I’ve long been concerned—at least since my time at the Brookings Institution five years ago—about the shortage of data on the prevalence of air conditioning in the United States. While the Census Bureau’s American Community Survey provides data on households’ access to heating, including the type of fuel used, no data is available on access to air conditioning, although extended and extreme heatwaves are becoming more common due to climate change.
Data on the presence of air conditioning in homes is available through the American Housing Survey, but this survey uses a much smaller sample. As a result, data is only available for regions and certain larger states and metro areas, not for smaller geographies, such as counties or even Census tracts. Unfortunately, surveys with large enough samples to provide accurate data at the county level or smaller are rare, so if something is not surveyed in the American Community Survey, it is likely to be surveyed at that spatial resolution by anyone else.
Census Bureau Local Air Conditioning Estimates
Yesterday, the Census Bureau released an experimental dataset, Local Air Conditioning Estimates, providing estimates of the share of US households with access to air conditioning at home at geographies down to the Census tract level. This dataset is exciting, but it’s important to keep in mind that it isn’t based on direct surveys of the prevalence of air conditioning, which would have required adding questions to the American Community Survey. Instead, it is based on small area estimation of air conditioning prevalence using a model built on data about housing units and households, as well as geographic data on the average July wet bulb globe temperature and other climate data.
The main predictors for whether households were likely to have air conditioning at home were:
- average July wet bulb globe temperature
- state-level percentage of air conditioning usage from the US Energy Information Administration’s Residential Energy Consumption Survey
- climate zones from the US Department of Energy Building America program
- being in a coastal county
Based on the analysis performed by the team that created this model, the results seem to be generally fairly accurate, but there are some caveats:
- The model did not work well for Alaska and needed to be specially tuned for this state.
- The model underestimated the prevalence of air conditioning in Seattle substantially (by 10 percentage points) compared to the American Housing Survey
- The fact that none of the dominant variables represent details of housing stock or household demographics/economic situations makes me a bit suspicious of how reliable these results are likely to be in especially low-income or low-quality-housing areas, particularly in larger states.
Playing with the Data
When I saw that this dataset existed this morning, I immediately had to give it an initial look. I wrote a short R script to assemble a shapefile of US counties with their modeled percentages of households with air conditioning, along with NOAA county-level estimates of the annual number of 90°F days for years in the 2020’s and 2050’s based on SPP 3-7.0 (middle path) climate projections. This latter dataset is unavailable for Alaska and Hawaii, so I decided to limit my analysis to the contiguous US.
Below, you can see three maps of the contiguous US that I created: the first shows the estimated share of households with air conditioning based on the Local Air Conditioning Estimates dataset. The latter two show NOAA-estimated numbers of 90°F days annually for the 2020’s and 2050’s, with hashed lines over counties where fewer than 90% of households are currently estimated to have air conditioning.
While it is true that counties with lower prevalence of air conditioning do seem to have fewer 90°F days, as one might expect, there are six counties where the Census estimates fewer than 90% of households have air conditioning, but that currently six counties have more than 75 days with high temperatures of 90°F or higher annually:
- Inyo County, CA: 68% of households with air conditioning; 104 90°F days.
- Crowley County, CO: 85% of households with air conditioning; 86 90°F days.
- Kiowa County, CO: 90% of households with air conditioning; 83 90°F days.
- Otero County, NM: 87% of households with air conditioning; 79 90°F days.
- Solano County, CA: 84% of households with air conditioning; 79 90°F days.
- Guadalupe County, NM: 82% of households with air conditioning; 75 90°F days.
I’ve posted the R script I wrote, along with the shapefile I produced to generate these maps; feel free to download them, but please credit me if you use them in your own work!