Incentive based emergency demand response effectively reduces peak load during heatwave without harm to vulnerable groups

The study addresses how to reduce peak electricity load during summer heatwaves amid rising electrification and climate change. Traditional supply-side expansion to meet short-lived peaks is costly and carbon-intensive. Demand response offers demand-side flexibility; yet, in regulated electricity markets like China, it remains unclear whether incentive-based emergency demand response (EDR) can elicit meaningful conservation without harming heat-vulnerable populations. The research question is whether households will curtail use during peak hours under incentive-based EDR in heatwaves, and how responses differ for vulnerable groups (low-income, households with children, and elderly). The purpose is to estimate causal impacts of EDR access and rebate coverage on peak demand, examine heterogeneity, and assess persistence over repeated implementations to inform policy in regulated markets.

Demand response (DR) includes price-based measures (e.g., time-of-use, critical peak pricing) and incentive-based measures (e.g., EDR). Prior pilots in California, Washington D.C., and the Netherlands show price-based DR can achieve sizable and stable demand reductions, enhanced by enabling technologies. However, such programs may impose financial burdens on low-income or heat-vulnerable groups and require deregulated markets and costly smart equipment, limiting scalability in partially monopolized systems like China. Existing empirical evidence on incentive-based EDR largely comes from institutional reports in advanced economies with limited large-scale trials and uncertain applicability to regulated markets. Literature also documents that low-income households face energy insecurity, elderly individuals have narrow safe temperature ranges, and children are sensitive to energy insecurity; price-based DR can exacerbate trade-offs (heating vs. eating). Incentive-based EDR may avoid added financial burden by providing rebates, but responses of vulnerable groups remain unclear.

Design: Large-scale field pilot in southwestern China during persistent heatwaves (daily max >35 °C) in July–August 2019. Six EDR treatment days between July 18 and August 21, 2019; peak period targeted 8:00–9:30 pm. Sample: 205,129 households with high-speed power line communication (HPLC) meters recording 15-minute usage; 53,129 rural and 152,000 urban. Survey subsample n=7,774, with 41.2% households with children and 38.5% with elderly. Randomization and groups: Clustered randomization by community to assign permission to apply for the EDR program. Three groups: EDR group (received message and confirmed participation), no-reply group (received message but did not confirm), and no-notification group (received no message). Assignment winners comprise EDR + no-reply; control is no-notification. Incentive: For participating households, electricity use during 8–9:30 pm on treatment day compared to same period on the prior day (benchmark). Savings paid at $0.143/kWh (≈¥1/kWh), exceeding increasing block residential tariffs. Outcomes: Electricity usage during on-peak times; electricity conservation computed as treatment day minus benchmark day. Causal identification:

• Intent-to-treat (ITT): Difference-in-differences (DID) comparing assignment winners vs. control to estimate effect of being selected by random assignment; spillover assessed by comparing no-reply vs. no-notification.
• Local average treatment effect (LATE): Instrumental variables two-stage least squares using random assignment as instrument for EDR rebate coverage (participation), estimating causal effect among compliers.
• Heterogeneity: Difference-in-difference-in-differences (DDD) for urban vs rural, households with children, and households with elderly; also examined temperature interactions and price heterogeneity.
• Matching: Dynamic time warping (DTW) matching leveraging 36 months of monthly and 15-minute pre-trial usage patterns to strengthen parallel trends; also DID within matched samples.
• Robustness: Parallel trends tests, Heckman two-step selection corrections, placebo tests; clustering of standard errors at household-group level; inclusion of household and time fixed effects and controls (weather, monthly average use, household characteristics). Sustainability: OLS models with treatment phase indicators and participation counts to assess persistence over six repeated interventions. Ancillary measurement: Smart sockets in 15 households to infer appliance-level curtailment behaviors.

• ITT (assignment to EDR access): Selection by random assignment reduced on-peak electricity use by 0.0155 kWh per household (p<0.001), a 1.02% reduction relative to the control mean (Supplementary Table 10). No evidence of spillover among no-reply households (coef. -0.0024 kWh; p=0.149).
• Rebate coverage (IV LATE): EDR rebate coverage decreased on-peak usage by 0.1145 kWh (p<0.001), a 7.32% reduction relative to the control mean of 0.04 kWh (Table 2; Supplementary Table 11). Matching (DTW) yielded consistent effects (e.g., -0.1113 kWh; p<0.001).
• Probability and magnitude of savings: EDR coverage increased the probability of saving behavior by 3.68x; average conservation 0.1297 kWh vs. control mean 0.0352 kWh (Supplementary Table 9).
• Heterogeneity: • Urban vs. rural: Urban households saved an additional 0.0545 kWh (p<0.001), corresponding to a 2.19% peak-load reduction; both urban and rural showed diminished temperature sensitivity under EDR. • Households with children: No significant effect (coef. -0.0138; p=0.940). • Households with elderly: Larger savings of 0.1638 kWh (p=0.026).
• Sustainability over repetitions: Treatment effects increased with participation count (coef. 0.0406 per additional participation; p<0.001); largest phase effect in the fifth treatment (coef. 0.1326; p<0.001). Spillover showed slight conservation after the second phase. Urban slightly more affected by repetition than rural (urban coef. 0.0407 vs. rural 0.0381; both p<0.001).
• Appliance-level actions (from smart sockets, n=15): Indicative curtailments included reduced operation of refrigerators (≈36–66 min), televisions (≈38–55 min), washing machines (≈23–33 min), and short air-conditioner off-intervals, with larger durations among elderly and urban households.
• Price heterogeneity: Households facing higher average/marginal electricity prices responded more strongly to EDR (Supplementary Note 11).
• Equity: Incentive-based EDR imposed no additional financial burden on vulnerable groups since rebates are paid only for realized savings.

The findings directly answer the research question: incentive-based EDR induces statistically and practically significant reductions in residential peak demand during heatwaves in a regulated market context without adding financial burden to vulnerable populations. The causal ITT and LATE estimates show that both broader access and actual rebate coverage reduce on-peak usage, with larger effects among participants. No spillover from messaging alone suggests monetary incentives are the primary driver. Heterogeneity analyses reveal greater conservation potential in urban households (likely due to higher appliance ownership and baseline use) and strong, positive responses from households with elderly members, while households with children prioritize comfort and study needs, showing no significant response. Repeated implementations enhance effects rather than attenuate them, implying learning-by-doing and planning behaviors (e.g., pre-cooling, appliance scheduling) and potential gradual investments in efficient devices. Policy implications include using EDR to meet peak reduction targets, relieve supply pressure, and defer costly, carbon-intensive peaker investments. Targeting and equity considerations are crucial: although EDR is financially benign for low-income groups, higher savings in urban areas may lead to disproportionate rebate flows; program design should consider balanced benefits and protections for populations with limited conservation potential. The stronger responses under higher electricity prices suggest complementarities with tariff structures.

This large-scale randomized EDR pilot in southwestern China demonstrates that incentive-based EDR effectively reduces residential peak loads during heatwaves, achieves up to 7.32% reductions among covered households (and 1.02% at access scale), and does so without imposing financial burdens on vulnerable groups. Urban households and households with elderly members exhibit greater savings, while households with children show no significant response. Repeated interventions amplify effects rather than causing decay. Methodologically, the study contributes robust causal estimates using random assignment, IV-LATE, DDD heterogeneity analysis, and DTW matching. Future research could: (i) examine long-term health and comfort outcomes across vulnerable groups; (ii) test varying incentive magnitudes, messaging, and automation technologies; (iii) explore integration with dynamic pricing or real-time control in regulated markets; (iv) assess scalability across seasons, regions, and grid conditions; and (v) evaluate cost-effectiveness and equity impacts under alternative targeting or tiered incentive designs.

• External validity: Results pertain to southwestern China during a single summer heatwave period and fixed peak hours (8–9:30 pm); generalizability to other regions, seasons, or market structures may be limited.
• Limited number of treatment phases: The near-absence of decay with repetition may reflect only six trials; longer horizons are needed to assess persistence and potential fatigue.
• Self-selection and compliance: Although addressed via random assignment, IV-LATE, matching (DTW), and Heckman corrections, unobserved time-varying factors may remain.
• Measurement scope: Appliance-level behaviors inferred from smart sockets in a small sample (n=15) may not be representative.
• Data sharing constraints: Individual HPLC data are subject to NDA, limiting external replication with microdata.
• Equity considerations: Greater urban savings may concentrate rebates; while financially non-burdensome, program design should ensure benefits for households with limited conservation potential and guard against possible health risks from over-curtailment among elderly participants.