Optimizing Electricity Demand Response: Evidence from China

The intervention involves sending text messages to residents on the day before high-demand days, encouraging them to reduce their electricity consumption.

The experiment is randomized across four dimensions:
First, comparing the effectiveness of moral suasion messages, peak-time rebate offers (PTR), a combination of both, or no intervention;
Second, within the PTR group, varying the rebate amount from 1 to 3 yuan per kilowatt-hour;
Third, within the moral suasion group, alternating messages that highlight either the social benefits or the social costs of reducing peak electricity usage;
Fourth, adjusting the frequency of intervention days to either a low or high frequency throughout the summer.

2024-07-04

2024-08-31

Electricity consumption during peak hours

Whether users opt out of our text messages (binary)

The intervention involves sending text messages to residents on the day before high-demand days, encouraging them to reduce their electricity consumption.

We answer these questions in the context of an 150,000-household randomized experiment in Chongqing, one of the largest cities in China, the country with by far the largest electricity demand in the world. Our experiment involves sending text messages to people on the mornings of hot summer days asking them to reduce electricity use. We randomize four aspects of these messages.
First, we randomize whether we send moral suasion messages vs. peak-time rebate offers (“PTR”) vs. a mix of both vs. neither. In the PTR conditions, households are paid for reducing use relative to their baseline. To avoid an inefficient kinked incentive structure, the payment is made at the end of the summer based on the sum of the net reductions across peak days.
Second, within the PTR condition, we randomize the amount of the PTR between 1 and 3 yuan per kilowatt-hour (kWh).
Third, within the moral suasion condition, we randomize across households two moral suasion messages: a “positive” message (focusing on the social benefits of peak electricity conservation) vs. a “negative” message (focusing on the social costs of peak electricity consumption).
Fourth, we randomize the frequency of peak days between an expected [10] or [20] over the summer, which we call the low- and high-frequency groups.

randomization done in office by a computer

the randomization of text messages is at the household-by-day level

No

150,000 households.

150,000 households.

Pure Control Groups: 20,000
Low Frequency, Positive Moral Suasion, Low PTR: 2580 households
Low Frequency, Positive Moral Suasion, High PTR: 1290 households
Low Frequency, Positive Moral Suasion, PTR Control: 21934 households
Low Frequency, Negative Moral Suasion, Low PTR: 2580 households
Low Frequency, Negative Moral Suasion, High PTR: 1290 households
Low Frequency, Negative Moral Suasion, PTR Control: 21934 households
Low Frequency, Moral Suasion Control, Low PTR: 2580 households
Low Frequency, Moral Suasion Control, High PTR: 1290 households
Low Frequency, Moral Suasion Control, PTR Control: 21934 households

High Frequency, Positive Moral Suasion, Low PTR: 1720 households
High Frequency, Positive Moral Suasion, High PTR: 860 households
High Frequency, Positive Moral Suasion, PTR Control: 14623 households
High Frequency, Negative Moral Suasion, Low PTR: 1720 households
High Frequency, Negative Moral Suasion, High PTR: 860 households
High Frequency, Negative Moral Suasion, PTR Control: 14623 households
High Frequency, Moral Suasion Control, Low PTR: 1720 households
High Frequency, Moral Suasion Control, High PTR: 860 households
High Frequency, Moral Suasion Control, PTR Control: 14623 households