|UFDC Home||myUFDC Home | Help|
This item has the following downloads:
1 IDENTIFYING THE EFFEC TS OF ENERGY EFFICIENT HOUSES ON ENERGY COMSUMPTION By ELOISE FRANCESCA AKA A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE UNIVERSITY OF FLORIDA 2009
2 2009 Eloise Francesca Aka
3 To my mother Beatrice Aka in Tunis (Tunisia) my husba nd Kenneth Watkins and loving family in Abidjan ( Cote dIvoire) and in the United States
4 ACKNOWLEDGMENTS I wish to express appreciation to my supervisory committee chairman, Dr. Carmen Carri n Flores, for her lasting patience, con s tant support and encouragement throughout all the phases of the study and graduate program. Credit is also due to Dr. Alfonso Flores -Lagunes member of the supervisory committee, for his constructive criticisms and suggestions in the direction of the study. Much valuable assistance in collecting the data was provided by Mr. Nicolas Taylor and Ms Megan Silbert Mr. Michael Dix also assisted in the preparation of descriptive maps In addition, much appreciated help was provided by the staff members of the Gainesville Energy Efficient Communit ies and the GeoPlan Center at the University of Fl orida. To these and others, I am deeply grateful.
5 TABLE OF CONTENTS page ACKNOWLEDGMENTS .................................................................................................................... 4 LIST OF TABLES ................................................................................................................................ 7 LIST OF FIGURES .............................................................................................................................. 9 LIST OF ABBREVIATIONS ............................................................................................................ 10 ABSTRACT ........................................................................................................................................ 11 CHAPTER 1 INTRODUCTION ....................................................................................................................... 12 2 LITERATURE REVIEW ........................................................................................................... 17 2.1 Energy Efficiency Programs ................................................................................................ 17 2.1.1 The Energy Star Program ........................................................................................ 18 2.1.2 Energy Star Homes .................................................................................................. 20 2.2 Determinants of Energy Consumption ................................................................................ 21 2.3 Conclusion ............................................................................................................................. 25 3 METHODOLOGY ...................................................................................................................... 26 3.1 Cost Minimization Behavior ................................................................................................ 26 3.2 The Difference in Difference Estimator ............................................................................. 28 3.2.1 Defin ition and Applications .................................................................................... 28 3.2.2 Mechanics of the DID Estimator ............................................................................ 29 3.2.3 Assumptions and Limitations .................................................................................. 30 3.3 Time Series and Cross Sectional Models ............................................................................ 33 3.3.1 Definitions and Applications ................................................................................... 33 3.3.2 The Pooled Or dinary Least Squares Model ........................................................... 35 3.3.3 Fixed Effects Model ................................................................................................ 37 3.4 Selection of Sample .............................................................................................................. 38 3.4.1 Location of the Study .............................................................................................. 38 3.4.2 Characteristics of the Subdivisions ......................................................................... 39 3.4.3 Energy consumption data ........................................................................................ 41 3.4.3 Other Data ................................................................................................................ 41 3.4.4 Data Cleaning ........................................................................................................... 42 3.5 Conclusions ........................................................................................................................... 42 4 ENERGY ANALYSIS USING DIFFERENCE IN DIFFERENCE ESTIMATOR ............... 45 4.1 Characteristic of Sample ....................................................................................................... 45 4.2 Energy Consumption Analysis ............................................................................................. 46
6 4.2 .1 Electricity Consumption .......................................................................................... 46 4.2.2 Gas Consumption ..................................................................................................... 50 4.2.3 Total Consumption .................................................................................................. 51 4.3 Conclusions ........................................................................................................................... 52 5 ENERGY PANEL DATA ESTIMATION ................................................................................ 62 5.1 Description of the Datasets ................................................................................................... 62 5.2 Estimations and Empirical Results ...................................................................................... 64 5.2.1 Monthly O LS Models and Results .......................................................................... 64 5.2.2 Pooled Ordinary Least Squares Estimation and Results ....................................... 68 5.2.3 Panel Data Estimations and Results ....................................................................... 70 5.3 Difference -in -Difference Estimations and Results ............................................................. 73 5 .4 Conclusions ........................................................................................................................... 78 6 CONCLUSIONS ......................................................................................................................... 90 6.1 Overview of the Research .................................................................................................... 90 6.2 Limitations of the Study ....................................................................................................... 91 6 .3 General Conclusions and Policy Implications .................................................................... 91 LIST OF REFERENCES ................................................................................................................... 94 BIOGRAPHICAL SKETCH ............................................................................................................. 97
7 LIST OF TABLES Table page 4 1 Description of the Neighborhoods ........................................................................................ 55 4 2 Variable Names and Definition ............................................................................................. 55 4 3 Descriptive Statistics of Variables ........................................................................................ 55 4 4 Results of Model 1 Bill Year 2006 ...................................................................................... 55 4 5 Resul ts of Model 1 Bill Year 2000 ...................................................................................... 56 4 6 Average Energy Consumptions (EC) .................................................................................... 56 4 7 Results of Electricity Consumption Analysis using Sample 1 ............................................ 57 4 8 Results of Model 1 Bill Year 2000 ...................................................................................... 58 4 9 Results of Model 1 Bill Year 2006 ...................................................................................... 58 4 10 Results of Natural Gas Consumption Analysis using Sample1 .......................................... 59 4 11 Results of Model1Bill Year 2000 ....................................................................................... 60 4 12 Resul ts of Model1Bill Year 2006 ....................................................................................... 60 4 13 Results of Total Consumption Analysis using Sample 1 ..................................................... 61 5 1 Variables Description ............................................................................................................. 81 5 2 Descriptive Statistics .............................................................................................................. 82 5 3 Monthly OLS Estimation Results ......................................................................................... 82 5 4 Descriptive S tatistics of Conventional Non-ES Homes ...................................................... 83 5 5 Descriptive Statistics of ES Homes ....................................................................................... 83 5 6 Parameter Estimates, Standard Errors, and P -va lues of Pooled OLS Estimation using Annual PTSCS Data ............................................................................................................... 84 5 7 F Statistics and P values of Autocorrelation Test ................................................................ 85 5 8 Parameter Estima tes, Standard Errors, and P -values of Interaction Term Variable in FE Estimation after Accounting for Autoregressive Disturbances AR (1) ........................ 85
8 5 9 Parameter Estimates, Standard Errors, and P -values of values of Interaction Term Variable in FE Estimation without Accounting for Autoregressive Disturbances AR (1) ............................................................................................................................................ 85 5 10 Parameter Estimates, Standard Errors, and P -values of the DID Estimation using Total Annual Electricity Consumptions ............................................................................... 86 5 11 Parameter Estimates, Standard Errors, and P -values of the DID Estimation using Summer Annual Electricity Consumptions .......................................................................... 87 5 12 Parameter Estimates, Standard Errors, and P -values of the DID Estimation using Winter Annual Electricity Consumptions ............................................................................. 88 5 13 Parameter Estimates, Stan dard Errors, and P -values of the DID Estimation using Other Annual Electricity Consumptions ............................................................................... 89
9 LIST OF FIGURES Figure page 3 1 The Cost Minimizing Probl em (CMP) ................................................................................. 43 3 2 Map of Alachua County, Florida created using Alachua County GIS parcels data. .......... 43 3 3 Map of all five subdivisions in A lachua County, Florida using Alachua County GIS parcels data ............................................................................................................................. 44 4 1 2005 Average Household Energy Usage Expenditures in Florida created using 2005 Residential Energy and Consumption Survey ...................................................................... 54
10 LIST OF ABBREVIATIONS AECC American Council for EnergyEfficient Economy AR Auto Regression CMP Cost Minimization Problem DID Difference -in -Difference DOE Department of Energy EC Energy Consumption EPA Environmental Protectio n Agency ES Energy Star FE Fixed Effect FEMP Federal Energy Management Program GHG Green House Gas GISS Goddard Institute for Space Studies GLS Generalized Least Squares GRU Gainesville Regional Utilities HERS Home Energy Rating System IPCC Intergovernm ental Panel for Climate Change kWh Kilowatt hours NASA National Aeronautics and Space Administration OLS Ordinary Least Squares PTSCS Pooled Time Series Cross -Section RESNET Residential Energy Services Network TSCS Time Series Cross -Section US U nited S tat es
11 Abstract of Thesis Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Science IDENTIFYING THE EFF ECTS OF ENERGY EFFICIENT HOUSES ON ENERGY COMSUMPTION: By Eloise Francesca Aka August 2009 Chair : Carmen Carri n -Flores Major: Food and Resource Economics As environmental protection and conservation have become prominent policy issues in light of global warming, various proposals have been made regardin g how to decrease greenhouse emissions, a key factor in raising global temperatures (IPCC, 2007) For example, Energy Star is a joint program of the U nited S tates Environmental Protection Agency (EPA) and the Department of Energy (DOE) that provides strict energy efficient guidelines which make an Energy Star home have a better performance in terms of energy use over time. By better performance, we mean lower energy consumption at the meter. Thus those houses which are rated as Energy Star will use less en ergy than those houses which lack this rating. The first Energy Star home was built in Gainesville, Florida in 1997. Since then, the number of Energy Star homes built in Gainesville has steadily risen. Although energy savings of at least 15% are ratified f or new homes, little is known of the performance of these homes over time. Results suggest that over time there is an increase of energy consumption of houses rated as Energy Star. We conclude that these guidelines are effective in reducing energy consumpt ion when the house is built but they may not hold energy savings over time.
12 CHAPTER 1 INTRODUCTION It is no secret today that the face of our planet is rapidly changing. In fact, according to the Intergovernmental Panel on Climate Change (IPCC) Annual report 2007, climate change or global warming due to increased greenhouse gas (GHG) emissions in the process of fossil fuel based energy production (coal, oil and natural gas) is one of the most significant environmental challenges that we a re facing as a global society. According to the Earth Policy Institute since 1970, the earths average temperature has risen by 0.8 degrees Celsius, or nearly 1.4 degrees Fahrenheit. During this span, the rise in temperature each decade was greater than in the precedi ng one. The same climate studies conducted by the Goddard Institute for Space Studies (GISS) show that the last two decades of the 20th century's were the hottest in 400 years and possibly the warmest for several millennia Industrialization, deforestat ion, and pollution have greatly increased atmospheric concentrations of water vapor, carbon diox ide, methane, and nitrous oxide which are all greenhouse gases that help trap heat near Earth's surface In other words, Humans are pouring carbon dioxide into the atmosphere much faster than plants and oceans can absorb it according to the IP CC Based on the 2005 Residential Energy Consumption Survey, out of the four most populated states, Florida has the highest average electricity consumption per household with 15,862 kilowatts ( k Wh) followed by Texas (15,149 kWh), California (6,992 kWh) and New York (6,882 kWh). According to the IPCC Annual Report 2001, with such temperatu re changes are associated lost of densely populated areas, such as low lying coastal regions, glacier retreat Arctic shrinkage an altered patterns of agriculture. Secondary and regional effects include extreme weather events, an expansion of tropical diseases changes in the timing of seasonal patterns in ecosystems and an overall tremend ous economic impact
13 Since our environment is being threatened by significant environmental and economic consequences due to climate ch ange there is an apparent need for environmental protection and conservation efforts. Over the years, several energy efficiency policies have been initiated in the effort of reducing greenhouse gas emissions. One of those initiatives, the Energy Star (ES ) program was crea ted in 1993 as a joint program of the U nited States (US) Environmental Protection Agency (EPA) and the US Department of Energy (DOE) in the effort of reduc ing energy cost and protect ing the environment through the implementation of energy efficient product s and practices. Computers and monitors were the first labeled products Today the EPA has extended the label to cover new homes commercial and industrial buildings During the past years of execution of this voluntary program, the EPA has published several annual reports reflecting the performance of this initiative. In fact, the EPA 2001 report states that so far 80 billion kilowatt hours of energy have been saved; over 57,000 Energy Star rated homes were built which corresponds to savings of more than $15 million annually from lower energy costs. Although EPA reports claim significant energy savings and despite the theoretical as well as practical interest in energy efficient policies, a very slim body of the academic literature examines the performance of such programs over time. Most studies in the current literature investigate the relationship between implementing energy efficiency policies, energy consumption and cost effectiveness with a positive effect of such programs in reducing energy consumpt ion as well as increasing energy savings. However, little is known about performance over time. This study will therefore help to further our understanding of efficiency programs by examining specifically how such initiatives like the ES program actually p erform over time. We do so by identifying the effect of the E S rating on energy consumption in Gainesville, Florida. After performing a Difference in Difference
14 estimation (DID) using monthly energy consumption data of 5664 single -homes for the years 2000 and 2006, results suggest that over time there is an increase of energy consumption of houses rated as Energy Star of roughly 7.59% These results lead to the conclusion that energy efficiency guidelines are successful in reducing energy consumption when t he house is built but they do not maintain energy savings over time These results are even more compelling knowing that in terms of market penetration, Gainesville is one of the most productive markets for ES homes nationwide. In fact, the first ES home w as built in 1997. Since then, the number of ES homes built in Gainesville has steadily risen (Smith and Jones 2003). In fact, as in 2006 a bout 769 homes have been certified under the Energy Star program in Alachua County, which can include homes inside a nd outside the Gainesville city limits (Crabbe, 2006) Energy Star certified home builders in the area include well known companies such as Atlantic Design Homes and G.W. Robinson. In addition, this study is the f irst to employ a DID estimator as a tool t o evaluate the performance of the ES rating. It also includes four different control groups in the analysis in order to avoid the limitations of this evaluation method. Since the initial DID use s only the years 2000 and 2006, the sample size is expanded t o cover bill years from 1996 to 2008 with the intention of strengthening the results by capturing any existing trends in consumption. With this new yearly time series cross -sectional data created, the study adopts additional econometric methods such as poo led Ordinary Least Squares (OLS) and fixed effect models. Only yearly electricity consumption is analyzed and divided into a total, winter, summer and other consumption (i.e. off season consumption). With the pooled OLS estimation, ES homes consume 34.695 k Wh more during the winter and 51.374 kWh more during the summer than nonES homes. The results are statistically significant ES homes
15 consumption decreases by 18.54 kWh and 1.846 kWh respectively for the total and other consumption. However, these result s are not significant. A downfall of the pooled OLS model is that is ignores heterogeneity between the houses and leads to inaccurate conclusions. The panel data models on the other hand, incorporate a time trend variable that capture s a time effect in the relationship between energy efficient construction and household electricity consumption, while controlling for house specific effects. With the fixed effect models, house s rated ES consume 89.63kWh of electricity more than traditional houses. During the summer these houses also consume 74.320 kWh more than traditional houses and 34.22 kWh more during the off season months. Consumption decreases by 61.57 kWh for ES homes during the winter This change can be explained by the seasonality of electricity cons umption It is important to note that all of these results are statistically significant at the 5% significance level with a P -value of 0.000. Finally, another DID performed using the yearly time series cross sectional data reveals that ES homes consumes 4 .989 kWh more than traditional home during the winter, 12.086 kWh more during the summer and 13.13 kWh more in total consumption. Yet, ES house s consume 3.942l kWh less than traditional homes during the off -season months Winter and off season results are not statically significant. The main finding is that when using a longer time series cross -sectional data, ES houses on average consume more electricity that traditional homes, especially during the summer months. This difference in levels of consumption between the two types of homes also increases over time. The balance of the thesis is organized as follows : Chapter 2 presents the literature review. Chapter 3 discusses the conceptual framework, methodology and data used in the study. Chapter 4 presents energy consumption analysis using a DID e stimation for the years 2000 and 2006. In Chapter 5 a pooled cross sectional data and fixed effect estimation are performed as well as
16 another DID estimation using yearly electricity consumption Finally, concluding thoughts and future research follow in chapter 6
17 CHAPTER 2 LITERATURE REVIEW The purpose of this chapter is to review the current literature on the effect of energy efficiency initiatives and policies on energy consumption, homeownership costs, public investments, land use patterns and regulations. In order to do so, a working definition of energy efficiency program is require d In addition, predictors of energy cons umption are also outlined. In this chapter, section 2.1 provides a working definition of energy efficiency and describes an example of an energy efficiency program of interest in this study implemented in the US: the Energy Star program. Section 2.2 summarizes the impact of energy efficiency efforts on homeownership costs, investment and land use patterns Predictors of household energy consumption are also discussed. Finally, section 2. 3 presents the conclusion of the literature review. 2.1 Energy Efficiency Programs According to the American Council for Energy -Efficient Economy (AECC), an E nergy Efficient Program is an organized effort to try to encourage and facilitate customer implementation of energy efficiency improvements (residential and business) By energy efficiency improvements the author mean s measures that result in producing t he same or better levels of amenities such as light, space conditioning, heating, motor drive power, etcwhile using less energy (Kushler, 2009). The main idea of Energy Efficiency as a utility system resource is that utility systems need to have adequate supply resources to meet customer demand. Faced with an increasing energy demand, one can increase supply of energy resources; reduce customer demand or a combination of both in order to keep the system in balance. With the reality of rising greenhouse em issions due to human activity associated with climate change threatening our environment a nd
18 economy, it is literally in all cases today vital for the overall welfare of society to reduce consumer demand for energy consumption, especially demand for elect ricity and natural gas. While Load Management programs only seek to reduce peak demand during specific, limited time periods, by temporarily restraining electricity use or shifting usage to other time periods, Energy Efficiency programs seeks to reduce ene rgy demand at all times, reduce total energy consumption, lessen consumption of natural resources, decrease greenhouse gas emissions associated with energy consumption and thus trim down national energy import and energy dependence As a result according t o the American Council for EnergyEfficient Economy, the following factors are key element in the implementation of an Energy Efficiency program: Public information, education and training Economic incentives for consumers (i.e. rebates, tax credit) Quality control, monitoring and evaluation With the new administration in place putting forwards a new Stimulus Package, the government is furthermore active in the implementation of sound, cost -effective energy management and investment practices to enhan ce the nation's energy security and environmental stewardship according to the DOE's Federal Energy Management Program (FEMP) Thus t his study focus on a specific example of an Energy Efficient program: the Energy Star program, which is discussed in more details in the following sub-section. 2.1.1 The Energy Star P rogram Energy Star is a term that includes a wide array of programs, all designed to promote energy efficient investments. In fact, the Energy Star initiative began with a limited agenda in the early 1990s, after the 1992 Energy Policy Act directed by the EPA The idea was to implement a program that will identify and designate particularly energy efficient products and provide
19 estimates of the relative energy efficiency of those products. In ad dition, this legislation was designed to reward the most energy efficient products with positive advertising, thereby encouraging consumers to buy those products and other manufacturers to improve the energy efficiency of their own products. The Energy Sta r designation is completely voluntary and has been used by manufacturers as a selling point. Currently, the EPA and D OE jointly run the voluntary labeling program. The program started with only computers and monitors and, by 1995, expanded to include addi tional office products and residential heating and cooling equipment. In 1996, EPA partnered with DOE to add other product categories to the labeling program. In the following years, the Energy Star voluntary labels were extended to cover a wide array of p roducts, with over 35 product categories, including: major appliances, office equipment, home electronics, commercial and industrial equipment, lighting, plumbing even new homes and commercial and industrial buildings. The definition of qualifying Energy S tar products is different for each product category, but tends to include only the most efficient products on the market a small fraction of the total market. This is not always the case, however. The vast majority of computers, monitors, copiers, faxes, VCRs, TVs, and exit signs are Energy Star -qualified. In addition to the Energy Star voluntary labeling program, Energy Star also encompasses a variety of public -private partnerships, many of which began as separate programs and were moved under the spons orship of the Energy Star program in the late 1990s. For instance, the EPA Green Lights Program was started in 1991 to advance the adoption of energy efficient lighting systems in industrial and commercial facilities through information and demonstration a ctivities. Similarly, the EPA Climate Wise program was created in the mid -1990s to provide information and assistance to industrial and commercial facilities to identify and implement
20 greenhouse gas emissions -reducing activities. These programs joined the Energy Star umbrella of programs in the late 1990s due to their similarity in mission to the core Energy Star mission. Other programs include: the Green Power partnership encouraging organizations to buy renewable energy, the Combined Heat and Power partne rship between the government a nd industry, and Energy Star Home Sealing, which helps homeowners improve the energy performance of their homes during remodeling and renovation. By 2001, Energy Star facilitated partnerships between the government and over 7, 000 public and private sector organizations based on the EPA 2003 Annual Report The following sub -section describes the structural characteristics that make a house Energy Star qualified. 2.1.2 Energy Star Homes As mention ed earlier, the Energy Star progr am extends to residential properties. In fact, an Energy Star rated home is expected to perform better than a conventional home due to an improved home envelope which includes: Energy efficient home sealing (insulation and air sealing), Energy efficient roof products, and Energy efficient windows, doors, ducts and skylights With such features energy consumption is therefore reduced due to the lessening of air leakages under doors, through roofs, windows and so forth. These physical attributes of a hom e in return translate into a Home Energy Rating System (HERS) score of 86 ( or better ) out of a 100 under the old rating system. In fact following the ES rating system, t he old HERS Score is a system in which a rated home is compared to a nother home of the same size and shape, built to the specifications of the HERS Reference Home ( with a HERS Score of 80). However, based on the new HERS rating set up by the Residential Energy Services Network (RESNET) every additional point in the old HERS score increase s a houses energy efficiency by 5% compared to
21 the HERS Reference Ho use In other words, the lower the HERS Index, the more efficient the home is. As result, u nder the new rating system which started in July 1st 2006, the minimum requirement for a home to b e rated Energy Star is an in dex of 85 in climate zone 1 through 5 (i.e. cooler climates) and an index of 80 for climate zones 6 though 8 (i.e. hotter climate) The following sub-section outlines the determinants of energy consumption in the current literat ure 2. 2 Determinants of E nergy C onsumption This section discusses several studies that investigated the relationship between efficient energy initiatives and homeownership cost savings It also outlines the impact s of such initiatives on property value and corporate behavior. In addition, impacts of land use regulations on energy consumption are described. Finally the predictors of energy usage at the household level are discussed Colton (1995) found that energy efficient investment in a home has the eq uivalent effect of reducing initial price of the home from 1.5% to 8% depending on the location. These improvements have the potential to reduce operating costs and improve overall affordability for low income and first time home buyers. Nevin and Watson ( 1998) investigated the effect of energy efficiency on property value and cost savings Despite limited data and the difficulty of identifying consistent energy saving variables, the authors found a positive impact of energy efficiency on housing prices. Sm ith and Jones (2003) analyzed the impact of energy efficient house construction on homeownership costs and property value Their results suggest that, due to an energy saving of $180 per year for the average Energy Star home, operating costs are reduced an d homeownership is thus made more affordable. In fact, residents can afford an additional $2,255 worth of mortgage. In addition, housing value increases by $4500 per unit due to green upgrades.
22 Since DeCanio and Watkins (1998) investigated the impact of En ergy Star Green Lights program on investment in energy efficient equipment. The authors found that voluntary programs such as the Green Lights program can potentially create energy saving investment improve corporate performance and reduce pollution. Yet, organizational and institutional factors can hinder investment. Also using data on the Energy Star Green Lights and other voluntary labeling programs, Howarth et al. (2000) developed a similar model According to their study, energy efficiency programs ar e effective in generating energy savings In fact, they reduce market failures caused by imperfect information and bounded rationality. Pig g (2002) examined the differences in energy use between participants in the Wisconsin Energy Star Homes program in 1999 and 2000. The study reveals that on average, participants of the program use 9% less natural gas in comparison to conventional new homes. This is due to reduced air leakage in Energy Star homes. Wisconsin Energy Star Homes programs participants also co nsume between 3% and 11% less electricity than non -participants .However the observed difference is not statistically significant The rating system used in study captures well actual heating consumption but slightly over -predicts heating use on average. G illingham, et al. (2004) reviewed literature on a broad range of existing non transportation energy efficiency policies including appliance standards, financial incentives, information and voluntary programs, and government energy use. Their results sugges t that these programs are likely to have collectively saved up to 4 quads of energy annually, with appliance standards and utility demand-side management likely making up at least half these savings (Gillingham et al. 2004).
23 Glaeser and Kahn (2008) studi ed the relationship between greenhouse gas emissions associated with household energy consumption (from private and public transportation, home heating and household electricity usage) and new urban development in 66 major metropolitan areas within the Uni ted States. The results of the study show that in general households located in the central cities have significantly lower emissions than the ones living in the suburban area, with lowest emissions in California and highest emissions in Texas and Oklahoma In addition, as land use regulations are stricter (which is the case in central cities) carbon dioxide emissions decrease. As a result, while current land use regulations restrict new urban sprawl in the cleanest areas (central cities) it has the unin tended effect of pushing new construction toward higher emissions areas (suburban areas) and thus increasing greenhouse emissions as a whole. Guerin, et al. (2000) reviewed numerous energy studies conducted since 1975 in order to identify occupant predict ors of household energy-consumption behavior and energyconsumption change. The study revealed that household consumption is affected by three main factors: occupant characteristics, occupant attitudes and occupant actions. The most recurrent factors inclu de age, income, education homeownership, desire for comfort, weather and incentives. Yohanis et al. (2007) used a sample of 27 houses in United Kingdom and investigated how occupancy and characteristics of the dwelling affect domestic electricity use. Th is study suggests that type of dwelling, its location, ownership and size, as well as household appliances, attributes of the occupants (number of residents, income, and age) and occupant pattern have different but significant effect on electricity consump tion. For instance, there is difference of 24% to 30% in consumption level between detached and terraced homes .There is a positive impact of floor area on electricity consumption. In addition, electricity consumption per
24 individual decreases as the number of residents increases. This is especially true for large houses with fewer occupants. The review of the literature on energy consumption identifies the main elements that affect level of households or firms energy consumption Of these elements are incl uded energy efficiency initiatives structural features of a house, characteristics its occupants, their behavior and actions and finally land use regulations. Researches show that energy efficiency programs such as the Energy Star program or the Energy St ar Green Lights program reduce households or firms energy usage. Consequently such programs contribute in the decrease of homeownership costs while raising property value, affordability and corporate productivity. However, the majority of those studies are comparative studies that use little to no statistical tools in order to control for confounding factors influencing the results. As a dwelling get bigger in size ( square footage for instance), energy consumption increases. Socio -economic characteristics s uch as income, education and age of its occupants also have a significant effect on the amount of energy consumed by the household. Finally as land regulations get stricter in the central cities, restricting energy consumption, consumption and therefore pollution levels are higher in newer development located in the ou ter cities where regulations are looser. The review of the current literature shows that energy conversation and environmental protection represent a major area of interest for economists. Ind eed, the environmentally conscious efforts of government are manifested by the implementation energy efficiency policies However, despites the growing popularity of such green initiatives, academic research in this field is limited. For instance, there are very few studies focused on evaluating the performance of energy effici ency policies over an extended period of time
25 2. 3 Conclusion This chapter defined the term energy efficiency program by summarizing the objectives of such an initiative and provi ding an illustration of it implemented here in the Unites State s: the Energy Star program The effects of energy efficient behavior on private cost savings, energy consumption and corporate investments was discussed, as well as impacts of land use regulati ons and urban development pattern Finally d eterminants of household energy usage were discussed While individual energy efficient behavior and governmental efforts are critical in protecting the environment in the light of climate change, there is a very limited academic literature on the implementation and performance of such initiatives overtime. This constitutes a limitation of the current literature. The available literature on energy efficiency programs is certainly biased towards short term energy use reductions and cost saving In fact, it focuses more on the implementation of energy efficiency measures rather than evaluating the per formance of these programs overtime. This study therefore stands to question whether or not energy efficiency program s work and if so, is this efficiency maintained overtime For this reason, it is necessary to examine the performance of these programs as time goes by. Doing so will better guide polic i es aimed at protecting the environment through the reduc ing energy con sumption and lessen ing of greenhouse emissions. This examination will achieve that goal by pointing out up to date challenges and changes faced by such efforts It will also suggest adequate policy response to such changes
26 CHAPTER 3 METHODOLOGY In this chapter the conceptual framework and econometric methods used for the study are discussed. To do so definitions and applications from the current literature are provided. The objective of this chapter is also to define economic predictive model s that all ow for the determination of the impact of energy efficiency programs on household energy consumption overtime Section 3.1 discusses the cost minimization behavior as the conceptual framework for the research Section 3. 2 describes the DID estimation its application in the literature and its mechanisms. Section 3. 3 introduces the different types of time series cross section al model s used in the study Section 3. 4 describes the location of the study and the different types of data used to perform the analy sis. Finally, the conclusions from the chapter are drawn in section 3. 5 3.1 Cost M inimization Behavior Following previous studies in the literature, the underlying economic framework of this resea rch is the cost minimization approach. The basic idea of t he cost minimization problem (CMP) is to explain how firms f ind input combinations that minimize production cost given the quantity of outputs (Varian, 1992). The CMP can be stated as follows: Min w *x with x x ) q, (3 1) w here x i s the nonnegati ve vector of inputs w is the vector of input prices, f(x) is the production function and q is the amount of outputs. The optimized value of the CMP is then given by the cost function c (w,q) which is referred to as the cost curve on Figure 3 1 Conditiona l on the fact that the output level q is produced the optimal set of inputs choices denoted x*(w,q) is obtained at the point of tangency where the constant slope of the cost curve is equal to the slope of the isoquant : the cost minimization point on Figur e 3 1 At this point the
27 optimal bundle of inputs (x1 *, x2 *) minimizes the cost of producing q outputs. See Figure 3 1 The CMP is analyzed by using the Lagrang e multiplier method. -q) (3 2) 0 ) (* i ix x f w (3 3) a nd f (x*) = q (3 4) In this specific study the economic agent considered is a homeowner. The economic goal of that individual is to make a rational decision that will minimize his homeownership or operating costs which include energy cost denoted Ce. This framework a lso assumes that there are only two possible types of house s available to the homeowner in the market: a house that is ES rated or a traditional non ES house. Based on the current literature (Y ohanis et al. 2007; Smith and Jones, 2003; Pigg, 2002) one knows that t he homeowners energy costs Ce are a function of the type of the house he decides to choose (ES vs. non.ES), additional structural characteristic of the house such size, location, etc denoted H and weather conditions denoted W. The CMP can be therefore stated as follows: Min Ce (ES H, W ) s.t. B (3 5) where Ce (ES H,W ) is the homeowners energy cost function and the constraint B is the homeowners available income or his energy cost budget for his household. Assuming that the homeowner makes r ational decisions and has full information in the market, the expectation is for him to choose a house that will minimize his operating expenditures in terms of reduced energy consumption. This will be materialized in a reduced energy bill. The homeowner i s thus expected to choose a house that is ES. Since pricing methods of utility consumption can be problematic and are often subject to change s this study adopts a cleaner version of consumption such as direct energy consumption. By doing this, th e study a void the problem of measuring the effects of ES program on these pricing schemes. Having this economic framework in mind, the rest of this chapter discusses the
28 econometric methods adopted in this research. The goal is to model how the ES rating of a house impact s household energy consumption and therefore dictates homeowners behavior 3. 2 The Difference -in -Difference Estimator 3. 2 .1 Definition and Applications The main objective of this study is to identify the effect of the Energy Star rating on energy consumption in Gainesville, Florida This section describes the framework used to evaluate the Energy Star programs performance over time: the Difference -in Difference (DID) Estimator The DID estimation is often used in empirical economic research in ord er to assess the effects of public interventions and other treatments of interests in the absence of purely experimental data. The common objective of evaluation studies is to estimate the average impact of a treatment on some outcome variable of interest (Abadie, 2005). In this study, the treatment is the Energy Star rating of homes and the variable s of interest are household energy consumption (EC) and property sale price There are many applications of a Difference -in Difference estimation in the litera ture, especially in the area of labor economics. For example, Car and Kruger (1994) evaluated the effect of an increase in the minimum wage (the treatment) on employment (the outcome) in the Fast Food Industry in New Jersey versus Pennsylvania. The D ID est imator shows a small significant increase of 0.59% in employment in New Jersey where the minimum wage increased. Another illustration of the D ID estimation is Meyer et al. (1995) paper which examined the effect of an increase in workers compensation on t ime out of work in Kentucky versus Michigan. Their before and after analysis using Difference in Difference estimation shows that time out of work increased by approximately 50 percent for those eligible for higher benefits and remained the same for those whose benefits were constant.
29 More recent studies by Angrist a nd Krueger (1999), Abadie (2005) and Athey and Imbens (2006) utilizing DID estimations are at the fringe of econometrics. In fact, they relax the general assumptions that the conventional estim at ion hinges on. The a uthors then investigate the performance of the estimation under these new assumptions. For instance, Abadie s study in 2005 examines the event when the outcomes of the treated versus the control group do not follow a parallel trend in the absence of a treatment This is explained by the fact that the selection for treatment is influenced by individual transitory shocks on past outcomes ( Abadie 2005). In a D ID estimation, in addition to a treatment and an outcome defined, two distinc t groups as well as two time periods are also specify The two groups are defined by their treatment status: the treatment group referred to as t reated group and the control group not subject to the treatment studied referred to as the non t reated group. T he first time period corresponds to the initial time period (let say T = T0) in the study and the second period corresponds to another chosen time period (T = T1). The following sub -section discusses the mechanics of the DID as well as how the outcome of i nterest is modeled. 3. 2 .2 Mechanics of the DID Estimator The outcome interest Yi is modeled by the flowing equation: Yi = 0 + 1 Treatmenti + 2 Ti+ (Treatmenti*Ti) + i (3 6 ) 012 i is a random unobserved error term which includes all determinants of Yi that the model omits. The dependen t variable, Yi represents the variable of interest. The explanatory variables of the model are defined as follow: Treatmenti is a dummy variable which will take the value of one if there the observation is treated and zero if the observation is not treated. Ti is the time variable
30 identifying each time period defined in the analysis (T = 0 or T =1 ). ( Treatmenti*Ti) is the interaction term between the two previous explanatory variables, Treatmenti and Ti. The purpose of this evaluation is to find a good est av ailable. The model is run using OLS estimation under the assumptions that the generalized standards hold. The parameter of interest in this analysis is associated with the interaction term (Treatmenti*Ti)gives the true effect to the treatment It corresponds to the difference over time of the average difference in the dependent variable between the two groups. = [ Yi (t reat ment group, T1) Yi( t reat ment group, T1] [Yi( control group,T0) Yi( control group,T0)] (3 7) The other coefficients have the following interpretation: 0 = constant term baseline 1 = treatment group specific effect accounting for average permanent differences between the treatment and the control group 2 = time trend common to both treatment and control groups: captures the changes in the dependant variable due to time i = error term containing any significant explanatory var iables not included in the model 3. 2 3 Assumptions and Limitations The main criteria for a good estimator in that this estimator is unbiased. In other words, the estimate of will be correct on average Mathematically, the expected value of the estima tor E [ ] is equal to or E [ ] = (3 8) Consequently, the results of the DID estimat ion hinges on the following assumptions: The model is correctly specified The error term is on average equal to zero: E[ i] = 0 (3 9)
31 The error term is uncorrelated with the other variables in the equation: i Treatment ) = 0 (3 10) i Ti) = 0 (3 11) and i Treatment Ti) = 0 (3 12) The last assumption or parallel trend assumption is the most critical (Abadie, 2 005).Therefore, if any of the assumptions mentioned above do not hold, is b iased. The main issue when using a DID estimation is the failure of the parallel trend assumption. For instance, if cov (i, Treatmenti *Ti) = E [ i (Treatmenti Ti)], the dependent variable for the t reatment group a nd the control group follows a different trend Ti. If the control group have a time trend of Ti control and the treatment group has a time trend of Ti treatment one will have in this case : E [ ] = (Ttreatment+ ) Tcontrol = + (3 13) This problem is common in several program evaluation studies causing DID estimators to be biased (Wooldridge, 2007) However, econometricians such as Meyer (1995) and Wooldridge (1995) proposed simple ways to avoid these issues. O ne approach is to collect more data on other time periods before and after the treatment occurred This helps capturing any pre existing trend differences between the two groups Another solution to having a biased difference in -difference estimator is to find other control groups which can indicate additional underlying trends that the researcher might not be aware of. In this study, th e second approach is adopted as information on four different controls groups is gathered in order t o avoid this limitatio n of the DID estimation Based on the literature, a dditional ways to make sure that the DID estimator obtained in not biased is to test for heteroskedasticity using a White Test or a Breush Pagan test In addition, testing for
32 autocorrelation using a Durbi n -Watson test will also help generate an accurate estimator. The following subsection discusses how to test for auto correlation. Testing for Autocorrelation : B y definition, autocorrelation is a violation of the following statement: conditional on the expla natory variables, the error term i n the two different time periods T=0 and T= 1 are uncorrelated: Corr ( UT =0 UT =1/ x ) = 0 (3 1 4 ) If the errors UT =0 UT =1 are correlated across time, there is serial correlation also known as autocorrelation, which is alwa ys a potential issue for regressions using time series data (Wooldridge, 2003). In other words, errors associated with adjacent observations are correlated, and errors for observations which are far apart are not. The most common form of serial correlati on is called the first order serial correlation in what case errors in one time period are directly correlated with errors in the next time period. As it is the case with heteros k edasticity, if serial correlation is present, the least squares estimators w ill still be unbiased, however they are no longer B.L.U.E. In addition, in the case of positive serial correlation estimates of the standard errors will be lower than they should be There is therefore a downward bias .It will cause the confidence interva ls to be small er than they really are and one will sometimes reject the null hypothesis when it should have been accepted Finally, serial correlation will cause the value of R2 to be higher than it should be, and estimates of the error variance will be s maller than they should be. The Durbin Watson test is a widely used method of testing for autocorrelation The test statistic denoted d has a value that lies between 0 and 4. A value of 2 indicates there appears to be no autocorrelation. If the DurbinWats on statistic is substantially less than 2, there is evidence of positive serial correlation In fact, small values of d indicate successive error terms are, on average, close in value to one another, or positively correlated. Large values of d indicate
33 suc cessive error terms are, on average, much different in value to one another, or negatively correlated 3.3 Time S eries and C ross-S ectional M odel s 3. 3 .1 Definition s and Applications In conducting economic research, panel, cross -sectional and time series dat a represent the most widely type of information used for policy analysis or program evaluation at the micro or macroeconomic level. Indeed, with the increa sed a v a ila bility of such inform a tion over time researchers are able to conduct more complete and accurate economic analysis. This type of data not only allows the study of differences between subjects but also the study of these differences over time. Panel data also known as longitudinal data or time series cross -sectional data (TSCS) refers to a da ta set where a large number of cases, units, people firms, cou nties, etc ( denoted i with i = 1,, N ) are observed at two or more time periods ( denoted t with t = 1,, T ). Examples of the use of panel data include the National Election study in Politic al Science where over 2000 individuals are observed over a three years time period. In fact, political scientists have been using panel data for over fifty years (Adolph, et al.2005, Beck and Katz (1995), Garrett (1998)). Another example of panel data is t he 1997 National Longitudinal Survey of Youth, where a representative sample of approximately 9,000 young individuals between the ages of 12 and 16 years old were repeatedly surveyed over several years on employment behavior and educational experience. In this study the data available is a pooled time series and cross -sectional data An independently p ooled time series and cross -sectional data (PTSCS) differ s from a typical panel data in that fact that it is either dominated by time or just has fewer unit s relative to the number of time periods. Examples of PTSCS data include studies of countries, states, individuals
34 observed over periods of time that are longer compared to the number of units in the sample. According to Wooldrigde (2006) a PTSCS data is obtained by pooling a random sampl e of units, individuals, firms, countries, etc i = 1, 2 N from a large population at different points in time t = 1, 2,, T usually different years but not necessarily In other words, a PTSCS data can be thought as group s of independent cross sections of all units i at time t piled on top of each other T he PTSCS also differs from a typical panel data since it does not follow the same units or individuals across time. I n addition, in a PTSCS the distances between eac h time period t does not have to be identical within each cross section and finally there can be variables that are constant for each unit i across time. This type of data applies to this study since the data used is composed of monthly house hold energy c onsumptions observed across a 13 -year s time frame. In addition, at the cross sectional level, the sample of houses selected and household energy consumptions vary across time. At the time series level, household samples and their energy consumption levels change as well However the house structural characteristics such as number of bedrooms and bathrooms affecting energy consumption are time invariant The panel is unbalanced. According to the literature, t here are many advantages from using PTSCS data wh ich include increasing sample sizes, analyzing the effect of time or determin ing whether relationships have changed over time. In fact by pooling random samples drawn from the same population, but at different points of time, we can get more precise esti mators and test statistics with more power (Wooldridge, 2003). Using PTSCS data therefore allows for an explicit and dynamic comparison of groups by examining changes and differences over time between the groups. However, complications in using PTSCS data include accounting for heterogeneity across sections as well as within section serial correlation, which compromise estimation results.
35 As a result, particular models have been developed in order to effectively analyze such data. The following subsection discusses the mechanics o f a general pooled cross section model, its assumptions and limitations. 3.3.2 The Pooled O rdinary Least S quares M odel A basic and general PTSCS model can be defined as followed: Yit Xit+ it with i = 1, 2,, N and t = 1, 2, T (3 1 5 ) Where Yit is dependent variable, Xit it is the error term it./ Xit) = 0 (3 1 5 ) it./ Xit) = 2 (3 1 7 ) This model assumes that the data has rectangular structure where N units are observed for the same time periods T. This model relies on the following assumptions : a ll the usual Ordinary Least Squares (OLS ) assumptions hold t he constant is the same across all units i and the effec t of any given explanatory variable X on the dependent variable Y is constant across observations assuming that there are no interactions in within each X s (Cameron and Trivedi, 2005). The constant intercept and constant effects of X on Y assumption are essential to specify the model. Violation of these two assumptions will result in biased e s timates due to variation of the intercept and change of the slope across units, over time or both. This general model can therefore take various forms. For instance, in case there are different intercept s at the unit level the initial model becomes: 1 N it iitit iYX (3 18) Assuming that the slope is the same for each unit but the intercepts are different over time, the model can be written: 1 T it titit tYX (3 1 9 )
36 Intercepts can also very over time and across unit. The main idea is that in case the data is created by either model, a homogeneous intercept is estimated instead and there is a risk of biased results. The flip side of this situation is to h ave a constant intercept but the effects of X on Y change across time, across units or both. On one hand, in case there is a variation in slope across units: 1 N it iititYX (3 20) On the other hand, i f there is a variation in slope over time, t he model is written: 1 T it titittYX (3 2 1 ) Nevertheless it is important to not e that all of these previous models assume that the error it is homoskedastic and serially uncorrelated across time. These events are very un likely to happen with TSCS models since there is heterogeneity across units and over time in reality In fact, Yit it+ it with stochastic errors it wher e E (it. / Xit) = 0 (3 2 2 ) it i,t 1+Ut (Ut meets all classical assumptions ). (3 2 3 ) In this particular study, the most restrictive model estimated b ased on the data available is the following pooled OLS model: Yit i+ it. ; i= 1, 2,, N; t= 1, 2,,T (3 2 4 ) Yit represents the amount of energy consumption at the household level that fluctuates with time is constant for all houses and across time. Xi represents the vector of house characteristics that affect energy consumption These exp lanatory variables are time invariant and therefore differ from the initial general model (321) defined earlier where Xit varies with time. Examples of such variables include number of bedrooms, bathrooms, age of the house, etc which are fixed across time it represents the error term including all relevant explanatory variables not accounted for in the model. i is individual houses in the sample and t represents the years of consumption.
37 In order to obtain consistent parameters estimates, OLS assumes that the idiosyncratic error tem it is uncorrelated with Xi. Estimating this model with OLS ignores the heterogeneity in homes and thus generates estimates that are unbiased and consistent in terms of the slopes but inefficient since the standard errors are si gnificantly underestimated. The correlation with units errors therefore needs to be accounted for. Based on the current literature, i n order to obtain efficient estimates Generalized Least Square (GLS) estimation can be performed Other approaches includ e estimating fixed and random effect, dynamic panel models or panel models for non normal dependent variables. This study adopts as remedy a fixed effect method that not only account s for a time invariant individual specific effect and but also a time effect. The following subsection describes the fixed effect s (FE) model in more details. 3.3.3 Fixed Effects Model Following the basic framework in the literature, the standard FE model is: Yit i i t+ it ; i= 1,2,,N; t= 1,2,,T. (3 2 5 ) The model assu mes that: E ( it/Xit, i) = 0 (3 2 6 ) Var it/Xiti) = Var it2 for all t=1,.., T (3 2 7 ) and Cov iti,t 1 /Xiti) = 0. (3 2 8 ) The main advantage of a FE analysis is that it controls for any unmeasured time invariant differences between units : i i captures time -constant but unit -varying effect that is not accounted for in the previous restrictive pooled OLS model (3 23) and is therefore part of the error term. i is also called unobserved or fixed effect, which help to remember that i is fixed i is also known as the unobserved or individual effect. In addition pooled OLS which does not account for this i is biased i is correlated with the eitit and
38 Xit are uncorrelated. A m ajor limitation of the FE estimation resides in the fact that only time varying effects will be identified. In order to overcome this limitation, for the purpose of this study time -varying variable s are incorporated into the analysis in order to capture a time trend. The following appropriate FE model is th us defined: Yit i + i+ i t+ t+ it, (3 2 9 ) The new explanatory variables in model are Cit and Dt Indeed Cit rep resents the interaction term between the Energy Star status of a house and the bill years. Dt represents the vector of dummy variables representing years of consumption. Since Cit captures the unit constant but time variant effect, indicates variation in Yit that is due to the effect of time. As it it it i,t1+Ut the model is tested for autocorrelation. With this model, the unobserved fixed effect can be differenced out: Yi,t 1 = i + i+ i t 1+ t 1+ i t 1 (3 30) Yi = Yit Yi,t 1 = (Cit Ci t 1) + (Dt t 1 ) + ( it, i t 1) (3 3 1 ) Following this model, while controlling for the specific features of each house, the change s energy consumption over the years are e xplained by change in the ES feature overtime, changes occurred during the specific year and consumption pattern in the previous year. Based on the data available, for the purpose of this study this FE is considered the most appropriate estimation method. The following section describes the location of the study as well as the subdivisions of houses included in the sample. 3. 4 Selection of Sample 3. 4 .1 Location of the S tudy Located 50 miles from the Gulf of Mexico, 85 miles south of the Georgia state line and 67 miles from the Atlantic Ocean Alachua County is located in North Central Florida (See Figure
39 3 2 ). It extends over 977 square miles and includes the municipalities of Archer, Alachua, Gainesville, Hawthorne, High Springs, LaCrosse, Micanopy, Newbe rry, and Waldo. The population of Alachua County is approximately 247,000. According to the Alachua County Property Appraisers 2008 Annual Report, t he population of the city of Gainesville is approximately 120,000. Demographically, Gainesville ranks 15th among Florida's most populous cities; more than 27 percent of its population consists of individuals between the ages of 25 and 44 years. Over 55,000 of that population are students attending the University of Florida the fourth largest public land grant university in the country. The city of Gainesville, where the subdivisions are located, is characterized by a pleasant sub -tropical climate year round with mild winter averages in the upper 50's to mid 60's while warm and humid summer temperatures in the upper 80's and lower 90's. Average annual rainfall is around 35 40 inches With Paynes Prairie State Park Preserve 7,000 acre San Felasco Hammock Preserve State Park, several rolling hills, short climbs, sinkholes and upland forests Gainesville offers numerous environmental amenities for outdoor activities including hiking biking and bird watching. The Gainesville area is also well known for its numerous world class springs an d rivers less than an hour drive which suitable for scuba diving, kayaking and snorkeling. 3. 4 2 Characteristics of the S ubdivisions All homes selected for the study are single -family homes. The selection was done at the subdivision level since there is a large concentration of Energy Star houses in the Mentone Subdivision. This subdivision therefore represents the treatment group in the DID estimation performed in the following section. All ES homes in Mentone have been built by Atlantic Design Homes. A s an Energy Star homes certified builder since 1997, Atlantic Design Homes has designed and built over 600 of
40 Gainesvilles most elegant homes in subdivisions like Haile Plantation, Richmond, Mentone, and Town of Tioga (Atlantic Design Homes 2007). Other subdivisions include Burberry and Ridgemont in Gainesville Florida In addition to more than 20 years in the construction industry, Atlantic Design Homes was also been recognized by the EPA as National Builder of the Year, in 2000. The company has received several awards over the years including the Grand Aurora Award for Best Community in the Southeastern United States, a local Ethics in Business Award and Alachua Countys only designated Clean Builder of the Year. Atlantic Design Homes has not only built more Energy Star homes th an any other builder in the Gainesville area; it has also recently become certified by the Florida Green Building Coalition thanks to its environmentally conscious practices ( Atlantic Design Ho mes 2007). Four other single builder subdivisions with similar size, single story non -ES homes were also selected. These homes have multiple builders and are considered conventional homes representing the control groups in the study They are located in t he following subdivisions: Broadmoor, Capri, Eagle Point and Stillwind. The Gainesville area is representative of the population of homes since similar houses with similar structural and environmental friendly features are also being built in several loc ations in the country. There are no specific conditions that apply to this region, making it a special case. Indeed, similarly ho t climatic conditions can be found in different parts of the country including Arizona California or Texas, as well as a natio nwide demand for green construction. F igure 3 3 provides a visual illustration of where the subdivisions are located within Alachua County. In addition based on location, each subdivision is zoned to a specific school by the Alachua County Public Schools Zoning Department For the purpose of the study, only middle schools are considered, three different middle schools in total. In fact, the subdivisions of
41 Mentone and Stillwind are zoned to Kanapaha middle school ; the subdivisions of Eaglepoint and Broadmo ore are zoned to FortClark middle school and the Capri subdivision is zoned to Westwood middle school The following subsection describes the energy data used for the study. 3. 4 .3 Energy c onsumption d ata The data on energy consumption which includes elect ricity and natural gas was collected by the local utility company, Gainesville Regional Utilities (GRU) for all of the homes except for the electric consumption of the Mentone subdivision. In fact, Clay Electric provided the monthly electricity usage for h ome s located in Mentone. This data reflects monthly consumption measured in kilowatts for electricity and thermos for natural gas measured at the meter. Water consumption levels were also observed but not used in the study. 3. 4. 3 Other D ata E ach homes data set also includes the conditioned/heated area number of bedrooms number of bathrooms year built and Energy Star status of the home provided by the Alachua County Property Appraiser office. Houses are identified by their tax parcel identification num ber, street address as well as subdivision. All homes are singe -builder, single story and single family homes. All houses are equipped with central air conditioning and heating system. Most houses do not include individual swimming pools and other features of the homes including ventilation systems, sealing penetration are not observable. With similar construction style and comparable size, the comparison of homes is done at the subdivisions level : Energy Star homes versus traditional homes (Smith and Jones 2003). Average monthly temperatures used to control for the seasonality of energy consumption was gathered using the Florida Automated Weather Network database
42 The ideal data for this study would also include socio -demographic data on households such as income level, education level, age, number of residents, household structure and so forth (Guerin, et al., 2000 and Yigzaw, et al., 2007) as well as occupants action and behavior such as thermostat setting (Pig g 2002) Data on occupant behavior and a ttributes would therefore add another dimension to the study by painting a more accurate picture how not structural features of a home ( i.e Energy Star upgrades) but also characteristics and attitude of its residents affect energy consumption of that sp ecific home. Additional information can also include environmental and locational characteristics of the properties such as tree coverage in the area, distances to main roads, parks, lakes, shopping plaza and downtown center. 3. 4. 4 Data Cleaning The county appraiser data, containing structural characteristics of the houses and GRU data containing the energy consumption information were merged in order to create new energy data sets. Each new dat a set was examined and correct for anomalies in consumption and structural features In fact only positive energy consumptions were retained as 3005 observations were dropped (2943 observation for no consumptions and 62 observations for negative billed consumptions) 2460 additional observations were dropped due to m issing billed consumption and 46 0 observations for missing bill year. 764 observations made up of houses with no bathrooms and bathrooms were also deleted from the energy data set. For the panel data energy analysis only electricity consumption was retained resulting in 205, 890 observations dropped. 3.5 Conclusions In this chapter the methodology used in the study was discussed. First, the economic framework the cost minimization problem was presented. Then the Difference in Difference estimation pooled OLS and FE methods were discussed. Finally, a description of the location of the study was provided. The models discussed will be estimated using energy consumption data,
43 structural characteristics of the homes in the sample and seasonality After d escribi ng the different types of data used, cleaning up process of the data is discussed. Figure 3 1 The C ost M inimizing P roblem (CMP). Figure 3 2 Map of Alachua County, Florida created using Alachua County GIS parcels data.
44 Figure 3 3. Map of all five su bdivisions in Alachua County, Florida using Alachua County GIS parcels data
45 CHAPTER 4 ENERGY ANALYSIS USING DIFFERENCE IN DIFFERENCE ESTIMATOR The objective of this chapter is to develop a model that can be used to identify the impact the Energy Star f eature of a home on household energy consumption which includes electricity, natural gas and total consumption. To do so, five different models are developed and results are summarized. The D ID framework adopted allow s for an evaluation of the Energy Star program overtime and the formulation of policy implications A limitation however in the implications of this analysis resides in the fact that only two time periods are considered. Another limitation is the unobservable pre treatment or post treatments ch aracteristics of the sample between the treated and the untreated that could affect energy consumption. In most program evaluations, the research can observe how the subject behaves at the end of the treatment. It is not feasible in this study. In fact, on ce a house has been built ES for instance ther e is no way to ob serve it under any other conditions. To solve this issue the study just considers the ES program as a natural experiment and attempt to control for any confounding factors outside the ES ratin g that affect the outcome, household energy consumption. Section 4.1 describes the sample. Section 4.2 describes the predictive models and summarized the results for the electricity, natural gas and total consumption analysis Finally conclusions are drawn in section 4. 3 4 .1 Characteristic of S ample This section describes the sample of homes that was used in the initial estimation of the D ifference in Difference estimator using electricity natural gas and total energy consumption which includes both ele ctricity and natural gas usage, for the years 2000 and 2006. A total of 568 4 homes were selected from both types of subdivision (conventional versus Energy Star rated). In fact, the sample includes 1 104 observations for homes rated E nergy S tar located in the Mentone subdivision and 4 560 observations NonES homes located in the four conventional
46 following subdivisions: Broadmoor, Capri, Eagle Point and Stillwind. This sample is referred to as sample 1 and is described in T able 4 1 As describe in T able 4.1 t he houses in Mentone have the smallest area s with an average of 1670.27 square feet. The biggest heated/cooled areas are found in the houses located in Broadmoor with an average of 2778.76 square feet. Houses in Capri, Eagle Point and Stilllwind have he ated /cooled areas that vary between 1778.67 to 1858.82 square feet. Table 4 2 provides a definition of the variables used in the analysis of energy consumption patterns observed for each type of house. Electricity, natural gas and total (combined electrici ty and natural gas consumption) consumptions for the bill years 2000 and 2006 (t ) are used for the analysis. The ES status of the house is also observed as well as the months of consumption. Table 4 3 provides the descriptive statistics of the variables. A s described in T able 4 3 on average, houses in the sample consume about 939.5 kWh of electricity, 25 units of natural gas and 1678.6 units of total consumption. In the following section, an energy analysis is performed using electricity, natural gas and t otal consumption data for 2000 and 2006. 4.2 Energy C onsumption A nalysis Results from the 2005 Residential Energy and Consumption Survey show that on average, when it comes to the energy expenditures, households in Florida spend a large percentage of their budget (35.2%) on air -conditioning. On the other hand only 5.1% of their budget is attributed to space heating. This is can be explained by the very hot temperatures during the summer time and mild winter temperatures which characterize this part of the c ountry. 15.5% is attributed to water heating, 7.5% to refrigerators 36.7% to other appliances and lighting. See Figure 4 1. 4.2.1 Electricity C onsumption In this section, five different models are run in order to evaluate the impact of the Energy Star rati ng energy consumption using 2000 and 2006 monthly electricity consumption.
47 A simple model is used in a nave analysis that estimates the effect of the Energy Star rating on emery consumption: Electric 2006 0 1ES (4 1) The results are summarized in T able 4 4. The constant corresponds to an average annual electricity consumption of 1027.139 kWh in 2006. The coefficient of Energy Star, which corresponds to the difference in average electricity use betwee n conventional homes and ES homes, shows that Energy Star homes consume on average 132.139 kWh less than the non Energy Star homes. With significant estimates (t statistics greater than five in absolute value), one can strongly reject the hypothesis that t he average electricity consumption for homes rated E nergy S tar and those without the rating is the same. Since this last estimation does not imply that the Energy Star rating is causing lower electricity consumption in 2006 the same regression is run for the year 2000 One can expect a lower consumption since the houses are newer. The following regression in then executed: Electric 2000 0 1ES (4 2 ) The results are summarized in T able 4 5 Base on this nave analysis, the average annual electricity co nsumption was 953.197 kWh in 2000. Homes rated Energy Star consumed on average 1 88.748 kWh less than conventional homes. The difference is also statistically different. Nevertheless, it will be incorrect or misleading to conclude that homes that are rated Energy Star have been saving less and less electricity as time goes by or as houses are aging or they have actually increased electricity consumption over the years. The answer lies in the examination of how the coefficient on the Energy Star variable chan ged between 2000 and 2006. In fact the difference in energy savings was much larger in 2000 than in 2006 (i.e. 1 88.748 versus 132.139 kWh ). The difference in the two coefficients is 56.609 kilowatts : 1= 1 88.748 (132.139) = 56.609
48 This is the estimate of the effect of the Energy Star rating on energy consumption between 2000 and 2006. In empirical economics, it is known as the Difference-in Difference estimator (DID). It is equal to the difference overtime in the average difference of electricity consu mption with the two building upgrades. In order to test whether the DID estimator is statistically different from zero; its standard error is obtained by using the following regression analysis defined in Model 2. Model 2 introduces the effect of time and the interaction term between the time effect and the ES effect as discussed earlier in the methodology section The following regression is then executed: E lectrici = 0 + 1 ESi + 2 Ti+ (ESi*Ti) + i (4 3 ) with DID = = [Electric (ES,06) E lectric (ES,00)] [E lectric (Non ES,06) E lectric (Non -ES,00)] (4 4) 0 = constant, baseline : average consumption of a NonES home 1 = treatment group specific effect accounting for permanent differences between ES and Non -ES homes ( or co nsumption due to ES rating) : 1 = E lectric (ES,00) E lectric (Non ES,00) 2 = time trend common to both groups: captures consumption changes due to time in all houses from 2000 to 2006: 2 = E lectric (Non -ES,06) -E lectric (Non -ES,00) i = error term Under the parallel trend assumptions, one can determine the expected value of the average consumptions for each group at both time periods The expect ed values are summarized in T able 4 6 This table is valid for electricity, natural gas and total energy consumption a nalysis. Since energy consumption is seasonal and varies from month to month, it is important to control for
49 monthly fixed effects. Dummy variables (noted Feb through Dec) are therefore included in Model 2 These dummy variables will take the value of one if the reading goes with a certain month. This is model 3. Electrici = 0 + 1 ESi + 2 Ti+ (ESi*Ti(Feb, Mar, Apr, May, Jun, Jul, Aug, Sep, Oct, Nov, Dec) + i (4 5) An additional fourth model (Model 4) not only includes monthly fixed effects but also heated/cooled area of the house measured in square feet. Electrici = 0 + 1 ESi + 2 Ti+ (ESi*Ti (Feb, Mar, Apr, May, Jun, Jul, Aug, Sep, Oct, Nov, Dec) i (4 6 ) There are two good reasons for including control variabl es in our estimation. On one hand, the energy consumption in 2000 may be different from the consumption observed in 2006. If so it is important to control for characteristics that might have been different .On the other hand, even if the average housing ch aracteristic are the same for both years, including control variables can greatly reduce the error term variance, and in return shrink the error of our DID estimator (i.e. ).When we introduce the monthly fixed effects we can see that o ur DID estimator is more significant than before ( p -value of 0.062 versus 0.015). Finally, Model 5 is logarithm model which provide s an approximate percentage effect on energy consumption. In this last model which is of main interest, 100 corresponds to the approximate percentage increase in energy consumption due to an ES rating. Log (Electrici ) = 0 + 1 ESi + 2 Ti+ (ESi*Ti) + Sep, Oct, Nov, Dec) i (4 7 )
50 Based this analysis, there is an increase in electric consumption of about 7.59% for ES homes due to the ES rating. It is furthermo re important to note that this coefficient is statistically significant or different from zero. The results of the analysis are summarized in the T able 4 7 4.2.2 Gas Consumption The same analysis is performed using monthly gas consumption for the years 2 000 and 2006. Once again five separates models are defined in the attempt of quantifying the effect of the Energy Star rating on natural gas usage over time. Model 1: Gas2000 0 1ES (4 8) and Gas2006 0 1ES (4 9) The results are summarized in the T ables 4 8 and 4 9 Table 4 8 shows that on average houses consumed 27.204 units of gas while ES homes consumed about 5.737 units less in 2000. In 2006, houses on average consum ed 24.524 units of gas while ES homes consumed about 3.970 units less in 2006 based on Table 4 9. The results are statistically significant. The difference in the two coefficients is 1.767 units of gas Model 2 is also defined as followed: Model 2: Gasi = 0 + 1 ESi + 2 Ti+ (ESi*Ti) + i (4 10) Similarly to the electricity consumption analysis, Model 2 for the natural gas data is tested for the presence of heteroskedasticiy using the the Breush -Pagan test There is statistical evidence for the presence of heteroskedasticity in Model 2. As a result the rest of the natural gas analysis is conducted using hereteroskedasticity robust standard errors models. The following models are run: Model 3: Gasi = 0 + 1 ESi + 2 Ti+ ay, Jun, Jul, Aug, Sep, Oct, Nov, Dec) + i (4 11)
51 Model 4: Gasi = 0 + 1 ESi + 2 Ti+ (ESi*Ti i (4 12) Model 5: Log (Gasi ) = 0 + 1 ESi + 2 Ti+ (ESi*Ti b, Mar, Apr, May, Jun, Jul, i (4 13) The findings of Model 2 through 5 are summarized in T able 4 10. Based this analysis, there is an increase in electric consumption of about 10.09% for ES homes due to the ES rating (see table of results). It is also important to note that this coefficient is statistically significant or different from zero. 4.2.3 Total Consumption The same analysis is performed using monthly total consumption including both electricity and natural gas for the years 2000 and 2006 in the attempt of quantifying the effect of the Energy Star rating on total energy consumption. The following regressions models are being applied and results are reported in Tables 4 11 and 4 12: Model 1: Total2000 0 1ES (4 14) and Total2000 0 1ES (4 15) Table 4 11 shows that on average houses consumed 1749.808 units of total energy while ES homes consumed about 355.808 units less in 2000. Based on Table 4 12, i n 2006 houses on average consumed 1745.893 un its while ES homes consumed about 249.607 units less in 2006. The results are statistically significant. The difference in the two coefficients is 106.201 units of total energy. The following model is defined as follow: Model 2: Totali = 0 + 1 ESi + 2 Ti+ (ESi*Ti) + i (4 17) Model 2 for the total consumption data is also tested for the presence of heteroskedasticiy using the Breush -Pagan test There is statistical evidence for the presence of heteroskedasticity in
52 Model 2 at the 5 % significance level As a result the rest of the gas analysis is conducted using hereteroskedasticity robust standard errors models The following additional models are run: Model 3: Totali = 0 + 1 ESi + 2 Ti+ Sep, Oct, No v, Dec) + (4 18) Model 4: Totali = 0 + 1 ESi + 2 Ti+ (ESi*Ti i (4 19) Model 5: Log (Totali) = 0 + 1 ESi + 2 Ti+ (ESi*Ti i (4 20) The results are summarized in T able 4 13. Based this total consumption analysis, there is an increase in total energy consumption of about 7.16% for ES homes due to the ES rating. This coefficient is statistically significant or different from zero as well. 4.3 Conclusions The application of the difference in difference estimator using sample1 provides evidence that energy consumption is affected by the presence of the Energy Star upgrades in a house. However, the most significant finding is that energy consumption actually increases over time due to those upgrades in a house that is rated Energy Star. In fact a house rated Energy Star consumes 7.59% more electricity, 10.09% more natural gas and 7. 16% more in total energy consumption due to the Energy Star rating compared to a conventional house. The results are statistically significant. These findings suggest that the energy savings advertized by the Energy Star program are not sustainable over ti me. Major limitations of these findings include the lack of socioeconomic information on the occupants of the homes that can affect levels of energy observed (Guerin, et al., 2000 and Yohanis, et al., 2007). In addition, the lack of information on pre trea tment conditions of ES homes limits the analysis.
53 Nevertheless, the current analysis suggests important implications for both policy makers and homeowners. Indeed, these results are definitely in favor of promoting the construction of energy efficient home s, as they reduce energy consumption and thus greenhouse gas emissions. Yet, although energy efficient upgrades work, they need to be updated and maintained throughout the years, as they deteriorate with time. Policy makers should therefore encourage or ev en mandate the execution of periodic performance tests on ES homes in order to ensure the houses still meet the qualifications of energy efficient houses and are still protecting the environment. These results, on the other hand show that homeowners can be mislead over the years into thinking that their energy bill will remain low due to the fact that their houses are ES qualified. Homeowners will hence benefit from these periodic performance tests as they will help ensure that ES upgrades are still efficie nt thus minimizing still energy and homeownership cots overtime. Given the evidence of short -term performance of the Energy Star upgrades in a house in terms of energy savings, the next step of the study is to expand this energy analysis to a larger sample of homes and a longer time period. Additional features of the homes affecting energy consumption will also be incorporated in the analysis in order to pint a full picture and make more adequate policy recommendations.
54 Figure 4 1 2005 Average Househol d Energy Usage Expenditures in Florida created using 2005 Residential Energy and Consumption Survey
55 Table 4 1. D escription of t he N eighborhoods Subdivisions Mentone Broadmoor Capri Eagle Point Stillwind Number of monthly 1104 1368 1272 1388 1104 Mean heated/cooled 1670.27 2778.76 1858.8 2 1778.6 7 1824.17 Table 4 2. Variable N ames and D efinition Variable s Definition s Dependent Variables Electric Electricity consumption in kWh Gas Natural gas consumption in thermos Total Co mbined electricity and natural gas consumption Independent Variables ES i Energy Star dummy variable ( =1 if Yes,=0 if No) T Bill year dummy variable (=1 if 2006, =0 if 2000) (T*ES ) i Interaction term between ES and T hdtarea Heated/co oled area in square feet month Vector of dummy variables indicated month of c onsumption Table 4 3. Descriptive Statistics of Variables Variables Mean Std.Dev. Min Max Dependent Variables Electric 939.5 470.5 118 2929 Gas 25 24.6 1 378 T otal 1678.6 750.9 287 12669 Independent Variables ES i 0.2 0.4 0 1 T 0.5 0.5 0 1 (T*ES )i 0.1 0.3 0 1 hdtarea 1804.7 299.8 1051 2762 month 6.5 3.5 1 12 Table 4 4. Results of Model 1 Bill Y ear 2006 Variables Estimates P values Const ant 1027.139 0.000 (11.415) ES 132.139 0.000 (25.415)
56 Table 4 5 Results of Model 1 Bill Year 2000 Variables Estimates P values Constant 953.197 0.000 (11. 887) ES 1 88.748 0.000 (26.952) Table 4 6 Average Energy Consumptions ( EC) T Non Es homes ES homes T EC = 0 EC = 0 + 1 T EC = 0 + 2 EC = 0 + 1 + 2 +
57 Table 4 7 Results of Electricity Consumption A nalysis using Sample 1 Model 2 Model 3 Model 4 Model5 Dependent variable Electric i Electric i Electric i Log (Electric i ) Constant ( 0 ) 953.2 *** (12.632) 647.058*** (20.042) 102.990 (40.902) 5.813 (0.038) ES i ( 1 ) 188.748*** (20.772) 188.748*** (16.310) 138.168 (16.337) 0.146 (0.019) T i ( 2 ) 74.077*** (17.358) 74.077*** (23.286) 74.077 (13 .651) 0.138 (0.027) (ES i *T i ) ( ) 56.609* (30.275) 56.609* (23.286) 56.609** (22.884) 0.0759 (0.026) Monthly fixed effects No Yes Yes Yes Heated/cooled area No Yes 0.288 (0.019) 0.00029 (0.00017) N 5664 5664 5664 5664 R 2 0.0189 0.03750 0.4019 0.4819
58 Table 4 8 Results of Model 1 Bill Y ear 2000 Variables Estimates P value Constant 27.204 0.000 (0.557) ES 5.737 0.000 (1.202) Table 4 9 Results of Model 1 Bill Y ear 2006 Variables Estimates P value Constant 24.524 0.000 ( 11.415) ES 3. 970 0.000 (1.071)
59 Table 4 10. Results of Natural Gas Consumption Analysis using Sample1 Model 2 Model 3 Model 4 Model5 Dependent variable Gas i Gas i Gas i Log (Gas i ) Constant ( 0 ) 27.204 *** (0.583) 57.449*** (1.129) 40.439 *** ( 1.681) 3.333 *** ( 0.047) ES i ( 1 ) 5.737*** (0.1.05) 5.737*** (0.612) 4.156 *** ( 0.594) 0.14 7 *** (0.022) T i ( 2 ) 2.680*** (1.388) 2.680*** (0.493) 2.680 *** ( 0.484) 0. 093 *** ( 0.014) (ES i *T i ) ( ) 1.767** (1.388) 1.767** (0.854) 1.767 ** (0.838) 0.1009 (0.031) Monthly fixed effects No Yes Yes Yes Heated/cooled area No Yes 0. 009 ( 0.0007) 0.000 3 *** (0.00021) N 5664 5664 5664 5664 R 2 0.0084 0.5954 0.6089 0.6635
60 Table 4 11. Results of Model1Bill Y ear 2000 Variables Estimates P value Constant 1749.808 0.000 (17.832) ES 355.808 0.000 (40.392) Table 4 12. Results of Model1Bill Y ear 2006 Variables Estimates P value Constant 1745.893 0.000 (15.367) ES 249.607 0.000 (34.808)
61 Tabl e 4 13. Results of Total Consumption Analysis using Sample 1 Model 2 Model 3 Model 4 Model5 Dependent variable Total i Total i Total i Log (Total i ) Constant ( 0 ) 1749.753 *** (18.833) 647.058*** (20.042) 1288.47*** (67.544) 7.126*** (0.035) ES i ( 1 ) 3 55.808*** (43.822) 355.808*** (37.159) 258.932*** (25.323) 0.164*** (0.017) T i ( 2 ) 3.859* (24.731) 3.859* (21.583) 3.859* (20.853) 0.048*** (0.026) (ES i *T i ) ( ) 106.201** (43.822) 106.201*** (37.159) 106.201*** (0.031) 0.0716*** (0.024) Monthly fixed effects No Yes Yes Yes Heated/cooled area No Yes 0.553*** (0.031) 0.00029 (0.00016) N 5664 5664 5664 5664 R 2 0.0230 0.264 0.3123 0.3133
62 CHAPTER 5 ENERGY PANEL DATA ES TIMATION The objective of this chapter is to develop a predictive model of ener gy consumption that can be utilized to identify how environmentally friendly upgrades such as Energy Star rating, structural characteristics of a house and environmental factors such seasonality influence changes in household energy usage. In this section only electricity consumptions are of interest since the area of study (Alachua County, Florida) is primarily a warm climate This energy consumption model hinges on the premise that energy consumption of a household reflects the impact of environmental conditions as well as physical attributes of the house. This study takes a similar approach as previous researches in the literature suggesting that energy efficient investments result not only in energy cost savings but also in environmental protection in te rms of greenhouse gas emission and pollution reduction (DeCanio and Watkins, 1998; Howarth et al. 2000; Pig g 2002; Gillingham, et al. 2004). However, this study goes one step further by evaluating performance over an extended period of time. Section 5 .1 provides a description of the data sets used for the analysis In the following section s the empirical models and results are presented In fact, s ection 5 2 discusses the OLS models and results. Section 5 .3 focuses on fixed effects estimation, statistica l tests and findings. Section 5 .4 presents Difference -in Difference estimation and results. Finally conclusions are made in section 5 .5 5 .1 Description of the D atasets Monthly electricity consumptions measured in kWh (97, 515 observations ) and measured at the meter w ere collected by Gainesville Regional Utilities For the purpose of the study only positive monthly consumptions ranging from 1 k W h to 7,121 k W h have be retained for the analysis Consumption years in the sample go from 1996 to 2008. Data on na tural gas and water
63 consumption were also obtained. However, only electricity data is used for the current analysis. Following the Smith and Jones (2003) analysis, i n order to account for the impact of climate variation on electricity consumption, the obse rved billed consumption is divided into three main categories : summer consumption, winter consumption and other consumption. The categories are created using average temperature s in Alachua County, Florida as reported by the Florida Automated Weather Netwo rk. The summer months extend from May through Au gust. During th es e summer months, average temperature ranged from a minimum of 72 F to a maximum of 80.25F averaging at 79.10F During the winter months (November through February) temperatures range from a l ow of 51F to a high of 59.75F averaging at 58.04F During the remaining months of March, April September and October temperatures range from a low of 60F to a high of 80F averaging at 70.08F The sample includes only single -residential family homes built between 1973 and 2005 with heated areas ranging from 1051 square feet to 4523 square feet. The number of bedrooms ranges from 2 to 5 bedrooms and the number of bathroom ranges from 2 to 4.5 bathrooms. Since the main objective is to measure the impact of E nergy Star rating on household energy consumption, the comparison is made at the subdivision level with Energy Stars homes located in the Mentone subdivision. The rest of the homes located in the other four subdivisions are considered traditional and do not possess an y energy saving upgrades. From this monthly data, a time series data set is created using STATA. In fact, the monthly information is collapsed and aggregated at the yearly level This new pooled TSCS data obtained contains 7793 observations an d 956 groups 13 time dummy variables corresponding to each bill year are also created Yearly consumption is divided in summer consumption (i.e. summer_conit), winter consumption ( i.e. winter_conit), other consumption (i.e. other_conit) and finally total c onsumption (i.e. total_conit). The TSCS data is used later on in this chapter to
64 identify a time trend in the impact of ES upgrades on household energy consumption showing how ES upgrades and savings hold overtime. In addition to ES status of the house and the heated/cooled area, both data datasets also include additional descriptions of the house such as number of bedrooms and bathrooms and the year of construction (i.e. effective year) indicating the age of the house. The following section introduces the analytical framework as well as empirical econometric model s adopted for the analysis and discusses the findings 5 .2 Estimation s and E mpirical R esults 5 .2.1 Monthly OLS M odel s and R esults In order to estimate the impact of energy saving investments such a s Energy Star upgrades on the level of household electricity consumption the following econometric framework is used: Ci = f ( hi, ri ,si (5 1) w here Ci is the consumption level of each individual ith home hi is a vector of structural characteristics associated with the house, ri is the energy efficiency rating or status of the house, si is a vector representing the seasonality parameter vector s to be estimated For the purpose of this study, using the available panel data, the following empirical model is then defined and parameters are estimated using regular OLS estimation1: billed_coni t 1 effyri2 bedsi 3 bathsi 4 htdareai5 htdarea2 i+ 6 ESi+ 7 8 winter + i (5 2) As defined in T able 5 1, the dependent variable billed_ coni t is month ly electricity consumption measured in kWh The independent variables provided by the Alachua County Appraiser Office are as follow: effyri (effective year) represents the year built of the house. It is therefore used to account for the age of the property. As effyri increases, the house is newer and 1 All OLS models are tested and corrected for the presence of heteroskedasticity using robust standard errors.
65 one therefore expects th e energy consumption to decrease due to the newer construction knowing that older constructions retain less energy. The variable ESi represents a dummy variable which will equal to 1 if a house is rated Energy Star and 0 is otherwise. Base on the literatur e, one expects ES rated home to consume less energy due to reduced leakage in ES homes (Pig g 2002) Bedsi and bathsi are the number of bedrooms and bathrooms in the home. h tdareai is the size of total heated area in the home measured in square feet. As th ese variables increase, which indicate a bigger home, on expects the energy consumption to also increase. htdarea2 i is the squared value of the h tdareai variable to keep linearity of the variable. Summer and winter are dummy variable s capturing the season ality of energy consumption. Summer is equal to 1 if consumption was billed during that time frame or equal to 0 otherwise. Winter is equal to 1 if electricity consumption was billed during that t he winter season or equal to 0 otherwise. It is expected to see an increase in consumption during the summer time and a decrease during the winter since air -conditioning system is mainly used during the summer for cooling purpose. i is a residual capturing the errors. Table 5 2 which provides the d escriptive s tatistics of the variables shows that the houses in the sample were built between 1973 and 2005. They include on average 2 to 5 bedrooms and 2 to 4.5 bathrooms. The heated are a s range from1051 to 4523 square feet. The average consumption in the sample is 944.067 kWh. Based on t he results of the electricity consumption model presented in table 5.3, t he adjusted R2 is 0.2127. All variables are statistically significant at the 5% level. Results show that a house rated Energy Star consumes on average 51.428 kWh less in terms of electricity consumption every month than a house rated non -Energy Star. This corresponds to a yearly savings of about 617.136 kWh in electricity consumption Knowing that the average monthly
6 6 consumption for the entire sample is about 944.07 kWh Energy Star updates reduce that consumption by about 5.45%. It corresponds to a saving of about $5.50 each month and $ 66.00 per year in energy expenditure. Since it costs about $1200.00 to upgrade a home to Energy Star Standards (Smith and Jones), it will take a little over 18 months for home owners to capitalize their investments in terms of energy costs. Compared to previous studies such as the Smith and Jones (2003 ), the current results are much more conservative. In fact while comparing traditional versus Energy Star rate homes, the Smith and Jones analysis (2003) found a 16% electricity savings in 2000 and a 10% savings in 2001 for Energy Star homes. On the other hand the results of the current study are more comparable to other studies such as the one by Pigg (2002) in terms of electricity savings. The explanatory variables were square footage, number of individuals in the home and participation the program (Yes=1 No= 0). The data available actually indicated a 4% savings (400 kWh per year). However, the electricity consumption difference was not statistically significant and the study could therefore not come to the conclusion that Wisconsin Energy Star homes act ually use less energy than non Energy Star comparable home. As results in T able 5 3 show wh en using a longer time period ( from 1996 to 2008 ) for electricity consumption instead of only two years 1999 and 2000 for Pigg (2002) or 2000 and 2001 for Smith and Jones (2003), the current analysis show s a significant decrease in energy savings due to the Energy Star rating This finding begs the question knowing how well energy savings attributed to the Energy Star program hold over a long period of time. Th e analysis shows a negative correlation between the year built of a house and the level of electricity consumption, as expected. In fact as a house is newer by one year electricity consumption decreases by 13.136 kWh every month. This corresponds to a sav ing of 157.632
67 kWh each year or 1.4% reduction In other words as the house ages and construction equipment slowly deteriorate as time goes by, each additional year will increase electricity consumption by 13.136 kilowatts every month or 157.632 kWh each year There is a positive relationship between the number of bedrooms, number of bathrooms and heated area of a house and monthly electricity consumption. Indeed, every additional bedroom increases the monthly electricity consumption by 59.588 kWh, 715.06 kWh yearly or 6.31% Every additional bathroom increases electricity usage by 35.720 kWh each month which corresponds to 428.64 kWh yearly or 3.78% Since houses generally have more bedrooms than bathrooms, the larger impact of bedrooms than bathrooms on energy usage was expected. In addition, bedrooms are generally larger than bathrooms thus require more energy usage. As anticipated as heated area increase, consumption increases An a dditional square foot in heated area results in a monthly increase of 0 .0530 k Wh in electricity consumption or an average 0.636 kWh a year or 0.0056% During the summer months, electricity consumption increases by 272.149 kWh per month, about 28.83%. This is predictable knowing that summers in North Florida are very warm and air conditioning system constitute s 35.2% of average annual energy expenditure per household. On the other hand, during the winter months, electricity consumption decreases by 265.378 kWh per month or 28.11%. I n fact, due to Florida temperate winter, space heating only constitutes 5.1% of yearly energy expenditure per household. Like previous studies, the current analysis included physical attributes of homes (heated area, year built, number of bedrooms and number of bathrooms) as well as Energy Star statu s in predicting energy use level. Yet the analysis did not contain any information on occupants such as number of residents, age, income, education level and so forth as well as occupant behavior
68 such as average thermostat setting points (Pigg 2002). The l ack of such information limits the findings and conclusions of the research. In the following sub section uses the same explanatory and dependent variables. Instead of a regular OLS model, the next analysis uses a pooled OLS model and the PTSCS data genera ted. 5 .2.2 Pooled O rdinary Least Squares Estimation and R esults U sing the monthly consumptions aggregated at the yearly level, a po o led OLS analysis is conducted using the four types of electricity consumption as the dependent variables : t otal_conit, summ er _conit, total_conit and other _conit The sample includes 7050 observation for non-ES homes and 743 observations for ES homes. Conventional homes were built between 1973 and 2005. ES homes were built later from 1997 to 2005. Both groups shared very simil ar structural characteristics (number of barrooms, bathroom and heated area). The d escriptive s tatistics of conventional homes are summarized in Table 54 .Table 5 5 summarized descriptive s tatistics of ES homes. Based on T able 5 4 and 5 5, the average tot al annual electricity consumption for ES homes is 857.5 kWh versus 905.7 kWh for non-ES homes The average annual summer electricity consumption for ES homes is 364 kWh versus 339.9 kWh for non -ES homes On average, ES homes consume 206.8 kWh versus 268.1 kWh for nonES homes during the winter On average, ES homes consume 286.8 kWh versus 297.7 kWh for non ES homes during other months Based on tables, ES homes consume less electricity. Using the PTSC data, the following restrictive pooled OLS models are e stimated2: (Model 1 ) pooled: total_conit 0 1 ESi2 effyri3 bedsi+ 4 b athsi + 5 htdareai6 htdareai2it (5 3) 2 All models are tested for heteroskedasticity and corrected for it using heteroskedasticity robust standard errors.
69 (Model 2) pooled: summer_conit = 0 1 ESi2 effyri3 bedsi4 b athsi + 5 htdareai+ 6 htdareai2it (5 4) (Model 3 ) pooled: winter _conit = 0 1 ESi2 effyri3 bedsi4 b athsi + 5 htdareai6 htdareai2it (5 5 ) (Model 4 ) pooled: other_conit = 0 1 ESi2 effyri3 bedsi4 b athsi + 5 htdareai6 htdareai2it (5 6) The results summarized in T able 5 6 show t hat the variable of interest ESi is statistically significant only in the summer months and winter months at the 5% significance level with a P value of 0.000. During the rest of the year it does not play a statically significant role in predicting household yearly electricity consumption (P -value = 0.197 for the total consumption and P -value= 0.734 for the other consumption). This finding shows the importance that weather plays in predicting the effect of ES rating on electricity consumption. The summer es timation shows that consumption will actually increase by 34.695 kWh due to these rating of a house. This can be in part explained by the extremely hot temperatures of the region during the summer time. It is also during this time of the year that resident s heavily rely on air conditioning systems for cooling purposes. In the contrary, the winter estimation reveals that electricity consumption decreases by 51.374 kWh due to the ES rating of the house. This result is also somewhat anticipated since less air condition systems are turn on during the winter. However, this model does not account for heterogeneity in the houses and produce a single constant intercept with underestimated standard errors. In order to account for the house specific effect, a FE model is estimated. In addition, a time trend is also included in the analysis. The following section discusses the estimation. Since this new variable is time sensitive, it will allow for the evaluation the effect s of the ES features on electricity consumption over time.
70 5 2. 3 Panel Data Estimation s and R esults In the FE model s house characteristics of a house which are time invariant cannot be identified. As a remedy, a new variable of interest, the interaction term between the ESi dummy and the billyeart: ( ES*billyear )it. A series of time dummy variables (Dt) are also created for each billing year going from 1996 to 2008. A total of 13 binary variables (D96, D97, D98,, D08) are included in the estimation Each dummy takes value of 1 if billing occurred duri ng the specific year; otherwise it is equal to 0. Since these variables are time sensitive they are also identified. The dependent variables are still four types of electricity consumptions. Using xtreg fe command in S T ATA, t he following models are esti mated as FE models3: (Model 5)FE: total_conit 0 it 1D972D983D994D005D01+ 6D027D038D049D0510D0611D0712D08it (5 7) (Model 6)FE: summer_conit 0 it 1D972D983D994D005D01+ 6D027D038D049D0510D0611D07+ 12D08it (5 8) (Model 7)FE: winter_conit 0 it 1D972D983D994D005D01+ 6D027D038D049D0510D0611D0712D08it (5 9) (Model 8)FE : other_conit 0 it 1D972D983D994D005D01+ 6D027D038D049D0510D0611D0712D08it (5 10) The parameter in the equations tells whether the ES rating of a house helped increase or decrease household electricity consumption over time and by how much. Each individual j ( j = 1 13) corresponds to the increase or decrease in electricity consumption specific that particular year only. 0 it respectively represent the constant and error terms. 3 All models are tested for heteroskedasticity and corr ected for it using heteroskedasticity robust standard errors.
71 Now that the models have been described, a joint significance and autocorrelation tests are carried out in order to ensure the use the appropriate modeling strategy before diving into the discussing of the results. Joint S ignificance Test R esults : Consider then, the hypothesis: Ho: 1= 2= 3== 13= 0 (5 11) Ha: Ho is not true where the null hypothesis ( Ho) indicates that the time dummy variables have no effect on the level of household electricity consumption. The altern ative hypothesis ( Ha) is that the time dummy variables are not jointly equal to zero. There are k = 13 restrictions that are being tested. Using STATA, F statistic obtained is 228.89 and the P value is 0.000. Based on the results of the joint significance test, the decision is to reject Ho. The time dummy variables are statistically and jointly not equal to zero at the 5% significance level They are therefore statistically significant in explain ing the variation in electricity consumption over time. This result confirms the importance of a time trend in examining electricity consumptions over time. Next, a n autocorrelation test is carried out and the results are presented Serial C orrelation Test R esults : Since most TSCS datasets are a very likely to be pl agued by the presence of serial correlation, all four FE models are tested for it and results are discussed. In case there is the presence of autocorrelation the error term is: it i,t 1+Ut where Ut meets all classical assumptions. The following hypotheses are therefore tested using the Durbin -Watson Test assuming that all X variables are strictly exogenous with = 0.05, the significance level. Consider the hypothesis: Ho: = 0 (5 12) Ha: (5 13)
72 where the null hypothesis (Ho) is no serial correlation in the error term and the alternative hypothesis ( Ha) is that there is autocorrelation Based on the results of the test summarized in Table 5 7 the decision is to reject Ha since the P values of all four models are 0.000 and less than = 0.05.There is statistical evidence that the error terms are serially correlated at the 5% significance level. In other words, the level of electricity consumption at time period t is not only determined by the changes in consumption that occurred during this time period but also by the memory of the level of consumption in the last time period t 1. This is defined as an autoregressive process of order one or AR ( 1 ). As mention ed previously in the study, if not accounted for, autocorrelation will affect OLS results: standard errors will be too small; test statistics will be too big and thus OLS will create a false confidence in results. Since the time dummy variables are jointly significant and there is evidence of serial correlation, ( Model 5)FE through (Model 8 )FE described earlier are run as FE models using the xtregarfe STATA comma nd. The results from the FE estimation which are summarized in T able 5 8 show that the variab le of interest (ES*billyear)it increases the level of total annual electricity consumption by 89.63 kWh Because of the time effect, ES rating increases annual electricity usage by 74.320 kWh for the summer consumption. This result is also statistically s ignificant with a P value of 0.000. Over time, annual winter consumption decreased by 61.57kWh while other consumption increased by 34.222 kWh due to the ES rating. It is important to note that all th ese result s are statistically significant at the 5% sign ificance level with a P -value of 0.000. When not corrected for serial correlation, the FE estimations generate the results summarized in T able 5 9. T he variable of interest (ES*billyear)it increases the level of total
73 annual electricity consumption by 11.609 kWh ES rating increases annual summer electricity consumption by 12.044 kWh Over time, annual winter consumption increases by 3.904 k Wh while other consumption decreases by 4.338 kWh due to the ES rating. All th ese result s are statistically significan t at the 5% significance level. The FE estimation tells a similar story of the effect energy efficient features of a house being dependent on the climatic conditions: more during the summer season than the winter season due to hotter temperatures Based o n this FE model, ES rating of a house actually increases electricity consumption as time goes by. This phenomenon can be explained by the gradual degradation over an extended period of time of energy efficiency features built into the house. The examination of such time trend introduced in this research is lacking in the current literature, as most studies only focus on the stationary effect of the ES rating which does not allow for the estimation of effect on time on ES performance. This positive time trend therefore shows that on average, from one year to the other, ES rating increases electricity consumption. In summary, the current analysis shows that changes in electricity consumption are explained by: Structural characteristics of the house accounted for as unit -specific effect Changes in consumption pattern occurred during a given year which can be explained for instance by changes in climate (i.e. summer, winter or off -seasons) Changes in consumption pattern which occurred in the previous year Chang e s in the performance of ES upgrades due to gradual degradation over time In the following section a nother DID estimation is performed and compared to the initial one conducted in chapter 4. 5 3 Difference -in -Difference Estimation s and R esults Based on the previous analysis in previous sections there is a statistically significant difference in the level of electricity consumption from one year to the other due to ES rating. The
74 following models examine how that difference actually behaves overtime. A DID estimation is therefore carried out with the parameter of interest bei ng T his parameter corresponds to the change over time in the difference in electricity consumption due to the ES rating. T his estimation differ s from the previo us one in chapter 4 in the sense that the dependent variables are here all electricity consumptions ( summer, winter and off -season consumptions ) instead of electricity, natural gas and a combination of electricity and natural gas. The current estimation al so differs from the previous one since it is carried out using billed years from 1996 to 2008 instead of only two years (2000 and 2006). By adding data on other time periods, the study hopes to capture any underlying existing trends and confounding factors affecting the results (Meyers, 1995). 3 models are specified. The first model (Model 9 ) DID include s the ES dummy variable as well as time dummy variables as explanatory variables : (Model 9) DID: total_conit 0 1 ESi1D972D983D994D005D016D02+ 7D038D049D0510D0611D0712D08it (5 14) Similarly to the DID estimation conducted in chapter 4 the constant term represents the average yearly electricity consumption for conve ntional or non -ES house. 1 associated with the ES dummy variable represents the change in electricity consumption due to the treatment specific effect: ES rating of the house. Each j (j= 1,, 13) represents the va riation in electricity consumption specific to that year It corresponds to the time trend impacting household electricity consumption common to both types of houses In the second model, (Model 10) DID, the interaction term between the ES dummy and the b ill year is added: (ES *billyear ) it. Associated with this variable captures the change overtime in the difference in electricity consumption between ES house and non -ES houses.
75 (Model 10)DID: total_conit 0 1 ESi1D972D98+ 3D994D005D016D02+ 7D038D049D0510D0611D0712D08it it (5 15) In the last model, (Model 11)DID, structural characteristics the houses are added in the attempt to explain better the variations in consumption. They include effective year, number of bedrooms, bathrooms and heated area. (Model 11)DID: total_conit 0 1 ESi1D972D983D994D005D016D02+ 7D038D049D0510D0611D0712D08it 2 effyri3 bedsi4baths 5htdar eai6 htdareai 2i (5 16) 2 ,, 6 represent the change in household s electricity consumption due the structural characteristics of a house. The models are run4 using OLS estimation and the results are summarized in the Table 5 10 using total consumption data. The average consumption for traditional homes is 862.281kWh and ES homes consume about 17.481kWh less base d on (model 9)DID. The results are statistically significant at the 5% level. As additional explanatory variables are added in to the analysis R2 increases from 0.0849 to 0.1824. The results of ( Model 11) DID show that total electricity consumption increased by 13.133 kWh due to the impact of the ES features overtime. The results are statistically signif icant at the 5% significance level Knowing that the average total electricity consumption is 910.010 kWh, ES qualified homes consume about 1.46% more energy than conventional homes. This result is much more conservative than the previous results in chapte r 4 ( i.e. 7.59%). The following 3 models are also estimated using summer consumption data and the results are summarized in T able 5 11. 4 All models are tested for heteroskedasticity and corrected for it using heteroskedasticity robust standards errors.
76 (Model 12)DID : summer_conit 0 1 ESi1D972D983D994D005D016D02+ 7D038D049D0510D0611D0712D08it (5 17) (Model 13)DID : summer_conit 0 1 ESi1D972D983D994D005D016D02+ 7D038D049D0510D0611D0712D08 r)it it (5 18) (Model 14)DID: summer _conit 0 1 ESi1D972D983D994D005D016D02+ 7D038D049D0510D0611D0712D08it 2 effyri3 bedsi4 5 htdareai+ 6 htdareai 2it (5 19) The results show tha t summer electricity consumption increased by 12.086 kWh due to the impact of the ES features overtime. The results are statistically significant at the 5% significance level with a P -value of 0.003. Since the average summer consumption in the sample is 342.21 kWh, it corresponds to a 3.53% increase for ES qualified homes. The average consumption for traditional homes is 376.425 kWh and ES homes consume about 36.385 kWh more based on (model 12)DID. The results are statistically significant at the 5% level. As additional explanatory variables are added in to the analysis R2 increases from 0 .4435 to 0.4863. The lager effect of the rating can be explained by the hot temperatures during the summer months and which reinforced the seasonality of consumption. The same analysis is carried out using winter consumption data and the results are summarized in T able 5 12. The 3 models are : (Model 15)DID : winter_conit 0 1 ESi1D972D983D994D005D016D02+ 7D038D049D0510D0611D0712D08it (5 20) (Model 16)DID : winter_conit 0 1 ESi1D 972D983D994D005D016D02+ 7D038D049D0510D0611D0712D08 r)it it (5 21)
77 (Model 17)DID: winter_conit 0 1 ESi1D972D983D994D005D016D02+ 7D038D049D0510D0611D0712D08it 2 effyri3 bedsi45 htdareai6 htdareai 2it (5 22) The results of (Model 17) DID show that winter electricity consumption increased by 4.989 kWh due to the change in the impact of the ES features overtime. Since the average winter consumption in the sample is 262.23 kWh, it corresponds to a 1.90% increase for ES qualified home s Yet, t he results are not statistically significant with a P -value of 0.123. The average consumption for traditional homes is 203.144 kWh and ES homes consume about 68.705 kWh less based on (model 1 5 )DID. The results are statistically significant at the 5% level. As additional explanatory variables are added in to the analysis R2 increases from 0.2825 to 0.3436. The smaller effect of the rating can be explained by the cooler temperatures during the winter months. The following 3 models are run using othe r consumption data and the results are summarized in T able 5 13: (Model18)DID: other_conit 0 1 ESi1D972D983D994D005D016D02+ 7D038D049D0510D0611D0712D08i (5 23) (Model 19)DID : other_conit 0 1 ESi1D972D983D994D005D016D02+ 7D038D049D0510D0611D0712D08it it (5 24) (Model 20)DID: other_conit 0 1 ESi1D972D983D994D005D016D02+ 7D038D049D0510D0611D0712D08it 2 effyri3 bedsi4baths 5htdareai 6 htdareai 2it (5 25) The results of (Model 20) DID show that other electricity consumption decreased by 3.942 kWh due to the impact of the ES features overtime. Since the average other consumption in the sample is 296.65 kWh, it corresponds to a 1.32% increase for ES qualified homes However,
78 these re sults are not statistically significant with a P -value of 0.135. The average consumption for traditional homes is 282.717 kWh and ES homes consume about 14.837 kWh more based on (model 18) DID. The results are statistically significant at the 5% level. As additional explanatory variables are added into the analysis R2 increases from 0.1693 to 0.2361. 5 .4 Conclusions This chapter presented predictive model s of household electricity consumption in order to f identify the effect of its energy efficiency statu s (Energy Star rated or not), physical attributes of the house and climatic conditions on levels of household energy usage First, the two different types of energy datasets used in this part of the study were described. Secondly, a simple OLS and pooled O LS estimations were carried out Then a FE and DID analysis were performed. Results from joint significance and autocorrelation test s were also outlined. Finally, findings from the estimation s were presented as well as their interpretations and implication s in terms energy consumption and energy savings outcomes Monthly electricity consumption from 1996 to 2008 is estimated as function of structural characteristics, Energy Star status and seasonality of consumption. The findings show that Energy Star rate d homes consume about 5.45% less electricity each month than traditional homes. These results are much more conservative than previous studies in the current literature that estimated saving of 3% all the way to 16% at the household level. The study theref ore shows that a short time frame (for instance two years in the current literature) does not allow for a true assessment of the effectiveness of the Energy Star program in term of electricity savings at the household level. Over a longer time period, ener gy savings shrink significantly (i.e. by about half) as well thus in effectiveness in greenhouse gas emission reduction, energy security and environmental protection. The study also shows that physical characteristics of the house, for instance number of bedrooms, bathrooms and heated area positively affect household electricity
79 consumption, as predicted. Environmental factors such as seasonality (winter and summer months) have a significant influence on monthly electricity consumption. Indeed, as the tempe rature rises during the summer so does electricity consumption as households heavily rely on air conditioning systems to cool off. In the contrary, during the winter months, as climate cools down, consumption decreases since household use less on air conditioning. The short -comings of the model rise in the fact that it does not include information on occupants of the homes such social characteristics (income, education levels, number of occupants, household composition, and so forth) or occupant behavior s uch as thermostat setting. As a result, the low predictive power of the model (21.24%) can be improved by including such explanatory variables. There is a very slim body of literature on the impact of energy saving investments such as the Energy Star progr am on household energy consumption over an extended period of time. The current research consequently contributes to this body of literature since it allows for the evaluation of an energy efficiency initiative, the Energy Star program over time. Results from the study mainly address the issue of sustainability of energy savings generate by such voluntary programs and therefore motivate a serious examination of the following questions: how much energy is actually saved over time by voluntary energy efficie ncy programs at the household level? Are those savings sustainable in the longrun? Are the investments actually worth it for the average homeowner participating in the Energy Star program? To what existent are such voluntary programs protecting the planet at the household level? In the attempt to answer the previous questions, the original monthly data was collapsed and aggregated at the yearly level as TSCS data using STATA. With a dataset that is time sensitive, the evaluation of the ES program over time is made more accurate.
80 The pooled OLS estimation shows that during summer months, electricity consumption rises by 34.695 kWh for homes rated ES. Consumption decreases by 51.374 kWh during the winter for ES homes. Both results are statistically significa nt. When considering total consumption and other consumption, levels of energy usage decrease respectively by 18.54 kWh and 1.846 kWh. These results are however not statistically significant. Since this estimation does not take in account house specific ef fect, FE estimation corrected for autocorrelation is carried out accounting for not only heterogeneity at the house level but also a time effect. Indeed the interaction term (ES*billyear ) it captures the time trend in the impact of ES ratin g on electricity consumption. All estimates are statistically significant. ES homes have a higher consumption overall for total consumption (89.63 kWh more), summer consumption (74.320 kWh more) and other consumption (34.22 kWh more) than non ES except during the winter when ES homes consumed about 61.57 kWh less than traditional homes. This shows that the effect s of the energy efficiency upgrades do change over time and they are not stagnant as assumed by the current literature. The changes can be attributed to the natur al degradation of construction over the years. The 13 time dummy variables created to indicate the bill years are all joint ly statistically significant Finally, the DID estimation revea ls a similar conclusion In fact, with ES homes consuming 13.133 kWh more in total consumption or 1.46% 12.086 kWh more in summer consumption or 3.53% and 4.989 kWh or 1.90% more in winter consumption than non ES except for the off season months when ES homes consumed about 3.942 kWh or 1.32% less than traditional homes The results during the winter and off season are not statistically significant. The DID conclusion however shows how the changes observed previously with the FE model change over time: they increase overtime revealing that the ES upgrades become less
81 and less efficient in conserving energy to a point that they actually increase household energy consumption. Keeping the main research question in mind, determining if energy efficient program such as the ES program saves energy and if the savings are sustain able over time, important lessons can be learned from this study. In fact, this study shows that if not maintained or updated, energy efficient features built into ES qualifie d homes can result in an increase of energy consumption at the household level ov er time. As these upgrades degrade overt time, the effect of ES rating on energy consumption in negatively affected. ES houses could be actually consuming more energy and thus hurting more the environment instead of protecting it. Some policy implications include the risk of investing into short term solutions instead of really assessing longterm performance of such initiatives in order to make adequate changes. As suggested earlier, a policy recommendation can be a mandatory periodic evaluation of ES home s to ensure that energy savings are still maintained. In the following the main conclusions and policy implications of the study are outlined Table 5 1. Variables D escription Variable Variable s Description billed_con Monthly billed elec tricity consumption in kWh ES Energy Star dummy effyr Year built of the house beds Number of bedrooms baths Number of bathrooms htdarea Heated area in the house in square feet summer Dummy variable indicating summer consumption ( 1=Yes ,0 =No) winter Dummy variable indicating winter consumption ( 1=Yes ,0=No)
82 Table 5 2 Descriptive Statistics Variables Obs. Mean Std. Dev. Min Max billed_con 97515 944.0672 548.574 1 7121 effyr 97515 1998.495 2.179315 1973 2005 ES 97515 0.087299 0 .282275 0 1 beds 97515 3.274286 0.46689 2 5 baths 97515 2.116521 0.320358 2 4.5 htdarea 97515 1850.127 342.7187 1051 4523 Table 5 3. Monthly OLS E stimation Results Variables Estimates P value effyr 13.136 0.000 (0.779) ES 51.428 0.000 (5.513) beds 59.588 0.000 (4.441) baths 35.72 0.000 (6.895) htdarea 0.053 0.000 (0.063) htdarea 2 0.000056 0.001 (1.65E 06) summer 272.149 0.000 ( 4.329) winter 265.378 0.000 (3.583)
83 Table 5 4 Descriptive S tatistics of Conv entional Non -ES Homes Variables Obs. Mean Std. Dev. Min Max total_con_m 7050 905.7 375.5 6 3333.8 sum_con_m 7050 339.9 213.0 0 1368.6 winter_con_m 7050 268.1 169.6 0 2051 other_con_m 7050 297.7 140.8 0 1141.8 effyr 7050 1998.5 2.3 1973 2005 beds 7050 3.3 0.5 2 5 baths 7050 2.1 0.3 2 4.5 htdarea 7050 1869.1 357.6 1051 4523 htdarea2 7050 3621507 1457106 1104601 2.05 E +07 Table 5 5 Descriptive S tatistics of ES Homes Variables Obs. Mean Std. Dev. Min Max total_con_m 743 857.5 387.8 23 3557.2 sum_con_m 743 364 174.5 0 1242.3 winter_con_m 743 206.8 112.6 16.5 1124.5 other_con_m 743 286.8 145.5 0 1319 effyr 743 1998.5 2.2 1997 2005 beds 743 3.2 0.4 3 4 baths 743 2.1 0.3 2 3 htdarea 743 1790.7 328.9 1329 2911 htdarea2 743 3314436 1332876 17 66241 847392
84 Table 5 6. Parameter Estimates, Standard Errors, and P values of Pooled OLS E stimation using A nnual PTSCS Data Models ( Model 1 ) pooled ( Model 2) pooled ( Model 3 ) pooled ( Model 4) pooled Explanatory variables Estimates P value Estimates P value Estimates P value Estimates P value Constant 28963 0.000 9834.77 0.000 14470.11 0.000 4658.12 0.001 ( 3627.498) ( 2274.653) ( 1955.712) ( 1425.936) ES i 18.524 0.197 34.695 0.000 51.374 0.000 1.846 0.734 (14.362) (6.719) (4.518) (5.435) Effyr i 14.444 0.197 4.941 0.000 7.229 0.000 2.273 0.001 ( 1.811 ) ( 1.137 ) ( 0.977 ) ( 0.712 ) beds i 52.67 0.00 14.336 0.035 22.977 0.000 15.356 0.001 ( 12.032) ( 6.803) ( 5.429) ( 4.552) bathsi 18.724 0.303 0.814 0.935 19.419 0.026 0.119 0.986 ( 18.185 ) ( 9.961 ) ( 8.746 ) ( 6.879 ) htdarea i 0.35 0.014 0.249 0.000 0.048 0.519 0.053 0.195 ( 0.142) 0.067 0.075 0.04 htdareai 2 1.62E 05 0.661 -3.80E 05 0.038 1.05E 05 0.59 9.35E 06 0.375 ( 3.70E 05) ( 1.74E 05) ( 1.94E 05) ( 1.50E 05) R 2 0.0966 0.0389 0.0761 0.0635
85 Table 5 7. F Statistics and P -values of A utocorrelation T est Models F statistic P value (Model 5) FE 867.154 0.000 (Model 6) FE 437.143 0.000 (Model 7) FE 952.973 0.000 (Model 8) F E 981.422 0.000 Table 5 8. Parameter Estimates, Standard Errors, and P values of Interaction Term Variable in FE Estimation after Accounting for Autoregressive Disturbances AR (1) Variable (ES*billyear)it Value of Rho Models Estimates P value (Model 5) FE 89.63 0.000 0.5474 (8.24) (Model 6) FE 74.32 0.000 0.3642 (3.3147) (Model 7) FE 61.57 0.000 0.5133 (3.428) (Model 8) FE 34.222 0.000 0.4737 (2.860) Table 5 9. Parameter Estimates, Standard Errors, and P values of values of I nteraction Term Variable in FE Estimation without Accounting for Autoregressive Disturbances AR (1) V ariable (ES*billyear) it Models Estimates P value (Model 5) FE 11.609 0.003 (3.949) (Model 6) FE 12.044 0.000 (2.126) (Model 7) FE 3.904 0.032 (1.823) (Model 8) FE 4.338 0.007 (1.595)
86 Table 5 10. Parameter Estimates, Standard Errors, and P -values of the DID Estimation using Total Annual Electricity Consumptions Dependent variable total_con it Models ( Model 9) DID ( Model 10) DID ( Model 11) DID Explanatory Estimates P value Estimates P value Estimates P value Constant 862.281 0.000 866.759 0.000 39991.93 0.000 (12.826) (12.936) (3709.649) ES i 17.481 0.007 34860.97 0.007 263000.03 0.035 (16.862) (12878.79) (12442.44) (Es*billyear) it No 17.392 0.007 13.133 0.035 (6.431) (6.213) D t Yes Yes Yes House features i No No Yes R 2 0.0849 0.0858 0.1824
87 Table 5 11. Parameter Estimates, Standard Errors, and P -values of the DID E stimation using S ummer A nnual E lectricity Consumptions Dependent variable summer_con it Models (Model 12) DID (Model 13) DID (Model 14) DID Explanatory variables Estimates P value Estimates P value Estimates P value Constant 376.425 0.000 379.901 0.000 15550.41 0.000 (5.700) (7. 076) (1717.59) ESi 36.385 0.000 27046.39 0.001 24165.41 0.003 (6.658) (8142.163) (12442.44) (Es*billyear) it No 13.519 0.001 12.086 0.003 (4.065) (4.051) Dt Yes Yes Yes House features i No No Yes R 2 0.4435 0.4463 0.4863
88 Table 5 12. Para meter Estimates, Standard Errors, and P -values of the DID E stimation using W inter A nnual E lectricity C onsumptions Dependent variable winter_con it Models (Model 15) DID (Model 16) DID (Model 17) DID Explanatory Estimates P value Estimates P value Estimates P value Constant 203.144 0.000 204.819 0.000 11149.95 0.000 (5.122) (4.543) (1531.77) ES i 68.705 0.000 13119.77 0.048 10055.91 0.121 (5.983) (6643.96) (6484.65) (ES*billyear) it No 6.514 0.049 4.989 0.123 (3.315) (4.051) D t Yes Yes Yes House features i No No Yes R 2 0.2825 0.2844 0.3436
89 Table 5 13. Parameter Estimates, Standard Errors, and P -values of the DID E stimation using O ther A nnual E lectricity C onsumptions Dependent variable other_con it Models (Mo del 18) DID (Model 19) DID (Model 20) DID Explanatory Estimates P value Estimates P value Estimates P value Constant 282.717 0.000 282.038 0.000 10291.57 0.000 (4.688) (5.173) (1507.63) ES i 14.837 0.007 5305.195 0.324 7921.286 0.1 34 (5.477) (5379.202) (5286.957) (ES*billyear) it No 2.640 0.325 3.942 0.135 (2.685) (2.639) D t Yes Yes Yes House features i No No Yes R 2 0.1693 0.1708 0.2361
90 CHAPTER 6 CONCLUSIONS This chapter outlines general conclusions and pol icy implications of the research. Section 6 .1 presents an overview of the research. Then section 6 .2 discusses the limitation of the study. Finally, policy implications as well as future steps are summarized in Section 6 .3. 6 .1 Overview of the Research Wit h a current U.S. greenhouse gas emissions increasing each year, policy makers are face with the overwhelming challenge of climate change. Energy efficiency policies or voluntary programs such as the Energy Star program is one way the current government att empts to conserve energy and preserve our environment. This study investigate s the effect of the Energy Star rating on energy consumption in Gainesville, Florida The main research question is therefore to determine if energy efficient upgrades save energy and if so, if this efficiency maintained overtime. Difference in Difference pooled OLS and FE estimations are adopted as tool s to evaluate the performance of the Energy Star rating (the treatment) overtime on household energy consumption (the outcome). T wo groups (ES homes and Non -ES homes) are defined and a 13 years time period is used. Dummy variables capturing seasonality of energy consumption and bill years are created. Energy consumption of homes initially include d electricity consumption measured i n kilowatts and natural gas consumption measured in thermos. The study focuses later on electricity consumption due the hot climate of the region of study. In recent years, EPA reports claimed significant energy savings due to the execution of the Energy S tar strict energy efficiency guidelines. However, t he different empirical models used in this study reveal unexpected findings. In fact estimate s show an increase energy consumption for homes rated Energy Star as a result of the Energy Star rating in Gain esville, Florida overtime from a minimum of 4.989 kWh per year to a maximum of 89.63 kWh depending on the model
91 The latest result is believed to be the most accurate prediction since it results from the most appropriate modeling approach for the data avai lable. These results are statistically significant. These findings challenge the longterm performance of the Energy Star program. Indeed, they lead to the conclusion that Energy Star guidelines are effective in reducing energy consumption when the house i s built but the se guidelines do not sustain energy savings over time. 6 .2 Limitation s of the S tudy A limitation in the study is the lack of data on occupant behavior or occupant actions (Guerin, et al., 2000) such as thermostat setting. In fact, there i s lack of information on whether residents of ES rated homes tend to set their thermostat higher during the summer and lower during the winter in a way that expected energy savings are reduced (Smith and Jones 2003). Characteristics of the occupants such as age, income, number of occupants, education and so forth that explain energy consumption are also not observable (Guerin, et al., 2000, and Yohanis et al., 2007). Adding more time invariant variables that affect energy consumption such as tree coverag e can also prov ide a better measure to the effect of time on energy consumption. Incorporating these elements to the current information available would make the ideal data for this research and produce stronger results 6 .3 General Conclusions and Policy Implications This analysis has important policy implications. In fact, it forces policy makers to look clo ser at the long term effects of energy efficiency programs in addition to the short -tem impacts. For instance, without long -term examination of progra ms such as the Energy Star program, there is the risk of attributing energy efficiency tax credits to home owners that are actually polluting more. In fact, the energy efficiency tax credits were initially established in 2006 as a result of the Energy Polic y Act of 2005. The main objective of energy efficiency tax credits is to encourage environmental stewardship by monetarily rewarding individuals or
92 organization s t hat invest in environmental friendly equipment or construction. Since then, energy efficiency tax credits have evolved. Starting at 10% of the cost or up to $500 in 2006 and 2007, credits have increased to a significant 30% or up to $1,500 today. Federal efforts such as the Economic Stabilization Act of 2008 signed by former President Bush and the American Recovery and Reinvestment Act of 2009 under the new Obama administration are example s of policy initiatives reinforcing the implementation of e nergy efficiency tax credits. Federal tax credits for Energy Efficiency includes installation of energy-efficient windows and doors, insulation, roofs, heating and cooling equipment such as water heater ( non-solar) and biomass stoves. Such features are built into ES qualifies homes. Other tax credit s such as the Residential Renewable Energy Tax Credits reward consumers who have installed solar energy systems (including solar water heating and solar electric systems), small wind systems, geothermal heat pumps, and residential fuel cell and microturbine systems (ENERGY STAR, 2007) In addition to federal tax incentives, some consumers will also be eligible for utility or state rebates, as well as state tax incentives for energy -efficient homes, vehicles and equipment As a result, if periodic evaluation s are not conducted and long -term performance not assessed the attribution of energy efficient tax credits can simply turn into a marketing tool for homebuilders or energy efficient equipment manufacturers. This will in return hinder the purpose of the initiative which is to encourag e behavior that actually decreases GHG emissions and thus protects the environment. Future works of this study will consist in incorporating more time varying and spatial explanatory variables which characterize each house and thus influenc e household energy consumption. For ins tance, change in tree coverage in the area Appraised home values or sale prices of the homes can be incorporated to see how ES features are being capitalized by
93 homeowners. Indeed, th e sale transaction information can be used t o proxy the changes in house hold composition over time. It will help measure how many times the home has changed owner. This will therefore account for the fact that different household types have different characteristics and habits and therefore different energy use levels On one hand, the results of this study can be used as a tool to market energy efficient home upgrades to home buyers in terms of reduced homeownership operating cost due to energy savings. On the other hand, it sheds some light on the need for constant evaluatio n and assessment of the performance of such initiatives over time Frequent evaluation will in return inspire more up to date and relevant policies aimed at promoting efficient energy usage and environment protection. The major implication from this study is that houses rated ES need to be evaluated periodically in order to ensure that the energy efficient upgrades built into the house are still efficient since they gradually degrade overtime. Another extension of this study can thus be to estimate the poi nt at which an ES house loses its energy saving and environmentally friendly capabilities. As described in the new Stimulus Package, environmental protection and natural resource conservation have an important place on the new administration s agenda. It accordingly follows that a proper evaluation of energy efficiency programs such as the Energy Star program is crucial. Indeed, this study shows that it constitutes a vital step toward meeting governmental goal of implementation of sound, cost -effective en ergy management and investment practices to enhance the nation's energy security and environmental stewardship ( U S D OE, 2009).
94 LIST OF REFERENCES Abadie, A. (2005) Semiparametric difference -in -differences estimators Review of Economic Studies 72, 1 19. Adolph, C ., D M. Butler and Sven E. Wilson. (2005). Which Time Series Cross -Section Estimator Should I Use Now? Guidance from Monte Carlo Experiments Paper presented at the 2005 Annual Meeting of the American Political Science Association,Washingt on, D.C. Alachua County Property Appraiser (2008). Annual Report 2008 S ite Retrieved July 10, 2009, from http://acpafl.org/pdf/annual_report2008.pdf Angrist, J.D. and Krueger, A.B. (1999). Empirical strategies in labor economics Handbook of Labor Economics vol. 3A, ed.O. Ashenfelter and D. Card. Amersdam: North Holland Athey,S.C. and G.W. Imbens, (2006) Identification and I nference in N onlinear D ifference in D ifference M odels Econometrica 74, 43198. Atlantic Design Homes (2007). Energy Efficie ncy: Energy Star & Green Building Site Retrieved July 11, 2009, from http://www.atlanticd.com/energy_star.html Ashenfelter, O (1978). "Estimating the Effect of Training Programs on Earnings," Review of Economics and Statistics Beck, N and J N. Katz. (1 995) What To Do (and Not To Do) with Time -Series CrossSection Data. American Political Science Review 89: 634 47 Cameron A. C and P K. Trivedi (2005, May). M icroeconometrics: Methods and Applications Cambridge University Press New York Card, D. and A.B Krueger. (1994) Minimum wages and employment: a case study of the fast food industry in New Jersey and Pennsylvania American Economic Review 84, 77293. Cho, S H, J.M. Bowker, and W.M. Park. (2006) Measuring the Contribution of Water and Gree n Space Amenities to Housing Values: An Application and Comparison of Spatially Weighted Hedonic Models Journal of Agricultural and Resource Economics 31(3): 485507. Colton, R. (1995). Energy E fficiency as a Credit E nhancement: Public U tilities and the A ffordability of F irst T ime H omeownership Unpublished report, Fisher, Sheehan, and Colton, Public Finance and General Economics Crabbe Nathan. Eco -F riendly and A ffordable? The Gainesville Sun. September 18, 2006. Newspaper on line. Available from ht tp://www.gainesville.com/article/20060918/LOCAL/209180319?Title=Eco-friendly and affordable. Accessed July 10th 2009.
95 DeCanio, S.J., and W Watkins. (2008) Investment in Energy Efficiency: Do the Characteristics of Firms Matter? Review of Economics and Statistics 80 (1 ): 95107. Earth Policy Institute (2008). Rising Temperatures and its Effects Site Retrieved July 10, 2009, from http://www.earth -policy.org/Books/PB3/PB3ch3_ss2.htm Energy Information Administration. ( 2005). Residential Energy Consumpti on Survey (RECS 2005) Site. Retrieved July 11, 2009, from http://www.eia.doe.gov/emeu/recs/ ENERGY STAR. (2004). Protecting the Environment -Together: EPA 2003 Annual Report Site Retrieved July 10, 2009, from http://www.energystar.gov/ia/news/downloads/annual_report_2003.pdf ENERGY STAR. (2006). What is a HERS Rating Site. Retrieved July 10, 2009, from http://www.energystar.gov/index.cfm?c=tax_credits.tx_index ENERGY STAR. (2006). Federal Tax Credit for Energy Efficiency Site Retrieved July 12, 2009, from http://www.energystar. gov/index.cfm?c=bldrs_lenders_raters.nh_HERS Epsey, M., and K. Owusu -Edusei (2001). Neighborhood Parks and Residential Property Values in Greenville, South Carolina Journal of Agricultural and Resource Economics .33.3: 487 492. Garrett, G (1998) Pa rtisan Politics in the Global Economy. New York: Cambridge University Press Gillingham, K N and K Palmer. (2004) Retrospective Examination of Demand Side Energy Efficiency Policies Discussion Paper Resource for the Future Glaser E.L., and M.E. Kahn ( August, 2008) The Greenness of Cities: Carbon Dioxide Emissions and Urban Development Working Paper 14238, National Bureau of Economic Research. Guerin.D.,B.L.Yust and J.G. Coopet. (September 2000) Occupants Predictors of Household Energy B ehavior and Consumption Change as Found in Energy Studies Since 1975 Family and Consumer Sciences Research Journal Vol.29, No.1: 4880 Howarth, R B., B M. Haddad, and B Paton. (2000) The Economics of Energy Efficiency: Insights from Voluntary Partic ipation Programs Energy Policy 28 9 : 477486.91. Intergovernmental Panel on Climate Change (2001). Climate Change 2007: Synthesis Report Site. Retrieved July 11, 2009, from http://www.ipcc.ch/ipccreports/tar/vol4/index.php?idp=0 Intergovernmental Panel o n Climate Change (2007). Climate Change 2007: Synthesis Report Site Retrieved July 10, 2009, from http://www.ipcc.ch/pdf/assessment report/ar4/syr/ar4_syr.pdf
96 Kushler, Martin. (2009, February 5). Utility Sector Energy Efficiency: Current Status, Key Cha llenges, and Potential Role in a Carbon Constrained Word, Presentation to the 36th Annual PURC Conference (2009) American Council for an Energy -Efficient Economy, Gainesville, Florida Meyers, B.D. (1995) Natural and quasi -experiments in economics J ournal of Business & Economics Statistics 13 9: 15161. Meyers, B.D., W.K Viscussi and D.L. Durbin. (1995) WorkersCompenstaion and Injury Duration: evidence from natural experiment American Economic Review 85, 32240. Nevin, R., and G. Watson (1998) Evidence of rational market valuations for home energy efficiency The Appraisal Journal 66: 401408. Pigg, S., (2002 ). Energy Savings from Wisconsin Energy Star Homes Program Research Report 211 1. Energy Center of Wisconsin. Smith, M., and P. Jone s. (2003, September ) The Impact of Energy Efficient House Construction on Homeownership Costs: A Comparative Study in Gainesville, Florida Family and Consumer Science Research Journal Vol.32, No.1, 76 98. U S Department of E nergy. ( 2009). Federal Energy Management Program Site Retrieved July 11, 2009, from http://www1.eere.energy.gov/femp/ Varian H R. (1992). Microeconomic Analysis 3rd ed. W W Norton & Co Inc Wooldridge, J. (2006). Introductory Econometrics, A Modern Approach, 3rd. ed. Michigan S tate University. Thomson South -Western Co. Wooldridge, J. ( 2007). Whats New in Econometrics? Lecture10: Difference in Difference Estimation. NBER. Summer Institute 2007. Yohanis. G.Y, J.D.Mondol, A. Wright and B.Norton. (2 007). Real life energy use in the UK: How occupancy and dwelling characteristics affect domestic electricity use. School of the Built Environment, University of Ulster, Newtownabbey BT37, Northern Ireland, UK
97 BIOGRAPHICAL SKETCH Eloise Francesca Aka was born in 1984 in Abidjan, Cote dIvoire (West Africa). The middle child of three children, she graduated from Lycee Blaise Pascal High School, Abidjan in 2002. She moved to the United States in August 2002 and earned her A ssociates of A rts degree in a ccounting from Santa Fe Community College in December 2004. Eloise then transferred to the University of Florida and earned her B achelor of S cience degree in food and resource e conomics in May 2007. During her undergraduate years, Eloise interned as a Management Consultant with East Gaine sville Development Corporation, volunteered at S h ands Hospital and represented the University of Florida at the first annual Florida International Leadership Conference in February 2007. After graduation, she also interned in the summer of 2007, working wi th IFAS professors on investigating consumers willingness to pay for locally grown fruits and vegetables in Gainesville, FL She was admitted in August 2007 into the Food and Resource Economics G raduate P rogram. Her areas of interest were i nternational d e velopment and e nvironmental e conomics A s a graduate Research Assistant, Eloise worked on a grant sponsored the Center for International Business and Research analyzing EUREPGAP certification for Florida growers. During her second year of graduate studies she provided consulting services to the Gainesville Energy Efficient Communities by analyzing energy consumption data for the Gainesville area and making energy efficient policy recommendations. S he also served as a Teaching Assistant of Agribusiness M arketing and Econometrics courses. She received her Master of S cience in f ood and resource e conomics in August 2009. Eloise plans to pursue a career as natural resource economist in the public utility sector a private organization or an international institution