 was hotter by most counts than 2023 and killed 37 fewer people. Maybe the 31-day unbroken streak in 2023 did more damage than a larger pile of hot days split into shorter runs. Maybe something somebody tried actually worked. The county reports don't assign the difference to any cause, and I'm not going to do it for them.
The 2024-to-2025 drop reads more simply: 33 fewer days above 110, 16 fewer nights above 90, 178 fewer deaths. A milder summer killed fewer people.
What happened underneath the totals is the part worth your time.
The profile that wouldn't move
These tables come from the appendix data in all three surveillance reports. Where a report disagrees with itself — the 2024 edition's housing-type graphic shows percentages that don't match its own appendix tables — I've used the appendix counts, since those add up to the reported totals.
Where people died
| 2023 | 2024 | 2025 | |
|---|---|---|---|
| Indoor | 156 (24%) | 138 (23%) | 111 (26%) |
| Outdoor | 480 (74%) | 467 (77%) | 313 (73%) |
| Unknown | 9 (1%) | 3 (0.5%) | 6 (1%) |
The total dropped by a third between 2023 and 2025. The indoor/outdoor split held within a few points. The small "unknown" group in each year accounts for the gap between the indoor-plus-outdoor counts and the annual total.
Among those who died indoors
| 2023 | 2024 | 2025 | |
|---|---|---|---|
| In an RV, trailer, or mobile home | 38 (24%) | 36 (26%) | 25 (23%) |
| AC unit present in the residence | 88% | 88% | 94% |
| AC not functioning (where present) | 85% | 70% | 72% |
| Aged 50 or older | 95% | 89% | 90% |
RVs, trailers, and mobile homes accounted for between 23 and 26 percent of indoor deaths every single year. Both the 2024 and 2025 reports point out that those housing types make up roughly 5% of county structures. Five-to-one overrepresentation, three years running.
Then the air conditioning. Nearly every one of these residences had a unit in it. In 70 to 85 percent of those cases the unit wasn't running. The equipment was on the premises and did nothing. A county health department has now documented that same finding three years in a row, and none of the three reports says why the units weren't working — whether the compressor was dead, the landlord hadn't come, the power was shut off, or the resident couldn't afford to run it. The surveillance system records that the cooling failed. It does not record how, which means it also can't tell anyone which of those failures to go fix.
Who died
| 2023 | 2024 | 2025 | |
|---|---|---|---|
| Male | 78% | 78% | 74% |
| Aged 50 or older | 392 (61%) | 359 (59%) | 253 (59%) |
| Experiencing homelessness | 291 (45%) | 299 (49%) | 208 (48%) |
| Cardiopulmonary comorbidity | 46% | 46% | 47% |
Unhoused people made up somewhere between 45 and 49 percent of the dead in all three years. Their share went up a little while the total came down. The heart-and-lung comorbidity rate went 46, 46, 47.
The reports don't break deaths out by neighborhood, zip code, or heat zone. Whether the same stretches of the county produce the same share of bodies year after year is a question the published surveillance simply cannot answer.
Substance involvement, with a necessary disclosure
The original 2023 report put 65% of deaths in the substance-involved column, counting any mention of drugs or alcohol on the death certificate. The 2024 report changed the definition to a specific list of diagnostic codes for overdose or acute poisoning, and told readers in plain language not to compare its number against earlier ones. The 2024 and 2025 appendices then went back and applied the new definition to 2023, which gives us a series that can actually be compared:
| 2023 (revised) | 2024 | 2025 | |
|---|---|---|---|
| Substance-involved | 363 (56%) | 349 (57%) | 236 (55%) |
| Stimulants (of substance-involved) | 321 (88%) | 309 (89%) | 194 (82%) |
Anybody who set the original 65% next to the following year's 57% and concluded something had improved was comparing two different measuring sticks. Measured the same way both years, the rate sat still. This is the sort of seam that one season's coverage sails right past and that reading the sequence catches.
One genuine shift
Indoor deaths discovered when somebody went to check on the person: 76% in 2023, 67% in 2024, 48% in 2025. Of everything in these tables, that's the number that actually moved. It might mean outreach is reaching people before they're in trouble. It might mean the population dying indoors has changed. It might be a difference in how the cases got classified. The reports don't explain it and I haven't found any coverage that asks. I'm flagging it because it stands out against a background of everything else holding still.
What three years of stability means
One year's report gives you a snapshot. Three give you a profile that barely twitches: the ratio of indoor to outdoor, the manufactured-housing overrepresentation, the dead air conditioners, the share who were homeless, the substance rate, the ages, the comorbidities.
A milder 2025 killed the same kinds of people in the same kinds of places. Fewer of them.
That doesn't prove the interventions did nothing. The 2023-to-2024 comparison, where the toll dropped a bit despite equal or worse heat, leaves honest room for outreach or cooling access having pulled some deaths off the board. But the 2025 figure can't be read as evidence that any institution learned anything until you've accounted for the weather first. And a death profile that holds this steady across three summers of very different severity suggests that whatever is being done is working around the edges of conditions that haven't changed.
Philadelphia: the same fifty-one schools
Maricopa's data tracks who dies. Philadelphia's school records track something else entirely — which buildings can't operate in hot weather — but the longitudinal question is identical. Does the list change, or does it just get re-announced?
The School District keeps a roster of buildings without adequate cooling. When temperatures spike, those schools dismiss early or go remote. Each event gets covered as weather.
| Season | What happened |
|---|---|
| 2023 | 90 schools early-dismissed June 2; 74 schools early-dismissed Sept. 5-6, expanded to 86 for Sept. 7-8 |
| 2024 | 63 schools early-dismissed Aug. 27-28 |
| 2025 | No heat-dismissal event located in public record; district reported 54 schools with inadequate cooling at opening |
| 2026 | 57 schools shifted to remote May 20; 52 schools on heat-dismissal schedule in June and September |
The September 2026 list ran 52 schools. I checked those names against the published rosters from 2023 and 2024.
Every one of the 52 had appeared on at least one 2023 heat-dismissal list. Fifty-one of the 52 appeared on both a 2023 list and the August 2024 list. The lone exception, Jay Cooke, was on the June 2023 list but not the one from August 2024.
Fifty-one schools, three years, same list.
The standing count of inadequately cooled buildings has come down: 63 in August 2024, 54 when school opened in August 2025, 52 in September 2026. Eleven buildings in two years. The district had set a goal of cooling every building by 2027. In August 2026 a spokesperson told Chalkbeat that the plan had changed, and cooling work would now be folded into larger modernization projects spread over the coming decade.
In April 2026 the Board approved a 10-year, $3 billion facilities plan covering 169 campus modernizations, with HVAC in there alongside new construction, accessibility, bathrooms, lighting, and kitchens. The district committed $1.1 billion from its current borrowing and said it would go looking for the other $1.9 billion. Maintenance officials told WHYY that the electrical retrofits most of the remaining schools need run $1 million to $1.5 million a building. So cooling the last 52 is a $52-to-$78 million job sitting inside a $3 billion plan with $1.9 billion in unfunded asks and no published date by which the dismissal roster hits zero. I could find no public document that reconciles this school by school — which cooling projects got finished, what each cost, and whether finishing one took that school off the next summer's list.
I wrote about these dismissals earlier this year, working from the September 2023 count of 86 schools and the 2027 deadline that was still on the books then. One season's data can tell you a protocol exists. Three seasons of names tell you the protocol governs a fixed set of buildings, and that calling it an emergency response has become a figure of speech.
What the data can't tell you
Maricopa's Heat Relief Network puts out seasonal maps of cooling locations — more than 200 in 2023, more than 124 in 2024, more than 200 in 2025 — but those counts lump together cooling centers, hydration stations, shade structures, and donation drop-offs, so the year-to-year comparison doesn't mean much. There's utilization data for county-funded partners in 2023 and for Phoenix's extended-hour sites in 2024 and 2025, but they cover different things, so no regional trend line can be drawn. Whether cooling capacity grew, shrank, or held steady while the death totals swung is not answerable from what's been published.
And the Sky Harbor temperature record measures an airport. Not the blocks where people die, not the inside of a trailer with a dead compressor, not the sidewalk where somebody goes down. Station data is what we've got. It isn't what kills anybody.
What repetition reveals
Both datasets lead to the same question. If a summer as brutal as 2023 shows up in 2027, does Maricopa County bury 645 people again? Are the same 51 Philadelphia schools still sending kids home at noon?
Nobody has published the analysis that would settle the first one. A death profile that holds this steady across three summers points toward conditions still fully in place, but the 2023-to-2024 numbers keep the question genuinely open. Each report describes its own year. No one is under any obligation to describe the sequence.
The Philadelphia question is easier, because the district answered it. It moved the cooling deadline from 2027 out into the next decade. Absent some surprise, those schools will be on the list.
In both cities the annual cycle works the same way. A number lands. It gets set against last year's number, and the comparison produces a story: worse, better, about the same. Then the number gets filed and everybody waits for the next summer to make another one.
Read in sequence, though, the numbers show the same profile repeating across three summers. Who is exposed, in what housing, with what broken equipment, with what bad heart and lungs, found by what means — all of it has been sitting in public documents for three years, and the comparison takes nothing more than a table and some arithmetic.]
-
Indoor deaths, nonfunctioning AC: Maricopa's reports document that air conditioning was present but broken in 70–85% of indoor heat deaths across three years, but a CDC cooling-center evaluation found that 31% of visitors cited transportation and 14% cited inaccessible hours as barriers, suggesting that for many indoor decedents the operative failure may not have been "go to a cooling center" but private-room equipment breakdown, immobility, or cost.
-
Houston's new cooling ordinance: Houston eliminated a rental exemption allowing window screens instead of refrigerated air in August 2026, but the enforcement pathway begins with a tenant-generated 311 complaint and no cooling-specific compliance dashboard, response-time distribution, or observed room-temperature outcome has been published.
-
Heat dismissals and household cost: Earlier WHYY reporting at Spring Garden School documented parents leaving work, paying for transport, and arranging supervision during 2023 heat dismissals, but the district has not published a 2026 estimate of household work loss, transportation expense, or replacement-care cost created by early-dismissal events.
-
Cooling-center utilization gaps: The Maricopa region's Heat Relief Network publishes seasonal maps mixing cooling centers, hydration stations, and donation sites, but available utilization data covers different program subsets each year, so whether the network's actual cooling capacity grew, shrank, or held steady alongside falling death totals remains unanswerable from the public record.