Patricia Miletich was 70 and lived in a retirement RV community in El Mirage, on the west edge of Maricopa County, Arizona. She died in June 2024. She had memory problems and had missed utility payments before, though that was not what happened this time. According to an autopsy report obtained by The Guardian, her electricity was connected. Her air conditioning unit was in place and moving air. The air coming out of the vents measured between 109 and 117 degrees Fahrenheit.
On August 31, three days before I wrote this, a team led by researchers at Yale published a peer-reviewed study in Nature Health estimating that residential air conditioning averts roughly 5,267 heat-related deaths a year in the contiguous United States. The estimate comes from about 30 million death records from 2010 through 2020, matched against county-level estimates of how much electricity households used for cooling. In that framework, Maricopa County — where nearly every home has AC equipment and cooling consumption runs high — reads as well protected.
Patricia Miletich lived inside that protection.
What the study measured
The central variable is not whether a household owns an air conditioner. It is an estimated average of annual household electricity consumed for cooling, calculated at the county level by a machine-learning system called FusionACS, which stitches together federal energy and housing surveys to produce a number for each county. Nobody read a meter in these homes. A single 2015 estimate was applied across the whole 2010–2020 period, with a 2020 estimate used as a check.
Counties with higher modeled cooling consumption showed a weaker relationship between extreme heat days and deaths. The researchers took the difference between the deaths that occurred and the deaths their model predicted without that cooling. The 5,267 figure is real, population-scale protection, and it matters for understanding how infrastructure shapes survival across a country.
The paper is three days old. No formal commentary or methodological critique has appeared in the journal's record yet, which is what you'd expect this soon after publication. The limitations on the table are the ones the authors put there themselves. Senior author Kai Chen makes the distinction plainly in Yale's summary: knowing that AC protects a particular person is a different question from estimating how many deaths it prevents across a population. This study answers the second question. It was built to answer the second question.
Heat, though, kills one person at a time, in a particular room, on a particular afternoon.
What a county average can't see
A county's modeled cooling-electricity value folds together how many homes have AC, how many run it, how efficient the equipment is, how hot the climate is, and how long people leave it on. The number doesn't separate those things. It doesn't register whether a unit works. It doesn't know whether a household could afford to run it through July, or whether the power stayed on during the afternoon the grid strained, or which room got cooled, or whether the person who died was in that room.
For a county's cooling statistic to translate into someone surviving a heat wave, a whole sequence has to hold at once. The equipment has to function, and the household has to afford the electricity to run it. The utility has to keep service connected; the grid has to hold at peak demand. The cooled space has to be where the vulnerable person actually spends the day. Any one of those can fail on its own. The county number picks up none of the failures.
Maricopa County, where both measurements exist
If you were looking for a place where county-level AC statistics say safety and room-level conditions say something else, Maricopa is where you'd go. A 2020 field study found 95 percent central AC across the county. The Census Bureau's Local Air Conditioning Estimates put Arizona statewide at 98 percent of occupied housing units with some form of cooling.
The county recorded 430 heat-related deaths in 2025. One hundred eleven of them happened indoors. In 94 percent of those indoor deaths, air conditioning equipment was present in the dwelling. In 72 percent of the AC-present cases, the equipment wasn't working.
The 2025 report publishes only that share. The fuller breakdown comes from 2024, when the county classified every indoor heat death by AC status. That year 138 people died indoors, with AC present in 88 percent of the cases. Among those homes, 70 percent had equipment that was broken, operating improperly, or couldn't be switched on even though the electricity was live. Another 18 percent had working equipment that wasn't turned on. Nine percent had no electricity at all.
Of 138 indoor deaths, AC equipment was present in 88% of cases. Among those: 70% nonfunctioning, 18% not turned on, 9% no electricity.
I looked at Maricopa's indoor heat deaths earlier this year through the question of nighttime heat and the recovery window that stops working when temperatures don't fall. The Nature study raises something adjacent: what happens when a county's aggregate cooling capacity becomes a variable in a national mortality model.
Maricopa does something most jurisdictions don't. It investigates indoor heat deaths, records whether the AC worked, and publishes the classification. That's why the distance between the county's number and the room's temperature is visible here at all. The 5,267 figure is assembled from death records and modeled electricity consumption. In most of the counties feeding that model, no one is checking the equipment.
The equipment
Nonfunctioning AC is the largest documented category in Maricopa's indoor deaths, and it clusters. In 2025, RVs, trailers, and mobile homes accounted for 23 percent of indoor heat deaths while making up roughly 5 percent of the county's housing. Cooling systems in those structures are smaller, harder to get repaired, and more easily overwhelmed by the heat pressing in from outside. Patricia Miletich lived in an RV community.
The 2020 Phoenix field study found the same dynamic with lower stakes. Researchers put temperature sensors in 46 air-conditioned homes across the metro area. Every household reported using their AC. Indoor temperatures still varied a lot, according to building age, insulation, which way the windows faced, where the thermostat was set, and whether the cooled part of the home was where the resident actually sat. Even in a city where cooling is near-universal and actively running, the county's average and the room's temperature were two different measurements.
The electricity bill
Arizona's federal LIHEAP profile for FY2024 reports 37,554 households receiving help statewide. LIHEAP — the Low Income Home Energy Assistance Program — is the main federal program for households that can't cover their utility bills, and those 37,554 households represent 6 percent of the Arizonans who qualified. Ninety-four percent of the eligible population got nothing, not because they were turned down on the merits but because the money ran out. The program's application page says assistance isn't guaranteed and is handed out first-come, first-served.
The households that do get in receive an average cooling benefit of $945, against a standard benefit ceiling of $640. The National Energy Assistance Directors Association projects an average Arizona summer electric bill of $1,060 for 2026. So even inside the program, the arithmetic leaves a gap the household covers.
A working AC unit in a home that can't afford to run it through the summer is, for the purpose of getting through August, the same as a broken one. The county's cooling estimate doesn't distinguish between them. A household that runs the AC in June and shuts it off when the bill lands still contributed cooling electricity to the county average, even though its August went uncooled.
The power connection
Katherine Korman was 82. Her utility, APS, remotely disconnected her service for nonpayment on May 13, 2024, a day when the local high reached about 99 degrees. She was found dead six days later. The medical examiner ruled the death accidental, from chronic alcohol use, with cardiovascular disease and heat stress contributing. The case ended in a $7 million settlement and a policy change: APS now pauses residential shutoffs whenever the next day's forecast hits 95 degrees, on top of the June 1 through October 15 shutoff ban already required under state regulation.
Salt River Project, which serves a different slice of the same metro area, runs its own rules. SRP suspends disconnections through July and August, and during National Weather Service Extreme Heat Warnings. June, September, October, and any hot day that doesn't trigger a formal warning fall under different protection. Two utilities, one county, two different sets of days on which a customer's power can legally be cut in dangerous heat.
Which of those rules applies to you depends on which side of a service boundary your house sits on. The county-level cooling estimate has no way of knowing that, or whether the power was on the afternoon the temperature spiked.
What the plateau tells us
The study's authors list their own constraints. Cooling estimates are county-level rather than household-level. One 2015 estimate stands in for a decade. Cooling centers and municipal heat-action plans, which may deserve some of the credit for lower mortality, aren't separated out. The net health effect of AC after accounting for the pollution from generating the electricity isn't calculated. These are honest caveats attached to a carefully built population study, and the 5,267 figure almost certainly reflects something true about cooling infrastructure and heat deaths at scale.
The researchers also found that the protective effect flattens out above median usage. More cooling electricity stops reducing mortality past a certain point. That flattening lines up with what Maricopa's death records show from the other side. Where a county already has substantial cooling infrastructure, the deaths that remain are shaped by the conditions the average passes over — whether the unit in one particular home works, whether the bill is payable in August, whether the service stayed connected through the worst week.
The study's public code repository substitutes fake county values for the real ones, so you can't look up the number the model assigned to Maricopa. Whatever it is, it's an average across 4.5 million people. Katherine Korman is inside it, and so are the 111 people who died indoors in 2025 in homes where, almost without exception, an air conditioner was physically present.
The chain running from a county's cooling statistic to a person's survival is not something a national model can evaluate. A household can check some of it: whether the unit actually cools the air it's pushing around, whether the worst week of summer is affordable, which shutoff policy applies when the temperature spikes. The county average doesn't answer any of those.
Forty-eight percent of Maricopa's indoor heat decedents in 2025 were found during welfare checks. Someone had to go looking.
Aggregate statistics overstating individual protection is not unique to air conditioning; the same distance opens up wherever population-level numbers are used to assess something that succeeds or fails one household at a time. Cooling is unusually visible because Maricopa County documents what most places don't. Across the three thousand-odd counties in the study's dataset, nobody is checking whether the AC worked.
- Nights that don't cool: A global analysis found the hottest nights warming faster than the hottest days — 0.32°C per decade versus 0.27°C — which matters because nighttime is when bodies and buildings are supposed to recover from daytime heat.
- Cooling centers and the clock: A Maricopa County assessment found that 14% of cooling-center visitors said centers weren't open when they could access them, and a broader survey of large U.S. jurisdictions found nearly half operated cooling centers only during daytime hours.
- Low-energy cooling, unproven rooms: A controlled chamber study showed a combined evaporative cooling system using about 230 watts could lower air temperature by 2.8°C, but the test used a thermal manikin in one room, not people in occupied homes or humid climates.
- When adaptation costs compete: Associated Press reporting on the Nature study documented families during 2026 heat waves choosing between medication and running an air conditioner, a tradeoff the mortality model's county-level electricity variable cannot register.