What the data shows
Across 92 states in India, Brazil and Mexico, heat and household spending matter about equally for whether a home has air-conditioning, and they work together. Hot states where households have money are the most air-conditioned. Hot states where they don't have less AC on average than states that barely need it.
The household models show the same thing more clearly. For families spending less than about $1,800 a year in India, or $3,000 in Mexico, living somewhere hotter makes no statistically detectable difference to whether they own AC. In Brazil the effect at that level is very small.
What we got wrong
Our pitch, written before we looked at the data, predicted that income would explain air-conditioning far better than heat. Across all states it doesn't: the two standardized coefficients are the same (0.41 and 0.41). The countries differ. In India, spending matters and heat on its own does not (r = −0.08). In Brazil it is the other way round (r = 0.54 for heat, 0.01 for spending), and Mexico is in between. A heat × spending interaction term in the state regression is positive (0.21, p = 0.054), which is only suggestive with 92 states. In the household data the pattern is clear: the confidence bands for poor and rich households don't overlap.
The 2040 projections
The projections are scenarios, not forecasts. The study takes each surveyed household, moves its income along one of five Shared Socioeconomic Pathways (SSPs) and its climate along one of two emission pathways (RCP 4.5 or 8.5), and asks the fitted model how likely the household is to own AC in 2040. Population grows as well, so the number of households rises. Across the ten combinations, between 85 and 181 million households in India, Brazil and Mexico would still have no air-conditioning in 2040; the middle scenario gives 134 million. In every scenario the households without AC are concentrated in the bottom spending deciles.
Not all of these households need AC. If we count only states projected to have at least 100 wet-bulb cooling degree days a year, the 2040 figure falls to 55 to 131 million, almost all of it in India. The threshold is our choice, and the notebook lets you change it.
The electricity cost
If poorer households close the gap the way richer ones did, by buying air conditioners, electricity use goes up a lot. In the same projections, median household electricity use in India grows by 130 to 154% by 2040, depending on the decile, and by up to 49% in Brazil and 53% in Mexico. The International Energy Agency's 2018 report The Future of Cooling expects global energy demand for air-conditioning to triple by 2050. Air conditioners also release the heat they remove outdoors, which in dense cities raises the temperature for everyone else.
Shade, ventilation, building design and fans all reduce heat with less energy, and some surfaces need none at all. Raman et al. (Nature, 2014) built a passive radiative cooler that reflects sunlight and radiates heat to the sky; it stayed 4.9 °C below air temperature in direct sunlight without a power supply. None of these replace air-conditioning during a humid heatwave, and for many of the households in this story, no form of cooling is affordable yet.
Limitations
Owning an air conditioner doesn't mean using it, and many owners ration electricity. The surveys were taken in different years (2012 to 2017), so "today" means the latest survey, not 2026. Indonesia was part of the original study, but its microdata are not public, so it is missing here. State-level correlations describe states rather than people. The household models address that, but they are observational and cannot rule out every confounder. Cooling degree days measure exposure to heat. They say nothing about age, health, housing quality or outdoor work, which determine how dangerous that heat is.
Data and methods
Sources. All data come from the replication files of Pavanello, F. et al. Air-conditioning and the adaptation cooling deficit in emerging economies. Nature
Communications 12, 6460 (2021), doi:10.1038/s41467-021-26592-2, published at github.com/Energy-a/comparative_paper_natcomms (GNU GPL v3.0;
commit 69a0125, accessed 2026-10-14). The study combines
household surveys (India NSS, Mexico ENIGH, Brazil POF) with climate data. We use its published
output files and did not re-estimate any model.
Measures. Heat exposure is the state mean of annual cooling degree days computed on wet-bulb temperature with a 24 °C base (1970–2016 for "today", 2021–2060 for projections). Spending is household expenditure per person in 2011 US dollars, the study's income proxy. AC share is the survey-weighted share of households owning an air conditioner. One state without a climate value is excluded (92 remain).
Statistics. State-level relationships use Pearson correlations within each country and an OLS regression of AC share on heat and log spending, all three standardized within country (n = 92; R² = 0.21). The hot/poor split uses each country's median state; shares are household-weighted; the hot-poor vs hot-rich comparison uses a two-sided Mann–Whitney U test (p = 0.009). Household-level curves are the study's logit average marginal effects with 95% confidence intervals. Decile values are medians across states of the state × decile mean, following the study's Figure 5. Households without AC are the study's state-level counts summed by country.
Reproducibility. The pipeline is notebooks/cooling_gap_pipeline.py;
raw files and provenance are in notebooks/raw/pavanello2021/. The notebook linked
above re-derives every number in this story from data/the-cooling-gap.json.
Data source: Pavanello et al. (2021), Nature Communications 12, 6460 — replication output data · doi:10.1038/s41467-021-26592-2 · replication files GPL-3.0 · IEA, The Future of Cooling (2018) · Raman et al., Nature 515, 540–544 (2014) · Last updated: 2026-10-14