SPATIAL ENERGY INEQUALITY AND THE PURSUIT OF JUST DECARBONIZATION: EXAMINING ENERGY BURDEN, EFFICIENCY, AND COMMUNITY STRUCTURE IN ITHACA’S SECTOR A Thesis Presented to the Faculty of the Graduate School of Cornell University In Partial Fulfillment of the Requirements for the Degree of Master of Science by Yanbei Liu December 2025 Copyright © 2025 by Yanbei Liu ALL RIGHTS RESERVED This thesis is submitted in partial fulfillment of the requirements for the degree of Master of Science in Systems Engineering at Cornell University. No part of this thesis may be reproduced, stored in a retrieval system, or transmitted in any form or by any means—electronic, mechanical, photocopying, recording, or otherwise—without the prior written permission of the author. ABSTRACT Urban decarbonization efforts increasingly recognize that achieving climate goals requires addressing uneven energy costs and efficiency conditions across communities. Yet fine-scale empirical evidence—especially for small and mid-sized cities—remains limited. This study develops a scalable analytical framework to evaluate neighborhood-level energy burden and residential energy efficiency, integrating principal component analysis, clustering, spatial hotspot detection, and spatial regression. Applied to Ithaca, New York, a city pursuing carbon neutrality by 2030, the framework identifies five neighborhood types and reveals pronounced spatial clustering shaped by income disparities, rental concentration, and building age. A key finding is that higher energy efficiency does not always correspond to lower energy burden; in several tracts, increased electrification without efficiency upgrades heightens cost stress. This decoupling underscores the need to consider both affordability and technical efficiency quality in local decarbonization planning. The results provide tract-level insights that can inform more equitable and place-sensitive energy transition strategies. Keywords: energy burden; residential energy efficiency; spatial analysis; energy justice iii BIOGRAPHICAL SKETCH Yanbei Liu earned her Bachelor of Engineering in Chemical Engineering from Tsinghua University, where her academic and research interests centered on emerging decarbonization technologies. During her undergraduate studies, she explored the integration of blockchain, Internet of Things (IoT) systems, and digital-twin frameworks to support low-carbon industrial transitions. She also developed a strong interest in modeling chemical reactions related to decarbonization pathways, focusing on how computational methods can optimize energy efficiency and reduce emissions in traditional chemical processes. She is currently completing her Master of Science in Systems Engineering at Cornell University, with a minor in Electrical and Computer Engineering. At Cornell, her work emphasizes data analytics, systems modeling, and the application of quantitative methods to support sustainable energy planning. Her graduate training combines technical engineering methods with interdisciplinary systems thinking, enabling her to analyze complex urban energy systems and evaluate the socioeconomic dimensions of decarbonization. Her research interests span spatial equity in energy systems, data-driven modeling of energy burden and energy efficiency, and the design of technology-enabled decarbonization strategies for cities. She is particularly interested in how digital tools—such as high-resolution data models, electrification simulations, and digital twins—can support equitable and effective climate- transition planning. iv ACKNOWLEDGMENTS I would like to express my deepest gratitude to my advisor, Professor Semida Silveira, for her patient guidance, insightful feedback, and unwavering support throughout this research. Her mentorship has greatly expanded my perspectives on energy policy, energy efficiency, and the broader challenges of sustainable transitions. I am especially grateful for her encouragement to explore innovative ideas and her continuous push to think critically and rigorously. I am also sincerely thankful to my minor advisor, Professor Lang Tong, whose teaching introduced me to essential concepts in electrical and computer engineering. His instruction on power markets and the structure of the U.S. energy system has been invaluable to my academic development and has profoundly shaped the analytical foundation of this thesis. Their expertise, encouragement, and generosity have made this work possible, and I am truly grateful for the opportunity to learn from both of them. I would additionally like to thank my family—Li Xiong, Tao Liu, and Jiaming Liu—for their unconditional love and encouragement. Their support has always been my strongest source of strength. I am also grateful to my friends, Fanxin Gong and Yihan Wang, whose companionship made the cold and lengthy winters in Ithaca feel much warmer. Their presence and support have sustained me through the most challenging periods of graduate study. To all who have accompanied me on this journey, thank you. This work would not have been possible without you. v TABLE OF CONTENTS 1. INTRODUCTION ................................................................................................................................ 1 1.1 BACKGROUND AND MOTIVATION ........................................................................................................ 1 1.2 PROBLEM STATEMENT ......................................................................................................................... 3 1.3 RESEARCH QUESTION AND OBJECTIVES ............................................................................................. 3 1.4 THESIS STRUCTURE OVERVIEW .......................................................................................................... 4 2. LITERATURE REVIEW .................................................................................................................... 4 2.1 ENERGY JUSTICE: FOUNDATIONS, PRINCIPLES, AND RELEVANCE .................................................... 5 2.2 ENERGY POVERTY, ENERGY BURDEN, AND THEIR PRIMARY DETERMINANTS ................................ 7 2.2.1 Conceptualizing Energy Poverty in Developed Contexts ............................................................ 7 2.2.2 Drivers of Energy Poverty: Income, Housing, and Prices ......................................................... 8 2.3 RESIDENTIAL ENERGY EFFICIENCY: MEASUREMENT, DRIVERS, AND POLICY IMPLICATIONS ....... 9 2.3.1 Measurement Approaches ............................................................................................................ 9 2.3.2 Drivers of Energy Efficiency...................................................................................................... 10 2.3.3 Efficiency, Electrification, and Affordability ............................................................................ 10 2.4 SOCIOECONOMIC AND HOUSING DETERMINANTS OF ENERGY OUTCOMES .................................... 11 2.5 SPATIAL ANALYTICAL APPROACHES IN ENERGY RESEARCH .......................................................... 12 2.6 RESEARCH GAP: THE NEED FOR FINE-SCALE, CONTEXT-SPECIFIC ANALYSIS IN SMALL CITIES 13 3. DATA AND STUDY SCOPE ............................................................................................................ 14 3.1 STUDY SCOPE .................................................................................................................................. 14 3.2 DATA SOURCE ................................................................................................................................. 15 3.3 VARIABLE CONSTRUCTION ............................................................................................................ 16 3.3.1 Data Structure ....................................................................................................................... 16 3.3.2 Socioeconomic and Housing Characteristic ........................................................................ 17 3.3.3 Energy Burden Proxy (EBP) ................................................................................................ 18 3.3.4 Energy Efficiency Proxy (EFP) ........................................................................................... 18 3.4 DATA PRE-PROCESSING ................................................................................................................. 19 3.5 DATA QUALITY AND LIMITATIONS ................................................................................................ 20 3.6 REPRODUCIBILITY AND SOFTWARE ............................................................................................... 20 4. METHODOLOGY AND RESULTS ................................................................................................ 21 4.1 ANALYTICAL FRAMEWORK............................................................................................................ 21 4.2 MODULE 1 – SOCIOECONOMIC STRUCTURE.................................................................................. 23 4.2.1 Correlations ................................................................................................................................ 23 4.2.2 PCA Eigenvectors ....................................................................................................................... 26 4.2.3 K-means clustering ..................................................................................................................... 33 4.2.4 Summary ..................................................................................................................................... 37 vi 4.3 MODULE 2 – ENERGY BURDEN (EBP) ........................................................................................... 37 4.3.1 Spatial Analysis ..................................................................................................................... 37 4.3.2 Association with Socioeconomic Factors ............................................................................. 38 4.4 MODULE 3 – ENERGY EFFICIENCY PROXY (EFP) ........................................................................ 41 4.4.1 Association with Socioeconomic Factors .................................................................................. 42 4.4.2 Spatial Distribution of Energy Efficiency ................................................................................. 44 4.4.3 Relationship Between Energy Efficiency and Energy Burden ................................................ 45 4.4.4 Summary ..................................................................................................................................... 46 4.5 SYNTHESIS OF FINDINGS................................................................................................................. 47 5. DISCUSSION ..................................................................................................................................... 48 5.1 POLICY IMPLICATIONS: TARGETED STRATEGIES FOR ITHACA’S ENERGY TRANSITION ........... 48 5.2 METHODOLOGICAL LIMITATIONS ................................................................................................. 52 5.3 FUTURE DIRECTIONS ...................................................................................................................... 53 6. CONCLUSION ................................................................................................................................... 54 REFERENCES ............................................................................................................................................. 57 vii LIST OF FIGURES FIGURE 1. OVERVIEW OF ITHACA CENSUS TRACTS LEVEL DATASET: SOCIOECONOMIC FACTORS, ENERGY BURDEN PROXY, AND ENERGY EFFICIENCY PROXY .............................................. 17 FIGURE 2. SCREE PLOT OF PERCENTAGE OF EXPLAINED VARIANCES BY 8 PRINCIPAL COMPONENTS ........................................................................................................................ 27 FIGURE 3. CORRELATION CIRCLE VIA PRINCIPAL COMPONENTS ANALYSIS (FIRST TWO DIMENSIONS ONLY) ............................................................................................................... 28 FIGURE 4. COORDINATES OF ITHACA CENSUS TRACTS DATA IN 8 PRINCIPAL COMPONENTS SPACE ............................................................................................................................................... 32 FIGURE 5. SOCIOECONOMIC FACTORS DISTANCE HEATMAP BETWEEN 15 CENSUS TRACTS IN ITHACA .................................................................................................................................. 33 FIGURE 6. FIVE CLUSTERS RESULTS AFTER K-MEANS CLUSTERING USING ITHACA CENSUS TRACTS LEVEL SOCIOECONOMIC FACTORS DATA ................................................................ 34 FIGURE 7. SCATTERPLOT MATRIX BETWEEN 8 PRINCIPAL COMPONENTS USING ITHACA CENSUS TRACTS LEVEL SOCIOECONOMIC FACTORS DATA ................................................................ 36 FIGURE 8. ENERGY BURDEN HOTSPOT MAP VIA ITHACA CENSUS TRACTS LEVEL DATA ............ 38 FIGURE 9. REGRESSION RESULTS OF 8 SOCIOECONOMIC FACTORS AND ENERGY BURDEN PROXY FROM R .................................................................................................................................. 40 FIGURE 10. FIVE CLUSTERS MAP USING SOCIOECONOMIC FACTORS DATA AND ENERGY BURDEN HOTSPOT MAP COMPARISON................................................................................................. 41 FIGURE 11. OLS RESULTS BETWEEN 8 SOCIOECONOMIC FACTORS AND ENERGY EFFICIENCY PROXY .................................................................................................................................... 42 FIGURE 12. ENERGY EFFICIENCY PERCENTILE MAP USING ITHACA CENSUS TRACTS LEVEL DATA ............................................................................................................................................... 45 FIGURE 13. BIVARIATE REGRESSION MODEL RESULTS BETWEEN ENERGY BURDEN PROXY AND ENERGY EFFICIENCY PROXY ................................................................................................. 46 viii LIST OF TABLES TABLE 1. 2021-2023 AMERICAN COMMUNITY SURVEY 5-YEAR DATASET LIST .......................... 15 TABLE 2. PEARSON CORRELATIONS BETWEEN 8 SOCIOECONOMIC FACTORS IN ITHACA CENSUS TRACTS DATA ........................................................................................................................ 23 TABLE 3. EIGENVALUE AND VARIANCE CAPTURED BY 8 PRINCIPAL COMPONENTS .................... 26 TABLE 4. LOADINGS OF 8 PRINCIPAL COMPONENTS WITH RESPECT TO 8 SOCIOECONOMIC FACTORS IN ITHACA CENSUS TRACTS DATA ........................................................................ 30 TABLE 5. CONTRIBUTIONS OF 8 PRINCIPAL COMPONENTS WITH RESPECT TO 8 SOCIOECONOMIC FACTORS IN ITHACA CENSUS TRACTS DATA (%) ................................................................. 31 TABLE 6. ITHACA CENSUS TRACTS DATA EXPRESSED BY 8 PRINCIPAL COMPONENTS ................ 32 TABLE 7. RAW DATA AND COLOR-DIFFERENTIATED CLUSTERING OUTCOMES ACROSS FIVE IDENTIFIED GROUPS .............................................................................................................. 36 TABLE 8. ANOVA BETWEEN 8 SOCIOECONOMIC FACTORS AND ENERGY BURDEN PROXY ........ 40 TABLE 9. ANOVA RESULTS BETWEEN 8 SOCIOECONOMIC FACTORS AND ENERGY EFFICIENCY PROXY .................................................................................................................................... 43 TABLE 10. ANOVA RESULTS BETWEEN ENERGY BURDEN PROXY AND ENERGY EFFICIENCY PROXY .................................................................................................................................... 46 TABLE 11. A CLUSTER-SPECIFIC POLICY DESIGN BASED ON SOCIOECONOMIC CHARACTERISTICS AND ENERGY PROFILES IN ITHACA’S FIVE CLUSTERS .......................................................... 49 ix LIST OF ABBREVIATIONS Abbreviation Full Term EFP Energy Efficiency Proxy EBP Energy Burden Proxy ACS American Community Survey PCA Principal Component Analysis OLS Ordinary Least Squares 1 1. Introduction This study investigates how socioeconomic and housing characteristics shape the spatial distribution of energy burden and energy efficiency in the City of Ithaca, New York. As cities accelerate electrification and building decarbonization to address climate change(Ithaca, 2019), the distributional consequences of these transitions have become a central concern in urban sustainability research((2023); (IEA), 2024; Awolesi et al., 2024; Bullard, 2015; Cui & Cao, 2024; Heffron & McCauley, 2018; Hoffman et al., 2021; Jenkins et al., 2016; Johansson et al., 2012; Nations, 2015; Thomas et al., 2022; V., 2021; Yang et al., 2024). Ensuring that all communities can access affordable and efficient energy services is a critical component of a just energy transition(Awolesi et al., 2024; Bouzarovski & Simcock, 2017; Bullard, 2015; Heffron & McCauley, 2017, 2018; Hoffman et al., 2021; Jenkins et al., 2016; McCauley, 2017; McCauley et al., 2019; Sovacool, 2015; Sovacool et al., 2017; Sovacool & Dworkin, 2015). The increasing attention to energy equity(Hoffman et al., 2021; Sovacool et al., 2017; Sovacool & Dworkin, 2015; Walker et al., 2014; Wei et al., 2023) reflects a growing recognition that decarbonization policies must not only reduce emissions but also prevent the deepening of socioeconomic disparities(Wang & Zhou, 2023; Wei et al., 2023). Against this backdrop, the present research examines fine-scale patterns of energy burden and efficiency within a small but socioeconomically diverse urban context, with the aim of informing equitable local climate policy. 1.1 Background and Motivation 2 Globally, energy equity has emerged as an essential dimension of sustainable development. Although technological advances and climate policies have expanded pathways toward low-carbon energy systems, the benefits of these transitions are often unevenly distributed. Households with lower incomes or those living in older, inefficient buildings typically face disproportionately high energy burdens and encounter greater obstacles in adopting efficient technologies(Al Kez et al., 2024; Hernández, 2016; Homsy & Kang, 2025; Memmott et al., 2021; Reames, 2016). As a result, energy transitions risk reinforcing existing forms of inequality unless they explicitly account for socioeconomic and spatial heterogeneity(Memmott et al., 2021). This challenge is particularly salient in Ithaca, a small city in upstate New York that has committed to achieving carbon neutrality by 2030(Ithaca, 2019). Electrification of heating systems, widespread building retrofits, and renewable energy integration are central to Ithaca’s decarbonization strategy. However, the city’s neighborhoods differ markedly in income levels, tenure characteristics, housing age, and demographic composition. Student-dominated rental districts coexist with older working-class neighborhoods and newer suburban areas, producing a mosaic of energy needs and vulnerabilities. Because the transition to electric heating in cold climates can increase electricity expenditures when buildings remain thermally inefficient, understanding local patterns of energy burden and efficiency becomes essential for designing equitable climate policies. Focusing jointly on these two dimensions provides an opportunity to evaluate the extent to which structural characteristics influence both the cost and technological configuration of household energy use. 3 1.2 Problem Statement Despite growing scholarly attention to energy inequality, empirical analyses that integrate socioeconomic, housing, and energy indicators at a fine spatial scale remain limited, especially in small or mid-sized cities. Existing studies often rely on metropolitan-level or regional data that obscure neighborhood-level disparities. In Ithaca, where decarbonization efforts are advancing rapidly, there is a lack of tract-level research that evaluates how socioeconomic patterns interact with energy burden and energy efficiency. As a result, policymakers have limited evidence on whether electrification and retrofit programs may impose uneven financial impacts or deliver unequal benefits across communities. This absence of localized, data-driven analysis creates a critical gap in understanding how Ithaca’s transition toward carbon neutrality intersects with questions of energy equity and spatial justice. 1.3 Research Question and Objectives To address this gap, the thesis examines the relationship between socioeconomic characteristics, energy burden, and energy efficiency across Ithaca’s census tracts. The central research question guiding the analysis is: ‘How do socioeconomic and housing characteristics jointly influence both energy burden and energy efficiency across Ithaca’s census tracts, and how does this variation support equitable policy making?’ To answer this question, the study seeks to characterize spatial disparities in socioeconomic and housing conditions, quantify the distribution of energy burden, evaluate the variation in 4 heating-related energy efficiency, and model the associations between these variables. By integrating these components, the research aims to identify tract-level community typologies that can inform differentiated, equity-oriented strategies within Ithaca’s broader decarbonization agenda. 1.4 Thesis Structure Overview The thesis is organized into six chapters. Chapter 2 reviews existing scholarship on energy burden, energy equity, socioeconomic determinants of energy use, and spatial methods in energy research, highlighting gaps relevant to small urban contexts. Chapter 3 outlines the study area, data sources, and the construction of key variables, including the tract-level proxies for energy burden and energy efficiency. Chapter 4 presents the analytical methodology and empirical results, drawing on principal component analysis, clustering, spatial hotspot detection, and regression modeling. Chapter 5 synthesizes the findings, examines their implications for equitable electrification policy in Ithaca, and discusses methodological limitations and directions for future research. Chapter 6 concludes the thesis by summarizing the main contributions and reflecting on their relevance for both local decision-making and broader energy equity debates. 2. Literature review The transition toward low-carbon energy systems has elevated questions of equity, affordability, and justice to a central position within contemporary energy research(Awolesi et al., 2024; Bouzarovski & Simcock, 2017; Bullard, 2015; Heffron & McCauley, 2017, 2018; Hoffman et al., 2021; Homsy & Kang, 2025; Jenkins et al., 2016; McCauley, 2017; McCauley et al., 2019; 5 Mushafiq et al., 2023; Sovacool, 2015; Sovacool et al., 2017; Sovacool & Dworkin, 2015; Yang et al., 2024). As cities attempt to decarbonize the residential sector—often the largest and most heterogeneous component of urban energy consumption—the distribution of energy costs and the accessibility of energy-efficient housing emerge as core concerns(Bird & Hernández, 2012; Chen et al., 2021; Cui & Cao, 2024; Fowlie et al., 2018; Reames, 2016; Rosenow et al., 2017; Sovacool, 2015). The literature consistently demonstrates that energy affordability and residential energy efficiency are not merely technical or economic issues but are deeply intertwined with socioeconomic structures, housing market dynamics, and historically embedded forms of inequality. A systematic review of these themes is therefore essential to establish the theoretical foundations and empirical motivations of this study. This chapter synthesizes the major conceptual frameworks related to energy justice and energy poverty, evaluates the determinants of household energy burden and residential energy efficiency, and critically examines the expanding application of spatial analytical methods within this domain. The review culminates by identifying a significant gap in the existing scholarship—namely, the lack of fine-grained, spatially explicit analyses of energy burden and efficiency in small and mid-sized U.S. cities—and thereby establishes the conceptual rationale for the empirical analysis conducted in subsequent chapters. 2.1 Energy Justice: Foundations, Principles, and Relevance The global pursuit of sustainable development has underscored the dual imperative of expanding access to affordable, clean energy while simultaneously reducing greenhouse gas emissions(Johansson et al., 2012). These objectives, reflected explicitly in Sustainable 6 Development Goals 7 and 13, require confronting longstanding tensions between energy affordability, technological transitions, and social justice (United Nations, 2015)(Nations, 2015; UNDP, 2000). Energy is widely recognized as a foundational input into human development, economic participation, and well-being; yet the processes of extraction, production, distribution, and consumption across the energy life-cycle remain fraught with distributive conflicts and asymmetries of power (Sovacool & Dworkin, 2015)(McCauley et al., 2019; Sovacool, 2015; Sovacool et al., 2017; Sovacool & Dworkin, 2015). Historic controversies surrounding fossil fuel extraction, biofuel production, hydropower projects, wind farms, and nuclear energy illustrate that energy systems are embedded within broader social and political structures that shape their impacts and acceptance. The emergence of energy justice as a coherent analytical framework reflects the need to conceptually integrate these concerns. Building upon earlier environmental justice activism but substantially systematized only in recent years, energy justice introduces principles of distributive, procedural, and recognition justice—along with increasingly discussed notions of cosmopolitan and restorative justice—to evaluate the fairness of energy systems past, present, and future. Distributive justice addresses how the costs and benefits of energy systems are spatially and socially allocated; procedural justice considers the inclusiveness and transparency of decision- making processes; recognition justice emphasizes the importance of acknowledging marginalized groups and historically disadvantaged communities within these processes (Jenkins et al., 2016)(Jenkins et al., 2016). Together, these principles provide a normative and empirical 7 framework for examining how energy transitions may reproduce, transform, or alleviate entrenched inequalities. Given the accelerating pace of low-carbon transitions, the urgency of energy justice has only increased. Established fossil-fuel infrastructures coexist with emergent renewable and electrified systems, generating complex patterns of technological coexistence and institutional lock-in. As Sovacool and Dworkin (2015)(Sovacool, 2015; Sovacool & Dworkin, 2015) argue, these transformations simultaneously mitigate some injustices while producing new ones—often in ways not yet fully visible to policymakers. Against this backdrop, understanding residential energy affordability and efficiency as questions of justice rather than mere engineering or economic optimization is both analytically and ethically essential. 2.2 Energy Poverty, Energy Burden, and Their Primary Determinants 2.2.1 Conceptualizing Energy Poverty in Developed Contexts Energy poverty is widely defined as the inability to secure necessary energy services for material and social well-being (Boardman, 1991; Bouzarovski, 2014)(Boardman, 1991; Bouzarovski & Simcock, 2017). Although the fundamental concept is broadly agreed upon, its operationalization varies across national and disciplinary contexts. In developing countries, energy poverty is predominantly framed as a lack of access to basic energy infrastructure and clean fuels (UNDP; Pachauri & Spreng, 2011)(UNDP, 2000). In contrast, in developed countries the focus shifts toward affordability—specifically, whether households can meet adequate energy needs without incurring disproportionate financial stress (Hernández, 2016)(Aguilar-Hernandez et al., 8 2018; Bird & Hernández, 2012; Hernández, 2016). This distinction reflects differences not only in energy systems but also in social expectations and cultural norms regarding energy use. In the United States, energy poverty is most commonly measured through the notion of “energy burden,” defined as the proportion of household income spent on energy bills (Drehobl & Ross, 2016; Reames, 2016)(Ahmed et al., 2025). Although energy burden is straightforward to calculate, it captures several underlying dimensions simultaneously: household income, fuel prices, climatic conditions, and, critically, the energy performance of residential buildings. As such, energy burden has become a widely used proxy for energy-related economic vulnerability. 2.2.2 Drivers of Energy Poverty: Income, Housing, and Prices The “energy poverty triangle”—comprising low household income, poor building energy efficiency, and high energy prices—offers a widely cited framework for understanding the structural determinants of energy poverty (Bouzarovski, 2014)(Bouzarovski & Simcock, 2017). Empirical studies consistently find that income remains the dominant predictor of energy burden (Memmott et al., 2021)(Memmott et al., 2021). However, income alone cannot account for substantial observed variation. Housing tenure and structural quality are equally influential: renters face the “split incentive” problem, wherein landlords have limited incentive to invest in efficiency upgrades while tenants bear the cost of higher utility bills (Bird & Hernández, 2012)(Bird & Hernández, 2012). Building age, particularly in pre-1980 housing stock, is strongly correlated with poor insulation, inefficient heating systems, and higher thermal loads (Levinson, 2014)(Walker et al., 2014). Energy prices, shaped by market regulation, fuel mix, and climatic conditions, further 9 interact with these factors. Climate intensifies these dynamics by directly influencing heating and cooling demand. In cold regions such as the Northeastern United States, heating needs represent a significant share of household energy expenditure, making building envelope quality and heating system efficiency especially consequential. 2.3 Residential Energy Efficiency: Measurement, Drivers, and Policy Implications 2.3.1 Measurement Approaches Residential energy efficiency (EE) can be measured at multiple spatial and conceptual levels. Engineering-based estimates evaluate building envelope performance, heating-system efficiency, and thermal characteristics; however, such data are rarely available for large samples. Consequently, regional and national analyses often rely on frontier-based econometric approaches such as Data Envelopment Analysis (DEA) and Stochastic Frontier Analysis (SFA). DEA, introduced by Charnes, Cooper, and Rhodes (1978)(Charnes et al., 1978), is a non-parametric method suitable for multiple-input, multiple-output scenarios, widely adopted in evaluating energy efficiency across regions and sectors. SFA, as developed by Aigner, Lovell, and Schmidt (1977)(Aigner et al., 1977), enables simultaneous estimation of inefficiency and random error. In recent years, DEA and SFA have been applied to evaluate industrial and regional energy efficiency across numerous national contexts, including China (Chu et al., 2019; Tao et al., 2016)(Chu et al., 2019) and Europe (Aguilar-Hernandez et al., 2018)(Aguilar-Hernandez et al., 2018). Although these techniques have traditionally been applied at macro- or meso-scales, their 10 logic underscores the multidimensional nature of energy efficiency and highlights the limitations of relying solely on single indicators such as fuel type or heating technology when examining neighborhood-level variation. 2.3.2 Drivers of Energy Efficiency Studies broadly categorize the determinants of energy efficiency into technical, structural, and institutional factors (Doytch & Narayan, 2016; He et al., 2020)(Amowine et al., 2020; Wang & Zhou, 2023). Technical drivers include heating-system technology, insulation levels, and the rate of technological innovation. While many scholars argue that advancements in energy-efficient technologies improve overall EE, others highlight the rebound effect, whereby efficiency gains induce increased energy consumption, partially offsetting expected reductions (Gillingham et al., 2016)(Gillingham et al., 2016). Structural factors include housing stock composition, urban density, and fuel mix. Institutional factors—such as environmental regulations, incentives, and market design—can decisively shape household-level energy choices and investment behavior (Haider & Mishra, 2021)(Fowlie et al., 2018). 2.3.3 Efficiency, Electrification, and Affordability Given the accelerating electrification of heating systems as part of climate mitigation strategies, the relationship between electrification and affordability has received substantial scholarly attention. Electrification does not guarantee reductions in household expenditures. In buildings with poor thermal performance, electric resistance heating may increase energy costs, intensifying financial stress (Rosenow et al., 2017)(Rosenow et al., 2017). Conversely, high- 11 efficiency heat pumps, when paired with adequate building envelopes, can deliver meaningful cost savings, particularly in moderate climates. Studies evaluating U.S. Weatherization Assistance Programs provide further evidence that the financial benefits of efficiency interventions are highly heterogeneous and contingent on local building characteristics (Fowlie et al., 2018)(Fowlie et al., 2018). This underscores the importance of integrating building characteristics and socioeconomic contexts when estimating potential impacts of electrification. 2.4 Socioeconomic and Housing Determinants of Energy Outcomes Energy outcomes are shaped not only by technological and economic variables but also by the social and spatial organization of housing markets. Income remains the most robust predictor of energy burden, yet its influence is mediated by housing conditions and tenure. Renters disproportionately occupy older and less efficient housing units, contributing to systematically higher energy expenditures relative to income (Reames, 2016; Bird & Hernández, 2012)(Bird & Hernández, 2012; Reames, 2016). The prevalence of student populations—particularly in university towns such as Ithaca—complicates interpretations of income-based metrics, as student households often report low incomes despite occupying relatively energy-intensive or poorly insulated rental units. Housing age is another strong determinant. Empirical studies repeatedly document that older buildings exhibit significantly worse thermal characteristics and heating-system efficiency (Levinson, 2014)(Walker et al., 2014). Combined with the limited ability of renters to influence building upgrades, this creates entrenched patterns of spatial inequity across neighborhoods. 12 Demographic factors, including age composition, household size, and race/ethnicity, have also been shown to correlate with energy insecurity and energy burden (Memmott et al., 2021)(Memmott et al., 2021). The mechanisms range from differential access to efficient housing to disparities in risk avoidance behaviors and heating practices. 2.5 Spatial Analytical Approaches in Energy Research The spatialization of energy affordability and efficiency has become a distinct and rapidly developing area within energy justice research. As Reames (2016)(Reames, 2016) demonstrates through a census-tract analysis of Detroit, energy burden exhibits pronounced spatial clustering that reflects deeper socioeconomic and institutional structures. Subsequent regional analyses by Bhattacharya et al. (2021)(Kola-Bezka, 2023) confirm that such clustering is a consistent pattern across U.S. metropolitan areas. Spatial analytical tools—such as Local Indicators of Spatial Association (LISA), Getis–Ord Gi*, and spatial econometric models including spatial lag (SAR), spatial error (SEM), and spatial Durbin models (SDM)—enable researchers to identify spatial dependence, quantify spillover effects, and distinguish between local and neighboring determinants of energy outcomes (Anselin, 1995; Elhorst, 2014; LeSage & Pace, 2009)(Anselin, 1988, 1995; Anselin et al., 2008; LeSage, 2009). These tools address a key methodological challenge: conventional regression models assume independence across observational units, an assumption violated when adjacent neighborhoods share infrastructure, socioeconomic characteristics, or diffusion processes such as technology adoption. 13 In parallel, dimensionality reduction and clustering techniques—especially principal component analysis (PCA) and k-means clustering—have been employed to develop socioeconomic and housing typologies. Such typologies capture latent neighborhood structures and enable integrated evaluation of energy outcomes across multifactor community profiles (Spielman & Logan, 2013)(Spielman & Logan, 2013; Spielman & Singleton, 2015). Their growing adoption underscores the methodological shift toward holistic, spatially explicit modeling approaches. 2.6 Research Gap: The Need for Fine-Scale, Context-Specific Analysis in Small Cities Despite the breadth of existing research, several important gaps remain. First, empirical analyses of energy burden and efficiency continue to focus disproportionately on large metropolitan regions, leaving smaller cities—where demographic compositions, rental markets, and housing age distributions may differ markedly—relatively understudied (Bednar & Reames, 2020)(Reames, 2016). Second, studies that simultaneously integrate socioeconomic typologies, spatial clustering, and both energy burden and energy efficiency indicators remain rare. Most scholarship isolates either affordability or efficiency, limiting the ability to understand how the two interact within heterogeneous urban neighborhoods. Third, small cities undergoing rapid decarbonization—such as Ithaca, with its ambitious 2030 carbon neutrality commitment(Ithaca, 2019)—require granular, place-specific evidence to design equitable policy interventions. Without such evidence, electrification and energy- efficiency programs risk exacerbating existing inequalities rather than alleviating them. The absence of tract-level analysis that jointly examines socioeconomic structures, heating fuel types, 14 and spatial dependencies constitutes a notable gap in both the academic literature and policy- relevant research. This study addresses these limitations by applying a multi-method framework—combining PCA-based community classification, K-means clustering, Getis–Ord hotspot analysis, and spatial econometric modeling—to examine how socioeconomic and housing characteristics jointly shape energy burden and energy efficiency across census tracts in Ithaca. In doing so, it contributes both methodological and empirical insights to ongoing discussions on equitable urban decarbonization. 3. Data and Study Scope 3.1 Study Scope The empirical analysis focuses on the City of Ithaca, located in Tompkins County, New York State, USA. Ithaca provides an exemplary case for examining energy equity within small urban communities pursuing rapid decarbonization. The city has committed to achieving carbon neutrality by 2030 and exhibits obvious spatial heterogeneity in socioeconomic and housing characteristics. The study area encompasses 15 census tracts that collectively capture the diversity of Ithaca’s urban fabric—from student-dominated neighborhoods near Cornell University to older residential districts and newly developed suburban zones. The region experiences a cold and prolonged winter, where heating demands are a major component of household energy consumption. 15 This climatic context makes Ithaca particularly relevant for analyzing how socioeconomic vulnerability, energy burden, and heating efficiency interact across space and time. 3.2 Data Source This study integrates multiple publicly available datasets from the U.S. Census Bureau’s American Community Survey (ACS)(Bureau, 2024) and TIGER/Line geospatial products. The datasets span the 2021–2023 ACS 5-Year Estimates, which provide the most granular socioeconomic and housing data available for small geographic units. All datasets were filtered to include only census tracts within the City of Ithaca and merged using the unique tract-level geographic identifier (GEOID). The full list of variables and ACS table identifiers is provided in Table 1. Table 1. 2021-2023 American Community Survey 5-Year Dataset List ACS Table Index Table Name B19013 median household income B19001 Household Income B01003 Total Population B25003 Housing tenure: renter vs. owner B15003 Education B25040 house heating fuel number B25035 year structure built B17001 Poverty Status B01002 Median age B05003 Sex by Age by Nativity and Citizenship Status B25132 Monthly Electricity Costs B25133 Monthly Gas Costs B25135 Annual Other Fuel Costs Source: American Community Survey 5-year estimates, 2019–2023, https://www.census.gov/programs- surveys/acs 16 Spatial boundary files for census tracts were obtained from the TIGER/Line Shapefile (Tompkins County, NY) and cross-validated with the Tompkins County Open Data Portal to ensure spatial consistency and coverage alignment. 3.3 Variable Construction 3.3.1 Data Structure To facilitate a clear analytical framework, all variables were organized into three thematic groups. First, socioeconomic factors were compiled to support the spatial disparities analysis, encompassing eight indicators that describe demographic and economic conditions. Second, an Energy Burden Proxy was constructed to approximate the proportion of annual household income allocated to energy expenditures, providing an indirect measure of energy insecurity. Third, an Energy Efficiency Proxy was developed to characterize the adoption of electric and solar heating technologies as a reflection of household-level efficiency patterns. Together, these components offer a comprehensive foundation for assessing the multidimensional drivers of energy vulnerability. 17 Figure 1. Overview of Ithaca Census Tracts Level Dataset: Socioeconomic Factors, Energy Burden Proxy, and Energy Efficiency Proxy 3.3.2 Socioeconomic and Housing Characteristic To characterize structural disparities among tracts, eight socioeconomic and housing variables were incorporated, including: • Educational attainment (% of population with college degree or higher); • Median household income; • Poverty rate; • Median age; • Housing tenure (% renter households); • Median year structure built; • Population density; and 18 • Gender composition. These indicators collectively represent economic capacity, demographic composition, and built-environment quality—factors expected to influence both energy burden and efficiency. 3.3.3 Energy Burden Proxy (EBP) Energy burden is defined as the share of household income spent on energy. Due to data availability, this study constructs an Energy Burden Proxy (EBP) at the tract level as: 𝐸𝐵𝑃𝑡 = 𝑇𝑜𝑡𝑎𝑙 𝐸𝑛𝑒𝑟𝑔𝑦 𝐶𝑜𝑠𝑡𝑠 𝑀𝑒𝑑𝑖𝑎𝑛 𝐻𝑜𝑢𝑠𝑒ℎ𝑙𝑑𝑒 𝐼𝑛𝑐𝑜𝑚𝑒𝑡 × 100% 𝑇𝑜𝑡𝑎𝑙 𝐸𝑛𝑒𝑟𝑔𝑦 𝐶𝑜𝑠𝑡𝑠 = 𝐴𝑛𝑛𝑢𝑎𝑙 𝑂𝑡ℎ𝑒𝑟 𝐹𝑢𝑒𝑙 𝐶𝑜𝑠𝑡𝑠 + (𝑀𝑜𝑛𝑡ℎ𝑙𝑦 𝐺𝑎𝑠 𝐶𝑜𝑠𝑡𝑠 + 𝑀𝑜𝑛𝑡ℎ𝑙𝑦 𝐸𝑙𝑒𝑐𝑡𝑟𝑖𝑐𝑖𝑡𝑦 𝐶𝑜𝑠𝑡𝑠) ∗ 12 where the numerator represents the total annualized energy expenditures, including other- fuel costs as well as monthly gas and electricity costs converted to yearly values, while the denominator captures the tract-level median household income. The EBP therefore approximates the financial pressure imposed by overall energy spending on local households and serves as a tract-level indicator of energy affordability. 3.3.4 Energy Efficiency Proxy (EFP) Given Ithaca’s cold climate and the central role of residential heating in overall energy consumption, a proxy for household-level energy efficiency (EFP) was derived from the ACS table B25040: House Heating Fuel. This table reports, for each census tract, the distribution of households by their primary heating fuel. For example, in tract 1 where 53.5% of households 19 mainly rely on utility gas and 43.7% primarily rely on electricity for heating, these values represent the shares of households using each fuel type. Building on this structure, the EFP is defined as the combined proportion of housing units using electricity or solar energy as their primary heating source: 𝐸𝐹𝑃 = 𝑃𝑒𝑙𝑒𝑐𝑡𝑟𝑖𝑐𝑖𝑡𝑦 + 𝑃𝑠𝑜𝑙𝑎𝑟 This variable captures the degree of electrification and the penetration of renewable heating technologies within each tract. While it does not measure technical efficiency directly (e.g., insulation or appliance performance), it reflects a tract’s progress toward cleaner and potentially more efficient energy sources. 3.4 Data Pre-Processing All tabular and spatial datasets were harmonized and merged at the census-tract level using the tract identifier (GEOID). The principal data processing steps implemented in R are summarized below: 1. Data cleaning and alignment: Variable names were standardized across ACS years, and missing or zero-income records were flagged for review. Tracts with incomplete records for key variables were inspected and, if necessary, excluded from the corresponding year’s analysis. 2. Normalization and transformation: Continuous variables were standardized (z-score normalization) prior to principal component 20 analysis (PCA) and K-means clustering to remove unit dependencies and enable distance- based comparisons. Log transformations were applied in regression where appropriate. 3. Panel assembly and neighborhood definition The three cross-sectional ACS extracts (2021–2023) were vertically concatenated into a tract × year panel for assessments of temporal persistence (e.g., hotspot frequency). Each year’s EBP and covariates were maintained as separate observations, which was subsequently used in the Getis-Ord Gi* hotspot analysis and spatial regression diagnostics. Adjacency was computed using polygon contiguity (Queen criterion), which treats any shared boundary or vertex as indicating neighborhood adjacency. 3.5 Data Quality and Limitations Several methodological considerations warrant mention. First, the ACS 5-year estimates represent rolling averages rather than true annual snapshots; temporal comparisons therefore capture medium-term trends rather than precise year-to-year changes. Second, the EFP is a proxy for energy technology composition, not direct thermal efficiency; it serves as a practical indicator of electrification and renewable adoption under data constraints. Finally, tract-level sampling errors and small population counts can introduce variability in estimates. These limitations are explicitly acknowledged in the interpretation of statistical results and in policy recommendations. 3.6 Reproducibility and Software 21 The entire workflow is implemented in R, and the primary packages used include sf, dplyr, readr, FactoMineR, factoextra, spdep, ggplot2, and viridis. The processing and analysis scripts are modularized into R scripts files that (1) ingest raw inputs, (2) perform deterministic cleaning and derivation, (3) execute PCA/k-means and spatial analyses, and (4) produce figures and tables. These scripts, together with processed data artifacts (excluding any restricted files), are archived for reproducibility. 4. Methodology and Results 4.1 Analytical Framework The analytical framework is designed to examine how socioeconomic and housing structures influence the patterns of energy burden and energy efficiency across the City of Ithaca. The approach integrates multivariate data analysis, spatial scan statistic, and visualization in a modular workflow comprising three interrelated components. The first module establishes the structural context of Ithaca’s communities. Eight socioeconomic and housing variables, including education, income, tenure, housing age, poverty rate, sex, and median age, were standardized and subjected to Principal Component Analysis (PCA) to reduce dimensionality and extract orthogonal latent components representing the major socioeconomic gradients. The first five principal components explain more than 80 percent of total variance and form a low-dimensional “PCA space.” Within this space, K-means clustering identifies tract-level subgroups with similar profiles, effectively partitioning the city into typologies of social and 22 housing conditions. These typologies represent underlying structural inequalities that may predispose certain communities to higher energy vulnerability. The second module quantifies and maps the spatial distribution of energy burden. For each tract and ACS 5-year dataset, the Energy Burden Proxy (EBP) was computed as the ratio of annualized energy cost to median household income. Spatial dependencies among tracts were modeled using a Queen contiguity matrix, and the Getis-Ord Gi* statistic was applied to detect local clusters (hotspots and coldspots) of unusually high or low burden. The resulting z-scores reveal the geographic concentration of energy vulnerability and its persistence across years. To examine the association with socioeconomic drivers, a multiple linear regression was estimated for EBP as a function of tract-level socioeconomic and housing variables. The regression coefficients quantify how attributes such as income, housing age, education, or renter share contribute to the observed burden disparities. Together, the hotspot analysis and regression results provide both spatial and statistical perspectives on energy inequity. The third module introduces energy efficiency as an explanatory dimension of community resilience. An Energy Efficiency Proxy (EFP) was derived from ACS Table B25040, calculated as the combined proportion of households using electric or solar energy for primary heating. This proxy captures the degree of electrification and renewable adoption—key indicators of potential energy efficiency and low-carbon transition readiness. Regression models were estimated to assess how socioeconomic factors influence EFP, and a subsequent model evaluated the direct relationship between EFP and EBP, testing the 23 hypothesis that lower efficiency is associated with higher energy burden. EFP values were further ranked into quintiles and mapped to visualize spatial heterogeneity in heating technology adoption. In conclusion, the framework is sequential but interconnected. Module 1 reveals the socioeconomic disparity between different tracts. Module 2 relates those disparities to measurable energy-burden outcomes. Module 3 introduces a feedback mechanism, evaluating how local energy efficiency characteristics mediate or exacerbate energy stress. 4.2 Module 1 – Socioeconomic Structure 4.2.1 Correlations Prior to conducting the principal component analysis, pairwise Pearson correlations were computed among the eight standardized socioeconomic and housing variables (Table 2). The resulting correlation matrix reveals several consistent structural patterns within Ithaca’s neighborhood composition. Table 2. Pearson Correlations Between 8 Socioeconomic Factors in Ithaca Census Tracts Data (negative relationships are marked as red) Populat ion Median income Median Year Built Educati on Poverty rate Tenure( Owner rate) Sex(Fe male) Median Age Population 1.0000 0.1671 0.3079 0.2189 -0.3661 0.4409 0.1650 0.1476 Median household income 0.1671 1.0000 -0.1130 0.2819 -0.3369 0.7098 0.1649 0.1425 24 Median Year Structure Built 0.3079 -0.1130 1.0000 -0.3720 -0.1427 -0.2104 0.2622 0.0126 Education(h ighschool degree) 0.2189 0.2819 -0.3720 1.0000 -0.3126 0.4551 -0.0495 0.1185 Poverty rate -0.3661 -0.3369 -0.1427 -0.3126 1.0000 -0.2648 -0.1941 -0.5960 Tenure(Ow ner rate) 0.4409 0.7098 -0.2104 0.4551 -0.2648 1.0000 0.0923 0.4744 Sex 0.1650 0.1649 0.2622 -0.0495 -0.1941 0.0923 1.0000 0.0976 Median Age 0.1476 0.1425 0.0126 0.1185 -0.5960 0.4744 0.0976 1.0000 (1) Income–Tenure Relationship. The strongest positive association is observed between median household income and owner- occupied housing rate (r = 0.71), indicating that wealthier tracts are more likely to be characterized by home ownership rather than rental housing. This relationship reflects the housing market stratification common in small U.S. cities, where ownership aligns with higher and more stable income levels. (2) Poverty and Socioeconomic Indicators. Poverty rate exhibits moderate to strong negative correlations with median income (r = −0.34) and education attainment (r = −0.31), confirming that tracts with lower education levels and lower incomes experience higher poverty incidence. The inverse association with median age (r = −0.60) further suggests that younger populations are concentrated in lower-income or student-dominated neighborhoods. 25 (3) Education and Housing Characteristics. Educational attainment (percentage of residents with at least a high-school degree) correlates positively with owner-occupied tenure (r = 0.46) and weakly with median income (r = 0.28). Conversely, it correlates negatively with median year structure built (r = −0.37), implying that better-educated populations tend to reside in older housing stock—likely reflecting the concentration of long-established households or university-adjacent neighborhoods within the city’s core. (4) Population Density and Housing. Total population shows moderate positive correlation with owner rate (r = 0.44) and with median year built (r = 0.31). These patterns suggest that more densely populated tracts include a mixture of older, owner-occupied units and multi-family residences. The correlations are consistent with Ithaca’s urban morphology, where compact, historically developed tracts coincide with higher population densities. (5) Gender Composition. The female share displays weak correlations with most other variables (|r| < 0.3), indicating minimal spatial differentiation by gender at the tract level. (6) Multicollinearity Assessment. Several variables—especially income, poverty rate, and tenure—show interdependencies exceeding |r| > 0.3, warranting dimensionality reduction prior to subsequent regression or clustering analyses. These correlations justify the application of PCA to extract independent 26 components that capture the principal socioeconomic gradients without violating orthogonality assumptions. In summary, the correlation structure highlights two latent axes of variation across Ithaca’s census tracts: a socioeconomic prosperity gradient, defined by higher income, education, and ownership rates opposing poverty; and a housing-demographic gradient, reflecting contrasts between older, established owner neighborhoods and younger, renter-dominated areas. The PCA that follows quantifies these latent dimensions and provides orthogonal components for clustering and further spatial analysis. 4.2.2 PCA Eigenvectors The principal component analysis (PCA) was applied to the standardized socioeconomic dataset of 15 census tracts using the correlation matrix as input. Table 3 summarizes the eigenvalues, the proportion of total variance explained by each component, and the cumulative variance captured. Table 3. Eigenvalue and Variance Captured by 8 Principal Components Eigenvalue % of Variance Cumulative% comp 1 2.77 34.68 34.68 comp 2 1.64 20.50 55.17 comp 3 1.03 12.82 67.99 comp 4 0.90 11.20 79.19 comp 5 0.74 9.21 88.41 comp 6 0.56 6.95 95.35 comp 7 0.34 4.26 99.61 comp 8 0.03 0.39 100.00 27 The first three principal components have eigenvalues greater than 1 and together explain approximately 68% of the total variance (34.7% by Component 1, 20.5% by Component 2, and 12.8% by Component 3). The scree plot (Figure 2) exhibits a clear inflection, or “elbow point”, after the third component, indicating that the first three components capture the major socioeconomic variation across census tracts. Beyond this point, the marginal gain in explained variance drops sharply, suggesting that the remaining components contribute primarily to idiosyncratic noise rather than substantive socioeconomic differentiation. Figure 2. Scree Plot of Percentage of Explained Variances by 8 Principal Components The PCA correlation circle (Figure 3) visually supports eigenvectors’ interpretations. Variables such as income, education, and owner rate are tightly grouped in the positive direction 28 of Dim 1, whereas poverty rate points strongly in the opposite direction. The almost perpendicular orientation of median year structure built and poverty rate vectors illustrates their weak correlation and supports the interpretation of Dim 2 as an independent housing-age factor. Figure 3. Correlation Circle via Principal Components Analysis (First Two Dimensions only) Component 1 – Socioeconomic Prosperity Gradient (34.7%) The first principal component displays strong positive loadings for median household income (0.69), tenure (owner rate) (0.68), education (high-school degree) (0.57), and population (0.55), coupled with a pronounced negative loading for poverty rate (−0.70). This axis therefore represents a wealth–stability gradient, separating affluent, owner-dominated tracts from economically disadvantaged areas characterized by higher poverty rates and lower educational attainment. 29 In the PCA variable circle above(Figure 3), these variables form a coherent cluster pointing toward the right-hand side of Dim 1, signifying their shared contribution to this latent socioeconomic dimension. Tracts with high Dim 1 scores can thus be interpreted as socioeconomically secure neighborhoods, whereas those with negative scores represent economically stressed or transitional areas. Component 2 – Housing Age and Demographic Renewal (20.5%) The second principal component is dominated by a very high positive loading on median year structure built (0.87), indicating the relative newness of the housing stock, and moderate contributions from sex (female) (0.39) and population (0.39). It exhibits negative contributions from education (−0.48) and income (−0.19). Accordingly, Component 2 differentiates tracts according to housing age and demographic renewal, contrasting newer, higher-density, possibly student-oriented neighborhoods (positive Dim 2) with older, established tracts (negative Dim 2). The vector for Median Year Structure Built in Figure 4 is nearly vertical, confirming that this variable drives most of the variation along the Dim 2 axis, independent from income-related effects captured by Dim 1. Component 3 – Socio-Demographic Composition (12.8%) The third principal component captures more subtle demographic variation, with moderate positive loadings for median household income (0.41), poverty rate (0.42), and median age (not shown numerically in Table 4 but typically positive in Dim 3), and a smaller positive contribution from education. 30 This component reflects mixed demographic characteristics—potentially distinguishing between older, moderately affluent tracts with stable income and age structures and younger or student-dominated tracts that deviate from this pattern. Although Dim 3 contributes less variance than the first two components, it adds sociological nuance by capturing population-age interactions that are orthogonal to the main economic and housing axes. Table 4. Loadings of 8 Principal Components with respect to 8 Socioeconomic Factors in Ithaca Census Tracts Data Dim.1 Dim.2 Dim.3 Dim.4 Dim.5 Dim.6 Dim.7 Dim.8 Population 0.5536 0.3941 0.2558 - 0.6083 0.0391 0.2114 - 0.2327 0.0486 Median household income 0.6934 - 0.1879 0.4106 0.3193 0.2739 - 0.3623 - 0.0368 0.0755 Median Year Structure Built - 0.0717 0.8698 0.1041 - 0.1616 0.1774 - 0.2053 0.3572 - 0.0113 Education(hig hschool level) 0.5699 - 0.4812 0.0290 - 0.3129 - 0.4997 - 0.0829 0.2960 0.0269 Poverty rate - 0.7014 - 0.3068 0.4222 0.0126 0.2026 0.3817 0.2106 0.0665 Tenure(Owner rate) 0.8518 - 0.2358 0.2089 0.0410 0.2919 0.2529 0.1106 - 0.1102 Sex(Female) 0.2444 0.5317 0.3565 0.5014 - 0.4781 0.2236 - 0.0228 - 0.0003 Median Age 0.6161 0.1693 - 0.6563 0.2167 0.1562 0.2669 0.1143 0.0732 31 Table 5. Contributions of 8 Principal Components with respect to 8 Socioeconomic Factors in Ithaca Census Tracts Data (%) Dim.1 Dim.2 Dim.3 Dim.4 Dim.5 Dim.6 Dim.7 Dim.8 Population 11.05 9.47 6.38 41.29 0.21 8.04 15.89 7.67 Median household income 17.33 2.15 16.44 11.37 10.18 23.62 0.40 18.51 Median Year Structure Built 0.19 46.14 1.06 2.91 4.27 7.59 37.43 0.42 Education (highschool degree) 11.71 14.12 0.08 10.92 33.89 1.24 25.70 2.35 Poverty rate 17.73 5.74 17.38 0.02 5.57 26.21 13.01 14.34 Tenure(Owne r rate) 26.16 3.39 4.25 0.19 11.56 11.50 3.59 39.36 Sex(Female) 2.15 17.24 12.40 28.05 31.01 8.99 0.15 0.00 Median Age 13.68 1.75 42.01 5.24 3.31 12.82 3.83 17.37 This PCA space(Table 6) forms the analytical basis for the subsequent K-means clustering (Section 4.2.3), which classifies census tracts into internally homogeneous groups based on their coordinates in the principal component space. For completeness and to preserve minor but potentially meaningful structure, all eight components were retained in the K-means clustering step to ensure that the classification incorporated the full data variance. By using orthogonal components rather than raw variables, the clustering analysis eliminates multicollinearity and ensures that Euclidean distances among tracts reflect genuine multidimensional disparities rather than redundant information. 32 Table 6. Ithaca Census Tracts Data Expressed by 8 Principal Components Dim.1 Dim.2 Dim.3 Dim.4 Dim.5 Dim.6 Dim.7 Dim.8 1 -1.4654 0.8399 -1.2916 0.3704 -1.1592 -0.7894 0.8799 0.2397 2 -3.9765 1.9945 0.1796 0.8487 1.6717 -0.2443 -0.4942 -0.0457 3 -2.0329 -2.1953 0.1834 0.4993 -0.7499 0.2749 0.2984 0.2134 4 -1.3361 -1.3194 1.6379 0.0419 0.3141 1.0067 0.9038 -0.1754 5 -0.1576 1.4428 0.5636 -1.1836 -1.3226 -1.0316 -0.2632 -0.2183 6 1.2736 0.2672 -0.3114 -0.4513 0.3991 -0.3871 0.0950 -0.0356 7 2.0469 -0.1218 0.2550 0.0417 0.8602 -0.4092 -0.1578 0.3577 8 0.3668 -1.4221 -0.8611 -0.6795 0.2425 0.4834 -0.8120 -0.0332 9 -0.1299 -0.2052 -0.8543 1.4215 -1.4649 0.7168 -0.9443 -0.1596 10 2.4661 1.1096 -0.7729 0.7010 0.5092 1.0345 0.3349 -0.0779 11 0.2872 0.6655 -0.2314 -0.3319 0.1828 0.6334 -0.4953 0.2127 12 0.3043 1.5360 1.5897 -1.5250 -0.6341 0.9017 0.0697 0.1006 13 1.2338 -1.5537 1.7493 0.8337 0.1392 -1.3373 -0.5607 -0.0302 14 -0.7326 -1.5524 -1.6564 -1.7851 0.8024 -0.4183 0.2558 -0.1632 15 1.8523 0.5142 -0.1794 1.1981 0.2095 -0.4342 0.8900 -0.1851 Figure 4. Coordinates of Ithaca Census Tracts Data in 8 Principal Components Space 33 4.2.3 K-means clustering To identify structurally similar neighborhoods, a K-means clustering procedure was applied to the full set of eight standardized principal components. This approach groups census tracts according to multidimensional similarity in their socioeconomic and housing characteristics, as revealed by the PCA. Euclidean distances between tracts were computed in the standardized principal-component space. The resulting distance heatmap (Figure 5) illustrates pairwise dissimilarities: darker red tones along the diagonal indicate clusters of high internal similarity, while lighter blue cells mark greater differentiation. This pattern suggested approximately four to five internally homogeneous groups, providing an empirical basis for selecting k = 5. Figure 5. Socioeconomic Factors Distance Heatmap between 15 Census Tracts in Ithaca 34 The model produced a configuration with five clusters of sizes 1, 2, 5, 2, and 5 tracts respectively. The between-cluster to total sum-of-squares ratio was 64.2%, signifying that most of the total variance in socioeconomic conditions was explained by differences between clusters rather than within them—a satisfactory level of partition quality given the small sample size (n = 15 tracts). The spatial arrangement of the five clusters is displayed in Figure 6, where each census tract is colored according to cluster membership and labeled by its tract index (1–15). Figure 6. Five Clusters Results after K-means Clustering using Ithaca Census Tracts Level Socioeconomic Factors Data According to the subgroups visualizations in the multidimensional scatterplot matrix (Figure 7): 35 Cluster 1 (teal) — A single tract with low median income (~$45K), high poverty rate(49%), extremely low ownership (1.6%), newest built houses(1991) and very young population (21.5 years). Represents Collegetown and campus-adjacent rental districts—student-dominated and economically precarious despite newer housing stock. Cluster 2 (orange) — Two tracts with moderate income, older homes, medium poverty (~20%), and modest ownership(40%), reflecting areas of older working-class neighborhoods with mixed tenure and gradual gentrification pressure. Cluster 3 (blue) — Five tracts with mixed income range($40k~60k), moderate poverty, mid-aged housing (1965–1984), diverse tenure. Defines transitional mixed-tenure neighborhoods—socially balanced and demographically heterogeneous. Cluster 4 (pink) — Two tracts characterized by moderate income but highest poverty (54– 68%), relatively low ownership, very young median age (20–22 years). These indicates vulnerable urban core and student-rental concentrations near downtown and campus. Cluster 5 (green) — Five tracts with highest income ($88k–155k), newer housing, high ownership(>50%), and older residents. Denotes affluent middle-class suburbs and South Hill area, relatively stable and homeowner-dominant. 36 Figure 7. Scatterplot Matrix between 8 Principal Components using Ithaca Census Tracts Level Socioeconomic Factors Data Table 7. Raw Data and Color-Differentiated Clustering Outcomes Across Five Identified Groups (Cluster 1: Teal; Cluster 2: Orange; Cluster 3: Blue; Cluster 4: Pink; Cluster 5: Green) Inde x Populatio n Incom e Year Built Educatio n Povert y Tenur e Sex Age Cluster 1 2006 40861 1981 0.977 0.173 0.037 0.520 32.9 3 2 2417 45595 1991 0.674 0.489 0.016 0.493 21.5 1 3 1655 56827 1938 1.000 0.536 0.226 0.483 22.3 4 4 2958 69083 1959 0.972 0.680 0.503 0.503 20 4 5 4857 61250 1984 0.998 0.047 0.191 0.539 19.5 3 6 4359 92335 1972 0.982 0.094 0.490 0.483 35.3 5 7 4588 12881 7 1965 0.976 0.124 0.579 0.492 36.8 5 8 4101 64783 1938 0.976 0.210 0.451 0.440 34 2 9 2721 52441 1938 0.941 0.132 0.291 0.586 35.3 3 10 4480 88594 1973 0.950 0.084 0.683 0.559 50 5 37 11 4751 65665 1965 0.921 0.222 0.385 0.515 35.1 3 12 6314 57201 1984 0.987 0.337 0.416 0.566 23.5 3 13 3266 15571 3 1947 1.000 0.193 0.540 0.514 19.4 5 14 3385 43883 1957 1.000 0.238 0.339 0.343 32 2 15 2972 11395 3 1978 0.987 0.082 0.589 0.550 40.4 5 4.2.4 Summary The clustering results underscore that economic prosperity and housing age are the principal dimensions structuring Ithaca’s social geography. Clusters 1 and 5 represent relatively secure, owner-occupied sectors, whereas Clusters 2 and 4 capture younger or economically stressed communities. Cluster 3 occupies an intermediate position, linking core rental zones and peripheral redevelopment areas. This five-group typology effectively translates the latent PCA dimensions into discrete, spatially interpretable neighborhood categories. 4.3 Module 2 – Energy Burden (EBP) 4.3.1 Spatial Analysis A local spatial autocorrelation analysis was performed using the Getis–Ord Gi* statistic on tract-level EBP values. EBP was calculated as the annualized total energy costs (annual other-fuel costs plus monthly gas and electricity costs × 12) divided by tract median household income, and mapped to census tracts. A Queen contiguity weights matrix defined neighborhood relations, and the resulting LocalG z-scores identify statistically clustered high (hotspots) and low (coldspots) EBP values. 38 The 2019~2023 LocalG map (Figure 8) reveals a pronounced hotspot in the southeastern tract (bright yellow), indicating a significant cluster of tracts with relatively high energy burden. Several small, higher-intensity pockets (orange) appear adjacent to the urban core, representing localized concentrations of elevated EBP. By contrast, the large tracts in the northern and western parts of the study area are coldspots (dark purple/blue), denoting significantly lower energy burdens among neighboring tracts. Figure 8. Energy Burden Hotspot Map via Ithaca Census Tracts Level Data 4.3.2 Association with Socioeconomic Factors To identify how socioeconomic and housing characteristics influence the spatial variation in household energy burden, a multiple ordinary least squares (OLS) regression model was estimated with the Energy Burden Proxy (EBP) as the dependent variable. The model 39 specification included eight standardized predictors: total population, median household income, median year structure built, education level, poverty rate, homeownership rate, female population share, and median age. All independent variables were standardized (z-scores) prior to estimation, enabling direct comparison of coefficient magnitudes as measures of relative influence. The model achieved a high explanatory power (R² = 0.907, Adjusted R² = 0.783, p=0.01298 < 0.05), indicating that the selected socioeconomic and housing variables collectively account for a substantial proportion of the observed cross-tract variation in energy burden. The regression results indicate that median household income is the only predictor approaching conventional statistical significance, displaying a negative association with the Energy Burden Proxy (β = −0.81, p = 0.051). This relationship aligns with expectations: higher- income tracts tend to allocate a smaller share of income to energy costs, thereby experiencing lower energy burden. A second variable—female share—also exhibits a marginally significant negative effect (β = −0.29, p = 0.072), suggesting that tracts with a larger proportion of female residents may have slightly lower energy burdens, although the underlying mechanisms remain unclear and the effect size is modest. All remaining predictors, including population size, median structure age, educational attainment, poverty rate, homeownership rate, and median age, show non-significant coefficients (p > 0.10). Their directions are generally consistent with theoretical expectations—for example, a positive sign for poverty rate and homeownership rate, and a negative sign for education—but 40 none provide statistical evidence of strong associations in this context. The lack of significance across most variables likely reflects the limited sample size (n = 15 tracts) and the relatively coarse spatial scale, both of which reduce statistical power and increase uncertainty in coefficient estimates. Figure 9. Regression Results of 8 Socioeconomic Factors and Energy Burden Proxy From R ANOVA results further confirm the dominance of income as the primary explanatory variable. The F-test for median household income is highly significant (F = 15.48, p < 0.01), whereas other factors contribute marginally to model variance. Table 8. ANOVA Between 8 Socioeconomic Factors and Energy Burden Proxy Df Sum Sq Mean Sq F value Pr(>F) z_Population 1 0.0003 0.0003 0.0014 0.9714 z_Median household income 1 3.4474 3.4474 15.4845 0.0077 z_Median Year Structure Built 1 0.3672 0.3672 1.6492 0.2464 z_Education(highschool degree) 1 0.0447 0.0447 0.2006 0.6700 z_Poverty rate 1 0.9695 0.9695 4.3547 0.0820 z_Tenure(Owner rate) 1 1.3600 1.3600 6.1087 0.0484 z_Sex(Female) 1 1.0607 1.0607 4.7644 0.0718 41 z_Median Age 1 0.0043 0.0043 0.0191 0.8945 Residuals 6 1.3358 0.2226 The spatial overlay of regression and clustering results (Figure 10) broadly supports the statistical conclusion that income stratification drives energy-burden patterns: tracts in Cluster 2,4—the socioeconomically vulnerable group characterized by lower incomes, older housing, and higher poverty—coincide with the principal energy-burden hotspots, while Clusters 5 generally align with coldspots, reflecting greater energy resilience. However, an important exception is observed in Cluster 5: tract index = 15 appears as an outlier, exhibiting an extremely high energy-burden LocalG (hotspot) despite the cluster’s overall moderate-to-high socioeconomic profile. This anomaly indicates that the relationship between income and energy burden is not strictly monotonic at the tract scale and that local factors can override cluster-level tendencies. Figure 10. Five Clusters Map using Socioeconomic Factors Data and Energy Burden Hotspot Map Comparison 4.4 Module 3 – Energy Efficiency Proxy (EFP) 42 4.4.1 Association with Socioeconomic Factors A multiple linear regression model was estimated to quantify how standardized socioeconomic and housing characteristics influence tract-level EFP values. The same eight predictors used in the previous energy-burden model were retained, each standardized (z-score) to allow coefficient comparability. The OLS results (Figure 11) reveal a strong model fit (R² = 0.916, Adjusted R² = 0.806, p < 0.01), indicating that socioeconomic and housing features collectively explain over 90% of the variance in energy-efficiency patterns across Ithaca’s tracts. Figure 11. OLS Results Between 8 Socioeconomic Factors and Energy Efficiency Proxy Among the predictors, education level exhibits a significant negative association with energy efficiency (β = −0.076, p = 0.022). This counterintuitive result suggests that tracts with higher proportions of high-school graduates—or, more generally, more formally educated populations— tend to have lower EFP scores. This likely reflects the geographic concentration of educated 43 residents in older housing near the city center (Cornell and downtown tracts) where electrification rates remain low, as heating systems are typically natural gas–based rather than electric or solar. The ANOVA results confirm this pattern and highlight additional influential variables: • Median household income (F = 21.16, p = 0.0037) • Median year structure built (F = 19.65, p = 0.0044) • Education attainment (F = 11.79, p = 0.0139) • Poverty rate (F = 9.89, p = 0.0200) The significance of median year built implies that newer housing is more likely to employ electricity or solar heating—consistent with Ithaca’s building-renewal trend and with national patterns where post-2000 constructions favor electric systems due to code efficiency requirements and climate targets. Conversely, the negative effect of education is likely a spatial artifact, as well-educated residents are concentrated in central, older neighborhoods surrounding Cornell University. Table 9. ANOVA Results Between 8 Socioeconomic Factors and Energy Efficiency Proxy Df Sum Sq Mean Sq F value Pr(>F) z_Population 1 0.0062 0.0062 1.6245 0.2496 z_Median household income 1 0.0805 0.0805 21.1636 0.0037 z_Median Year Structure Built 1 0.0747 0.0747 19.6498 0.0044 z_Education(highschool degree) 1 0.0448 0.0448 11.7876 0.0139 z_Poverty rate 1 0.0376 0.0376 9.8929 0.0199 z_Tenure(Owner rate) 1 0.0002 0.0002 0.0601 0.8145 z_Sex(Female) 1 0.0061 0.0061 1.5972 0.2532 z_Median Age 1 0.0007 0.0007 0.1904 0.6778 Residuals 6 0.0228 0.0038 44 4.4.2 Spatial Distribution of Energy Efficiency The spatial visualization in Figure 12 illustrates the distribution of EFP percentiles across Ithaca’s 15 tracts. Higher EFP scores (lighter orange and yellow tones) appear in the southern and southeastern tracts, particularly Tract 15 (South Hill)—the same area identified earlier as an energy-burden hotspot. This alignment suggests that higher electrification does not necessarily translate to lower energy burden. Instead, in colder climates such as Ithaca, electric heating—if based on conventional resistance systems rather than high-efficiency heat pumps—can increase electricity expenditures and thus elevate the local energy burden. In contrast, the northwestern and central tracts (dark purple areas) exhibit low EFP values, indicating continued reliance on fossil-based heating (natural gas, fuel oil). These zones largely overlap with Clusters 3 and 4 from the socioeconomic classification, representing older, lower- income neighborhoods near downtown Ithaca and Cornell University. While their low EFP may limit immediate electricity costs, it also reflects slower adoption of clean and efficient heating technologies. 45 Figure 12. Energy Efficiency Percentile Map using Ithaca Census Tracts Level Data 4.4.3 Relationship Between Energy Efficiency and Energy Burden To further explore the connection between energy system characteristics and affordability, a simple bivariate regression model was estimated to assess whether higher energy efficiency (EFP) corresponds to lower household energy burden (EBP). The model specification is as follows: 𝐸𝐵𝑃2023 = 𝛽0 + 𝛽1𝐸𝐹𝑃 + 𝜀 The regression results (Table 10) indicate a positive but statistically insignificant association between energy efficiency and energy burden (β = 0.513, p = 0.746). The model’s explanatory power is low (R² = 0.008, Adjusted R² = -0.068), with an overall F-statistic of 0.1099 (p = 0.746*). The ANOVA table confirms that variation in EFP explains only about 0.8% of the total variance in EBP across the 15 census tracts. 46 Figure 13. Bivariate Regression Model Results Between Energy Burden Proxy and Energy Efficiency Proxy Table 10. ANOVA Results Between Energy Burden Proxy and Energy Efficiency Proxy Df Sum Sq Mean Sq F value Pr(>F) EFP$`Energy efficiency` 1 0.0720 0.07200 0.1099 0.7455 Residuals 13 8.5178 0.65522 While the positive coefficient suggests that tracts with higher electrification and solar- heating adoption tend to exhibit slightly higher energy burdens, the lack of statistical significance indicates that the relationship is not robust when considered in isolation. 4.4.4 Summary The positive sign of the coefficient is contrary to the expected direction under an ideal efficiency–affordability tradeoff (where higher EFP would correspond to lower EBP). However, this finding aligns with field observations and the spatial results discussed earlier: in cold-climate cities such as Ithaca, electrification without concurrent efficiency improvements (e.g., 47 installation of modern heat pumps or building envelope upgrades) can raise electricity expenditures. Specifically, tracts such as South Hill (index 15) show high EFP but also marked as a hotspot, largely due to reliance on electric resistance heating in multi-unit rental buildings. This pattern suggests that, while electrification may signal progress toward decarbonization, it can intensify energy costs for residents when housing units remain thermally inefficient or when tenants lack control over heating infrastructure. Conversely, tracts with low EFP but low EBP—notably the Cornell University campus area (Tracts 5 and 6)—demonstrate that centralized heating systems (e.g., district steam or combined heat-and-power plants) can achieve low household energy burden even with low electrification rates. This underscores that technological configuration and institutional control of energy systems play a decisive role in shaping household energy outcomes. 4.5 Synthesis of Findings Integrating the three analytical modules yields a coherent narrative of energy equity in Ithaca: 1. Socioeconomic structure (Module 1) reveals deep inequality, with five distinct clusters ranging from affluent, owner-occupied neighborhoods to low-income, high-poverty tracts. 2. Energy burden analysis (Module 2) shows that income remains the strongest predictor of household energy stress, with persistent hotspots in lower-income, older-housing zones. 48 3. Energy efficiency modeling (Module 3) demonstrates that while newer, wealthier areas exhibit higher electrification (EFP), this does not necessarily reduce energy burden— particularly in South Hill, where electric heating is widespread. Taken together, these results imply that electrification without efficiency retrofitting risks reproducing or even deepening affordability inequities. Policy interventions in Ithaca should therefore target simultaneous progress on (1) building efficiency upgrades, (2) equitable access to efficient electric heating technologies, and (3) income-based energy assistance, ensuring that the transition to low-carbon systems also delivers social and economic benefits. 5. Discussion This chapter synthesizes the empirical findings from the previous analyses to propose actionable strategies for reducing energy burden and improving household energy efficiency across Ithaca’s neighborhoods. The discussion focuses on policy differentiation among the five socioeconomic clusters identified in Chapter 4, linking each cluster’s structural characteristics with the corresponding energy-burden (EBP) and energy-efficiency (EFP) profiles. It also addresses methodological constraints and potential directions for future refinement. 5.1 Policy Implications: Targeted Strategies for Ithaca’s Energy Transition The City of Ithaca—among the first in the United States to commit to full decarbonization through the Ithaca Green New Deal (2019)(Ithaca, 2019)—faces the dual challenge of reducing greenhouse gas emissions while maintaining energy affordability. The findings from Chapters 49 4.2–4.4 indicate that these objectives are spatially uneven: energy burden and efficiency levels vary systematically across census tracts according to income, tenure, and housing characteristics. Therefore, a one-size-fits-all policy approach is thus insufficient; instead, interventions should be geographically and socioeconomically differentiated. A cluster-specific policy design is shown below: Table 11. A Cluster-Specific Policy Design Based on Socioeconomic Characteristics and Energy Profiles in Ithaca’s Five Clusters Cluster Tracts Key Features Energy Profile Policy Suggestion Cluster 1 2 Lowest income and high poverty, new rental apartments, young residents Fully electrified and medium energy burden Rental efficiency upgrades and winter cost support for student tenants. Cluster 2 8,14 Older homes and residents, moderate income, medium poverty, relatively high ownership Moderate energy efficiency and highest energy burden. traditional weatherization, fuel-switching incentives, landlord retrofits. Cluster 3 1,5,9, 11,12 Mixed income, diverse tenure, mid- aged housing. Balanced performance, moderate EFP and EBP. Scalable electrification with envelope improvements; ideal demonstration zone for retrofit programs. Cluster 4 3,4 Highest poverty, old rental housing, young residents Lowest energy efficiency and high energy burden Direct assistance, tenant protection, comprehensive retrofits. Cluster 5 6,7,10 ,13,15 Highest income and low poverty, high ownership, newer homes, old residents Moderate efficiency and lowest energy burden Heat-pump conversions, submetering, and college–city partnership retrofits. Cluster 1 – Low-Income, Electrified Rental Zones 50 Cluster 1 consists of newly developed but predominantly rental-based housing with the lowest income, high poverty, and young residents, largely corresponding to the student-oriented neighborhoods near Cornell and Collegetown’s periphery. Despite being fully electrified, these areas show medium energy burden, reflecting the high operational cost of electric heating and limited tenant control over energy systems. Policy interventions should focus on rental efficiency upgrades—such as heat-pump water heaters, envelope insulation, and thermostat management—implemented through landlord incentive programs or mandatory rental standards. Short-term measures like winter cost support or seasonal subsidies for student tenants could mitigate temporary affordability pressures without discouraging electrification. The city’s Energy Efficiency Resource Standard (EERS) framework could also extend to campus-affiliated rentals to ensure continuous efficiency improvements in this transitory population. Cluster 2 – Aging Owner-Occupied Homes with High Energy Burden Cluster 2 covers tracts 8 and 14, characterized by older housing stock, moderate income, and mid-level poverty, with a relatively high share of homeownership. These districts exhibit moderate energy efficiency but the highest energy burden, indicating aging structures with persistent heating inefficiencies. For this group, policy emphasis should be placed on traditional weatherization programs, such as attic insulation, window sealing, and furnace upgrades. Coupling these measures with fuel-switching incentives—transitioning from oil or gas to air-source heat pumps—can reduce 51 operational costs while advancing decarbonization. A landlord retrofit incentive scheme, especially for small multi-unit properties, would help overcome split incentives in mixed-tenure environments. Outreach partnerships with local contractors could ensure technical assistance and cost transparency for homeowners. Cluster 3 – Mid-Income, Mixed-Tenure Transition Zones Cluster 3 includes five tracts (1, 5, 9, 11, 12) displaying diverse tenure types, mixed incomes, and mid-aged housing stock. Energy indicators reveal balanced efficiency and burden, suggesting relatively stable performance but room for improvement. These areas represent the most scalable zones for citywide decarbonization pilots. Policies should target whole-home retrofit demonstrations, integrating envelope improvements, smart metering, and electrification of heating and water systems. Because this cluster spans various residential forms—from small apartment buildings to detached homes—it offers an ideal testbed for scalable retrofit models and workforce training under Ithaca’s electrification initiative. Coordinated funding from state clean energy programs (e.g., NYSERDA’s Clean Heat initiative) could be pooled with municipal capital to accelerate participation. Cluster 4 – High-Poverty Core with Obsolete Housing Cluster 4 (tracts 3 and 4) represents the most socioeconomically vulnerable cluster, characterized by very high poverty, older rental housing, and younger residents. These neighborhoods show the lowest energy efficiency and elevated energy burden, highlighting the urgent need for intervention. 52 Given their structural and financial constraints, the immediate priority should be direct cost relief through emergency heating programs and bill credits, paired with tenant-protection policies to prevent displacement during energy retrofits. Over the longer term, comprehensive building rehabilitation—including deep energy retrofits and electrified heating—must be coordinated through public-private partnerships. These neighborhoods should be prioritized in the rollout of Ithaca’s district heating network, ensuring that efficiency improvements align with social equity goals. Cluster 5 – High-Income, Electrification-Ready Suburban Areas Cluster 5 comprises the city’s most affluent tracts (6, 7, 10, 13, 15), marked by high income, low poverty, strong ownership, and newer or well-maintained housing stock. These areas demonstrate moderate efficiency and the lowest energy burden, indicating capacity for self- financed electrification. Policy focus here should shift toward upgrading to next-generation heat pumps, building- level submetering, and whole-home electrification financing to achieve net-zero performance. For the anomalous index 15 tract, which shows unexpectedly high energy burden relative to income—likely due to large, energy-intensive homes—policies should encourage deep retrofits such as building envelope optimization and geothermal system integration. Collaborative models between Cornell, the city, and homeowner associations could support voluntary “net-zero neighborhood” programs, showcasing carbon-neutral residential exemplars. 5.2 Methodological Limitations 53 While the clustering framework provides a useful typology of Ithaca’s community energy conditions, several methodological limitations require further discussion. First, the analysis operates at the census tract level, which constrains spatial resolution and may mask intra-tract variability. Ithaca’s compact urban structure encompasses neighborhoods with distinct functional identities—student housing, faculty districts, and historic cores—whose internal heterogeneity is not fully captured in tract-level aggregation. Second, the proxy measure of energy efficiency derived from building age and median income cannot capture equipment-level efficiency or behavioral patterns affecting energy use. Future studies could integrate building-level audit data or smart meter information to improve measurement precision. Finally, the use of k-means clustering presupposes linear separability and equal variance among clusters, assumptions that may not hold in urban energy systems characterized by nonlinear interactions between socioeconomic and spatial variables. Advanced unsupervised methods such as Gaussian mixture modeling or self-organizing maps could provide more robust delineations of heterogeneous urban energy profiles. 5.3 Future Directions The analytical approach developed in this study can be extended in two complementary directions. At the macro scale, this framework could inform the design of a citywide “digital twin” for electrification planning, as demonstrated in initiatives such as RMI’s How Digital Twins Are Enabling City-Wide Electrification. Embedding socioeconomic-energy cluster data into a 54 dynamic simulation environment would enable planners to visualize retrofit impacts, forecast energy demand shifts, and identify optimal investment sequences for achieving Ithaca’s 2030 carbon neutrality targets. At the micro scale, future work could refine community-level analysis through higher- resolution spatial and demographic data, identifying energy priorities at the neighborhood or parcel level. Given Ithaca’s distinctive urban composition—where small, functionally diverse communities coexist within short distances—such localized modeling could identify community- specific priorities and equity considerations. Tailored strategies, such as cooperative solar programs for student neighborhoods or targeted incentives for aging faculty housing, could then be aligned with broader municipal decarbonization pathways. Ultimately, integrating detailed community profiles with predictive modeling tools will enable the city to identify optimal electrification pathways that balance equity, efficiency, and carbon reduction goals. 6. Conclusion This study presents an integrated analytical framework to examine the relationship between socioeconomic characteristics, energy burden, and energy efficiency at the neighborhood level in Ithaca, New York. By combining multivariate analysis, principal component extraction, and K- means clustering, the research identified distinct community typologies that reflect both structural and behavioral determinants of local energy inequities. The results demonstrate that socioeconomic segmentation—particularly in income, tenure, and housing vintage—strongly 55 conditions spatial patterns of energy performance and affordability. These findings provide a quantitative basis for developing differentiated, place-based energy policies. The analysis revealed five distinct clusters of census tracts across Ithaca. Low-income rental neighborhoods exhibited high energy burdens despite electrification progress, underscoring the persistence of structural inefficiencies and tenant-landlord split incentives. Older, moderate- income zones faced similar burdens due to aging housing stock and partial adoption of weatherization measures. Meanwhile, mid-income mixed-tenure areas displayed moderate energy outcomes and represent optimal sites for pilot retrofit and electrification programs. High-poverty clusters within the urban core demand direct support and comprehensive building rehabilitation, while affluent suburban and university-adjacent tracts show readiness for advanced electrification and decentralized renewable integration. These differentiated profiles emphasize that equitable decarbonization requires locally adaptive solutions rather than uniform interventions. From a methodological standpoint, this research demonstrates how clustering-based analytical frameworks can bridge socioeconomic and technical energy data to uncover actionable spatial patterns. While limited by small sample size and aggregate tract-level granularity, the approach provides a scalable foundation for data-driven energy planning. Future research could extend this framework both horizontally—to encompass a broader regional scale across New York State—and vertically—to achieve finer community-level resolution through real-time building energy data, digital twin simulation, and household-level feedback integration. Such 56 refinements would enable the identification of micro-priorities and the optimization of retrofit pathways tailored to each community’s socioeconomic and infrastructural context. In conclusion, this thesis contributes to Ithaca’s vision of achieving a just, science-based transition toward carbon neutrality. By linking statistical evidence with policy design, it reinforces the principle that decarbonization and social equity must advance concurrently. The insights gained here can inform municipal decision-making, guide local electrification investments, and serve as a transferable framework for other mid-sized cities seeking to operationalize equity-centered energy transitions through data analytics. 57 REFERENCES [1]. (2023), I. World Energy Transitions Outlook 2023: 1.5°C Pathway. International Renewable Energy Agency, Abu Dhabi, Volume 1. [2]. (IEA), I. E. A. (2024). World Energy Investment 2024. Rep. IEA. https://www.iea.org/reports/world-energy-investment-2024 [3]. Aguilar-Hernandez, G. A., Sigüenza-Sanchez, C. P., Donati, F., Rodrigues, J. F., & Tukker, A. (2018). Assessing circularity interventions: a review of EEIOA-based studies. Journal of Economic Structures, 7(1), 1-24. [4]. Ahmed, I. A., Effiong, E., Shanmugasundaram, L., Hummer, L., Orttung, R. W., & Shittu, E. (2025). Measuring the invisible: A framework for quantifying energy justice in urban communities in the United States. Energy Research & Social Science, 122, 103986. https://doi.org/https://doi.org/10.1016/j.erss.2025.103986 [5]. Aigner, D., Lovell, C. A. K., & Schmidt, P. (1977). Formulation and estimation of stochastic frontier production function models. Journal of Econometrics, 6(1), 21-37. https://doi.org/https://doi.org/10.1016/0304-4076(77)90052-5 [6]. Al Kez, D., Foley, A., Lowans, C., & Del Rio, D. F. (2024). Energy poverty assessment: Indicators and implications for developing and developed countries. Energy Conversion and Management, 307, 118324. https://doi.org/https://doi.org/10.1016/j.enconman.2024.118324 https://www.iea.org/reports/world-energy-investment-2024 https://doi.org/https:/doi.org/10.1016/j.erss.2025.103986 https://doi.org/https:/doi.org/10.1016/0304-4076(77)90052-5 https://doi.org/https:/doi.org/10.1016/j.enconman.2024.118324 58 [7]. Amowine, N., Ma, Z., Li, M., Zhou, Z., Yaw Naminse, E., & Amowine, J. (2020). Measuring dynamic energy efficiency in Africa: A slack‐based DEA approach. Energy Science & Engineering, 8(11), 3854-3865. [8]. Anselin, L. (1988). Spatial Econometrics: Methods and Models. Kluwer Academic Publishers. [9]. Anselin, L. (1995). Local Indicators of Spatial Association—LISA. Geographical Analysis, 27(2), 93-115. https://doi.org/https://doi.org/10.1111/j.1538-4632.1995.tb00338.x [10]. Anselin, L., Gallo, J. L., & Jayet, H. (2008). Spatial panel econometrics. In The econometrics of panel data: Fundamentals and recent developments in theory and practice (pp. 625-660). Springer. [11]. Awolesi, O., Salter, C. A., & Reams, M. (2024). A Systematic Review on the Path to Inclusive and Sustainable Energy Transitions. Energies, 17(14). [12]. Bird, S., & Hernández, D. (2012). Policy options for the split incentive: Increasing energy efficiency for low-income renters. Energy Policy, 48, 506-514. https://doi.org/https://doi.org/10.1016/j.enpol.2012.05.053 [13]. Boardman, B. (1991). Fuel Poverty: From Cold Homes to Affordable Warmth. Belhaven Press. [14]. Bouzarovski, S., & Simcock, N. (2017). Spatializing energy justice. Energy Policy, 107, 640-648. https://doi.org/https://doi.org/10.1016/j.enpol.2017.03.064 https://doi.org/https:/doi.org/10.1111/j.1538-4632.1995.tb00338.x https://doi.org/https:/doi.org/10.1016/j.enpol.2012.05.053 https://doi.org/https:/doi.org/10.1016/j.enpol.2017.03.064 59 [15]. Bullard, R. D. (2015). Environmental Justice in the United States. In J. D. Wright (Ed.), International Encyclopedia of the Social & Behavioral Sciences (Second Edition) (pp. 756- 762). Elsevier. https://doi.org/https://doi.org/10.1016/B978-0-08-097086-8.91013-4 [16]. Bureau, U. S. C. (2024). American Community Survey 5-year estimates, 2019–2023. https://www.census.gov/programs-surveys/acs [17]. Charnes, A., Cooper, W. W., & Rhodes, E. (1978). Measuring the efficiency of decision making units. European Journal of Operational Research, 2(6), 429-444. https://doi.org/https://doi.org/10.1016/0377-2217(78)90138-8 [18]. Chen, Z.-g., Kong, L.-j., Wang, M., Liu, H.-k., Xiao, D.-k., & Wu, W.-p. (2021). The Effect of Urban Spatial Form on Energy Efficiency: A Cross-Sectional Study in China [Original Research]. Frontiers in Energy Research, Volume 9 - 2021. https://doi.org/10.3389/fenrg.2021.792199 [19]. Chu, J., Wu, J., Zhu, Q., An, Q., & Xiong, B. (2019). Analysis of China’s regional eco- efficiency: A DEA two-stage network approach with equitable efficiency decomposition. Computational economics, 54(4), 1263-1285. [20]. Cui, H.-y., & Cao, Y.-q. (2024). Do smart cities improve energy efficiency? A test of spatial effects and mechanisms. Sustainable Cities and Society, 101, 105124. https://doi.org/https://doi.org/10.1016/j.scs.2023.105124 https://doi.org/https:/doi.org/10.1016/B978-0-08-097086-8.91013-4 https://www.census.gov/programs-surveys/acs https://doi.org/https:/doi.org/10.1016/0377-2217(78)90138-8 https://doi.org/10.3389/fenrg.2021.792199 https://doi.org/https:/doi.org/10.1016/j.scs.2023.105124 60 [21]. Fowlie, M., Greenstone, M., & Wolfram, C. (2018). Do Energy Efficiency Investments Deliver? Evidence from the Weatherization Assistance Program*. The Quarterly Journal of Economics, 133(3), 1597-1644. https://doi.org/10.1093/qje/qjy005 [22]. Gillingham, K., Rapson, D., & Wagner, G. (2016). The Rebound Effect and Energy Efficiency Policy. Review of Environmental Economics and Policy, 10(1), 68-88. https://doi.org/10.1093/reep/rev017 [23]. Heffron, R. J., & McCauley, D. (2017). The concept of energy justice across the disciplines. Energy Policy, 105, 658-667. https://doi.org/https://doi.org/10.1016/j.enpol.2017.03.018 [24]. Heffron, R. J., & McCauley, D. (2018). What is the ‘Just Transition’? Geoforum, 88, 74-77. https://doi.org/https://doi.org/10.1016/j.geoforum.2017.11.016 [25]. Hernández, D. (2016). Understanding ‘energy insecurity’ and why it matters to health. Social Science & Medicine, 167, 1-10. https://doi.org/https://doi.org/10.1016/j.socscimed.2016.08.029 [26]. Hoffman, J., Davies, M., Bauwens, T., Späth, P., Hajer, M. A., Arifi, B., Bazaz, A., & Swilling, M. (2021). Working to align energy transitions and social equity: An integrative framework linking institutional work, imaginaries and energy justice. Energy Research & Social Science, 82, 102317. https://doi.org/https://doi.org/10.1016/j.erss.2021.102317 [27]. Homsy, G. C., & Kang, K. E. (2025). Energy burden: Exploring the intersection of race, income, and community characteristics across the United States. Energy Research & Social Science, 127, 104207. https://doi.org/https://doi.org/10.1016/j.erss.2025.104207 https://doi.org/10.1093/qje/qjy005 https://doi.org/10.1093/reep/rev017 https://doi.org/https:/doi.org/10.1016/j.enpol.2017.03.018 https://doi.org/https:/doi.org/10.1016/j.geoforum.2017.11.016 https://doi.org/https:/doi.org/10.1016/j.socscimed.2016.08.029 https://doi.org/https:/doi.org/10.1016/j.erss.2021.102317 https://doi.org/https:/doi.org/10.1016/j.erss.2025.104207 61 [28]. Ithaca, C. o. (2019). Ithaca Green New Deal: A policy framework for achieving community- wide carbon neutrality by 2030. https://www.cityofithaca.org/DocumentCenter/View/10465/Ithaca-Green-New-Deal- Resolution [29]. Jenkins, K., McCauley, D., Heffron, R., Stephan, H., & Rehner, R. (2016). Energy justice: A conceptual review. Energy Research & Social Science, 11, 174-182. https://doi.org/https://doi.org/10.1016/j.erss.2015.10.004 [30]. Johansson, T. B., Patwardhan, A. P., Nakićenović, N., & Gomez-Echeverri, L. (2012). Global energy assessment: toward a sustainable future. Cambridge University Press. [31]. Kola-Bezka, M. (2023). Think Global Act Local: In search for ways to increase the engagement of local communities in energy transition. Energy Reports, 9, 1668-1683. https://doi.org/https://doi.org/10.1016/j.egyr.2022.12.143 [32]. LeSage, J., & Pace, R.K. (2009). Introduction to Spatial Econometrics (1st ed.). (Chapman and Hall/CRC). https://doi.org/https://doi.org/10.1201/9781420064254 [33]. McCauley, D. (2017). Energy justice: Re-balancing the trilemma of security, poverty and climate change. Springer. [34]. McCauley, D., Ramasar, V., Heffron, R. J., Sovacool, B. K., Mebratu, D., & Mundaca, L. (2019). Energy justice in the transition to low carbon energy systems: Exploring key themes in interdisciplinary research. Applied Energy, 233-234, 916-921. https://doi.org/https://doi.org/10.1016/j.apenergy.2018.10.005 https://www.cityofithaca.org/DocumentCenter/View/10465/Ithaca-Green-New-Deal-Resolution https://www.cityofithaca.org/DocumentCenter/View/10465/Ithaca-Green-New-Deal-Resolution https://doi.org/https:/doi.org/10.1016/j.erss.2015.10.004 https://doi.org/https:/doi.org/10.1016/j.egyr.2022.12.143 https://doi.org/https:/doi.org/10.1201/9781420064254 https://doi.org/https:/doi.org/10.1016/j.apenergy.2018.10.005 62 [35]. Memmott, T., Carley, S., Graff, M., & Konisky, D. M. (2021). Sociodemographic disparities in energy insecurity among low-income households before and during the COVID-19 pandemic. Nature Energy, 6, 186-193. https://doi.org/10.1038/s41560-020-00763-9 [36]. Mushafiq, M., Arisar, M. M., Tariq, H., & Czapp, S. (2023). Energy Efficiency and Economic Policy: Comprehensive Theoretical, Empirical, and Policy Review. Energies, 16(5). [37]. Nations, U. (2015). Transforming Our World: The 2030 Agenda for Sustainable Development. United Nations General Assembly., (SDG7/13 context). [38]. Reames, T. (2016). A community-based approach to low-income residential energy efficiency participation barriers. Local Environment, 21, 1-18. https://doi.org/10.1080/13549839.2015.1136995 [39]. Rosenow, J., Kern, F., & Rogge, K. (2017). The need for comprehensive and well targeted instrument mixes to stimulate energy transitions: The case of energy efficiency policy. Energy Research & Social Science, 33, 95-104. https://doi.org/https://doi.org/10.1016/j.erss.2017.09.013 [40]. Sovacool, B. K. (2015). Fuel poverty, affordability, and energy justice in England: Policy insights from the Warm Front Program. Energy, 93, 361-371. https://doi.org/https://doi.org/10.1016/j.energy.2015.09.016 https://doi.org/10.1038/s41560-020-00763-9 https://doi.org/10.1080/13549839.2015.1136995 https://doi.org/https:/doi.org/10.1016/j.erss.2017.09.013 https://doi.org/https:/doi.org/10.1016/j.energy.2015.09.016 63 [41]. Sovacool, B. K., Burke, M., Baker, L., Kotikalapudi, C. K., & Wlokas, H. (2017). New frontiers and conceptual frameworks for energy justice. Energy Policy, 105, 677-691. https://doi.org/https://doi.org/10.1016/j.enpol.2017.03.005 [42]. Sovacool, B. K., & Dworkin, M. H. (2015). Energy justice: Conceptual insights and practical applications. Applied Energy, 142, 435-444. https://doi.org/https://doi.org/10.1016/j.apenergy.2015.01.002 [43]. Spielman, S. E., & Logan, J. R. (2013). Using High-Resolution Population Data to Identify Neighborhoods and Establish Their Boundaries. Annals of the American Association of Geographers, 103, 67-84. https://doi.org/10.1080/00045608.2012.685049 [44]. Spielman, S. E., & Singleton, A. (2015). Studying Neighborhoods Using Uncertain Data from the American Community Survey: A Contextual Approach. Annals of the Association of American Geographers, 105(5), 1003-1025. https://doi.org/10.1080/00045608.2015.1052335 [45]. Thomas, M., DeCillia, B., Santos, J. B., & Thorlakson, L. (2022). Great expectations: Public opinion about energy transition. Energy Policy, 162, 112777. https://doi.org/https://doi.org/10.1016/j.enpol.2022.112777 [46]. UNDP, U. (2000). WEC (2000) World Energy Assessment: energy and the challenge of sustainability. United Nations Development Programme, New York. [47]. V., S. (2021). Grand Transitions: How the Modern World Was Made. Oxford University Press https://doi.org/https:/doi.org/10.1016/j.enpol.2017.03.005 https://doi.org/https:/doi.org/10.1016/j.apenergy.2015.01.002 https://doi.org/10.1080/00045608.2012.685049 https://doi.org/10.1080/00045608.2015.1052335 https://doi.org/https:/doi.org/10.1016/j.enpol.2022.112777 64 [48]. Walker, R., Liddell, C., McKenzie, P., Morris, C., & Lagdon, S. (2014). Fuel poverty in Northern Ireland: Humanizing the plight of vulnerable households. Energy Research & Social Science, 4, 89-99. https://doi.org/https://doi.org/10.1016/j.erss.2014.10.001 [49]. Wang, X., & Zhou, D. (2023). The underlying drivers of energy efficiency: a spatial econometric analysis. Environmental Science and Pollution Research, 30(5), 13012-13022. https://doi.org/10.1007/s11356-022-23037-1 [50]. Wei, T., Duan, Z., & Xie, P. (2023). Spatial disparities and variation sources decomposition of energy poverty in China. Journal of Cleaner Production, 421, 138498. https://doi.org/https://doi.org/10.1016/j.jclepro.2023.138498 [51]. Yang, Y., Xia, S., Huang, P., & Qian, J. (2024). Energy transition: Connotations, mechanisms and effects. Energy Strategy Reviews, 52, 101320. https://doi.org/https://doi.org/10.1016/j.esr.2024.101320 https://doi.org/https:/doi.org/10.1016/j.erss.2014.10.001 https://doi.org/10.1007/s11356-022-23037-1 https://doi.org/https:/doi.org/10.1016/j.jclepro.2023.138498 https://doi.org/https:/doi.org/10.1016/j.esr.2024.101320 1. Introduction 1.1 Background and Motivation 1.2 Problem Statement 1.3 Research Question and Objectives 1.4 Thesis Structure Overview 2. Literature review 2.1 Energy Justice: Foundations, Principles, and Relevance 2.2 Energy Poverty, Energy Burden, and Their Primary Determinants 2.2.1 Conceptualizing Energy Poverty in Developed Contexts 2.2.2 Drivers of Energy Poverty: Income, Housing, and Prices 2.3 Residential Energy Efficiency: Measurement, Drivers, and Policy Implications 2.3.1 Measurement Approaches 2.3.2 Drivers of Energy Efficiency 2.3.3 Efficiency, Electrification, and Affordability 2.4 Socioeconomic and Housing Determinants of Energy Outcomes 2.5 Spatial Analytical Approaches in Energy Research 2.6 Research Gap: The Need for Fine-Scale, Context-Specific Analysis in Small Cities 3. Data and Study Scope 3.1 Study Scope 3.2 Data Source 3.3 Variable Construction 3.3.1 Data Structure 3.3.2 Socioeconomic and Housing Characteristic 3.3.3 Energy Burden Proxy (EBP) 3.3.4 Energy Efficiency Proxy (EFP) 3.4 Data Pre-Processing 3.5 Data Quality and Limitations 3.6 Reproducibility and Software 4. Methodology and Results 4.1 Analytical Framework 4.2 Module 1 – Socioeconomic Structure 4.2.1 Correlations 4.2.2 PCA Eigenvectors 4.2.3 K-means clustering 4.2.4 Summary 4.3 Module 2 – Energy Burden (EBP) 4.3.1 Spatial Analysis 4.3.2 Association with Socioeconomic Factors 4.4 Module 3 – Energy Efficiency Proxy (EFP) 4.4.1 Association with Socioeconomic Factors 4.4.2 Spatial Distribution of Energy Efficiency 4.4.3 Relationship Between Energy Efficiency and Energy Burden 4.4.4 Summary 4.5 Synthesis of Findings 5. Discussion 5.1 Policy Implications: Targeted Strategies for Ithaca’s Energy Transition 5.2 Methodological Limitations 5.3 Future Directions 6. Conclusion References