Will LEDs Trigger Rebound Effects? Examining the Key Evidence
- Nov 3, 2014 7:00 pm GMT
- 769 views
By Harry Saunders and Jesse Jenkins
It was a pleasant surprise to see Rob Day’s thoughtful critique of a 2010 article on the energy consumption implications of new solid state lighting technologies – of which one of us (Saunders) was a co-author – evidently incited by a recent op-ed piece published in the New York Times by Michael Shellenberger and Ted Nordhaus. To our knowledge, no one, aside from the reviewers for the Journal of Physics, has probed these results so deeply. Refreshingly, Day reveals that he holds himself to an exacting standard of intellectual objectivity, the hallmark of scholarly discourse at its finest.
We took Day’s critique seriously and as a result, devoted many hours to reassessing the evidence and analysis in the Tsao et al. “Solid-state lighting: an energy economics perspective” paper, and it is in this spirit that we offer a reply to Day’s analysis. We first lay out a picture of the issue considered in the large, and then enumerate four specific responses to Day’s stated objections.
The authors of the “Solid-state lighting” paper were deeply disturbed, in the wake of our 2010 article, to see the popular media misinterpret our results (an opinion piece in The Economist being a primary culprit) portraying our findings as essentially a condemnation of energy efficiency improvements. So co-authors Jeff Tsao and Harry Saunders followed up with an explanatory article in Energy Policy, clarifying the meaning of our findings. In particular we were at pains to point out that lighting efficiency gains lead to improvements in economic welfare, and that “such welfare gains, and their associated energy use, should not be seen as simply continuing a pattern of wasteful energy use. All welfare-enhancing technologies (including energy-efficiency-enhancing technologies like solid-state lighting and, eventually, smart solid-state lighting) can directly benefit climate change mitigation (or adaptation) because they provide humanity with new approaches (and the wealth to implement both new and old approaches) to these problems.” Our insistent conclusion was that “[o]n the whole, we interpret our results to mean that improved lighting technologies should be pursued vigorously.”
As a matter of recorded fact, then, Tsao and Saunders do not condemn, but instead forcefully applaud lighting efficiency gains arising from new technologies. (For Jenkins’s views on rebound and efficiency, see “Are Rebound Effects a Problem for Energy Efficiency?”).
While the results of the Tsao et al. paper apply to lighting efficiency gains explicitly, they do point to a more general call for caution: to the extent energy demand forecasts currently relied on by policy makers ignore or understate such rebound effects (and they do), we have less time than is commonly believed to devise climate change mitigation solutions. What many of our colleagues seem to miss is that this holds true even if rebound does not rise to the level of “backfire.”
Responding to Day
Now to the specific points raised by Day:
1. On beta: consumption of light versus changes in GDP per capita and cost of light
Day’s first critique takes aim at our estimates of the ratio we call “beta,” which is most easily thought of as the fraction of income spent on lighting by the average individual. This ratio plays a key role in the solid-state lighting energy economics paper, which estimates this beta for various nations and at various stages of economic development.
Day points out that there is a 3x spread in beta (actually 3.3x statistically-speaking as stated at the top of page 273 in the “The World’s Appetite for Light” paper Day cites, and from which we reference data). Day is worried that this spread is so large as to render meaningless our claim that we see a fairly consistent beta throughout history and various levels of economic development.
To explain why this critique misses the mark, let’s unpack this beta ratio for a moment. Mathematically, beta is calculated as the ratio between per capita consumption of light and gdp/CoL, where CoL is the cost of light (in total costs per lumen-hour) and gdp is gross domestic product (GDP) per capita. All these components of beta change dramatically over time and over the span of the data, which encompasses three centuries, six continents and five lighting technologies. Indeed, per capita consumption of light varies by 5 orders of magnitude, gdp by an order of magnitude, and CoL by 4 orders of magnitude. Note also that luminous efficacy (lighting energy efficiency) changes over 600-fold across our data set.
In that context, even a 3.3x spread in the beta values we observe is in fact amazingly consistent. Instead of showing radical changes, beta barely budges at all. (Statistically speaking, the relatively small observed difference in beta is not significantly different from a constant beta given the more than five orders of magnitude difference in the independent/explanatory variable). This is what Day overlooks in his critique of our paper.
Of course we should expect to see some variations in lighting use among nations with different natural insolation patterns, at different stages of development, with different urbanization patterns, etc. Yet, when we are talking about global rebound effects, it is the extraordinary stability of beta across a global data set that matters, not the absolute values for individual nations. The beta we measure is not some “magic” constant, but over our data set — across several countries, continents, technologies, and over data spanning several orders of magnitude and hundreds of years – the beta measurement reveals a remarkable tendency for humans to allocate a relatively fixed portion of their income to lighting services. Apparently humans do not consume light in the same way they do ice cream.
2. On “saturation” effects: does demand for lighting saturate and decline over time?
Day also speculates that, over time and as incomes rise, consumption of lighting will saturate as households, commercial and industrial enterprises, and municipal governments run out of new uses for light, with the result that “rebound levels drop.”
(It is impossible to resist the temptation here to point out that more lights in the living room is not an absurd idea in the developing world, which will drive much of lighting demand for a long time; and to note that a stroll around any urban center in the industrialized world will give you an intense appreciation of human creativity when it comes to new uses for cheaper lighting.)
But before addressing this issue directly, it is worth reviewing the connection between the beta ratio estimated in the solid-state lighting paper and energy efficiency rebound effects. To give you the picture, think of it hypothetically this way: Assume beta is constant. This means the world spends a fixed fraction of its GDP on light despite changes in the cost of lighting and increases in wealth. Now suppose lighting efficiency doubles while GDP remains constant. This means the cost of light is now half (see equation (3) in the solid-state lighting energy economics paper). With a constant beta, consumers will respond by doubling their consumption of light (to keep the expenditure fraction fixed). This means they would, all else equal, also double their consumption of energy to power the light, except that the energy consumption per unit of light produced has fallen by half due to the efficiency improvement. The final result: actual light-associated energy consumption is unchanged – a case of 100% rebound.
It is not feasible, given our limited data set (and Day’s appetite for recent data), to develop definitive quantitative rebound estimates to explore how any saturation effects might have affected rebound in the recent past for, say, China. To carry out such an analysis with confidence, we would need a far richer time series data set for a specific nation to carry out this analysis.
Given these limitations, it is fair to say that a declining beta value (as we observe in China from 1993 to 2006 or the UK in the 20th century) indicates that rebound effects could be below 100 percent, and this could be a sign of some kind of saturation of demand for lighting (as would be expected from a gradual declining marginal utility of additional lighting consumption, for example). If so, the reverse is also true: if beta rises over time, as we observe in the UK from 1800-1900 as the British economy entered its prime development period, rebound effects are likely to exceed 100 percent, indicating backfire. Again, from a global climate perspective, the net balance of increasing and decreasing betas in individual nations is what matters. With much of the world still very much in the development stage, the precise balance between any saturation in mature economies and latent demand in rapidly developing economies is hard to determine.
Furthermore, we can develop further insights into this question of “saturation” by looking at the evolution of lighting demand in the UK. Researchers Roger Fouquet and Peter Pearson perform just such an analysis, using sophisticated econometric measurements to estimate price and income elasticities for lighting in the UK over the period 1750 to 2000. (Tsao et al. gratefully used the Fouquet and Pearson UK data series for the beta analysis, by the way).
To understand the relevance of Fouquet and Pearson’s analysis for rebound and saturation effects, it is important to first realize that it is the price elasticity of demand for lighting that matters for rebound analysis. That is, energy efficiency gains in the provision of lighting look to the user exactly like a reduction in the cost of lighting. What rebound analysis tries to do is look at the change in lighting-associated energy use given an efficiency gain, and then compare this to two counterfactual scenarios: the expectation that the efficiency gain will instead reduce energy use on a one-for-one basis (“zero rebound scenario”); and a different expectation that the efficiency gain will have no effect on energy use – i.e., will leave energy demand unchanged all else equal (“100% rebound scenario”).
To illustrate, suppose the price elasticity of lighting is measured to be -0.6 (that is, a 10% reduction in the cost of lighting results in a 6% increase in lighting use). Suppose we then invoke a doubling of lighting efficiency, which halves the cost of light. This means lighting use will increase by 0.5-0.6 = 1.515. But with double the efficiency, lighting-associated energy use will now be half this, or 0.758. The zero rebound scenario predicts lighting-associated energy use would instead drop to 0.5 to match the efficiency gain. The 100% rebound scenario predicts energy use would remain unchanged at 1.0. So rebound is (0.758 – 0.5)/(1.0-0.5) = 52%.
Fouquet and Pearson’s results are revealing. In each decade from 1750 to 1890, the price elasticity of demand for lighting in the UK was below -1.0, indicating that lighting efficiency improvements during this period led to a backfire in energy demand (i.e. greater than 100% rebound). The price elasticity then gradually declined in magnitude through the early 20th century (a saturation effect in action perhaps?). But then elasticity of lighting demand stabilized in the period 1950 to 2000 at around -0.6 to -0.8. In this range, lighting efficiency improvements would still trigger a 42% to 66% rebound given our data showing that luminous efficacy increased 4.6-fold from 1950.
In other words, even accounting for Day’s proclivity for recent data, we still see very large propensity for rebound to be ongoing in the UK, despite being in later stages of economic development.
Furthermore, Fouquet and Pearson’s analysis offers another caution: for economies in a rapid stage of development, it is entirely within the realm of reason to worry that lighting rebounds may mimic, at least for a time, the backfire conditions evident in the UK during the 19th century.
It is worth noting that the quality of the historical data series used in the Fouquet and Pearson paper is hard to verify. They have done a heroic job searching primary data sources from UK annals to develop this long time series of data, but we acknowledge there are still many possible sources of error. That means we do have to look at the apparent trends in lighting demand and price elasticity during any given period of UK history with some caution. Short-term trends may be an artifact of measurements errors. Or they may be entirely accurate. It’s hard to say for certain. What is clear however is the remarkable approximate consistency of the beta ratio observed in the Tsao et al. paper, which holds up despite the many orders of magnitude changes in per capita GDP and cost of light (as discussed above).
In summary, both the relatively constant long-term beta observed in the Tsao et al. paper and the still quite large rebounds indicated in the Fouquet and Pearson paper point to a reality that is a far cry from the popularly-held view that new lighting efficiency technologies will automatically and directly reduce lighting-associated energy consumption. For lighting, the safer assumption is that long-run energy demand rebound following lighting efficiency improvements is real and it is large, even in mature, developed economies. In rapidly developing economies, rebound appears even larger still. So yes, lighting efficiency gains will likely help with climate change mitigation, but not as much as so many apparently believe; nor will they help as quickly.
3. On the cost of light: capital versus energy costs
Day also takes aim at our assumption that the ratio between capital costs (i.e. cost of purchasing a lamp or light fixture) and energy costs (i.e. costs of electricity or fuel) over the life of a lighting source will remain pretty constant in the future, at an approximate ratio of 1/3. Day points out that modern LEDs with controls might be more capital intensive than a simple CFL or incandescent bulb.
In fact, though the ratio between capital versus operating cost of lamps likely varied somewhat with technology, it does not appear to have varied by much, including for what can be anticipated in the long-term for solid-state (LED) lighting. According to Day “anyone involved in the LED industry” knows that the “upfront cost barrier of more expensive LEDs” is one of the key barriers to adoption, but he mistakenly implies that this upfront cost means the capital/operating ratio is far higher for LEDs than for other lighting technologies. While an LED costs much more up front than a traditional bulb, it also lasts far longer, and therefore consumes a lot more energy over the course of its longer life. The real capital to energy ratios for LEDs and other lighting options thus differ by much less than Day implies in his article.
To further unpack that 1/3 capital to energy ratio from the paper a bit, the Tsao et al. paper assumes half of the capital cost is due to the bulb itself, while the other half is associated with the lighting fixture. In other words, the paper assumes the light bulb has a 1/6 capital to energy ratio.
The table above shows the current costs of real light bulbs available at your neighborhood Home Depot today, as well as the estimated cost of an LED in the future, should capital costs continue to decline. As we can see, Day is right in that for LED lighting the capital/operating cost ratio is higher than for CFLs or incandescent at its current early stage of technology development. But this will hardly remain the case. As anyone who shops at Home Depot can see, the capital cost of LED lighting is dropping incredibly quickly.
While the ratio between capital and energy costs for LED bulb today is currently closer to 0.3, if LED prices continue to fall as expected, it may soon approach 0.17 — i.e. very close to 1/6 (or .1667) ratio observed historically and used in the Tsao et al. paper. (Note total capital/energy ratios for halogen and CFL bulbs on the market today are also not far from 1/6).
In short, LEDs may be somewhat more capital intensive now than the assumption in the Tsao et al. paper, but they will likely not be for long. Within a decade, LED prices are likely to fall and the capital ratio will converge to the long-term estimate.
4. Short-run versus long-run estimates
Day also cites a German study that finds that short-term rebound from residential lighting retrofits there is small. Our study was precisely not concerned with short-term rebound—and of course, neither are climate policy makers.
As Tsao and Waide state on page 260 of the “World Appetite for Light” paper:
“Nearly all empirical studies of which we are aware focus on relatively short (months to years) time periods during which societal-use paradigms for an energy service are relatively static. It is only over longer (decades to centuries) time periods that radically new societal-use paradigms may be expected to emerge, with associated radical changes in consumption of that service (Rosenberg 1982). It is in fact these radically new societal-use paradigms that were envisaged in the first formulation of the rebound effect (Alcott 2005; Jevons 1906).”
When it comes to the decades-long task of mitigating climate change, very short-run surveys tell us little about the long-run rebound effects that will impact the trajectory of global energy demand, and thus the magnitude of the decarbonization challenge.
The bottom line: It is easy to argue that the burden of proof rests squarely on the shoulders of those who hold that new lighting technologies will greatly reduce energy consumption to explain why the future will be so much different from hundreds of years of recorded history of human behavior. The future might indeed differ from the past. We have no way of knowing. But the future might also closely resemble the past, and if it does, we have serious reasons to rethink and re-estimate the contribution of lighting efficiency to global climate mitigation efforts.
Harry Saunders earns his keep as a management consultant, helping executive teams improve their decision quality. But his hobby, and passion, is academic research – in energy and sustainability economics, evolutionary biology, and law of evidence theory. His wife and four children believe this is a ridiculous hobby.
Jesse Jenkins is a researcher, analyst, and writer with expertise in energy systems and climate change, electric power systems, energy policy, and innovation policy. He is currently a Digital Strategy Consultant and Featured Columnist at TheEnergyCollective.com. Jesse holds a MS in Technology & Policy from the Massachusetts Institute of Technology, where he is currently pursuing a PhD in Engineering Systems and researching the future of the electricity system.
Alcott B. (2005). Jevons’ paradox. Ecological Economics. 54(1).
Day, R. (2014). In defense of LEDs: why the rebound effect for solid-state lighting may be overblown. TheEnergyCollective.com.
The Economist (2010). Not such a bright idea. August 26, 2010.
Fouquet, R. and P. J.G. Pearson (2012) “The long run demand for lighting: elasticities and rebound effects in different phases of economic development.” Economics of Energy and Environmental Policy 1(1) 83-100.
Rosenberg N. (1982). Inside the Black Box: Technology and Economics. Cambridge University Press
Saunders, H.D. and Tsao, J.Y. (2012). Rebound effects for lighting. Energy Policy 49.
Schleich, S., Mills B., Dütschke, E. (2014). A Brighter Future? Quantifying the Rebound Effect in Energy Efficient Lighting. Fraunhofer ISI Working Paper
Shellenberger, M. and Nordhaus, T. (2014) The problem with energy efficiency, New York Times, October 8, 2014.
Tsao, J., Saunders, H.D., Creighton, J.R., Coltrin, M.E., Simmons, J.A. (2010). Solid-state lighting: an energy economics perspective. Journal of Physics D.
Tsao, J.Y. and Waide, P. (2010). The world’s appetite for light: empirical data and trends spanning three centuries and six continents. Leukos 6(4).
Tsao, J.Y. Crawford, M.H., Coltrin, M.E., Fischer, A.J., Koleske, D.D., Subramania, G.S., Wang, G.T., Wierer, J.J., Karlicek, R.F. (2014). Solid-State lighting: toward smart and ultra-efficient solid-state lighting. Advanced Optical Materials. 2(9).
Photo Credit: LEDs and Rebound Effects/shutterstock