The weather affects air conditioner purchases to fill the energy efficiency gap

The study addresses whether short-run temperature fluctuations can help overcome the energy efficiency gap in appliance purchases by increasing consumer adoption of Energy Star air conditioners. The energy efficiency gap—underinvestment in cost-effective energy-saving technologies—has been linked to behavioral anomalies such as myopic discounting and limited attention to energy efficiency attributes, leading consumers to underestimate operational cost savings. The authors posit that changes in the physical environment, particularly temperature, can reshape predicted future utility via mechanisms like projection bias and salience, thereby influencing the choice between efficient (Energy Star) and less efficient air conditioners. Understanding this linkage is important for policies targeting Net Zero emissions and consumer welfare through reduced energy costs and pollution.

Prior work shows ambient conditions affect economic choices: weather influences automobile purchases, solar adoption, clothing catalog orders, and college enrollment; poor air quality affects health insurance purchases, housing prices, and willingness-to-pay for pollution control. Behavioral mechanisms include projection bias—consumers overweight current conditions when forecasting future utility—and salience—heightened attention to certain attributes under particular contexts (e.g., heat). Energy efficiency research documents limited attention to operational costs and underweighting of future fuel economy, contributing to the energy efficiency gap. This paper extends the literature by examining whether short-run temperature deviations (in levels and relative to recent conditions) affect the likelihood of purchasing Energy Star air conditioners.

Data: Weekly transaction-level records of room/window air conditioner purchases from the Nielsen Retail Scanner Dataset, 2006–2019, linked to county-level weather from NOAA Global Surface Summary of the Day (GSOD). Key meteorological measures include weekly average temperature, cooling degree days (CDDs), heating degree days (HDDs), precipitation, wind speed, and relative humidity (and humidity squared). Socioeconomic covariates for heterogeneity analyses include county-level median income, education, age, racial composition, housing characteristics (rooms, owner/renter ratio), heating fuel shares, state-level electricity prices, and county-level climate attitude scores (beliefs/worry about climate change, support for CO2 regulation, Democratic vote share). Model: The primary specification is a generalized linear regression estimated via generalized least squares (GLS) with extensive fixed effects due to convergence issues in nonlinear models. The dependent variable is a dummy for whether a transaction is Energy Star certified. Core regressors are temperature-bin indicators based on weekly average temperature relative to a reference interval (20–22 °C) and, alternatively, continuous measures of weekly CDDs and HDDs. Controls include precipitation, wind speed, relative humidity and its square, and transaction price. Fixed effects include month and year fixed effects to capture time-varying confounders; clustering of standard errors at the store level, with robustness to alternative clustering levels (county, 3-digit ZIP, DMA). Robustness checks: (1) Control for contemporaneous weather in the transaction week to address temporal autocorrelation. (2) Include lagged weather controls at 1–3 weeks and aggregate monthly weather to assess harvesting effects; effects remain significant. (3) Drop price outliers (<$100) to ensure results are not driven by unusual pricing. (4) Test alternative reference temperature thresholds (16–18 °C; 65 °F for CDD/HDD); results persist. (5) Include quadratic terms for weather variables to assess nonlinearity; main effects hold. (6) Alternative fixed-effects structures and clustering levels; significance robust. (7) Falsification test using Energy Star telephones, which should be weather-insensitive; temperature coefficients are insignificant, supporting causal interpretation. Heterogeneity: Interact weekly CDDs/HDDs with background climate (historical CDDs/HDDs), state electricity prices, county income, education, median age, owner/renter ratio, rooms, electricity as heating fuel share, and climate attitudes to identify differential responsiveness. Estimation software: Stata 16 for regression; R (4.0.2) for data download/figures. Code and selected datasets available on GitHub.

• Temperature deviation and Energy Star purchases: Relative to the 20–22 °C reference, the probability of purchasing an Energy Star air conditioner increases as weekly average temperature deviates from this range. Increases: +>2% when 24–32 °C; +>5% when >32 °C. Colder weather also increases Energy Star purchases by about 0.7–3% (some models suggest a rise then decline at very low temperatures, consistent with substitution to other heating devices).
• Marginal effects around 21 °C: A 1 °C increase from 21 °C raises the probability of an Energy Star purchase by about 0.4%; a 1 °C decrease has a smaller magnitude change (about 0.2% in the opposite direction).
• Economic impact: A 1 °C increase (decrease) from 21 °C is estimated to reduce total national energy expenditure on air conditioning by $35.46 million ($17.73 million), corresponding to 0.13% (0.06%) of annual AC energy expenditures, via increased Energy Star adoption. Based on average annual US AC shipments (~4.44 million), this implies about 17,760 and 8,580 additional Energy Star purchases replacing non-Energy Star units for +1 °C and −1 °C deviations, respectively.
• Mechanisms: Patterns are consistent with projection bias and salience—deviations from comfort heighten attention to operating costs and increase valuation of energy efficiency.
• Heterogeneity: Larger responses where electricity prices are higher; where prior-year background CDDs are fewer (milder climates); mixed/non-monotonic patterns for background HDDs. Responses tend to be larger in counties with higher income and higher median age; education shows a positive but not always statistically significant difference; racial composition (share white) shows no significant effect. Housing and energy context matter: higher owner ratios correlate with stronger CDD responses; electricity as heating fuel share is associated with smaller HDD responses, suggesting central systems reduce sensitivity. Stronger pro-climate attitudes (belief in harm, worry, support for CO2 regulation, Democratic support) correlate with larger responses.
• Robustness: Results persist under alternative clustering, fixed effects, non-linear weather terms, exclusion of price outliers, alternative temperature thresholds, and lag structures; falsification with telephones shows no spurious weather effects.

Findings indicate that short-run temperature fluctuations significantly shift consumers toward Energy Star air conditioners, supporting the notion that ambient conditions influence attention and perceived future utility through salience and projection bias. These behavioral responses can partially close the energy efficiency gap at moments when temperature deviates from comfort, yielding measurable energy cost savings and emissions reductions. The heterogeneity results suggest policy targeting: greater responsiveness in high electricity price areas, milder historical climates, higher-income and older populations, and communities with stronger pro-climate attitudes. Policymakers and retailers could time and tailor non-price interventions (e.g., information campaigns, rebates, marketing) to periods of temperature deviations and to responsive subpopulations to amplify Energy Star adoption. Integrating consumer behavioral responses into climate damage assessments and mitigation planning can enhance the effectiveness of demand-side strategies as heat exposure intensifies.

The study demonstrates that deviations from a 20–22 °C weekly average temperature increase the likelihood of purchasing Energy Star air conditioners, with economically meaningful national energy savings. Evidence aligns with projection bias and salience mechanisms and remains robust across multiple specifications and tests. Policy and marketing strategies that leverage short-run weather fluctuations—particularly in areas with high electricity prices or pro-climate attitudes—can enhance the adoption of energy-efficient appliances and contribute to emissions reductions. Future research should incorporate finer temporal resolution, broader sales channels (including online), and disaggregate consumer segments (households vs. private entities) to refine estimates of energy savings and to design more targeted interventions.

• Temporal aggregation: Weekly aggregation may mask daily weather fluctuations that influence purchases.
• Coverage: Nielsen Retail Scanner data, while extensive, do not include all transactions (e.g., online sales) and are not exhaustive nationwide; potential spatial/temporal coverage patterns could bias estimates if correlated with weather (though considered unlikely).
• Aggregation level: Partial coverage complicates aggregating to county-level totals and quantifying total electricity saved.
• Buyer identification: Store-level sales lack identification of purchaser types; private entities may behave differently from households, limiting targeted policy implications.
• Product substitution at extremes: At very low temperatures, consumers may substitute to other heating/cooling systems (e.g., central systems, heat pumps), affecting generalizability of the AC-specific response.