Rob Stavins (Harvard) and Richard Newell (Duke) put together a most excellent workshop, which brought together one possible all-star team of energy economists to discuss the energy efficiency gap (the wedge between the cost-minimizing level of energy efficiency investment and the investment level actually realized). Instead of asking participants to engage in shameless self promotion (the standard format of academic workshops), we were asked to give a 20 minute summary of the theoretical and empirical literature on one possible proposed explanation for the energy efficiency gap. I was asked to speak about the concept of learning by using, which is not a topic I had thought about extensively and only spend about 60 seconds on in my 15 week intermediate microeconomics course.
Learning by using goes back to Rosenberg (1982) and is distinctly different from learning by doing. “Doing” refers to gains that are internal to the production process, where firms’ average costs of production drop due to learning. Learning by using refers to gains that are generated by the subsequent use of the product.
In the most general terms, Rosenberg argues that there are performance characteristics of a durable capital good that cannot be learned without prolonged experience with it. If you have a product involving complex interacting components or materials subject to stress in extreme environments, the outcome of the interaction cannot be precisely predicted ex ante. Learning by using helps determine the optimal performance characteristics of the capital good and of course, closely related, its optimal maintenance. The cleanest example of this was the development of jet engine powered airplanes and their optimal maintenance regimes.
The economic literature on learning by using’s contributions to energy efficiency improvements is very small. Learning by using generates two types of knowledge: embodied and disembodied. The first type is where early use of a product generates knowledge that leads to design modifications to improve the product. This is essentially a feedback loop. An excellent example of learning by using improvements can be found in American muscle cars. Some of the more recent V-8 engines switch off a few cylinders while coasting down the highway, resulting in pretty impressive fuel efficiency. Consumers only care about power when they need it. The new smart engines sense when you need the power and provide that big American V-8 (or V-10 if you have a Viper!) power. Another example is that while I was writing this post, I have gone to the refrigerator five times to see what is in there. I did not eat anything, but I just had to get a better understanding of my choice set. By the time I will have finished blog post number two, that door will have been opened easily ten times and let lots of warm air into the fridge. If we knew that this “Max factor” existed, we could suggest that a fridge which one could look into without opening the door might be a good idea (assuming that the two doors have the same insulation capacity).
The second type of knowledge is disembodied, where prolonged use of the piece of capital equipment leads to alterations in use that require no or little design changes. Experience generates increased productivity by changing the way one uses the piece of equipment. I now use what I thought was just a phone as a flashlight, a level and a decibel meter.
In the energy efficiency literature, learning by using has been described differently. In fact, in Gillingham et al’s. excellent survey they use learning by using for what others call “experience goods”. With an experience good, the consumer does not know her true valuation for the product until she has owned it or seen it in action. So she may not adopt some product that she would actually like to have, because her prior beliefs about her valuation are biased. If individual learning about product quality/value is transferable between households, adoption can lead to peer spillovers as other households correct their priors. So if you want to put this into a traditional adoption framework, the user of a new product or a more energy-efficient product generates a positive externality by getting others to want to adopt the process. I am getting this with my iPhone 5s: each time I use my fingerprint to log on, others want one. There is next to no literature on this in energy economics.
If we think of this phenomenon in the context of demand-side management programs these spillovers have been divided into free drivers and program spillovers. Free-drivers are non-adopters who install energy-efficient devices by learning about them from program participants. Program spillovers are additional purchases of energy-efficient devices by program participants. There is a very small and not very active literature in this area.
What papers should be written if any? On the experience goods side, I would like to see some quasi experimental studies quantifying spillover effects in energy efficiency programs. Judd Boomhower is working on what promises to be an excellent paper on refrigerator adoption in Mexico by neighbors of participants in a national rebate program. Second, I would like to see more work, not necessarily by economists, on how people decide to adopt more energy-efficient devices. Hunt Allcott is doing some interesting work along these lines.
Finally, I would like to see how machine learning is improving our energy efficiency. The Nest thermostat learns from our “using” and improves our energy efficiency. Maybe if people have limited attention to all things energy efficiency, machines can learn from our behavior and make more optimal decisions for us.
I walked away from this workshop most motivated and excited. There is much work to be done. So let’s get to it.
Maximilian Auffhammer is the George Pardee Professor of International Sustainable Development at the University of California Berkeley. His fields of expertise are environmental and energy economics, with a specific focus on the impacts and regulation of climate change and air pollution.