Internet-connected thermostats (CTs) control heating and cooling systems in about 30 % of North American homes, and capture half of annual thermostat sales. In 2017 the U.S. Environmental Protection Agency created a program to certify the performance of ENERGY STAR® Connected Thermostats. To demonstrate compliance with energy-saving criteria, thermostat vendors must submit performance summaries for a representative sample of up to 1250 homes. Vendors must then re-submit results from a new, representative sample every six months in order to maintain their ENERGY STAR certification. This procedure has created a unique record of each thermostat's long-term performance in response to changes in weather, customer demographics, building stock, and algorithms. Our analysis covers submissions from 13 different vendors, who submitted data up to 7 times over four years. We found that these semi-annual samples generated relatively stable trends for comfort temperatures and Heating, Ventilating and Air Conditioning (HVAC) runtimes over the study period. However, some vendors achieved consistently more energy-conserving comfort temperatures and shorter HVAC runtimes than others. The most recent submission runtimes averaged 700 hours for heating and 1,300 hours for cooling, but some vendors achieved runtimes as much as 17 % below the mean. This implies lower energy consumption is due in large part to more successful algorithms and control strategies used by those vendors.

At least five million American homes have Internet-connected thermostats. These devices improve comfort and reduce energy consumption using cloud-based algorithms. Every five minutes, they collect and transmit detailed operating information, including thermostat setpoints, actual indoor temperatures, occupancy, and HVAC operation. One thermostat provider established a program that enables customers to anonymously “donate” their data to researchers and more than 50,000 customers have opted in. This dataset represents the most comprehensive public data on home temperature preferences for North America and provides far more detail than any previous method based on surveys or monitoring programs. The data show in detail preferred temperatures while occupants are home, sleeping, and away. On average, these households lower their thermostats about 1 °C when they are asleep compared to when awake, though this average conceals both widespread constant operation and deeper setbacks. The peak usage of air conditioners in Texas was shown to precisely match the grid's systemwide peak. The connected thermostats also raise a survey research question: when should policymakers rely on a small sample of rigorously selected buildings instead of a huge, unrepresentative sample with detailed data? Many fruitful applications of this dataset will be constrained by privacy protections and reluctance of firms to share information.