A new study by Yoonjung Ahn, assistant professor of geography and atmospheric science at the University of Kansas, has opened the door on a fine-grained dataset of air conditioning usage. This most recent dataset is the first to include the entire United States. Recent research indicates that over 90% of Americans own some kind of air conditioning. Huge gaps still exist in the efficacy of systems and even the kinds of systems employed, notably between rural and urban locales.
The study is significant not only because it updates air conditioning data last collected by the U.S. Census Bureau in 1980, but because it employs advanced machine-learning techniques to fill historical gaps. Ahn aims to compile a dataset spanning from 1980 to present that can enhance understanding of how extreme heat affects different populations.
To develop this dataset, Ahn used machine-learning algorithms, particularly XGBoost, with an incredible 97-99% accuracy rate. The accuracy was inconsistent by the air conditioning type. Some units maxed out at 87%, with others going all the way to 97%. The research categorized AC types into four groups: central AC, other types (such as window or portable units), evaporative coolers, and households with no cooling systems.
Certain other variables, such as housing type, had over 10% missing and were therefore excluded. Then I applied a second machine learning model, XGBoost, to classify homes into four AC types,” Ahn said.
This dataset further illustrates the inequitable realities of AC usage between regions, showing that access to AC cannot be treated equally. In Florida, for example, almost one out of five households depend on non-electric forms of AC. Meanwhile, more than 95% of houses in suburban counties do. This study reinforces what we should already know about California and New Mexico, two states with extensive Hispanic representation. These areas have a high concentration of unique AC technologies, such as evaporative coolers.
Another key takeaway from Ahn’s research is the need to recognize and address socioeconomic and demographic disparities. “This dataset is important for understanding how people experience extreme heat as the climate warms,” she stated. The detailed data reveals critical insights that broader datasets cannot provide, particularly regarding vulnerable populations.
Despite 90% of Americans having access to some form of air conditioning, not all systems are created equal, Ahn noted. “Portable or evaporative units, for example, don’t cool homes well in humid regions.”
The real-world impacts of this research reach far beyond the ivory tower. Public health officials and policymakers can use these findings to identify areas lacking adequate cooling solutions and target support programs or energy subsidies. Ahn emphasized that understanding the distribution of AC types is essential for assessing which communities are most vulnerable to extreme heat.
Climate and heating type were the strongest predictors, by far, overall, she added. The Hispanic share of the residents was a key predictor for some types of ACs.
The study points to gaps in data completeness as a key challenge. Major metropolitan areas such as New York City showed much more severe patterns of missing data, making such predictions about AC distribution all the more difficult.
“