How many households in India, Brazil and Mexico own an air conditioner depends on the heat and
on what they can spend. Data on 276 million households, with projections to
2040.
Without AC today232M84% of households
Still without in 2040134M85 to 181M across scenarios
Using published output data from Pavanello et al. (2021), we examine what determines household
air-conditioning (AC) in 92 states of India, Brazil and Mexico, home to
276 million households. Across states, heat exposure (wet-bulb
cooling degree days) and spending per person are equally strong predictors of AC ownership
(standardized β = 0.41 and 0.41), contrary to our
expectation, written down before the analysis, that income would dominate. Their combination matters most: in states
hotter but poorer than their country's median, 8.2% of households own AC,
against 17.7% in hot, richer states (p = 0.009). Household-level models
show the marginal effect of heat on AC ownership is indistinguishable from zero below roughly
$1,800 (India) to $3,000 (Mexico) of annual household spending. Projected to 2040, between 85 and
181 million households would still lack AC across ten climate–socio-economic
scenarios (134 million in the middle scenario), concentrated in the bottom
spending deciles and, among hot states, overwhelmingly in India.
1 Introduction
Air-conditioning is the most direct private adaptation to heat, and its spread is a major driver
of future electricity demand. Whether a household can adopt it depends on both need (climate) and
means (income). This piece asks which of the two governs adoption in three large emerging
economies, and who remains without AC as incomes grow and climates warm.
2 Data and measures
All inputs are the replication output files of Pavanello et al. (2021), derived from national
household surveys (India NSS 2012, Mexico ENIGH 2016, Brazil POF
2017). Heat exposure is the state mean of annual cooling degree days on
wet-bulb temperature (24 °C base, 1970–2016). Spending is household expenditure per person (2011
US$). AC share is survey-weighted. Projections to 2040 combine two emission pathways (RCP 4.5,
8.5) with five Shared Socioeconomic Pathways; the middle scenario is RCP 4.5 / SSP2.
3 Results
Figure 1. Share of households with AC by state against heat exposure (left)
and spending per person (right). Dot area ∝ households. Within-country Pearson correlations,
heat / spending: India −0.08 / 0.44,
Brazil 0.54 / 0.01, Mexico
0.38 / 0.29. n = 92 states.
Source: Pavanello et al. (2021).
Neither driver dominates. In India, heat alone is uncorrelated with AC share (r =
−0.08, p = 0.655) while spending is (r =
0.44, p = 0.009); in Brazil the pattern
reverses. Pooling states with all variables standardized within country, heat and log spending
have equal coefficients (β = 0.41, p < 0.001; β =
0.41, p < 0.001; R² = 0.21).
Figure 2. States hotter and poorer than their country's median (red) against
hotter and richer states (dark). Household-weighted AC share: 8.2% vs
17.7% (cool states: 16.4% poorer, 15.2%
richer). Two-sided Mann–Whitney U on state shares, p = 0.009.
Figure 3. Household logit results. Left: predicted probability of AC
ownership by heat exposure, other covariates at means. Right: average marginal effect of +100
cooling degree days on AC probability (percentage points) by annual household spending; shaded:
spending range where the 95% CI includes zero in India or Mexico. Bands: 95% CI. Source:
Pavanello et al. (2021).
At the household level the interaction is unambiguous. For an Indian household spending about
$1,100 a year, 100 additional cooling degree days raise AC probability by
0.06 points; at $60,000 the effect is
7.2 points. A heat × spending term in the state regression is
positive but only marginally significant at this sample size (β =
0.21, p = 0.054).
Figure 4. AC share by national spending decile, today (dashed) and in 2040
(RCP 4.5 / SSP2, solid), median across states. Grey band: range across 10 scenarios.
Red: households still without AC in 2040. Poorest decile, today → 2040: India
0.8% → 8%, Brazil 2.6% →
2.3%, Mexico 1% → 0.5%.
Figure 5. Households without AC (millions), today and in 2040 across 10
scenarios (ticks; red: RCP 4.5 / SSP2). Left: all states. Right: states projected to
exceed 100 wet-bulb CDD. Totals: 232M today;
85–181M in 2040 (hot states:
55–131M).
4 Robustness
Scenario uncertainty. The 2040 count spans a factor of two across scenarios,
from 85M (RCP 8.5 / SSP5) to 181M
(RCP 4.5 / SSP3). Under every SSP the high-warming pathway yields fewer households
without AC, because heat raises adoption. The concentration of the gap in the bottom deciles
holds in all ten scenarios (Figure 4, band).
Need versus lack. Lacking AC is not a deficit where it is not hot. Restricting to
states above 100 projected CDD reduces the middle-scenario count from
134M to 92M, of which
89% are in India. The threshold is a judgement call; the
notebook reports the count for other cut-offs.
Ecological inference. State correlations (n = 92) can mislead about
households. The household-level marginal effects (Figure 3) are estimated on microdata with
controls for household size, education, tenure, region and (in Brazil and Mexico) housing
quality, and point the same way.
5 Discussion
Heat exposure and household spending jointly determine AC ownership, and below a spending
threshold heat has no detectable effect in India and Mexico. Projected income growth closes much of the gap by
2040, least for the poorest households. Closing it through air-conditioning raises household
electricity use substantially: by
130–154% for the median Indian household in the
middle scenario. The IEA (2018) projects global energy demand for air-conditioning to triple by
2050. Lower-energy measures, from building design to passive radiative cooling (4.9 °C below
ambient in direct sunlight; Raman et al., 2014), are complements rather than substitutes in humid
heat. Limitations: survey years differ (2012–2017); ownership
is not use; Indonesia, part of the original study, lacks public microdata.
Data availability
Dataset: data/the-cooling-gap.json in the story repository; pipeline: notebooks/cooling_gap_pipeline.py; raw inputs and provenance: notebooks/raw/pavanello2021/. Original data: Pavanello et al. (2021), Nature
Communications 12, 6460, doi:10.1038/s41467-021-26592-2; replication
files at github.com/Energy-a/comparative_paper_natcomms (GPL-3.0). A Jupyter
notebook reproducing every statistic can be opened in Google Colab via the "Notebook" link above.
Data source: Pavanello et al. (2021), Nature Communications 12, 6460 — replication output data · Last updated: 2026-10-14
