What Drives Energy Use? Prices, Policy, and Inefficiency
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Economics. The day in economics: theory, mechanisms and the evidence beneath them.
Ruth: I'm Ruth.
Warren: And I'm Warren, here with Iris.
Ruth: We're looking at the paper: "What Drives Energy Use? Prices, Policy, and Inefficiency".
Warren: Here's the gist of the paper. What drives cross-state differences in U.S.
Ruth: energy consumption?
Warren: We combine LMDI decomposition, stochastic frontier analysis, and variable-importance methods on a panel of 50 states plus DC over the 2006–2022 period The observed 12.8% decline in per capita energy use is driven almost entirely by intensity improvements Efficiency policies operate through two channels: lowering the frontier and compressing inefficiency above it The results suggest that pricing and regulation are the primary drivers of cross-state energy use differences.
Ruth: Decomposition of Energy Trends The paper utilizes Logarithmic Mean Divisia Index (LMDI) decomposition to analyze trends in U.S.
Title and authors: Ruth: So we're looking at this paper called "What Drives Energy Use? Prices, Policy, and Inefficiency". It tackles why per capita energy use varies so much across the fifty U.S. states and D.C. over the two decades from two thousand six to two thousand twenty-two <ref:2604.12112#pg1>.
Warren: That's right, Ruth, and it's a big question because people often think that geographic differences or climate are what cause these huge variations in how much energy folks use per person.
Iris: Exactly. This study tries to figure out how much of that variation is actually due to things like prices and policies versus those basic geographic or economic fundamentals the paper mentions on page three of this work <ref:2604.12112#pg2>.
Ruth: The core finding they present is that the overall decline in per capita energy use across all sectors was almost entirely driven by improvements in energy intensity, which means using less energy for the same amount of activity.
Warren: That intensity effect is what's making up a big chunk of that twelve point eight percent drop between two thousand six and two thousand twenty-two according to the paper on page one of this work.
Iris: It seems they break down how much each sector—commercial, industrial, residential, and transport—contributes to that intensity reduction, showing those percentages adding up to about sixteen percentage points of the total change in energy use >
Ruth: And then they move into the model where they use stochastic frontier analysis to look at the demand side versus supply side forces that shape this consumption.
Warren: They set up a frontier where one side is demand increasing with things like income and climate, and the other side is supply improving because of energy prices and efficiency policies >
Iris: That frontier equation involves an inefficiency index, which they call I sub s, that captures the gap between what's actually being used and what should be used at the frontier level >
Ruth: And then they use variable importance methods to see which forces—prices, efficiency policies, or GDP—are really moving this entire system around.
Warren: The results show that energy prices are identified as the most influential determinant of the frontier energy use, accounting for roughly twenty-six percent of the total variation through that frontier channel alone >
Iris: But they also find that state energy efficiency policies rank second on the frontier at about thirteen percent of variation, but they gain an extra six percentage points by operating through that inefficiency channel >
Ruth: That means efficiency policies are doing two things: they are lowering the actual limit of what's possible on the frontier, and they are also compressing how inefficient everyone is above that limit >
Warren: So when you look at the policy impact simulation, it suggests that if you raise average energy prices by ten percent in states with low prices, you could see a reduction in best-practice per capita energy use of about nine point seven percent >
Iris: That's a pretty strong number they found, but I wonder if that simulation holds up when we consider how people actually react to those price signals or if the policy implementation is as smooth as the model assumes >
Ruth: The paper points out that higher income is actually the strongest predictor of inefficiency, meaning richer states tend to operate further away from their efficiency frontier >
Warren: That makes sense because it suggests that in high-income states, we see more stuff like comfort-driven and rebound-related usage where people use more energy just because they can afford it or want a bigger house >
Iris: They interpret that large gap between observed and frontier energy as being a mix of technological inefficiency, adoption frictions, behavioral responses, and rebound effects rather than just one single distortion >
Ruth: So to wrap up this "What Drives Energy Use? Prices, Policy, and Inefficiency" paper we see that prices and efficiency policies are the dominant predictors of frontier energy use with nearly forty percent of that variation explained >
Warren: It's a clear signal that pricing and regulation are the primary drivers of how energy use differs from state to state >
Iris: I just want to add that their method is quite detailed in how they decompose the variance into frontier, inefficiency, and noise, which helps separate the structural issues from things like random shocks >
Ruth: So for listeners who only listen to this show the main thing is that we've got a pretty clear picture now of what drives these cross-state energy differences based on this study >
Warren: Next up we're going to look at another paper that deals with how social interactions affect choice modeling, which might give us some interesting context on those behavioral responses Iris mentioned earlier >
The paper's summary: Ruth: So, to recap, this paper is really showing us that when we look at energy use across different states over two decades, the biggest driver isn't just where you are geographically or how much money you make overall.
Warren: No, it’s not just geography or income level. It’s about the rules—the prices and the policies—that really steer how much energy a state uses compared to what it could use if there were no rules.
Ruth: Exactly, Warren. They're using this whole framework of frontier and inefficiency to show that those two things—prices and efficiency policies—are actually doing most of the heavy lifting in explaining why one state uses way more energy than another.
Warren: And the numbers back that up by showing that those factors account for nearly forty percent of the variation we see across all states in terms of what's on this frontier.
Ruth: That’s a big piece of information for policymakers because it tells us exactly where to focus our efforts if we want to encourage lower energy use. It supports using price signals and efficiency regulations as the main tools, not just trying to manage the basic economic growth variables.
Warren: The simulation they run is pretty telling too; they showed that if you raise average energy prices by ten percent in states with low prices, you could see a reduction in best-practice per capita energy use of about nine point seven percent.
Ruth: That’s a concrete projection for what those price signals actually do on the ground, and it suggests that regulation is also working through that inefficiency channel. They found that stronger efficiency policies are associated with substantially reduced inefficiency, which means they are both setting a higher bar and making sure everyone stays below it.
Warren: It’s interesting to see how they separate the effects; the frontier part of the variation is mostly driven by those big price signals, but then you get this other part from the efficiency policies that compresses everything above that limit.
Ruth: And there's this important caveat they raise about inequality; they found that higher income states tend to be further away from their efficiency frontier, which points toward comfort-driven usage and rebound effects being strong in richer areas.
Warren: So, for someone just listening to the show, the big picture is that energy use isn't just a function of local conditions but is heavily shaped by how pricing and regulation interact across states.
Ruth: Right. It suggests that if we want to see consistent reductions in consumption, we have to look at how those policy levers are interacting with those economic forces rather than just looking at the raw data points.
Warren: This sets up a good contrast for what's next; we've talked about how policy affects energy use, but now we need to see if people actually *choose* to follow those policies when they have options.
The paper's improvements: Tom: So, we're looking at how this research suggests ways to make it even better than what they did in the original paper about energy use across states.
Ruth: The main improvement they propose is using that dual-channel view more explicitly; instead of just showing the frontier and inefficiency separately, they want to clearly show how efficiency policies work through both channels at the same time.
Warren: That means showing exactly how a policy lowers the theoretical limit of what’s possible while also squeezing the inefficiency gap above that limit.
Iris: I think that's crucial for identification because it lets us pinpoint whether a change in energy use is due to shifting the frontier or just better behavior within that frontier.
Ruth: Right, and they also suggest using those variable importance methods—the LMDI decomposition and the Random Forest permutation importance—to get a clearer picture of which variables are truly driving that variation across states.
Warren: That helps because we know energy prices are key to the frontier variation, but we can also see how policy itself contributes differently to both channels through those methods.
Iris: The paper suggests that by using these techniques, we can move beyond just saying policies *affect* consumption and start modeling precisely *how* they affect the structure of energy demand across different regions.
Ruth: It implies that future studies should focus on integrating those decomposition methods directly into the frontier modeling to get a more granular understanding of how much is structural inefficiency versus policy impact.
Warren: And from a quantitative standpoint, it means we can test if our current assumptions about price elasticity and efficiency gains hold up when we look at the actual dynamics across different states.
Iris: So, the implication for researchers is to move toward models that can handle those complex interactions between the demand side forces and supply side improvements in a more transparent way.
Ruth: It also sets up some interesting avenues for future work, like testing if these policy effects hold true when you introduce more dynamic elements into the frontier equation over a longer time horizon.
Warren: It's basically saying, "Okay, we found the main drivers; now let's build a model that can actually simulate those policy levers interacting with those forces in real-time across different states."
Conclusion: Tom: So, to wrap up this paper, "What Drives Energy Use? Prices, Policy, and Inefficiency," the main conclusion is that energy prices and efficiency policies are the dominant predictors of frontier energy use across states.
Ruth: Right. They're showing that these two factors account for nearly forty percent of the variation we see between states in terms of what’s on that frontier.
Warren: It really hammers home the point that pricing and regulation are the primary drivers of how much energy people use differently from state to state.
Iris: I think it’s important because it provides a strong empirical basis for using price signals and efficiency regulations to actually promote lower energy consumption targets.
Ruth: Exactly, Iris. It supports using those tools rather than just trying to manage the basic economic growth variables in isolation.
Warren: And the simulation they ran is pretty telling too; if you raise average energy prices by ten percent in states with low prices, you could see a reduction in best-practice per capita energy use of about nine point seven percent.
Iris: That’s a concrete projection for what those price signals actually do on the ground, but we have to remember that the model assumes people react predictably to those prices.
Ruth: Right, and they also found that stronger efficiency policies are associated with substantially reduced inefficiency, which means they are both setting a higher bar and making sure everyone stays below it.
Warren: It’s interesting to see how they separate the effects; the frontier part of the variation is mostly driven by those big price signals, but then you get this other part from the efficiency policies that compresses everything above that limit.
Iris: And there's this important caveat they raise about inequality; higher income states tend to be further away from their efficiency frontier, which points toward comfort-driven usage and rebound effects being strong in richer areas.
Ruth: So, for someone just listening to the show, the big picture is that energy use isn't just a function of local conditions but is heavily shaped by how pricing and regulation interact across states.
Warren: It’s a clear signal that we need to look at those policy levers interacting with economic forces rather than just looking at the raw data points on paper.
Iris: I just want to add that their method is quite detailed in how they decompose the variance into frontier, inefficiency, and noise, which helps separate the structural issues from things like random shocks.
Ruth: So for those of you listening who only catch the highlights, remember this is "What Drives Energy Use? Prices, Policy, and Inefficiency," showing that prices and policies are huge drivers.
Warren: Next up we're going to look at another paper that deals with how social interactions affect choice modeling, which might give us some interesting context on those behavioral responses Iris mentioned earlier.
David Benatia, Remy Molini´e, Pierre-Olivier Pineau
HEC Montréal
econ.GN, q-fin.EC
Submitted: 2026-04-13
Updated: 2026-11-03
License: http://creativecommons.org/licenses/by/4.0/
Importance score: 84/100
The gist: What drives cross-state differences in U.S.
Key concepts
- LMDI Decomposition
- This method is used to break down changes in total per capita energy consumption into contributions from different end-use sectors (like commercial or residential) over time. It helps pinpoint exactly which sectors are driving the overall trend of energy use change.
- Stochastic Frontier Analysis (SFA)
- SFA creates a model to estimate an ideal 'frontier' of energy use based on factors like income and climate. The gap between actual state energy use and this frontier is measured as inefficiency, helping researchers separate technological limits from other causes.
- Frontier vs. Inefficiency
- The frontier represents the best possible energy use given current technology and conditions. Inefficiency is the gap between what states actually consume and this ideal frontier. The study shows that policies affect both: lowering the frontier (best practice) and reducing inefficiency above it.
- Pricing and Policy Channels
- Energy prices are shown to be a major driver, influencing the frontier by encouraging better energy use. Efficiency policies also work through two channels: they improve the frontier directly and reduce inefficiency, making them key tools for managing energy consumption differences across states.
Terminology
Summary
What drives cross-state differences in U. p. 1 What Drives Energy Use? Prices, Efficiency Policies, and
the Demand Frontier
David Benatiaa,∗
, Remy Molini´eb
, Pierre-Olivier Pineaub
aHEC Montr´eal, D´epartement d’Economie Appliqu´ee, 3000 Chemin de la ´
Cˆote-Sainte-Catherine, Montr´eal, H3T 2A7, QC, Canada
bHEC Montr´eal, D´epartement de Sciences de la D´ecision, 3000 Chemin de la
Cˆote-Sainte-Catherine, Montr´eal, H3T 2A7, QC, Canada
Abstract
What drives cross-state differences in U. per capita energy use fell 12. 8% between 2006 and 2022, driven almost
entirely by intensity reductions rather than activity or climate
• Energy prices and efficiency policies are the key drivers: nearly 40% of
cross-state frontier variation
• Efficiency policies operate through two channels: lowering the frontier and
compressing inefficiency above it
• GDP per capita works in the opposite direction: its main contribution is
to widen the gap between observed consumption and the frontier
• Ranking of drivers is robust across LMG and Random Forest importance
measures
arXiv:2604. We combine
LMDI decomposition, stochastic frontier analysis, and variable-importance methods on a panel of 50 states plus DC over the 2006–2022 period. The observed
12. Efficiency policies also operate through a
second channel by reducing inefficiency, adding a further 6 percentage points to
their total contribution. The results suggest that pricing and regulation are the
primary drivers of cross-state energy use differences. per capita energy use fell 12. 8% between 2006 and 2022, driven almost
entirely by intensity reductions rather than activity or climate
• Energy prices and efficiency policies are the key drivers: nearly 40% of
cross-state frontier variation
• Efficiency policies operate through two channels: lowering the frontier and
compressing inefficiency above it
• GDP per capita works in the opposite direction: its main contribution is
to widen the gap between observed consumption and the frontier
• Ranking of drivers is robust across LMG and Random Forest importance
measures
arXiv:2604.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Decomposition of Energy Trends
The paper utilizes Logarithmic Mean Divisia Index (LMDI) decomposition to analyze trends in U.S. per capita energy consumption across end-use sectors over the 2006–2022 period. p. 3 Our LMDI decomposition confirms this
pattern: the observed decline in U. per capita energy use between 2006 and
2022 is driven almost entirely by intensity reductions across all end-use sectors. The discrete LMDI
decomposition in Figure 2 shows that this decline is overwhelmingly driven by
reductions in energy intensity across all four end-use sectors. The intensity effects in the commercial, industrial, residential, and transport sectors contribute
about −5. 4, and −2. , respectively, so that aggregate intensity
improvements account for roughly 16 p. fuel extraction is associated with a marked rise in energy use. Figure 1: Per capita energy consumption in 2022 and change in per capita energy consumption
by state and for the world, 2006–2022. By contrast,
the intensity index exhibits a marked downward trend. It gradually falls from
1 in 2006 to close to 0. It gradually falls from
1 in 2006 to close to 0. 86 in 2020 and continues to decline to roughly 0. per capita energy use fell 12. 8% between 2006 and 2022, driven almost
entirely by intensity reductions rather than activity or climate
• Energy prices and efficiency policies are the key drivers: nearly 40% of
cross-state frontier variation
• Efficiency policies operate through two channels: lowering the frontier and
compressing inefficiency above it
• GDP per capita works in the opposite direction: its main contribution is
to widen the gap between observed consumption and the frontier
• Ranking of drivers is robust across LMG and Random Forest importance
measures
arXiv:2604.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Economic Framework and Model Specification
The study develops an economic framework based on stochastic frontier analysis (SFA) to decompose observed per capita energy use in state s as qs = q∗s / Is, where q∗s is a conditional energy-demand frontier. The frontier depends on two forces operating in opposite directions: demand side needs increasing with income and climate, and supply side best-practice efficiency improving with energy prices and policies. p. 0 per capita energy use fell 12. 8% between 2006 and 2022, driven almost
entirely by intensity reductions rather than activity or climate
• Energy prices and efficiency policies are the key drivers: nearly 40% of
cross-state frontier variation
• Efficiency policies operate through two channels: lowering the frontier and
compressing inefficiency above it
• GDP per capita works in the opposite direction: its main contribution is
to widen the gap between observed consumption and the frontier
• Ranking of drivers is robust across LMG and Random Forest importance
measures
arXiv:2604. , 2016, 2019). As with any index-based decomposition,
however, LMDI attributes all observed changes to a small set of pre-defined effects
without a residual term, so its scope for economic interpretation is inherently
limited. per capita energy use fell 12. 8% between 2006 and 2022, driven almost
entirely by intensity reductions rather than activity or climate
• Energy prices and efficiency policies are the key drivers: nearly 40% of
cross-state frontier variation
• Efficiency policies operate through two channels: lowering the frontier and
compressing inefficiency above it
• GDP per capita works in the opposite direction: its main contribution is
to widen the gap between observed consumption and the frontier
• Ranking of drivers is robust across LMG and Random Forest importance
measures
arXiv:2604. To our knowledge, no existing study
quantifies the relative contribution of specific explanatory variables to an estimated energy-demand frontier. We do so by applying both LMG and Random
Forest importance measures to the frontier and inefficiency components of our
SFA model.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Variable Importance and Contribution
The authors employ both Logarithmic Mean Divisia Index (LMG) decomposition and Random Forest permutation importance to quantify the relative contribution of specific explanatory variables to the estimated energy-demand frontier and inefficiency term. The point estimates for the variance decomposition show that the frontier accounts for nearly two thirds of the cross-state variation in log energy use, while noise is negligible. p. 1 A variance decomposition attributes 63% of cross-state variation in
log energy use to the demand frontier, 34% to inefficiency above it, and 3% to
noise. Within the frontier, energy prices account for roughly 26% of cross-state
variation and state efficiency policies for about 13%, while GDP and climate
together explain only around 10%. Within the frontier, energy prices account for roughly 26% of cross-state
variation and state efficiency policies for about 13%, while GDP and climate
together explain only around 10%. Efficiency policies also operate through a
second channel by reducing inefficiency, adding a further 6 percentage points to
their total contribution. Within the frontier, energy prices account for roughly 26% of cross-state
variation and state efficiency policies for about 13%, while GDP and climate
together explain only around 10%. Efficiency policies also operate through a
second channel by reducing inefficiency, adding a further 6 percentage points to
their total contribution.Where the paper says it
Where the paper says it
Where the paper says it
Policy Implications and Findings
The analysis confirms that energy prices and efficiency policies are the dominant predictors of frontier energy use, accounting for nearly 40% of cross-state frontier variation. The results suggest that pricing and regulation are the primary drivers of cross-state energy use differences. Specifically, a 10% increase in average energy prices is associated with approximately a 9. p. 5 We do so by applying both LMG and Random
Forest importance measures to the frontier and inefficiency components of our
SFA model. Conditional on year fixed effects, energy prices account for approximately 26% of cross-state frontier variation and state energy-efficiency policies
for about 13%, while GDP and climate together account for roughly 10%. fuel extraction is associated with a marked rise in energy use. Figure 1: Per capita energy consumption in 2022 and change in per capita energy consumption
by state and for the world, 2006–2022. Therefore, the aggregate 12. 8% decline in per capita energy use is more than
fully accounted for by within-sector intensity improvements, while changes in
activity and climate largely cancel out, providing little net contribution to the
observed reduction. per capita energy use fell 12. 8% between 2006 and 2022, driven almost
entirely by intensity reductions rather than activity or climate
• Energy prices and efficiency policies are the key drivers: nearly 40% of
cross-state frontier variation
• Efficiency policies operate through two channels: lowering the frontier and
compressing inefficiency above it
• GDP per capita works in the opposite direction: its main contribution is
to widen the gap between observed consumption and the frontier
• Ranking of drivers is robust across LMG and Random Forest importance
measures
arXiv:2604.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Equity Considerations
The paper notes that higher income is the strongest predictor of inefficiency (γY = 3.61), implying that richer states tend to operate further from their frontier. p. 0 per capita energy use fell 12. 8% between 2006 and 2022, driven almost
entirely by intensity reductions rather than activity or climate
• Energy prices and efficiency policies are the key drivers: nearly 40% of
cross-state frontier variation
• Efficiency policies operate through two channels: lowering the frontier and
compressing inefficiency above it
• GDP per capita works in the opposite direction: its main contribution is
to widen the gap between observed consumption and the frontier
• Ranking of drivers is robust across LMG and Random Forest importance
measures
arXiv:2604. We anchor the specification in a parsimonious economic framework in which
the frontier represents the minimum energy consumption consistent with a baseline level of energy services, conditional on state characteristics. The inefficiency
term is interpreted as a composite that absorbs technological and allocative inefficiency, adoption frictions, preference-driven usage, and rebound responses—a
framing that avoids attributing the entire gap to any single channel. We anchor the specification in a parsimonious economic framework in which
the frontier represents the minimum energy consumption consistent with a baseline level of energy services, conditional on state characteristics. The inefficiency
term is interpreted as a composite that absorbs technological and allocative inefficiency, adoption frictions, preference-driven usage, and rebound responses—a
framing that avoids attributing the entire gap to any single channel. We anchor the specification in a parsimonious economic framework in which
the frontier represents the minimum energy consumption consistent with a baseline level of energy services, conditional on state characteristics. The inefficiency
term is interpreted as a composite that absorbs technological and allocative inefficiency, adoption frictions, preference-driven usage, and rebound responses—a
framing that avoids attributing the entire gap to any single channel.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Conclusion
The main finding is that energy prices and efficiency policies are the dominant predictors of frontier energy use, accounting for nearly 40% of cross-state frontier variation. This finding provides direct support for energy and climate policies using price signals and energy efficiency regulation to promote lower energy consumption. p. 5 Conditional on year fixed effects, energy prices account for approximately 26% of cross-state frontier variation and state energy-efficiency policies
for about 13%, while GDP and climate together account for roughly 10%. These
rankings are robust across both decomposition methods, indicating that pricing
and regulation are the largest contributors to cross-state variation in frontier per
capita energy use.Where the paper says it
Acknowledgements
We acknowledge the funds made available by the Chair in Energy Sector Management (HEC Montr´eal) to support this research.
References
Aigner, D., Lovell, C. p. 4 1The distributional assumptions on inefficiency in these models trace back to Aigner et al. (1977) and Battese and Coelli (1995); for comprehensive treatments see Kumbhakar and Lovell
(2000) and Kumbhakar et al. 1The distributional assumptions on inefficiency in these models trace back to Aigner et al. (1977) and Battese and Coelli (1995); for comprehensive treatments see Kumbhakar and Lovell
(2000) and Kumbhakar et al.Where the paper says it
Where the paper says it
Allcott, H., Greenstone, M., 2012
Ang, B., 2005
Ang, B.W., 2015
Barrera-Santana, J., Marrero, G.A., Ramos-Real, F. p. 3 Ang
(2005) provides a practical guide to the Logarithmic Mean Divisia Index (LMDI)
to decompose changes in energy or emissions indicators into activity, structural,
and intensity components; subsequent applications to U. data at the national
and state levels show that most long-run reductions in energy intensity reflect
efficiency gains rather than structural change, with higher real energy prices
playing a central role (Metcalf, 2008). Ang
(2005) provides a practical guide to the Logarithmic Mean Divisia Index (LMDI)
to decompose changes in energy or emissions indicators into activity, structural,
and intensity components; subsequent applications to U. data at the national
and state levels show that most long-run reductions in energy intensity reflect
efficiency gains rather than structural change, with higher real energy prices
playing a central role (Metcalf, 2008).Where the paper says it
Where the paper says it
Barrera-Santana, J., Marrero, G.A., Ramos-Real, F.J., 2022b
Battese, G.E., Coelli, T. p. 4 1The distributional assumptions on inefficiency in these models trace back to Aigner et al. (1977) and Battese and Coelli (1995); for comprehensive treatments see Kumbhakar and Lovell
(2000) and Kumbhakar et al. 1The distributional assumptions on inefficiency in these models trace back to Aigner et al. (1977) and Battese and Coelli (1995); for comprehensive treatments see Kumbhakar and Lovell
(2000) and Kumbhakar et al.Where the paper says it
Where the paper says it
Battese, G.E., Coelli, T.J., 1995
Berg, W., Vaidyanathan, S., Jennings, B., Cooper, E., Perry, C., DiMascio, M., Singletary, J., 2020
Filippini, M., Hunt, L., 2011
Filippini, M., Hunt, L., 2012
Filippini, M., Hunt, L.C., 2015
Filippini, M., Hunt, L.C., 2016
Gerres, T., Chaves Avila, J.P., Llamas, P.L., San Rom´an, T.G., 2019
Gr¨omping, U., 2006
Gr¨omping, U., 2007
Gr¨omping, U., 2009
Heun, M.K., 2024
Jiang, R., Li, R., Wu, Q., 2019
Jiang, X.T., Dong, J.F., Wang, X.M., Li, R.R., 2016
Korsavi, S., Azari, R., Iulo, L., Mahdavi, M., 2025
Kumbhakar, S.C., Lovell, C.A.K., 2000
Kumbhakar, S.C., Wang, H.J., Horncastle, A.P., 2015
Metcalf, G.E., 2008
Orea, L., Llorca, M., Filippini, M. p. 3 A second strand addresses this limitation by using stochastic frontier analysis (SFA) to estimate aggregate energy-demand frontiers and to interpret deviations from the frontier as inefficient use. Filippini and Hunt (2011) estimate
an economy-wide frontier for 29 OECD countries, and Filippini and Hunt (2012)
apply a similar approach to U.Where the paper says it
Reuter, M., Patel, M.K., Eichhammer, W., 2019
Serrano-Puente, D., 2021
Strobl, C., Boulesteix, A.L., Kneib, T., Augustin, T., Zeileis, A., 2008
U.S. Energy Information Administration, 2023
World Bank, 2025
Wright, M.N., Ziegler, A., 2017
Appendices
Appendix A provides more detailed information about our dataset
Appendix D details the LMDI decomposition methodology
Appendix E describes variable importance implementation details
Appendix F provides description of units and transformations
Appendix G covers the ACEEE Scorecard construction and econometric considerations
Appendix H presents additional figures
Appendix I contains robustness checks
Table 5 summarizes the units and transformations applied to each variable in the SFA frontier equation.
Table 6 reports the stochastic frontier estimation results. p. 1 The results suggest that pricing and regulation are the
primary drivers of cross-state energy use differences. Keywords: Energy efficiency, Stochastic frontier analysis, LMDI
decomposition, Variable importance, Energy policy, US states
JEL: Q41, Q48, C24, D24
∗Corresponding author
Email addresses: david.Where the paper says it
Table 7 reports subset robustness key coefficient comparison. p. 4 The inefficiency
term is interpreted as a composite that absorbs technological and allocative inefficiency, adoption frictions, preference-driven usage, and rebound responses—a
framing that avoids attributing the entire gap to any single channel. Yet even
within this strand, existing studies report coefficient estimates and discuss their
signs without quantifying how much of the cross-state variation each determinant
accounts for.Where the paper says it
Figure 5 illustrates the empirical model and this dual-channel schematically
Figure 6 summarizes these results into a sunburst decomposition of variance.
Table 9 reports summary statistics for the five most and five least efficient states.
Figure 7 illustrates the LMG decomposition
Figure 8 illustrates the RF permutation importance.
Table 8 reports contribution of each variable. p. 2 Our
results challenge this view: energy prices and efficiency policies emerge as the
most important predictors of cross-state variation in frontier energy use, ranking
well ahead of GDP and climate. We quantify each variable’s contribution. Our
results challenge this view: energy prices and efficiency policies emerge as the
most important predictors of cross-state variation in frontier energy use, ranking
well ahead of GDP and climate. We quantify each variable’s contribution.Where the paper says it
Where the paper says it
Figure 9 plots mean SFA efficiency vs. p. 2 Cross-state differences in per capita energy consumption in enduse sectors across the 50 U. states and the District of Columbia over 2006–2022
are investigated, combining Logarithmic Mean Divisia Index (LMDI) decomposition, stochastic frontier analysis (SFA), and variable-importance methods. Cross-state differences in per capita energy consumption in enduse sectors across the 50 U. states and the District of Columbia over 2006–2022
are investigated, combining Logarithmic Mean Divisia Index (LMDI) decomposition, stochastic frontier analysis (SFA), and variable-importance methods. per capita energy use fell 12. 8% between 2006 and 2022, driven almost
entirely by intensity reductions rather than activity or climate
• Energy prices and efficiency policies are the key drivers: nearly 40% of
cross-state frontier variation
• Efficiency policies operate through two channels: lowering the frontier and
compressing inefficiency above it
• GDP per capita works in the opposite direction: its main contribution is
to widen the gap between observed consumption and the frontier
• Ranking of drivers is robust across LMG and Random Forest importance
measures
arXiv:2604. per capita energy use fell 12. 8% between 2006 and 2022, driven almost
entirely by intensity reductions rather than activity or climate
• Energy prices and efficiency policies are the key drivers: nearly 40% of
cross-state frontier variation
• Efficiency policies operate through two channels: lowering the frontier and
compressing inefficiency above it
• GDP per capita works in the opposite direction: its main contribution is
to widen the gap between observed consumption and the frontier
• Ranking of drivers is robust across LMG and Random Forest importance
measures
arXiv:2604.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Figure 10 presents geographic distribution of mean SFA efficiency scores. p. 2 states and the District of Columbia over 2006–2022
are investigated, combining Logarithmic Mean Divisia Index (LMDI) decomposition, stochastic frontier analysis (SFA), and variable-importance methods. A
common view in the policy debate is that differences in energy use across regions
largely reflect geographic and economic fundamentals, such as climate, industrial
structure, and income levels, leaving limited room for policy to close the gap.Where the paper says it
Table 2 reports baseline vs Mundlak adjustment coefficient comparison. p. 4 Filippini and Hunt (2016) extend this line by distinguishing persistent from transient efficiency in U. states,
using a Mundlak adjustment to separate time-invariant unobserved heterogeneity from genuine inefficiency.Where the paper says it
Table 3 reports summary statistics for the five most and five least efficient states.
Figure 9 plots mean SFA efficiency vs. Effs by state.
Figure 5 illustrates the empirical model and this dual-channel schematically
Figure 6 summarizes these results into a sunburst decomposition of variance.
Table 8 reports contribution of each variable.
Figure 7 illustrates the LMG decomposition
Figure 8 illustrates the RF permutation importance.
Table 9 reports summary statistics for the five most and five least efficient states.
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Variable Importance Ranking for Policy Impact: The improved AI system can rank policy drivers using both Log Mean Divisia Index (LMDI) and Random Forest permutation importance to determine which variables drive cross-state variation in frontier energy use, as shown in Section 5.3.
-
Frontier and Inefficiency Decomposition: The system can decompose the variance of log energy use into three components—frontier, inefficiency, and noise—to quantify how much of the gap between observed and conditional demand is due to technological inefficiency versus other factors like adoption frictions or rebound effects.
-
Policy Channel Analysis: The AI can analyze how efficiency policies operate through two distinct channels by examining their coefficients in both the frontier equation (βEff) and the inefficiency equation (γEff), allowing it to distinguish between shifting the minimum energy use and compressing the distribution of inefficiency above it.
-
Policy Recommendation Simulation: The system can simulate policy impacts by using pooled cross-sectional coefficients to illustrate potential magnitudes, such as how raising energy prices in low-price states could reduce frontier energy use by approximately 14.5% in those states.
Abstract
Why is U.S. energy use so high, even among rich economies? Using a 2006--2022 panel of the 50 states and DC, an LMDI decomposition documents a 12.8% national decline alongside substantial cross-state heterogeneity. To explain differences in energy-use levels across states, we estimate a microfounded stochastic-frontier model and adapt the LMG relative-importance method to the frontier and inefficiency components. Our approach assigns 56.3% of variation to the frontier, 42.2% to inefficiency, and 1.5% to noise. Prices and efficiency-policy environments are the two leading observed contributors, jointly accounting for 31%: prices operate mainly through the benchmark, whereas policy contributes through both components. Their leading ranks are stable across 21 specifications. Persistent low prices, uneven efficiency-policy environments, structural demand, and a sizeable composite inefficiency component help explain high U.S. energy use. Policy should therefore combine durable price signals, coherent efficiency programs, structural investment, targeted measures addressing avoidable energy use, and targeted household support.
Related papers
- Huge If True: Learning from the Size of the Surprise
- The Rebundling of Freelance Work in the AI Age
- Dynamic investment and coordination in teams: Theory and experimental evidence
- Does Frontier AI Strengthen Cyber Defense? Evidence from Project Glasswing
- Generative AI at Work
- What is Work For? Labor, Politics, and Meaning in the Age of AI