Do Places or People Matter More in Determining Household Carbon Emissions?

This study investigates whether household carbon emissions are primarily driven by household characteristics or by the places people live. Using a 20-year panel of Census and ACS data on over one million individuals and a mover design, the author estimates place effects for nearly 1,000 cities and 61,500 neighborhoods. The analysis shows that moving to a different city shifts household emissions by 85% of the origin-destination mean difference; neighborhood moves yield a 53–60% shift. Place effects explain 14–23% of overall emissions variation. A one standard deviation reduction in neighborhood place effects could lower household emissions by approximately 40%.

Full Citation and Link to Article

Lyubich, E. (2024). The role of people vs. places in individual carbon emissions. American Economic Journal: Applied Economics. Advance online publication. https://doi.org/10.1257/app.20220383

Extended Summary

Central Research Question

This study examines the extent to which individual household carbon emissions are determined by where people live (place effects) versus who they are (people effects). The central question is: to what degree do neighborhood or city characteristics causally shape a household’s residential energy use and transportation-related carbon emissions, compared to the role of household-level preferences and demographics? The research seeks to disentangle the contribution of urban form, infrastructure, and environmental factors from household sorting and behavioral differences. The answer informs whether urban planning and housing policies that alter “place” can lead to meaningful reductions in carbon emissions.

Previous Literature

Prior studies have highlighted large spatial differences in household carbon emissions, emphasizing the role of urban density, transportation infrastructure, and local climate (e.g., Jones and Kammen 2014; Ummel 2014; Green and Knittel 2020). However, those studies largely relied on static, observational data and could not separate the causal effect of place from self-selection: people who care about the environment may choose to live in low-emission cities. This paper builds on econometric techniques developed in labor economics and public finance that use mover designs to estimate place effects on wages, health, and mobility (e.g., Chetty and Hendren 2018; Finkelstein et al. 2016). It applies a two-way fixed effects model, previously used to estimate firm-level wage effects, to measure geographic effects on household carbon emissions for the first time.

Data

The study uses restricted-access data from the U.S. Census Bureau, linking 20 years of Decennial Census long-form and American Community Survey (ACS) microdata for over 17 million individuals and 12 million households. A panel is constructed for approximately one million individuals who are observed at least twice in different years, allowing for the identification of movers across places. Carbon emissions are estimated from household-reported residential energy expenditures (electricity, natural gas, and heating oil) and commuting behavior (commute distance, frequency, and mode of transportation).

To estimate emissions, the study combines Census data with:

• Utility price data from the Energy Information Administration.

• Emissions factors from the Environmental Protection Agency.

• Commute emissions from transportation mode and distance calculations.

• Local climate data from NOAA.

• Walk, bike, and transit scores from Walk Score.

The geographic analysis is conducted at two levels:

1. Core-Based Statistical Areas (CBSAs), approximating metropolitan regions (~1,000 units).

2. Census tracts, serving as proxies for neighborhoods (~61,500 units).

The sample includes approximately 250,000 mover households who change either city or neighborhood.

Methods

The core methodological approach is a mover design paired with a two-way fixed effects model. This design exploits variation in emissions for the same household before and after moving to different locations, controlling for all time-invariant household characteristics (αᵢ) and time-varying observables (e.g., income, household size). The specification is:

ln(CO₂ᵢₜ) = αᵢ + ψⱼ(ᵢₜ) + Xᵢₜβ + τₜ + εᵢₜ

Where:

• CO₂ᵢₜ is household i’s log carbon emissions at time t.

• ψⱼ(ᵢₜ) is the place (city or tract) effect.

• Xᵢₜ includes observed household characteristics.

• τₜ are year fixed effects.

Two decomposition strategies are employed:

1. Event Study Approach: Estimates how emissions change after a household moves, relative to the difference in average emissions between the origin and destination. If emissions converge toward the destination average, this suggests a causal place effect.

2. Variance Decomposition: Using a leave-one-out estimator (Kline, Saggio, and Sølvsten 2020), the study calculates the proportion of total variance in household emissions attributable to place effects versus household fixed effects.

The study tests for symmetric treatment effects (i.e., moves from high- to low-emission places reduce emissions by the same amount as reverse moves increase them) and for robustness to sorting based on unobservable characteristics.

Findings/Size Effects

The results show that place plays a substantial causal role in determining household carbon emissions.

1. Event Study Results:

   
   

   • When households move across cities, their carbon emissions shift by 85% of the average difference between the origin and destination city.

   • When households move across neighborhoods, their emissions shift by 53–60% of the neighborhood-level average difference.

   
   

   These findings indicate that place effects are strong, particularly across cities, though neighborhood-level sorting is more influential than at the city level.

2. Variance Decomposition:

   
   

   • City-level place effects explain 14–16% of the total variation in household carbon emissions.

   • Neighborhood-level place effects explain 22–23%.

   • Observable household characteristics explain another 40–45%, while the remainder is due to unobserved household fixed effects.

   
   

3. Policy Simulation:

   
   

   • Moving a household from a neighborhood one standard deviation above the national mean place effect to one standard deviation below would reduce household carbon emissions by 40% on average.

   • If all suburban and rural households lived in their nearest principal city (based on average place effects), emissions from residential energy and commuting would decline by 15%—comparable in magnitude to the projected emissions reductions from the 2022 Inflation Reduction Act.

   
   

4. Correlates of Place Effects:

   
   

   • Low-emissions places tend to have:

     
     

     • Higher population density

     • Smaller home sizes

     • Better access to bike and pedestrian infrastructure

     • Cleaner electricity generation (e.g., renewables over coal)

     • Milder climates

     
     

   • Walk Score, Bike Score, and Transit Score are all negatively correlated with place effects: more walkable and transit-rich neighborhoods produce lower emissions.

   
   

5. Sorting and Robustness:

   
   

   • No evidence was found of systematic sorting of households with changing environmental preferences. Symmetry in emissions changes across origin-destination directions supports the validity of the fixed-effects model.

   • Results are robust to controlling for household changes in income, household size, and homeownership.

Conclusion

The paper concludes that place matters: differences in local infrastructure, urban form, climate, and energy supply causally shape household carbon emissions. Even when controlling for a rich set of observable and unobservable household characteristics, geographic context accounts for a meaningful share of variation in emissions.

This has important implications for climate policy. While traditional carbon taxes and clean energy subsidies target the demand side (people), the findings suggest that place-based policies targeting the built environment—such as upzoning, public transit expansion, and clean energy infrastructure—can effectively reduce emissions at scale. Since these interventions affect all households in a location, they offer the potential for broad-based impact, especially in areas where individual behavior is constrained by the lack of low-emission options.

The study does not provide welfare or cost-benefit estimates of such policies, nor does it isolate the causal impact of specific place-based features (e.g., bike lanes or density). However, it lays the empirical groundwork for evaluating place-based climate policies as a complement to traditional economic instruments. It also informs the design of geographically differentiated climate policy, including questions around redistribution and environmental justice.

Overall, this work brings new insight to the longstanding debate on the role of individual behavior versus systemic constraints in climate action. It shows that carbon emissions are not solely the result of individual preferences or income but are deeply shaped by place—suggesting that where we live matters profoundly for the planet.