University of Southern California Dissertations and Theses

The interactions between housing and business

(USC Thesis Other)

The interactions between housing and business

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content 1 The Interactions between Housing and Business By Bingbing Wang A Dissertation Presented to the FACULTY OF THE GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirement for the Degree DOCTOR OF PHILOSOPHY URBAN PLANNING AND DEVELOPMENT August 2018 Dissertation Committee: Richard K. Green (chair) Raphael Bostic Gary Dean Painter 2 Dedicated To my families 3 Acknowledgements First and foremost, I would like to express my sincere appreciation and thanks to my advisor, Richard K. Green, for the continuous support of my Ph.D. study both academically and emotionally. His immense patience and knowledge lead me through the whole research and writing of the thesis. He has conveyed to me how good economic research is done. We met almost weekly to discuss my research and he could solve my problems on each of these meetings. He helped me on every little issue and confusions I have for the research and guided me through how to look for an idea, how to find the data, how to design the research, how to find the appropriate methods to deal with endogeneity, how to interpret the results properly, how to write academically and also how to interact and corporate with the other researchers. I really appreciate his dedication to cultivating students by contributing his time, ideas, and understanding, which make my Ph.D. study productive and pleasant. I would also like to thank my dissertation committee members Professor Raphael Bostic and Gary Painter for their guidance and comments to improve the dissertation. They took time to meet with me to discuss my papers and guided me through the research process. I am also grateful to Professor Rodney Ramcharan who met with me and helped on the research design and methodology. Professor Christian L. Redfearn and Professor Jorge De la Roca also helped me on the data and methodology. My peers and friends at USC provided their support and companion during the five years for the qualifying exam, scholarship application, the dissertation writing and job searching. 4 Finally, this dissertation is dedicated to my beloved family. My husband showed tremendous patience and support for my Ph.D. study. My parents also supported me for my research. I would not have been able to do this without the help and support from my families. 5 Abstract The research examines the relationships between housing and business. It firstly studies whether home-owning hurts employment and identifies the home-owning factors affecting home- owners’ job outcomes. Secondly, it studies whether home-ownership rates affect business development nearby. Thirdly, it measures the effect of the proximity to Walmart and Whole Foods on property values. These relationship is important in how we manage the relationship between housing and business development regarding the match and distance among different kinds of neighborhoods and business types. 6 Table of Contents Acknowledgements ....................................................................................................................................... 3 Abstract ......................................................................................................................................................... 5 Chapter 1. Introduction ............................................................................................................................... 11 Chapter 2. Housing Tenure and Unemployment ........................................................................................ 16 1. Introduction........................................................................................................................................ 16 2. Mechanisms ....................................................................................................................................... 19 3. Data, methodology and results........................................................................................................... 23 3.1. Macro relationship between employment growth and home-ownership rates ............................. 23 3.2. Employment outcome differences for home-owners and renters ................................................. 26 4. Different types of home-owners ........................................................................................................ 36 4.1. Unemployment probabilities ........................................................................................................ 38 4.2. Employment spell......................................................................................................................... 42 4.3. Mobility ........................................................................................................................................ 46 5. Conclusion ......................................................................................................................................... 50 6. References.......................................................................................................................................... 53 Chapter 3. The Effect of Home-owning on Business Development: Micro-level Evidence ...................... 58 1. Introduction........................................................................................................................................ 58 2. Theoretical framework ....................................................................................................................... 64 2.1. Individual firm location choice and home-ownership rates ......................................................... 65 2.2. Variation of the relationship with distance, income and industry types....................................... 67 3. Methodology ...................................................................................................................................... 72 4. Data .................................................................................................................................................... 74 5. Result ................................................................................................................................................. 83 5.1. The pooled sample results ............................................................................................................ 83 5.2. Results for neighborhoods of different income levels ................................................................. 88 5.3. The results for different industries ............................................................................................... 95 6. Conclusion ....................................................................................................................................... 106 7. References........................................................................................................................................ 108 8. Appendix.......................................................................................................................................... 111 8.1. K- means clustering method ....................................................................................................... 111 8.2. K-mean clustering mapping ....................................................................................................... 115 8.3. Detailed description of each industry ......................................................................................... 121 Chapter 4. The Effect of Walmart and Whole Foods on Nearby Property Values ................................... 125 1. Introduction..................................................................................................................................... 125 7 2. Methodology .................................................................................................................................... 127 3. Data .................................................................................................................................................. 130 4. Results ............................................................................................................................................. 132 4.1. The difference with the current literature ................................................................................... 135 5. Conclusion ....................................................................................................................................... 137 6. References........................................................................................................................................ 138 Chapter 5. Conclusions ............................................................................................................................. 140 8 List of Figures Figure 3-1: The spatial position of residents and business on Google Maps .............................................. 60 Figure 3-2: The research design for analyzing the home-owning impact on job counts ............................ 62 Figure 3-3: The position of the project and the houses on the periphery .................................................... 62 Figure 3-4: The prediction of the net impact of home-owning on business with respect to distance ......... 69 Figure 3-5: The prediction of the net impact of home-owning on business with respect to distance for different income levels ................................................................................................................................ 71 Figure 3-6: The illustration for the geographical relationship of the donut rings of different radii ............ 78 Figure 4-1: The effect of Walmart on nearby house prices: ..................................................................... 133 Figure 4-2: The effect of Whole Foods on nearby house prices: .............................................................. 134 9 List of Tables Table 2-1: Literature summary ................................................................................................................... 22 Table 2-2: OLS result on the relationship of employment growth rates and homeownership rates ........... 25 Table 2-3: GMM result on the relationship of employment growth rates and homeownership rates ......... 25 Table 2-4: Micro-summary statistics for PSID ........................................................................................... 27 Table 2-5: SIPP summary statistics ............................................................................................................ 28 Table 2-6: Tenure choice effect on employment outcomes: Part 1............................................................. 31 Table 2-6: Tenure choice effect on employment outcomes: Part 2............................................................. 32 Table 2-7: Tenure choice effect on mobility: Part 1 ................................................................................... 35 Table 2-7: Tenure choice effect on mobility: Part 2 ................................................................................... 35 Table 2-8: Mortgage’s effect on owners' unemployment probability ......................................................... 38 Table 2-9: Negative equity’s effect on owners' unemployment probability ............................................... 39 Table 2-10: The effects of wealth and attachment to community on unemployment probability for owners .................................................................................................................................................................... 41 Table 2-11: Mortgage’s effect on owners’ employment spell length ......................................................... 43 Table 2-12: Mortgage’s effect on owners’ unemployment spell length ..................................................... 43 Table 2-13: Negative equity’s effect on owners' annual employment and unemployment weeks ............. 44 Table 2-14: The effects of wealth and attachment to community on owners' annual employment weeks . 45 Table 2-15: The effects of wealth and attachment to community on owners' annual unemployment weeks .................................................................................................................................................................... 46 Table 2-16: Mortgage’s effect on owners’ mobility ................................................................................... 47 Table 2-17: Negative equity’s effect on owners’ mobility ......................................................................... 48 Table 2-18: The effects of wealth and attachment to community on owners' mobility .............................. 49 Table 2-19: Mortgage’s effect on owner’s mobility with SIPP data........................................................... 50 Table 2-20: Result summary ....................................................................................................................... 51 Table 3-1: Summary statistics for resident characteristics .......................................................................... 79 Table 3-2: Summary statistics for job characteristics ................................................................................. 81 Table 3-2.1: Job counts ............................................................................................................................... 81 Table 3-2.2: Job count change .................................................................................................................... 81 Table 3-2.3: Other job characteristics ......................................................................................................... 82 Table 3-3: Results on the cluster level of within .3 miles ........................................................................... 84 Table 3-4: Results for different radii ........................................................................................................... 86 Table 3-5: The instrument variable results ................................................................................................. 87 Table 3-6: The fixed effect results for the lower income group.................................................................. 90 Table 3-7: The fixed effect results for the higher income group ................................................................ 91 10 Table 3-8: The fixed effect results with lagged homeownership rates as the independent variable for the lower income group .................................................................................................................................... 93 Table 3-9: The fixed effect results with lagged homeownership rates as the independent variable for the higher income group ................................................................................................................................... 94 Table 3-10: Result summary for the pooled sample and samples of different incomes ............................. 95 Table 3-11: Descriptive statistics for the industry job counts at the cluster level ....................................... 96 Table 3-12: The block group photo with maximum job counts of each industry ....................................... 97 Table 3-13: The industry result for the lower income group: the positive effect part .............................. 100 Table 3-14: The industry result for the lower income group: the negative effect part.............................. 101 Table 3-15: The industry result for the higher income group: the positive effect part ............................. 102 Table 3-16: The industry result for the higher income group: the negative effect part ............................ 103 Table 3-17: The result summary for the industry...................................................................................... 105 Table 4-1: Sample statistics for the data ................................................................................................... 131 Table 4-2: The effect of Walmart in CA using the method in current literatures ..................................... 136 11 Chapter 1. Introduction My dissertation studies the interactions between housing and business. This interaction involves two directions: firstly, how housing market affects business? We address this issue through the perspective of home-owning. Specifically, my first essay is on whether homeowning negatively impacts employment. Homeowning reduces mobility. Reduced mobility constrains job search, and job search affects employment. My second essay is on whether areas with more homeowners are prone to oppose business development. If owners view the business projects as nuisance, they might want to oppose these projects built near their neighborhood. If homeowning indeed negatively affects employment and business, we should reconsider policies promoting homeowning, such as the preferential tax treatment of owning. The second direction of this interaction involves the business impact on housing markets. One reason owners might oppose business development is a perception that nuisance arising from business depresses home values. In the third essay, I test whether grocery stores, like Walmart and Whole Foods, do in fact lead to home value decrease. If there are negative impacts from homeowning on business and housing prices are lower nearby business, housing and business should not be mixed together. Then we might support the Euclidean zoning (the zoning that separates residential and business land use). In particular, Chapter 2 (the first essay) points out that home-owning might add frictions on home-owners’ job market search through higher transaction costs associated with the house sell or home-owners might be locked-in with negative equity. We examine the home-owning impact on personal job outcomes, like the unemployment probability, employment spell length, mobility and wages. We further investigate which home-owning features, such as the existence of 12 mortgages, negative equity, housing improvement and attachment to community, can explain home-owners’ job outcomes. Former literatures either use IV or simultaneous methods to deal with the endogeneity that home-owners might be intrinsically more successful in their jobs. We combine the individual fixed effects with the Hausman test and an IV method with whether the first two children are of the same sex to remove the endogeneity. Using Current Population Survey panel data from 1988 to 2013, Panel Study of Income Dynamics data from 1994 to 2013 and Survey of Income and Program Participation data from 2004 to 2013, the results are that job performances of home-owners are not inferior, with no higher risk of being unemployed or lower wages. However, home-owners are indeed less mobile and stay longer in employment. Therefore, the frictions generate immobility and longer employment spell, but do not harm personal job outcomes. For the home-owning features, the existence of mortgage prolongs the employment duration. The result contradicts the posit that home-ownership is detrimental for personal job performances but rather suggests that home-owning might benefit the economy by providing more stable labor pools. Chapter 3 (the second essay) examines the home-owning spillover effects on business. The positive spillover is that home-owners are more stable employees and they build pleasant amenities to protect their property values, which lead to favorable business developments. The negative spillover stems from the home-owners’ obstruction on business projects due to business nuisance, like noise, traffic or pollution, known as the NIMBY effect. Whether the positive effect outweighs the negative one depends on the relative strength of the two opposing forces. Three factors might contribute to the relative strength: the distance between the home-owners and business, the income of the home-owners, and the industry types of the business. The understanding of the relationship between home-owning and business development is essential for zoning policies to achieve social 13 optimal outcomes. The prediction is that the negative impacts occur at adjacent distances for the high income communties. There are two challenges to this analysis. The first is the research design, within which the spatial effect of home-owning on business with distance can be examined. The geomapping of business centers reveal that most business centers are surrounded by residents. I thus employ a K means clustering method from machine learning, which clusters the business block groups that are geographically close. Then I use K nearest neighborhood method to draw residential donut rings of different radii to the job cluster to study the home-owning impact. The second challenge is the endogeneity that the home-ownership rate change might be a result of the job count change. I incorporate multiple identification strategies. Firstly, I perform the fixed efffects (FE) for the panel data. But FE can’t control for reverse causation. Thus, secondly, I perform an Instrument Variable (IV) estimation with the Federal Housing Administration (FHA) loan limit divided by the median house price as the instrument for within .3 miles and the ratio of families with childrn under 18 as the instrument for 3-5 miles to test the endogeneity of the home-ownership rates. We also conduct analyses using the lagged home-ownership rates as the independent variables since the home- owning impact might take time to accumulate. Using the data of American Community Survey (ACS) five-year estimates and Longitudinal Employer-Household Dynamics (LEHD) at the block group level from 2009 to 2014, the empirical result identifies the adjacent distance of within .3 miles as a negative spillover point and the close disance of 3-5 miles as a positive one. Lower income groups exert positive spillovers for within .3 and 3-5 miles while higher income groups genertate negative impacts for .3-1 miles, suggesting that the positive spillover outweighs the negative one for lower income neighborhoods. Different industries also match with different neighborhoods. Service industries like Retail, Art 14 and Professional Services are found with positive home-owning impacts in higher income groups and job counts of Manufacturing, Real Estate and Car Rental and Leasing are positively correlated with home-owning in lower income groups. Chapter 4 (the third essay) investigates the impact of two grocery stores, Walmart and Whole Foods, on nearby property values. Potential home-owners might identify safe housing investment by the existence of quality retail in the neighborhood, like Whole Foods or Star Bucks. However, they might refrain from house purchase with a Walmart in proximity. I thus test the differential effects of Whole Foods and Walmart on nearby house prices. I employ an improved difference in difference method which differs from the traditional one. The traditional difference in difference method is essentially a fixed effect model controlling for treatment and time fixed effects and comparing the average treatment effects of the control and treatment groups. However, the individual observations can be very different in controls and treatments. This doesn’t meet the premise of the dif-in-dif method that the treatments and controls need to be similar. The optimal realization of the dif-in-dif method is to break the sample into multiple small groups within which the treatments and controls are similar. This improved dif-in-dif method realizes this by forming one group of controls and treatments for each property with a transaction record. The controls and treatments are selected within a small local area from the property to ensure their similarity. The controls include the properties farther to the store than the property under study and the treatments include the properties closer to the store than the property under study. The effect of the store on each house for a specific distance is then calculated using the difference of the property prices in the treatments and controls. Then the individual effects are smoothed with a nonparametric transformation, yielding a continuous function of the effects with respect to distances. 15 One drawback of this method, however, is that it posits that the geographically close properties are similar, which might not be true. The local proximity might ensure the similarity in land values and neighborhood features but the subject under study is the house prices. Therefore, I fix this by accounting for the housing heterogeneity through incorporating the housing feature differences into the calculation of the effects. The result shows that both the effects of Walmart and Whole Foods on house prices are positive, with a larger impact from Whole Foods. This supports the homeowners’ investment philosophies that favor the existence of Whole Foods but challenges the attitudes towards Walmart. Using the house transaction data from Data Quick and the store opening time and location of Whole Foods and Walmart collected online. The empirical results are as follows: Walmart exerts a net negative effect on house prices with a decrease of between 15% to 30%. The largest impact happens at the distance of around 2 kilometers (1.25 miles) for the first three years after its opening. Whole Foods produces a positive impact between 15% to 0% increase with the largest impact at the distance around 1.5 (.94 miles) kilometers for the first three years. Chapter 5 concludes the dissertation with summaries of the findings and implications for policies. 16 Chapter 2. Housing Tenure and Unemployment 1. Introduction Labor markets resemble those predicted by Adam Smith’s invisible hand. But the resemblance is less than perfect, however, because of a number of market imperfections (Rocheteau, Rupert, and Wright, 2007). The larger these imperfections are, the more deadweight loss the economy suffers. To the extent policymakers can recognize a channel for an imperfection, the better is the chance that they can address it. This article focuses on one imperfection in particular—search costs—and a particular channel contributing to these costs: home-owning. As such, it follows a series of papers, beginning with Oswald (1996), that look at the link between home-owning and unemployment, but it also makes novel contributions not appearing in previous work. Oswald (1996, 1997, and 1999) kicked off the debate about home-owning and employment with a controversial paper that found positive correlations between unemployment and lagged home-ownership across the 50 states in the U.S. and countries in the EU. This paper, while provocative, had little in the way of controls; it was particularly lacking in a mechanism for dealing with the fact that people who chose to be immobile would find home-ownership financially more rewarding, because they were more likely to be able to amortize the large transaction costs involved in buying a house. In a later paper with fixed effects, however, Blanchflower and Oswald (2013) continued to find that home-owning leads to unemployment. The connection between home-owning and employment is straightforward: if the fact of being a home-owner reduces mobility, home-owners will be at a disadvantage in the face of unexpected labor market shocks. Home-owners do face costs when disposing of their homes— 17 particularly when their houses are worth less than their mortgage balances-and so may face a labor market friction not faced by renters. This article firstly compares results between estimated macro-models of labor markets, using states as units of observation, with micro-models, using households from the Panel Study of Income Dynamics (PSID) and Survey of Income and Program Participation (SIPP) as units of observation. Deaton and Muellbauer (1980) showed that macro models based on aggregate data, because of their restrictions on household heterogeneity, will tend to have biased coefficients. We shall see that macro and micro models produce very different predictions about the impact of home-owning on unemployment. Third, it exploits panel data to try to control for the unobservables associated with home- owning. Home-owners are no doubt different in unobservable dimensions. For example, households choose home-owning knowing that the fixed costs of owning are high, and need time to be amortized. Hence when households become home-owners, they are identifying themselves as less mobile people—the home-owning does not cause the mobility, but is rather a product of it. We employ individual level fixed effects (FE) to deal with the unobservables associated with individuals. FE can remove the time-invariant unobservables across the individuals. However, the existence of time-variant unobservables can create identification problems. Another method commonly used in the literature is the IV method which addresses the endogeneity through creating exogenous variations. We follow the literatures and try the popular instruments: the state level home-ownership rates (van Leuvensteijn and Koning, 2004; Munch et al., 2006), the state level tax rates (Coulson and Fisher, 2009; Yang, 2015) and whether the first two children are off the same sex (Coulson and Fisher, 2009). But problems lie in the fact that firstly, IV result validity 18 depends heavily on the choice of instruments and secondly, Young (2017) finds that IV results suffer from being falsely significant and sensitive to outliers than OLS. Specifically, the first two lead to significant results but state level variables are not ideal instruments for individual level variations (Coulson and Fisher, 2009). In particular, the state level home-ownership rates are correlated with unemployment rates thus not uncorrelated with individual unemployment status. As for the state level tax rates, they might affect individuals’ working spirits. Therefore, their results suffer from reliability issues. The children instrument is the individual level instrument but is weak in our analysis with inflated and insignificant results. Thus, we eventually employ the FE model and check whether the FE successfully removed the endogeneity using Hausman test. These sets of investigations show at most a weak relationship between owning and labor market frictions. But having established that home-owning per se probably does little to influence labor markets dynamics, we then to see whether there are particular features of home-owning that affect the labor market. These include whether a home-owner has a mortgage, whether a home- owner has an underwater mortgage, wealth levels and measures of attachment to community. We find that with two exceptions, none of these has a discernible impact on whether one is unemployed or not, or whether one leaves employment for unemployment. The exceptions are that people who have a mortgage before the crisis are less likely to leave employment for unemployment and wealth accumulation predicts lower unemployment probability and shorter working hours. The remainder of this article contains four sections. In section II, we conjecture about mechanisms through which home-ownership might affect the labor market. In section III, we present a series of tests. First, we change the specification of the left-hand side variable from unemployment to employment growth, for tests using aggregate data. Healthy labor markets can have higher unemployment rates because of churning—in particular, the natural unemployment 19 rate can be higher in dynamic job markets, as people more frequently move from job to job in order to find better matches. Employment growth captures labor market dynamism in a way that the unemployment rate does not. Secondly, we do a series of tests using household level data. The tests examine differences between home-owners’ and renters’ unemployment probabilities, unemployment spells, employment spells, mobility for jobs and interstate moving probabilities. Thirdly, we study the effects of specific characteristics of home-owning, including presence of mortgage, negative equity, wealth accumulation and attachment to community, on employment outcomes. Section IV concludes the paper. 2. Mechanisms Aspects of search theory predict that home-owners might be at a disadvantage in the labor market; other aspects predict that home-owners might be at an advantage. Home-owners might find labor market search inhibited by high moving costs, high selling costs, and attachment to community. They also may live in communities with more stringent zoning, which in turn reduces employment opportunities. On the other hand, home-owning might help workers attach to good jobs. For example, Coulson and Fisher (2002) showed that employers like locating in cities where labor is less mobile, because the pool of available labor is more reliable. Hence, the relative immobility of home-owners might produce better job market outcomes. Home-owners are also wealthier than renters, in part because of wealth accumulated through home-ownership, and therefore might, in the presence of unemployment, have the resources to take the time to search for a good job match with higher wages (Green and Hendershott, 2001; Bloemen and Stancanelli, 2001). This means home-owners 20 might have longer unemployment spells (Goss and Phillips, 1997), but also get better jobs. Home- owners might stay longer and work harder in their job positions to meet burdens of mortgage payments (Goss and Phillips, 1997; Flatau, Forbes, Hendershott and Wood, 2003). They might also search harder for local jobs and have lower reservation wages for local jobs to avoid moving (Goss and Phillips, 1997; Munch, Rosholm and Svarer, 2006). On the other hand, firms are willing to pay more to workers to encourage long term employment (Farber, 1999), thereby avoiding the cost of turnover. Clearly, theoretical predictions about how tenure choice affects labor market outcomes are (1) dependent on the specific outcome being measured and (2) ambiguous. Blanchflower and Oswald (2013) posit that home-ownership creates negative externalities for the labor market. The reason is that home-owners tend to block the construction of the facilities that produce nuisances to local communities, but are beneficial for the broader economy. Home- owners also tend to engage in fiscal zoning, blocking out low-to-moderate income people on the grounds that they cost more in government service than they pay in taxes (Fischel, 2004). This behavior, however, prevents employers from having access to the full range of labor they need. Coulson and Fisher (2009) found that the unemployment probability is increased for both home- owners and renters when the regional home-ownership rate increases. The issue of the impact of home-ownership on labor mobility has become particularly germane in the aftermath of the financial crisis. A debate—related to but separate from the one discussed here—has developed regarding whether negative equity home-owners are more or less likely to move for job opportunities (Stein, 1995; Genesove and Mayer, 2001; Chan, 2001; Engelhardt, 2003). Moreover, during the aftermath of the financial crisis, some scholars hypothesized that some underwater home-owners were locked into their homes and were therefore unable to move to job opportunities in other places. Recent empirical evidence, however, is mixed 21 about whether underwater borrowers are more or less likely to move for a job (Ferreira, Gyourko, and Tracy, 2010; Schulhofer-Wohl, 2011; Mumford, 2013; Valletta, 2013; Coulson and Grieco, 2012). Table 2-1 provides a detailed literature summary. 22 Table 2-1: Literature summary Author Aggregate (home-ownership rates and unemployment rates) Less likely to be unemployed More likely to be unemployed Shorter unemployment spells Longer unemployment spells Higher wages Lower wages Externality Endogeneity correction method Country Data Oswald 1996 + no correction US and EU Goss and Phillips 1997 yes Heckman two step US 1986 PSID Nickell 1998 + no correction OECD countries Green 2000 + for the middle aged no correction US Statistical abstract of the US Green 2001 yes Heckman two step US PSID (1986- 1992) Coulson and Fisher 2001 yes yes yes no correction US CPS March and PSID (1992 wave) Coulson and Fisher 2009 - yes yes yes IV US 1990 CPS Supplement Leuvensteijn 2004 yes simultaneous equations Netherland Munch 2006 yes no simultaneous equations Denmark individual level Munch 2007 yes yes simultaneous equations Denmark individual level Garcia 2002 - simultaneous equations Spain individual level Laamanen 2013 + on unemployment spells (micro- level) yes yes (through job mismatch and crowding out consumptions) natural experiment with IV Finland individual level 23 3. Data, methodology and results Our empirical results are divided into three parts. First, we replace the unemployment rate with employment growth rates as the dependent variable in a macro empirical model to see whether home-owning affects employment growth in aggregate. We use the Current Population Survey Merged Outgoing Rotation Groups (CPS MORG) data for these estimates. Second, we estimate employment outcome differences of owners and renters using micro data, specifically the Panel Study of Income Dynamics (PSID) and Survey of Income and Program Participation (SIPP). Third, we study whether there are specific aspects of home-owning--presence of a mortgage, negative equity, attachment to community, and amount of equity--that contribute to any overall difference. 3.1. Macro relationship between employment growth and home-ownership rates Blanchflower and Oswald (2013) estimated that home-owning has a negative effect on employment and a positive effect on unemployment. We estimate a similar regression, except that instead of performing estimates in levels, we estimate the impact of owning on changes in employment, i.e., the employment growth. The right hand side variables include the lagged one period dependent variable and lagged one to lagged five period log home-ownership rates and other controls. We estimate the following model: t i t i k t i t i t i C HR EMGRTH EMGRTH , , 1 , 1 1 , 1 , ) log(           (1) We look at the relationship for 50 states and District of Columbia from 1988 to 2013. EMGRTHi,t is the employment growth rate in state i in period t. log(HRi,t-k) is the log home- ownership rates in state i in period t-1 through t-5. Ci,t is a vector of control variables in state i in period t, including state home-ownership rates, age dummies, sex, education, marriage, race, and 24 time and state fixed effects. The data used for the aggregate model are the annual means for each state of each period. We have 15 education dummies and 2 race dummies. The education dummies are from less than 1 st grade to doctorate degree 1 . The race dummies are white, black and others. The control variables are from the CPS MORG data. The employment data are from the U.S. Bureau of Labor Statistics. The home-ownership rates are from the U.S. Bureau of Census. All data are downloaded from Federal Reserve Economic Data- FRED - Federal Reserve Bank of St. Louis. Because we have lagged dependent variables as right-hand side variables, we use both OLS and GMM (Arellano-Bond estimation) to estimate equation (1). The results are in Tables 2-2 and 2-3. Of the 20 coefficients on the impact of ownership on employment growth, only two are different from zero at the 95 percent level of confidence. The probability of obtaining two significant coefficients out of 20 is 82 percent. These results are in contrast with Blanchflower and Oswald (2013)’s, which looks at the impact of tenure on levels in employment in their Table 6 2 . Our view is that the evidence of the impact of tenure on employment growth is pretty murky. In any event, Deaton and Muellbauer (1980) discussed problems of aggregation bias in macro-level regressions, so we now move on to household level data to look for evidence of the influence of housing tenure on employment outcomes. 1 The categories for the education dummies are: less than 1 st grade, 1 st -4 th grade, 5 th or 6 th , 7 th or 8 th , 9 th , 10 th , 11 th , 12 th grade NO DIPLOMA, high school graduate or diploma or GED, some college but no degree, associate degree-occupational/vocational, associate degree-academic program, bachelor’s degree, master’s degree, professional school, doctorate degree. This categorization is only for the data after 1992. For the data before 1992, we matched another education variable in CPS MORG with this categorization. 2 In their paper, the coefficients are significant. They use an OLS specification. 25 Table 2-2: OLS result on the relationship of employment growth rates and homeownership rates VARIABLES Employment growth: OLS 1988-2013 1984-2013 1985-2013 1986-2013 1987-2013 1988-2013 Emgrowthlag1 .632*** .661*** .657*** .640*** .588*** .631*** (.0216) (.0196) (.0199) (.0192) (.0199) (.0216) Loghomelag1 -.016 -.0300** (.0152) (.0130) Loghomelag2 .0121 -.0082 (.0198) (.0101) Loghomelag3 -.00862 .000284 (.0198) (.0096) Loghomelag4 .014 -.00699 (.0192) (.0095) Loghomelag5 -.0169 -.0102 (.0142) (.0095) Average effect -.00308 - - - - - Year dummies 24 28 27 26 25 24 State dummies 50 50 50 50 50 50 Other controls 20 20 20 20 20 20 Observations 1,275 1,479 1,428 1,377 1,326 1,275 R-squared .834 .802 .805 .825 .832 .834 Note: The dependent variable is the state level employment growth rate and the independent variables of interest are the lagged log of the state level homeownership rates. Other controls include fifteen educational dummies, one sex dummy, two race dummies, one marriage dummy and age dummy. Standard errors are clustered at the state level and are in parentheses. Significance level: *** p<0.01, ** p<0.05, * p<0.1 Table 2-3: GMM result on the relationship of employment growth rates and homeownership rates VARIABLES Employment growth: GMM 1988-2013 1984-2013 1985-2013 1986-2013 1987-2013 1988-2013 Emgrowthlag1 .555*** .596*** .597*** .585*** .517*** .555*** (.0364) (.0254) (.0246) (.0235) (.0383) (.0362) Loghomelag1 -.00818 -.0463** (.0220) (.0206) Loghomelag2 .0168 .00501 (.0237) (.0170) Loghomelag3 -.0113 .0193 (.0149) (.0177) Loghomelag4 .0162 .00919 (.0183) (.0159) Loghomelag5 -.00813 .000115 (.0184) (.0138) Average effect .00108 Year dummies 24 28 27 26 25 24 State dummies 50 50 50 50 50 50 Other controls 20 20 20 20 20 20 Observations 1,224 1,428 1,377 1,326 1,275 1,224 Note: The dependent variable is the state level employment growth rates and the independent variables of interest are the lagged log of the state level homeownership rates. Other controls include fifteen educational dummies, one sex dummy, two race dummies, one marriage dummy and age dummy. Standard errors are clustered at robust and are in parentheses. Significance level: *** p<0.01, ** p<0.05, * p<0.1 26 3.2. Employment outcome differences for home-owners and renters We next explore the effect of home-owning at the individual level. We use panel data to control for individual fixed effects and study five measures of employment outcomes: unemployment probabilities, unemployment spell length, employment spell length, wages and mobility. We use the annual PSID 3 data from 1994 to 2013 and the monthly SIPP data from 2008 to 2013. The PSID data are used to estimate the impact of tenure on unemployment probability, wages, and mobility. The SIPP data are superior for following spell lengths, because survey respondents are questioned annually; PSID respondents are surveyed biannually. The summary statistics for the PSID are in Table 2-4; the summary statistics for the SIPP are in Table 2-5. We have four different categories for the variables in Table 2-4. The first gives national averages based on census data. The remaining columns present summary statistics for different categories of heads surveyed for the PSID. Data in the first column are from 2012; data in the others are averages of the sample period of 1994 to 2011. 3 Only the heads are included in the analysis. 27 Table 2-4: Micro-summary statistics for PSID Variables (1) Census (2) Heads of both owners and renters (3) Heads of renters (4) Heads of owners Dependent variables: Unemployment rate 8.20% 6.58% 10.76% 4.25% Move 0 Not move - 74.54% 47.45% 87.29% 1 Job - 2.77% 5.80% 1.31% 2 Housing consumption - 13.67% 28.91% 6.51% 3 Involuntary move - 9.03% 17.84% 4.89% Annual median income 27519 35360 26000 43001 Independent variables Homeowners 65.40% 66.80% - - State home-ownership rates -65% 20.00% 25.70% 29.48% 26.36% 65%-70% 44.00% 25.28% 25.09% 23.96% 70%- 36.00% 49.02% 45.43% 49.69% Race White 72.41% 81.29% 70.09% 87.55% Black 12.61% 14.06% 24.05% 8.49% Asian 4.75% 1.85% 1.95% 1.79% Latino 16.40% 2.80% 3.91% 2.17% Married 58.95% 58.80% 38.57% 70.11% Male 49.20% 74.38% 60.33% 82.26% City size Large city (population>.25 m) - 67.33% 71.14% 65.22% Medium city (.02m-.25 m) - 15.43% 15.56% 15.37% Small city (<.02 m) - 17.23% 13.30% 19.41% Education High-school and lower 58.41% 70.12% 78.47% 65.50% Bachelor 18.88% 22.31% 17.26% 25.09% Master and above 10.38% 7.57% 4.26% 9.41% Age 18-25 9.9% 9.27% 21.08% 2.67% 25-35 22.16% 24.87% 35.27% 19.02% 35-45 22.16% 27.90% 23.66% 30.26% 45-55 24.29% 25.83% 14.81% 32.02% 55-65 19.69% 12.13% 5.18% 16.03% Homeowner attributes |Negative equity|>0 - - - 2.32% Has mortgages 65% - - 82.82% Note: The table reports the calculations of the PSID data based on weights. Data period is from 1994 to 2011 for every two years, except that for the “move” variable, which is from 2001 to 2011. The first column (1) is from the 2012 census data. Note from the summary statistics that the unemployment rates are lower for owner heads than their counterpart renters. This likely reflects omitted variables. Renters have higher mobility rates than owners, again perhaps reflecting omitted variables. Renters also have lower wages. 28 Renter household heads are far less likely to be married than owners, and are more likely to live in apartments and in large cities. Table 2-5: SIPP summary statistics Variables (1) Census (2) Whole Sample (3) Owners (4) Renters Dependent variables: Unemployment rate 8.20% 8.08% 7.35% 9.81% Unemployment spell (months) 1.76 1.61 2.12 Employment spell (months) 37.28 38.74 33.86 Move 1 Non-mover - 97.54% 98.98% 94.14% 2 .Moved, same county - 1.64% 0.67% 3.91% 3 .Moved, different county within same state - 0.49% 0.23% 1.12% 4 .Moved, different state - 0.33% 0.12% 0.83% Independent variables Homeowners 65.40% 70.25% - - State homeownership rates -65% 20.00% 26.86% 24.08% 33.43% 65%-70% 44.00% 42.80% 43.50% 41.16% 70%- 36.00% 30.34% 32.43% 25.42% Race White 72.41% 81.43% 85.18% 72.56% Black 12.61% 11.65% 8.54% 18.98% Asian 4.75% 4.01% 3.78% 4.54% Latino 16.40% 2.92% 2.49% 3.93% Married 58.95% 54.17% 61.76% 36.24% Male 49.20% 51.06% 50.76% 51.78% City size metro - 84.39% 83.27% 87.04% Education High-school and lower 58.41% 68.11% 64.73% 76.09% Bachelor 18.88% 21.00% 22.75% 16.86% Master and above 10.38% 10.89% 12.52% 7.05% Age 18-25 9.90% 14.50% 12.70% 14.50% 25-35 22.16% 22.70% 18.08% 22.70% 35-45 22.16% 23.31% 23.53% 23.31% 45-55 24.29% 24.16% 27.22% 24.16% 55-65 19.69% 15.33% 18.47% 15.33% Note: The table reports the calculations of the SIPP data based on weights. Data is the 2008 panel, covering 2008 to 2013. The first column (1) is from the 2012 census data. Testing the effect of tenure on individual employment outcomes has an endogeneity problem. Owners may have unobservables that lead to more success in job performance. The 29 people who are more able, more inclined to settle down, and have more permanent income, are also more likely to be home-owners. At the same time, more able people may be better at their jobs; and those who want to settle down may work harder to keep their local jobs or search harder for local jobs. This is particularly important in light of the fact that mortgage underwriting depends in part on job history (Goss and Phillips, 1997). As for mobility, those households that are more likely to move are also more likely to decide to be renters (Haurin and Gill, 2002). Per our former discussion, to address the endogeneity, we estimate equations with a fixed effect specification to exploit the panel data, and complement this with a Durbin–Wu–Hausman test on endogeneity to check whether the fixed effects purged the effects of unobservables. We also present results on unemployment probability and wages with Instrumental Variables (IV) method. The instrument we use considers whether the first two children of a household are of the same sex 4 . This instrument is used by Coulson and Fisher (2009) in their explorations on the relationship between tenure choice 5 and job outcomes. They argue that households with two children of the same sex are more likely to want another child, and therefore may be more prone to expand into a larger dwelling via home-ownership. At the same time, sex of the first two children is exogenous with respect to employment. We estimate the following models: t i j t i j t i j t i j r C OWN UE P , , 1 , , 1 1 , , 1 , , ) (        (2) ) exp( ) ( ) ( , , 2 , , 2 1 , , 2 0 , t i j t i j t i j i j C OWN t h t h        (3) 4 We only present the IV results with whether the first two children are of the same sex as the instrument because it is an individual level instrument. The IV results using the state level home-ownership rates and state level marginal tax rate will be provided upon request. 5 First stage regression shows significant results. 30 t i j t i j t i j t i j r C OWN MOVE P , , 3 , , 3 1 , , 3 , , ) (        (4) t i j t i j t i j t i j C OWN Wage , , 4 , , 4 1 , , 4 , ,        (5) Pr (UEj,i,t) is the probability of unemployment for individual j in state i in period t. hj,i (t) is the hazard rate for individual j in state i in period t. We consider both the hazard rate from unemployment to employment and the hazard rate from employment to unemployment. Pr(MOVEj,i,t) is the probability of move for individual j in state i in period t. OWNj,i,t-1 is the dummy of whether the individual j in state i in period t-1 is a home-owner or not. Cj,i,t is a set of control variables for individual j in state i in period t, including state home-ownership rates, marriage, number of children, sex, race, education, the “wife’s” wage larger than zero dummy, age, state and time fixed effect dummies. We are only able to get convergence on our IV models for equations (2) and (5), so we report uncorrected results, IV results, and panel results. Columns 1 and 2 in Table 2-6 summarize the unemployment probability results. We find under a simple logit that home-owners are less likely to be unemployed. After the fixed-effect correction, we find that the coefficient is still negative. The Hausman test suggests that the fixed effects remove the endogeneity. The IV result is not significant with large standard errors. 31 Table 2-6: Tenure choice effect on employment outcomes: Part 1 Unemployment probability Income (1)Logit (2) LOGIT_FE (3)IV_PROBIT (4)OLS (5)OLS_FE (6)FE_IV Homeownership -.033*** -.123*** .275 3.971*** 1.323*** 6.022 (.001) (.045) (5.486) (.117) (.155) (11.793) Controls Y Y Y Y Y Y Time 20 20 20 20 20 20 State 50 50 50 50 50 50 Observation 173,067 45,640 44,723 163,853 163,853 42,118 Note: We use the annual PSID data from 1994 to 2011 for the unemployment probability and income investigation. The coefficients are marginal effects for the unemployment probability in Columns (1), (2) and (3). (4), (5) and (6) are the income results. (2) and (5) control for individual fixed effects. (3) and (6) use the instrument of whether the first two children are of the same sex. Controls include state ownership rates, marriage, children number, sex, race, education, age, time and state fixed effects. Standard errors are clustered at the individual level, and they are in parentheses. Significance level: *** p<0.01, ** p<0.05, * p<0.1 We use the SIPP to estimate a hazard rate for leaving unemployment, following people who were unemployed at some period between May 2008 through July 2013. For some individuals, we have multiple unemployment spells within the time period of the data. We perform parametric Weibull regressions and adjust for both left truncation and right censoring problems. We also perform OLS regressions, where unemployment spell length is the left hand side variable. This will allow us to use fixed effects to test—and possible purge—endogeneity. Altogether, we have 1,141,215 observations, made up of 36,029 individuals, of whom 15,592 had an unemployment spell of one month or longer 6 during the study period. The period contained 49,894 unemployment transitions (from employment to unemployment). Many individuals had multiple transitions. For the hazard estimation from unemployment to employment, we also have 1,141,215 observations with 36,029 individuals. Among them there were 49,830 employment transitions (from unemployment to employment). 6 For the definition of being unemployed, we categorize the ones who are not working the full month as unemployed. 32 The results in columns 7-8 in Table 2-6 show that owners are only slightly more likely to get out of unemployment than renters. After taking into account fixed-effects, there is no difference in leaving unemployment between owners and renters. On the other hand, owners seem to be more likely to stay employed, although relative to the proportional hazard and OLS results, the fixed- effects results are attenuated. Overall, our results show that owners are no less likely to get out of unemployment, and slightly more likely to stay employed. Perhaps owners are more stable and committed; perhaps they are disciplined by having mortgages to pay back; or perhaps they are attached to their neighborhoods. Table 2-6: Tenure choice effect on employment outcomes: Part 2 Unemployment spell length Annual number of weeks unemployed Employment spell length Annual number of weeks employed (7) PH (8) OLS (9) OLS_FE (10) OLS_FE (11) PH (12) OLS (13) OLS_FE (14) OLS_FE Homeownership -.009*** -.020*** -.001 -.534*** .744*** 3.718*** .801*** 1.263*** (.003) (.001) (.002) (.087) (.042) (.125) (.179) (.380) Controls Y Y Y Y Y Y Y Y Time 61 62 62 11 61 62 62 2 State 50 50 50 50 50 50 50 50 Observation 1,141,215 1,166,114 1,166,114 98,432 1,141,215 1,166,114 1,166,114 27,389 Note: The monthly SIPP data from May 2008 to July 2013 is used for columns (7)-(9) and (11)-(13). The annual PSID data is used for (10) from 1994 to 2013 and for (14) from 2009 to 2013. (7) and (11) employ the proportional hazard (PH) rate model with Weibull distribution as the baseline hazard distribution. (8) and (12) use the OLS specification. (9), (10), (13), and (14) use the individual fixed effects specification. Controls include state ownership rates, marriage, children number, sex, race, education, age, time and state fixed effects. Negative coefficients mean that homeowners have shorter spell length/working weeks than renters. Standard errors are clustered at the individual level, and they are in parentheses. Significance level: *** p<0.01, ** p<0.05, * p<0.1 It may be the case that while home-owners are no less likely to find new employment than renters, and more likely to remain in their jobs, they may find themselves in worse matches than renters. We turn to household income to investigate job match quality: if owners get worse outcomes in job matching, they will have lower income, all else being equal. 33 Our OLS regression suggests that owners, if anything, make more money than renters, all else equal. But this might again reflect unobservables. When we do the fixed-effect regression, the coefficient on home-owning gets smaller, but remains significant and positive. But when we use a Hausman test on the fixed-effect model, we find that the fixed effects are not sufficient to eliminate correlation between unobservables and home-owning (p < .0001). We therefore attempt an IV estimate of the impact of home-owning on income, using Coulson and Fisher’s (2009) suggested instrument of whether a family has two children of the same gender. Unfortunately, the instrument is weak, and so, predictably, the standard error of the two stage estimate using IV with fixed effects is large. The coefficient on home-owning remains positive (and is larger than the other coefficient estimates), but not different from zero at standard confidence levels. We can say, however, that home-owning does not worsen matches in the labor market. One related work from Yang (2015) found that home-owners’ wage increase is lower than that of renters’ using the mortgage deduction and whether the sex of the first two children are the same as instruments. The major difference of our work with Yang (2015)’s is that the dependent variables are different. Yang (2015) studies the post-unemployment wage increase difference while we study the difference of income including wages for home-owners and renters regardless of whether they are employed or unemployed, post or prior. The results are not necessarily in contradiction since for example, before being unemployed, the wage for a home-owner is 1,000. After unemployment and being reemployed, the wage increases to 1,100. For a renter, before being unemployed, the wage is 800, then after unemployment and being reemployed, the wage increases to 1,000. This scenario fits the results of both papers if we assume the wage level is the same with income. 34 Home-ownership might reduce mobility owing to the high transaction costs associated with purchasing and selling a house. While we have shown that unemployment duration is not elongated by home-owning, testing whether home-owning reduces mobility is in itself a worthwhile exercise. We use the PSID data to test whether owners are less likely to move for work reasons than renters and the SIPP data to test whether they are less likely to move across different states. We use a multinomial logit regression (Table 2-7) to test whether owners are less likely to move for work (Column (1)). It is a multinomial logit, because we want to separate work-related moves from pure housing-consumption related moves (hence the three outcomes for the multinomial logit are stay in place, move for a job, and move for other reasons). We do find in the simplest model that owners are less likely to move for a job than renters. After taking into account fixed effects, the probability of owners moving for work reasons rises a little bit, but still is lower than the probability for renters. A Hausman test confirms that fixed effects solve the endogeneity problem. 35 Table 2-7: Tenure choice effect on mobility: Part 1 Mobility for work (1) MLOGIT (2) MLOGIT_FE (3) OLS_FE (4) LOGIT_FE Homeownership .099*** .119*** -.043*** -1.640*** (.006) (.010) (.002) (.088) Controls Y Y Y Y Time 10 10 10 10 State 50 50 50 50 Obs 109,159 66,628 109,159 9,282 Note: We use the annual PSID data from 1996 to 2013 for this estimation. The dependent variable is the moving for work reasons. Column (1) uses the multinomial regression with the base category as not moving, the of interest category as moving for work reasons and the third category as moving for other reasons like house and neighborhood consumptions or response to outside events. Column (2) controls for individual fixed effects. The coefficients are odds ratios for columns (1) and (2). Larger than one coefficients mean that owners are more likely to move compared to renters. Column (3) uses the linear regression controlling for individual fixed effects, with 1 as moving for work reasons and 0 as not moving plus moving for other reasons. Column (4) uses the logit specification with individual fixed effects. The result is the marginal effect. Controls include state homeownership rates, race, age, children number, marriage, sex, education, individual, state and year fixed effects. Standard errors are clustered at the individual level, and they are in parentheses. Significance level: *** p<0.01, ** p<0.05, * p<0.1 Table 2-7: Tenure choice effect on mobility: Part 2 Mobility across different states (5) MLOGIT (6) OLS_FE (7) LOGIT_FE Homeownership .115*** -.017*** -2.024*** (.009) (.001) (.088) Controls Y Y Y Time 61 61 61 State 50 50 50 Obs 1,067,835 1,067,835 36,855 Note: We use the monthly SIPP data from May 2008 to July 2013 for this estimation. The dependent variable is moving across different states. Column (5) uses the multinomial regression with the base category as not moving, the of interest category as move across different states and the third category as moving within the same county and across different counties. Coefficients are odds ratios. Larger than one coefficients mean that owners are more likely to move compared to renters. Column (6) uses the linear regression controlling for individual fixed effects, with 1 as moving across different states and 0 as not moving plus moving within the same county and across different counties. Column (7) uses the logit specification with individual fixed effects. The result is the marginal effect. Controls include state homeownership rates, race, children number, education, individual, state and year fixed effects. Standard errors are clustered at the individual level, and they are in parentheses. Significance level: *** p<0.01, ** p<0.05, * p<0.1 Column (5) shows that people are less likely to move to other states. The fixed effects model in (6) and (7) continues to predict lower mobility for home-owners. Combining the 36 employment spell results and the mobility results, we find that home-owners may be more committed to their jobs, but they are less likely to move for jobs and across different states. 4. Different types of home-owners So far, we have shown that at the aggregate level, the home-ownership rate has no significant effect on employment growth. At the individual level, owners are less likely to move than renters. Owners’ employment outcomes are not inferior, and owners have longer employment spells. In total, after we control for endogeneity, we have reason to believe that owners have more employment stability than renters but are less willing to move. The question is why. Some possible explanations are: a. Most owners have mortgages, and so need to meet a debt-service payment that motivates them to stay employed (Goss and Phillips, 1997; Flatau, Forbes, Hendershott and Wood, 2003). b. Some home-owners have negative equity. A number of papers have argued that negative equity has an impact on mobility (Stein, 1995; Chan, 2001; Donovan and Schnure, 2011; Blanchflower and Oswald, 2013; Valletta, 2013), but in some of them the impact is positive, and in others it is negative. c. Home-owners accumulate wealth through owning, and this wealth may allow owners to better handle life uncertainties or difficulties. Wealth is found to exert a positive impact on reservation wages and a small negative impact on employment probability (Blundell, Magnac and Meghir, 1997; Bloemen, 1995; Bloemen and Stancanelli, 2001). Homeowners are better able to move and 37 smooth the transition period from the sale of their houses (Green and Hendershott, 2001). d. Owners may have greater stakes in their communities. They are invested in their neighborhoods. They may have emotional attachments to their houses, neighbors and communities. This motivates them to find employment quickly and locally. One way to test this might be to examine the effect of years of being home-owners and the effect of house improvements on employment outcomes. To test those four points, we perform regressions for home-owners only. We estimate the equations below: t i j t i j t i j t i j r C X owner UE P , , 1 , , 1 1 , , 1 , , ) | (        (6) t i j t i j t i j i j C X owner t h , , 2 , , 2 1 , , 2 . ) | (        (7) t i j t i j t i j t i j r C X owner MOVE P , , 3 , , 3 1 , , 3 , , ) | (        (8) X 𝑗 ,𝑖 ,𝑡 −1 = Mortgage 𝑗 ,𝑖 ,𝑡 −1 /Negative Equity 𝑗 ,𝑖 ,𝑡 −1 /Wealth 𝑗 ,𝑖 ,𝑡 −1 /Attachment 𝑗 ,𝑖 ,𝑡 −1 Where j is the individual, i is the state and t is the period. Equation (6) uses the probability of unemployment as the dependent variable. Equations (7) and (8) use the employment hazard rates and mobility probability as their dependent variable respectively. Equations (6), (7) and (8) all use the same independent variables of interest Xj,i,t-1, which include: whether owners have a mortgage, negative equity levels and negative equity dummy, wealth levels with housing and without housing and their dummies, amount spent on house improvements, and years of being home-owners. Controls include marriage, sex, state home-ownership rates, number of children, race, education, age, year and state fixed effects. The explanatory variables of interest are in their 38 lagged one period log values. We run regressions for the whole period, for before the global financial crisis (i.e., before 2007), and during the financial crisis. 4.1. Unemployment probabilities First, we present the unemployment probability results from logit regressions. Table 2-8 shows the effect of presence of a mortgage on owners’ unemployment probabilities. The existence of a mortgage creates an unobservables problem: these who choose to have mortgages may behave differently in the job markets than those that do not, and so we perform fixed-effect regressions. Table 2-8: Mortgage’s effect on owners' unemployment probability Unemployment probability (1) Whole period (2) FE (3) Before 2007 (4) FE (5) After 2007 (6) FE Mortgage dummy -.013*** -.002 -.012*** -.0007 -.020*** -.166 (.001) (.048) (.001) (.056) (.003) (3.836) Controls Y Y Y Y Y Y Time 20 20 17 17 2 3 State 50 50 50 50 50 - Obs 109,442 18,698 92,660 13,511 16,734 1,288 Note: we use the annual PSID data from 1985 to 2013 for this logit estimation. Only owners are included. The dependent variable is the probability of unemployment. The independent variable of interest is the mortgage dummy. Column (1) is for the whole period, covering 1985 to 2013 using the logit specification. Column (2) is the logit with individual fixed effects. Columns (3) and (4) are for the period before 2007—pre-financial crisis. Columns (5) and (6) are for after 2007—post-financial crisis. The results are all marginal effects. Controls include state homeownership rates, age, marriage, number of children, sex, education, race, time and state fixed effects. Standard errors are clustered at the individual level and are in parentheses. Significance level: *** p<0.01, ** p<0.05, * p<0.1 In non-fixed-effect specifications, owners with mortgages are less likely to be unemployed for the whole period, and also for the periods before and during the financial crisis [Columns (1), (3) and (5) of Table 2-8]. But when we add household level fixed-effects, the impact disappears. Perhaps a selection criterion for mortgages via underwriting is driving the likelihood of continuing employment, which is why those with mortgages appear to have better employment outcomes. 39 The explanatory variable negative equity (Table 2-9) also presents identification issues, because debt lovers may have unobservables that make them more likely to be unemployed, 7 so we once again use fixed effects techniques. For most specifications (whole period and pre-2007), we find no evidence that negative equity increases the probability of unemployment. After 2007, however, those who are underwater on their mortgage are more likely to be unemployed. It may be that for the whole period and the pre-2007 period, time and state fixed effects are soaking up differences in negative equity. Data sparseness prevents us from estimating a post-2007 fixed- effect regression with state dummies, which may explain the apparently positive relationship between being an underwater borrower and being unemployed. Table 2-9: Negative equity’s effect on owners' unemployment probability Unemployment probability (1) Whole period (2) FE (3) Before 2007 (4) FE (5) After 2007 (6) FE Negative equity dummy .184* .217 .108 -.100 .208* .454* (.100) (.152) (.191) (.311) (.120) (.245) Controls Y Y Y Y Y Y Time 9 9 6 6 2 2 State 50 50 50 50 50 - Obs 53,948 6,343 37,070 2,580 16,749 1,287 Note: we use the annual PSID data from 1994 to 2013 for this logit estimation. Only owners are included. The dependent variable is the probability of unemployment. The independent variable of interest is the negative equity dummy. Column (1) is for the whole period, covering 1994 to 2013. Column (2) is the logit regression with individual fixed effect. Columns (3) and (4) are for the period before 2007—pre-financial crisis. Column (5) and (6) are for after 2007-—post-financial crisis. The results are all marginal effects. Controls include state homeownership rates, age, marriage, number of children, sex, education, population, race, time and state fixed effects. Standard errors are clustered at the individual level and are in parentheses. Significance level: *** p<0.01, ** p<0.05, * p<0.1 Table 2-10 shows the effects of the other home-owner characteristics on the probability of unemployment. In all cases here we use individual fixed-effect panel regressions. While house value per se does not contribute to the probability of unemployment, having positive equity has a 7 Donovan and Schnure (2011), Aronson and Davis (2011), and Valletta (2013) using different data find no evidence of a lock in effect. 40 marginally significant—and positive—impact on unemployment, with a ten percent change in home equity increasing the probability of unemployment by about 88 basis points. We should note, though, that this coefficient is significant at only the 90 percent level of confidence. Non-housing wealth, however, reduces the probability of unemployment. 41 Table 2-10: The effects of wealth and attachment to community on unemployment probability for owners Dependent variable: unemployment probability for owners Wealth Log (house value) -.012 (.040) Log (positive equity) .088* (.049) Positive equity dummy -.074 (.142) Wealth1 (excluding housing) -.028 (.033) Wealth1 dummy -.339*** (.102) Wealth2 (including housing) .003 (.040) Wealth2 dummy -.220 (.165) Attachment to community Log (improvement) .458 (.085) Improvement dummy -.006 (.148) Time for being homeowners 5-10 years .128 (.126) 10- years .225** (.104) Controls Y Y Y Y Y Y Y Y Y Y Time 20 9 9 5 9 9 9 9 4 4 State 50 50 50 50 50 50 50 - - - Obs 18,394 5,445 6,109 4,501 6,109 5,569 6,109 111 4,247 1618 Note: we use the annual PSID data from 1985 to 2013 for this logit estimation with individual fixed effects. Only owners are included. The dependent variable is the probability of unemployment. The independent variables of interest are the wealth and attachment to community. The wealth includes log of the house value, log of the positive equity value, positive equity dummy, log of the wealth value without housing and its dummy, and log of the wealth value with housing and its dummy. The dummy is whether the owner has non-negative wealth. The attachment to community includes log of the improvement value that the owner made to their property and its dummy, and the time of being homeowners. Controls include state homeownership rates, age, marriage, children number, sex, education, population, race, state and year fixed effects. Standard errors are in parentheses. Significance level: *** p<0.01, ** p<0.05, * p<0.1 As for attachment to community, we find that it matters, but only for those who have stayed in a community for a long time. Those who have remained in the same house for ten years or more 42 are more likely to be unemployed, reflecting perhaps a stickiness that is not picked up in the unobservables. 4.2. Employment spell We analyze the determinants of employment and unemployment spells similarly to how we analyzed the determinants of unemployment probability. We are attempting to determine why owners have longer employment spells than renters. We are interested in differences in the effects of mortgages, negative equity, wealth and attachment to community before and after the financial crisis. We use the 2004 and 2008 panels of SIPP to represent the periods before and during/after the crisis. Tables 2-11 and 2-12 show the impacts of having a mortgage obligation on employment and unemployment spell lengths. The first two columns in Tables 2-11 and 2-12 show that before the crisis, owners with mortgages were slightly more likely to remain employed, and no more or less likely to remain unemployed, than owners without mortgages. The third and fourth columns in both tables show that after the crisis, owners with mortgages behaved the same as owners without, except in the fixed effect regression for staying employed. The Hausman test of that regression shows, however, that the fixed-effects are not sufficient to purge the impact of unobservables. 43 Table 2-11: Mortgage’s effect on owners’ employment spell length Employment spell length (1) Before 2007 (2) FE (3) After 2007 (4) FE 8 Mortgage dummy .028*** .018*** -.737 .723* (.002) (.004) (.526) (.402) Controls Y Y Y Y Time 37 37 61 61 State 50 50 50 50 Obs 615,827 615,827 875,946 875,946 Note: we use the monthly SIPP data of the 2004 and 2008 panel for this estimation. Only owners are included. The dependent variable is the employment spell length. The independent variable of interest is the mortgage dummy. Column (1) and (2) are for the period before 2007—pre-financial crisis. Column (2) controls for individual fixed effects. Column (3) and (4) are for after 2007—post-financial crisis and (4) controls for individual fixed effects. Controls include state homeownership rates, age, marriage, number of children, sex, education, population, race, state and time fixed effects. Standard errors are clustered at the individual level and are in parentheses. Significance level: *** p<0.01, ** p<0.05, * p<0.1 Table 2-12: Mortgage’s effect on owners’ unemployment spell length Unemployment spell length (1) Before 2007 (2) FE (3) After 2007 (4) FE Mortgage dummy .0005 .0002 -.003 -.003 (.0003) (.0003) (.005) (.008) Controls Y Y Y Y Time 37 37 61 61 State 50 50 50 50 Obs 615,827 615,827 875,946 875,946 Note: we use the monthly SIPP data of the 2004 and 2008 panel for this estimation. Only owners are included. The dependent variable is the unemployment spell length. The independent variable of interest is the mortgage dummy. Column (1) and (2) are for the period before 2007—pre-financial crisis with OLS specifications. Column (2) controls for individual fixed effects. Column (3) and (4) are for after 2007—post-financial crisis and (4) controls for individual fixed effects. Controls include state homeownership rates, age, marriage, number of children, sex, education, population, race, state and time fixed effects. Standard errors are clustered at the individual level and are in parentheses. Significance level: *** p<0.01, ** p<0.05, * p<0.1 The effect of the presence of negative equity on employed and unemployed annual weeks is presented in Table 2-13. We use PSID data from 2009 to 2013. Our dependent variable is number of weeks employed or unemployed. Owners with negative equity are more likely to work longer 8 A Hausman test shows that this result still suffers from endogeneity. 44 than those with positive equity, before taking into account individual fixed effects. Once fixed effects are added, the impact disappears. Negative equity appears to have no impact on unemployed length. Table 2-13: Negative equity’s effect on owners' annual employment and unemployment weeks (1) Number of weeks employed (2) Number of weeks employed_FE (3) Number of weeks unemployed (4) Number of weeks unemployed_FE Negative equity dummy .644* -.158 -.163 .149 (.368) (.405) (.230) (.316) Controls Y Y Y Y State 50 50 50 50 Time 2 2 2 2 Obs 16,882 16,882 16,838 16,838 Note: we use the PSID data from 2009 to 2013 for this estimation. Only owners are included. The dependent variable is the number of weeks of employment and unemployment per year. The independent variable of interest is the negative equity dummy. Columns (2) and (4) control for individual fixed effects. Controls include state homeownership rates, age, marriage, children number, sex, education, population, race, time and state fixed effects. Standard errors are clustered at the individual level and are in parentheses. Significance level: *** p<0.01, ** p<0.05, * p<0.1 Table 2-14 shows the other variables’ effects, based on 2009 to 2013 data. Positive equity and wealth including housing predict shorter employment length. More housing improvements also predict fewer works employed per year. None of the wealth or attachment variables predict unemployment length (Table 2-15). Taken together, the empirical results show that wealth affects owners’ employment length more than they affect unemployment length. 45 Table 2-14: The effects of wealth and attachment to community on owners' annual employment weeks Dependent variable: annual number of weeks employed Wealth accumulation Log (house value) -1.032** (.465) Log (positive equity) -.599** (.250) Positive equity dummy .365 (.620) Wealth1 (excluding housing) .016 (.154) Wealth1 dummy -.076 (.540) Wealth2 (including housing) -.419** (.195) Wealth2 dummy -.694 (.744) Attachment to community Log(house improvement) -.312 (.400) Dummy improvement -1.551*** (.511) Time for being homeowners 2-5 years -1.526 (1.078) Controls Y Y Y Y Y Y Y Y Y Y Time 2 2 2 2 2 2 2 2 2 2 State 50 50 50 50 50 50 50 50 50 50 Obs 3,807 3,332 3,807 3,070 3,764 3,442 3,807 424 3,807 570 Note: we use the annual PSID data from 2009 to 2013 for this panel estimation with individual fixed effects. Only owners are included. The dependent variable is the annual number of weeks being employed. The independent variables of interest are the wealth and attachment to community. The wealth includes log of the house value, log of the positive equity value, positive equity dummy, log of the wealth value without housing and its dummy, and log of the wealth value with housing and its dummy. The dummy is whether the owner has non-negative wealth or not. The attachment to community includes log of the improvement value that the owner made to their property and its dummy, and their time of being homeowners. Controls include age, marriage, number of children, sex, education, population, race, state and year fixed effects. Standard errors are clustered at the individual level and are in parentheses. Significance level: *** p<0.01, ** p<0.05, * p<0.1 46 Table 2-15: The effects of wealth and attachment to community on owners' annual unemployment weeks Dependent variable: annual number of weeks unemployed Wealth accumulation Log (house value) -.036 (.079) Log (positive equity) .102 (.074) Positive equity dummy .289 (.225) Wealth1 (excluding housing) -.054 (.047) Wealth1 dummy -.244 (.167) Wealth2 (including housing) -.016 (.063) Wealth2 dummy -.120 (.280) Attachment to community Log(house improvement) .060 (.145) Dummy improvement .218 (.192) Time for being homeowners 2-5 years .540 (3.772) Controls Y Y Y Y Y Y Y Y Y Y Time 8 8 8 8 8 8 8 6 6 6 State 50 50 50 50 50 50 50 50 50 50 Obs 16465 11027 11886 10168 16329 11228 11886 1244 8984 1,786 Note: we use the annual PSID data from 1994 to 2013 for this panel estimation with individual fixed effects. Only owners are included. The dependent variable is the annual number of weeks being unemployed. The independent variables of interest are the wealth and attachment to community. The wealth includes log of the house value, log of the positive equity value, positive equity dummy, log of the wealth value without housing and its dummy, and log of the wealth value with housing and its dummy. The dummy is whether the owner has non-negative wealth or not. The attachment to community includes log of the improvement value that the owner made to their property and its dummy, and their time of being homeowners. Controls include age, marriage, number of children, sex, education, population, race, state and year fixed effects. Standard errors are in parentheses. Significance level: *** p<0.01, ** p<0.05, * p<0.1 4.3. Mobility We have already shown that home-owners are less likely to move than renters. In this section, we want to show how the home-owning characteristics might affect owners’ mobility. We use PSID data from 1985 to 2013 to investigate various mortgage features’ impacts on people’s 47 job-related mobility. We find that borrowers with mortgages and negative equity are no more less likely to move than others. Our result is consistent with Ferreira, Gyourko, and Tracy (2010), Schulhofer-Wohl (2011), Coulson and Grieco (2012) and Demyanyk et al (2017) that the negative equity has no negative impact on job related mobility. But borrowers with positive equity are less mobile. Paradoxically, owners with more equity are less mobile while owners with more valuable houses are more mobile. The other wealth variables are either marginally significant or not significant at all 9 . Length of housing tenure seems to predict mobility: those who live in a house for a long time are less likely to move. We acknowledge, however, that specification issues remain in looking at the relationship between length of housing tenure and mobility; we are not satisfied that we have purged correlated omitted variables in our estimates. Table 2-16: Mortgage’s effect on owners’ mobility Moving probability (1) Whole period (2) FE (3) Before 2007 (4) FE (5) After 2007 (6) FE Mortgage dummy 1.114*** -0.139 1.155*** -.152 .881*** -.852 (.101) (.165) (.111) (.187) (.248) (.690) Controls Y Y Y Y Y Y Time 20 20 17 17 2 2 State 50 50 50 50 50 - Obs 150,877 9731 128,142 7,543 19,284 369 Note: we use the annual PSID data from 1985 to 2013 for this logit estimation. Only owners are included. The dependent variable is the probability of moving: 1 presents moving for job reasons and 0 represents not moving and moving for other reasons. The independent variable of interest is the mortgage dummy. Column (1) is for the whole period, covering 1985 to 2013. Column (2) is the individual fixed effect result for the whole sample. Column (3) shows the result for the period before 2007—pre-financial crisis and (4) reports the individual fixed effect result for (3). Columns (5) and (6) are the results for after 2007—post-financial crisis. Controls include state homeownership rates, age, marriage, children number, sex, education, population, race, state and time fixed effects. Positive coefficients mean that people are more likely to move for jobs in the presence of mortgages. Standard errors are robust (except fixed effects) and are in parentheses. Significance level: *** p<0.01, ** p<0.05, * p<0.1 9 For wealthh2 dummy, the Hausman test shows that this is endogenous thus we could conclude anything from this. 48 Table 2-17: Negative equity’s effect on owners’ mobility Moving probability (1) Whole period (2) Before 2007 (3) After 2007 Log (|negative equity|) .992 1.022 .978 (.020) (.035) (.025) Controls Y Y Y Time 9 5 3 State - - - Obs 72,951 42,449 30,502 Note: we use the annual PSID data from 1994 to 2013 for this multinomial logit estimation. Only owners are included. The dependent variable is the probability of moving. The independent variables of interest are the log of the absolute value of negative equity. We only report the coefficients for job moves. Column (1) is for the whole period. Column (2) is for before 2007—pre-financial crisis. Column (3) is for after 2007—post-financial crisis. Coefficients are odds ratios. Larger than one coefficient indicates that people are more likely to move for jobs with larger negative equity. Standard errors are clustered at the individual level and are in parentheses. Significance level: *** p<0.01, ** p<0.05, * p<0.1 49 Table 2-18: The effects of wealth and attachment to community on owners' mobility Dependent variable: moving probability Wealth accumulation Log (house value) 1.594*** (.192) Log (positive equity) .616*** (.050) Positive equity dummy .571** (.136) Wealth1 (excluding housing) 0.953 (.058) Wealth1 dummy .653* (.144) Wealthh2 (including housing) 0.913 (.073) Wealthh2 dummy .190*** (.061) Attachment to community Log(house improvement) 1.051 (.252) Dummy improvement 2.192*** (.391) Time for being homeowners 2-5 years .376*** (.047) 5-10 years .870 (.135) Controls Y Y Y Y Y Y Y Y Y Y Time 7 7 7 7 7 7 7 - 6 5 State 50 50 50 50 50 50 50 - 50 - Obs 39,322 24,971 28,106 22,449 28,106 25,626 28,106 1,422 23,378 20803 Note: we use the annual PSID data from 1995 to 2013 for this multinomial logit estimation with individual fixed effects. Only owners are included. The dependent variable is the probability of moving. The independent variables of interest are the wealth and attachment to community. The wealth includes log of the house value, log of the positive equity value, positive equity dummy, log of the wealth value without housing and its dummy, and log of the wealth with housing and its dummy. The dummy is whether the owner has non-negative wealth or not. The attachment to community includes log of the improvement value that the owner made to their property and its dummy, and the time of being homeowners. We only report the coefficients for job moves. Controls include age, marriage, children number, sex, education, state and time fixed effects. Coefficients are odds ratios. Larger than one coefficient indicates that people are more likely to move for jobs with the increase of right hand side variables. Standard errors are in parentheses. Significance level: *** p<0.01, ** p<0.05, * p<0.1 50 We finally use the 2004 and 2008 panels of the SIPP to test whether owners with mortgages are more or less likely to move to other states. The results are in Table 2-19. Our view is that the results indicate there are no identifiable relationship between having a mortgage and interstate moves. Table 2-19: Mortgage’s effect on owner’s mobility with SIPP data Move across different states (1) MLOGIT (2) OLS_FE (3) LOGIT_FE Obs 2008-2013 3.340** .004 1.010** 820,791 (1.647) (.003) (.485) 2004-2006 .465 -.0006 .900 555,572 (.470) (.002) (.891) Note: we use the monthly SIPP data of the panels of 2004 and 2008 for this multinomial logit estimation. Only owners are included. The dependent variable is the probability of moving across different states. The independent variable of interest is the mortgage dummy. For (1) the other two categories are not move and moving for within the same county and the moving to different counties but within the same state. Coefficients larger than one mean that people are more likely to move in the presence of mortgages. For (2) and (3), the dependent variable is a 0 and 1 dummy with 1 for moving across different states and 0 for not moving and moving within counties and across counties. Controls include age, marriage, children number, sex, education, state and time fixed effects. Standard errors are clustered at the individual level and are in parentheses. Significance level: *** p<0.01, ** p<0.05, * p<0.1 Overall, therefore, we find no systematic evidence that ownership produces immobility; and hence, any differences in outcomes can not be explained via a mobility channel. 5. Conclusion In this article, we investigate whether home-ownership affects the probability of being unemployed, the durations of employment and unemployment, mobility and wages. We wish to test the effects at both the individual and aggregate levels. We summarize our results in Table 2- 20. 51 Table 2-20: Result summary First part FE FE Result IV Result Unemployment probability Not endogenous -.123*** .275 Income Endogenous 1.323*** 6.022 Employment spell length (SIPP) Not endogenous .801*** - Number of weeks employed (PSID) Not endogenous 1.263*** - Number of weeks unemployed (PSID) Not endogenous -.534*** - Mobility PSID Not endogenous .119***(MLOGIT) - Mobility SIPP Not endogenous -.017***(OLS) - Second part FE Time FE Result Unemployment probability Negative equity Not endogenous >=2007 .454* (no state FE) Log (positive equity) Not endogenous 1985- 2013 .088* Wealth1 dummy Not endogenous 1985- 2013 -.339*** 10+ owners Not endogenous 1985-2013 .225** (no state FE) Employment spell length (SIPP) Mortgage Not endogenous 2004-2007 .018*** Mortgage Endogenous 4.2% 2008-2013 .723* Number of weeks employed (PSID) Log (house value) Not endogenous 2011-2013 -1.032** Log (positive equity) Not endogenous 2011-2013 -.599** Wealth2 (including housing) Not endogenous 2011-2013 -.419** Improvement dummy Not endogenous 2011-2013 -1.551*** Mobility PSID Log (house value) Not endogenous 1995-2013 1.594***(Mlogit) Log (positive equity) Not testable 1995-2013 .616***(Mlogit) Positive equity dummy Not endogenous 1995-2013 .571**(Mlogit) Wealth1 dummy Not endogenous 1995-2013 .653*(Mlogit) Wealthh2 dummy Endogenous 8% 1995-2013 .190***(Mlogit) Improvement dummy Not endogenous 1995-2013 2.192***(Mlogit) Years of being owners 2-5 Endogenous 4.2% 1995-2013 .376***(Mlogit) Mobility SIPP Mortgage Not endogenous 2008-2013 1.010**(Logit) Owner housing is both a consumer good and an investment. The investment aspects of owning might affect people’s employment. The direction of the impact is uncertain. The wealth embedded in owner housing might provide resources that allow people to overcome the obstacles to moving, such as moving costs. On the other hand, transaction costs related to owner housing might inhibit moves. These costs might “trap” people in their jobs; we thus might see people stay 52 in jobs longer than they should, or accept jobs with a suboptimal wage. Identifying the impact of owning on mobility is, however, difficult, because individuals less prone to moving are more likely to become owners in the first place. Finally, owners whose houses have lost value might be more likely to stay in place because (1) they anchor on purchase prices and are averse to losses and/or (2) because they might have negative equity, and do not want to bear the cost of either a default or of raising equity to make the mortgage whole. In another line of reasoning, Blanchflower and Oswald (2013) argued that ownership produces NIMBYism, which in turn stunts economic development and employment growth. Overall, arguments about the impact of ownership, both at the individual and aggregate levels, appear to be tempests in a teapot. Empirically, we find that ownership rate does not affect employment growth. We find that at the individual level, owners are not more likely to be unemployed, have unemployment spells that are not longer than renters, and that their wages are not lower. Owners’ employment spells are slightly longer; owners are less likely to move. These results indicate that owning does not lock people in and their employment outcomes are not negatively influenced. In fact, the individual level home-owning effect might be positive. As for how different aspects of owning affect employment, we find: • Among presence of mortgage, negative equity, attachment to community and wealth, only wealth plays a clear role on unemployment probabilities. • Mortgages elongate employment spell length. This is consistent with the idea that mortgages are a commitment device for earning money. 53 • Wealth plays a large part in shortening peoples’ employed time, but no part in elongating their unemployed time. • Attachment to communities may be relevant to people’s employed time, but we have yet to find a convincing method for removing the relationship between confounders of attachment to community and employment behavior. • The presence of mortgage or negative equity does not affect the probability of a job-related move. We have estimated a large number of specifications relating home-owing, or aspects of home-owning, to employment outcomes. Our interpretation of the various results is that we have little, if any, evidence that home-owning has an impact on employment outcomes. To the extent it has an impact, it is largely positive. 6. References Aaronson, Daniel and Davis, Jonathan. 2011. “How much has house lock affected labour mobility and the unemployment rate?” Chicago Fed Letter, Number 290. Retrieved from https://www.chicagofed.org/publications/chicago-fed-letter/2011/september-290 Blanchflower, David G. and Oswald, Andrew J. 2013. “Does High Home-Ownership Impair the Labor Market?” Working Paper No. 19079, Retrieved from National Bureau of Economic Research website: http://www.nber.org/papers/w19079 Bloemen, Hans G. 1995. "The Relation between Wealth and Labour Market Transitions: An Empirical Study for the Netherlands." CentER Discussion Paper no. 9599. Blundell, Richard, Magnac, Thierry and Meghir, Costas. 1997. "Savings and Labor Market Transitions." Journal of Business and Economic Statistics, 153-64. 54 Chan, Sewin. 2001. “Spatial Lock-in: Do Falling House Prices Constrain Residential Mobility?” Journal of Urban Economics, 49(3), 567-586. doi:10.1006/juec.2000.2205 Coulson, N. Edward and Fisher, Lynn M. 2002. “Tenure Choice and Labour Market Outcomes,” Housing Studies, 17(1), 35-49. doi:10.1080/02673030120105875 Coulson, N. Edward and Fisher, Lynn M. 2009. “Housing Tenure and Labor Market Impacts: The Search Goes On,” Journal of Urban Economics, 65(3), 252-264. doi:10.1016/j.jue.2008.12.003 Coulson, N. Edward and Grieco, L.E. Paul. 2012. “Mobility and Mortgages: Evidence from the PSID,” Regional Science and Urban Economics, 43(1), 1-7. Deaton, Angus and Muellbauer, John. 1980. “An Almost Ideal Demand System,” American Economic Review, vol 70, no.3, pp.312-25. Demyanyk, Yuliya, Hryshko, Dmytro, Luengo-Prado, J. María, and Sørensen E. Bent. 2017. "Moving to a Job: The Role of Home Equity, Debt, and Access to Credit." American Economic Journal: Macroeconomics, 9(2): 149-81.DOI: 10.1257/mac.20130326 Donovan, Colleen and Schnure, Calvin. 2011. “Locked in the House: Do Underwater Mortgages Reduce Labor Market Mobility?” SSRN Journal. doi:10.2139/ssrn.1856073 Engelhardt, Gary V. 2003. “Nominal Loss Aversion, Housing Equity Constraints, and Household Mobility: Evidence from the United States,” Journal of Urban Economics, 53(1), 171-195. doi:10.2139/ssrn.1808954 Farber, Henry S. 1999. “Mobility and Stability: the Dynamics of Job Change in Labor Markets. In: Ashenfelter, O.C., Card, D. (Eds.),” Handbook of Labor Economics, vol. 3B Elsevier, Amsterdam, the Netherlands, pp. 2439– 2483. 55 Ferreira, Ferreira, Gyourko, Joseph, and Tracy, Joseph. 2010. “Housing Busts and Household Mobility,” Journal of Urban Economics, 68(1), 34-45. Fischel, William A. 2004. “An Economic History of Zoning and a Cure for its Exclusionary Effects,” Urban Studies, 41(2), 317–340. Flatau, Paul, Forbes, Matt, Hendershott, Patric H. and Wood Gavin. 2003. “Homeownership and Unemployment: The Roles of Leverage and Public Housing,” Working Paper No. 10021, Retrieved from National Bureau of Economic Research website: http://www.nber.org/papers/w10021 Genesove, David and Mayer, Christopher. 2001. “Loss-aversion and Seller Behavior: Evidence from the Housing Market,” Quarterly Journal of Economics, 116 (4), 1233–1260. Goss, Ernest P. and Phillips, Joseph M. 1997. “The Impact of Homeownership on the Duration of Unemployment,” The Review of Regional Studies, 27 (1), 9–27. Hans Bloemen, Elena Stancanelli. 2001. “Individual Wealth, Reservation Wages, and Transitions intoEmployment,” Journal of Labor Economics, University of Chicago Press, 19 (2), pp.400-439. Haurin, Donald R. and Gill, H. Leroy. 2002. “The Impact of Transaction Costs and the Expected Length of Stay on Homeownership,” Journal of Urban Economics, 51(3), 563-584. doi:10.1006/juec.2001.2258 Mumford, Kevin J. and Schultz, Katie. 2013. “The Effect of Underwater Mortgages on Unemployment,” Retrieved from http://www.krannert.purdue.edu/faculty/kjmumfor/papers/Underwater_and_Unemployed pdf Munch, Jakob R., Rosholm, Michael, and Svarer, Michael. 2006. “Are Homeowners Really 56 More Unemployed?” Economic Journal, 116(514), 991-1013. doi:10.1111/j.1468- 0297.2006.01120 Munch, Jakob R., Rosholm, Michael, and Svarer, Michael. 2007. “Home Ownership, Job Duration, and Wages,” Journal of Urban Economics, 63(1), 130- 145.doi:10.2139/ssrn.1147065 Oswald, Andrew J. 1996. “A Conjecture on the Explanation for High Unemployment in the Industrialized Nations: Part 1,” Working Paper. Coventry: University of Warwick, Department of Economics. Warwick economic research papers (No.475). Retrieved from http://wrap.warwick.ac.uk/1664/ Oswald, Andrew J. 1997. “Thoughts on NAIRU (correspondence),” Journal of Economic Perspectives, 11(1), pp. 227–228. Oswald, Andrew J. 1999. The Housing Market and Europe’s Unemployment: A Non-technical Paper,” Working paper. University of Warwick. Retrieved from http://www.andrewoswald.com/docs/homesnt.pdf Rocheteau, Guillaume, Rupert, Peter, and Wright, Randall. 2007. “Ináation and Unemployment in General Equilibrium.” Scandinavian Journal of Economics, 109(4), pp. 837-855. Schulhofer-Wohl, Sam. 2011. “Negative Equity Does Not Reduce Homeowners’ Mobility,” Working Paper No. 16701, Retrieved from National Bureau of Economic Research website: http://www.nber.org/papers/w16701 Stein, Jeremy C. 1995. “Prices and Trading Volume in the Housing Market: A Model with Down-Payment Effects,” The Quarterly Journal of Economics, 110(2), 379-406. doi:10.2307/2118444 Valletta, Robert G. 2013. “House Lock and Structural Unemployment,” Labour Economics, 57 25(2013), 86–97. Young, Alwyn. 2017. “Consistency without Inference: Instrumental Variables in Practical Application,” Working Paper. Yang, Xi. 2015. “The Effeccts of Home Ownership on Post-unemployment Wages,” Working Paper 58 Chapter 3. The Effect of Home-owning on Business Development: Micro-level Evidence 1. Introduction The study of the home-owning impact dates back to the Oswald Hypothesis (1996 and 1997) that the high unemployment rate in western countries is a consequence of high home- ownership rates. The underlying mechanism is that home-owing reduces peoples’ mobility through higher transaction costs. And the lowered mobility imposes a negative impact on employment outcomes (Green and Hendershott, 2001; Chan, 2001; Coulson and Fisher, 2001 and 2009; Ferreira, Gyourko, and Tracy, 2010; Schulhofer-Wohl, 2011). Besides this lowered mobility mechanism, Blanchflower and Oswald (2013) proposed another mechanism through which home-owning affects the business, the NIMBY effect. This impact was first proposed by Frieden (1979) that NIMBYism might be one major factor altering, delaying or stopping residential development. Later, Fischel (2001 and 2009) brought about the Home-voter Hypothesis that stringent zonings are the measures undertaken by home-owners to protect their property values. Glaeser and Gyourko (2017) also proposed that home-owners impose strict zoning codes to raise constructions costs thus increasing their property values. But none of the literatures carry out empirical studies to investigate whether this NIMBY effect exists. We therefore fill this gap by examining the effect of home-owning 10 on business development. And we are the first to test this impact. 10 The reason to study the home-owner effect is that they are the primary source of the residents’ spillover effects, as opposed to renters and landlords. For renters, their temporary stay and absence of property at stake make them not as concerned as the homeowners. Additionally, because of the renter status, they possess limited legal rights to obstruct the projects. And just because of their temporary stay, they do not contribute as much as the owners to the long-term development of the community. For the landlords, even though they also care about their property values and local amenity, the individual landlords might not be able to exert any effective obstruction since they don’t live there, or might be far away, and are not directly as influenced as the home- 59 Specifically, home-owning exerts both negative and positive spillovers on neighboring business. The negative one arises through the NIMBY effect (Blanchflower and Oswald, 2013) that home-owners tend to oppose the business built adjacent to avoid the traffic, noise, and property value decline through stringent zoning or protesting. For instance, the New York residents in the Upper East Side, seeking to protect their property values, protested the subway entrances near their building 11 . The positive spillover is that home-owners stabilize the local labor pool through their longer residential stay (McCable, 2013; Coulson and Fisher, 2001; Green and Wang, 2016). Additionally, due to their financial stake in the community, home-owners improve their housing and community to build better schools, create pleasant environment and achieve lower crime rates (DiPasquale and Glaeser, 1998; Rohe, Zandt and McCarthy, 2002; Dietz and Haurin, 2003). Besides, there exists a YIMBY (Yes, in my back yard) effect that some home-owners, instead of opposing the business, invite the business to locate near them (Stephens, 2017). Both the negative and positive spillover impacts vary with distance between residence and business, neighborhood income levels, and business types. The negative spillover primarily acts in adjacent distance while the positive one for farther distances; different income home-owners exert varying effects on business. Literatures found that home-owners’ positive impacts are stronger in top income quartile than in the bottom income quartile (DiPasquale and Glaeser, 1998). Further, Glaeser and Gyourko (2017) found that higher income home-owners are linked with stricter zoning codes; neighborhoods also differ in their response to business of varying types. This motivates us to 1) identify the optimal distance; and 2) find the appropriate mix of various types owners. For the company landlords, they might be more effective in keeping out nuisance creating business, but are not quite related to the labor stability. We thus focus on investigating the home-owning impact on business. 11 They formed a block association and hired lawyers and engineers to conduct transportation and environmental assessments to let others believe this entrance should and can go elsewhere. Besides this, the local residents in Spring Valley in Las Vegas protested the expansion of an existing asphalt mixing plant. They tried to lobby the County Commissioner to overturn the recommended approval of the plant. 60 of business and residence. These two are important in planning arrangements where we could avoid the frictions by positioning the right industries in the right neighborhoods with appropriate distances. Specifically, we first develop a partial spatial equilibrium model of firm location choice, which integrates home-owning spillover effects with varying distance, residence income and business types. This shed a new light on the need for theoretical modelling of the micro-level home-owning impact. Second, pursuant to the theory, we build an empirical spatial framework which translates the conceptualized relationship into a concrete and testable setting. This framework builds on the optimal research design where we place two sets of similar business and residence with home-ownership rates as the only difference to observe the home-owning impact. However, real geo-mapping (Fig. 3-1) shows that the business in yellow, the affected, are mainly clustered together and surrounded by residents in gray, the effect source. Figure 3-1: The spatial position of residents and business on Google Maps 61 Therefore, we group the business geographically close in the U.S. to form business clusters and then draw residents surrounding the cluster to constitute donut rings of different radii to the center (Fig. 3-2). This differs fundamentally from previous literatures studying the spatial relationship with distance, where the effect sources, like a factory, lies in the center and the affected, house prices, are located on the periphery 12 (Fig. 3-3). For the implementation of the framework, we firstly find the data with the lowest geographical level measurements of business and home- owning from American Community Survey (ACS) five-year estimates and Longitudinal Employer-Household Dynamics (LEHD) at the block group level 13 . Secondly, we identify the block groups with more job counts than housing units as the job-oriented block groups and thirdly cluster the geographically adjacent job oriented block groups with the K-means clustering method for the whole nation. Then we use a K-nearest neighborhoods method to draw the donut rings surrounding the clusters. For a detailed examination of the home-owning impact, we construct 6 donut rings with the radii of within .3, .3-1, 1-2, 2-3, 3-5, and 5-10 miles. 12 The reason is that for one particular house, it is affected most likely by only one factory with certain distances, like 3 or 5 miles. But for a business center, it is surrounded in adjacent distances by residence, thus being affected by multiple residence areas in a 360 degree of directions. 13 LEHD provides the block level data. To match with the ACS, we aggregate the data to block group level. 62 Figure 3-2: The research design for analyzing the home-owning impact on job counts Figure 3-3: The position of the project and the houses on the periphery Owner 1 Renter 1 Business 1 Business 2 Owner2 Renter3 Owner3 Renter2 Project 63 For endogeneity, there might be: 1) un-observables or omitted variables affecting the job jobs and at the same time correlating with home-ownership rates; 2) the reverse causation that home-owners choose the location with less business or existing home-owners move due to nearby business growth. This spatial pattern is discovered in Liu and Painter (2012)’s paper that immigrants’ residential locations follow employment opportunities. We incorporate two strategies to deal with the endogeneity. First, we use space, time and individual/cluster fixed effects (FE) to remove the compounding factors that don’t vary across time and correct for the time trends. Second, if we found significant results for FE, the concern is that the time-variant omitted variables and reverse causations can lead to biased results in FE. To address this, we employ the Instrumental Variable (IV) method to further test the result. We use two instruments for home-ownership rates: the FHA loan limit divided by the median house price for the adjacent rings and the ratio of families with children under 18 for farther rings due to the different underlying dynamics. As opposed to FE, the IV corrects for the endogeneity by creating exogenous variations, which could remove the effects from un-observables and reverse causation. For consistency check, we also perform the analyses using lagged home-ownership rates as the independent variable due to the reverse causation and the time lags for the home-owning effect to accumulate. Additionally, we introduce another measure of the home-owning impact, the density of home-owners, to validate the results. Using the data of ACS and LEHD Workplace Area Characteristics (WAC) at the block group level from 2009 to 2014, we construct a rich dataset describing the evolution of home- ownership rates and job counts within a spatial structure (varying distances and relative positions among each other) for a period with significant home-ownership rate shocks. After incorporating the 2010 Decennial Census geocodes, this detailed micro level data allows controlling for fixed locational features in a time varying context. Using this data, we provide new results on the spatial 64 effect of home-owning on business. We find that 1) home-owning has negative influences on business for within .3 miles and positive influence for 3-5 miles. The IV method provides an estimate of the home-owning impact equal to (-12.97%) 14 for within .3 miles and 14.62% for 3-5 miles; 2) when performing the analyses for differing income subgroups, the negative impact occurs from the higher income group for adjacent distances while the positive impact exists from the lower income group for all the distances, suggesting that positive spillovers, like stable labor and YIMBYism outweigh NIMBYism in lower income neighborhoods; 3) for the industry and neighborhood match, service industries like Retail, Art and Public Administration have positive home-owning impacts in higher income neighborhoods. The job counts of these two industries are at the same time negative related to home-owning in lower income groups. Job creating industries like Manufacturing, Real Estate and Car Rental and Leasing are discovered with positive home- owning impacts in lower income groups and the job counts of them are negatively related to home- owning in the higher income neighborhoods. The results are robust across all model specifications as well as when changing home-ownership rates to home-owner densities. The paper is organized as follows. Section 2 describes the theoretical framework and hypothesis. Section 3 explores the methodology and the identification strategies to deal with the endogeneity. Section 4 presents the data and Section 5 provides the empirical results. Section 6 concludes. 2. Theoretical framework 14 The interpretation is that if the home-ownership rate decreases from 66% to 65%, job counts increase from 1000 to 1012.9. 65 The theoretical framework helps us to acknowledge about the underlying working mechanisms, assess possible outcomes, evaluate factors contributing to various outcomes, and interpret the empirical results while being aware of possible alternatives and limitations. The theoretical model builds a connection between home-ownership rates and business development, which is measured by business job counts. The model does this by incorporating home-ownership rates into individual firm’s location choice where home-owing exerts impacts through their spillovers on local labor stability, local amenities and the friction cost for dealing with the obstruction from the NIMBY effect. After establishing the relationship of individual firm location choice and home-ownership rates, we sum each individual firm’s location decision to obtain the overall job counts, therefore building the relationship between job counts and home- ownership rates of a specific area. At the end, we analyze how this relationship varies with distances between business and residence, the residence income levels and business industry types. 2.1. Individual firm location choice and home-ownership rates We start from describing how individual firm chooses its locations. The classical location theory predicts that business chooses its location among just a few options (McFadden, 1974). With the production technology and product differentiation fixed, business, in an attempt to maximize profits, consider each location with its proximity to targeted customers, transportation access, agglomeration effects, availability of employees, wage levels, local amenities, land prices, local tax policies, and finally the friction cost from home-owners. We illustrate this process with a model where we index each individual firm with e = {1,…,E} and each location with i = {1,…,I}. Firms maximize their profits subject to their production function and total cost constraint, 66 , , , , ei e i e i e i q Max pq TC  (1) , , , , ( , ) e i i e i i i z i friction q f K Z TC Kv Z P C     (2) where qe,i denotes the output for establishment e, at location i, K represents the capital, and Zi are local attributes discussed before in location i, including labor 15 , land, tax, transportation, distance to CBD, agglomeration and neighborhood amenity. TCe,i is the total costs of establishment e in location i with v as the unit cost of capital, Pi,z as the unit cost of local attributes, and Ci,friction as the friction cost. With the fact that the cost of capital is homogeneous across locations, we omit the capital and substitute Eq. (2) into Eq. (1) to obtain the profit maximization equation: , , , () e i i i i z i friction i Max pf Z Z P C     (3) Home-owning on the one hand benefits the local attributes Zi through its positive spillovers. On the other hand, home-owning might harm business through its negative spillovers on the friction cost, Ci,friction. Specifically, home-owners stabilize the local labor pool, which works within a commutable distance between business and residence shown in Eq. (4). Secondly, home-owners improve local amenities for an effect area with certain distances as limits and different income neighborhoods might provide amenities of varying features, given in Eq. (5). Finally, the negative spillover/the friction cost depends on distance, income and industry types. The NIMBY effect acts in adjacent distances for the nuisance creating business. And the perceptions of whether an industry 15 Although labor mobility could make labor as a non-local input, part of the labor feature is still localized, which is related to the local agglomeration. And the wage is also related to different locations. The labor here signifies the local part of the labor feature. 67 creates nuisance, the response to business nuisance and success rates of obstruction all differ with different income neighborhoods specified in Eq. (6). Zi (labor stability) = f 𝑑𝑖𝑠𝑡𝑎𝑛𝑐𝑒 (𝑂𝑊𝑁 𝑖 ) (4) Zi (local amenity) = f 𝑑𝑖𝑠𝑡𝑎𝑛𝑐𝑒 ,𝑖𝑛 𝑐𝑜𝑚𝑒 (𝑂𝑊𝑁 𝑖 ) (5) 𝐶 𝑖 ,𝑓𝑟𝑖𝑐𝑡𝑖𝑜𝑛 = f 𝑑𝑖𝑠𝑡𝑎𝑛𝑐𝑒 ,𝑖𝑛𝑐𝑜𝑚𝑒 ,𝑖𝑛𝑑𝑢𝑠𝑡𝑟𝑦 (𝑂𝑊𝑁 𝑖 ) (6) 2.2. Variation of the relationship with distance, income and industry types Now we have built the relationship between individual firm’s location choice and home- ownership rates. We then analyze the features of the relationship as to whether the profit is positive or negative and how it varies with distance, income and business types. Before that, we assume that across the few locations to choose from, the only difference is the home-ownership rates. Thus, Zi only includes the labor and amenity. Another assumption is that the price of them Pi,z does not vary by distance, income and industry types 16 . Consequently, Eq. (3) can be think of as the net positive home-owning spillover (pf(Z) − 𝑍 𝑖 𝑃 𝑖 ,𝑧 ) minus the negative spillover Ci,friction, yielding the net home-owning impact. We next take partial derivative of Eq. (3) with respect to distance for all the locations to yield Eq. (7): , () friction iz C fZ pP D Z D D             (7) 16 This assumes that the price of the local amenities is a regional feature, which does not change with the distance, income or industry types within a local area, a micro-level area specified for discussion of the relationship between business and residence. 68 To shape the net impact curve, we need to sign Eq. (3) and Eq. (7). There exist three cases. The first is when (pf(Z) − 𝑍 𝑖 𝑃 𝑖 ,𝑧 )<0 in Eq. (3)- the business does not match with the neighborhood and it is a non-preferred business. In this case, we have negative net impacts or profits. The second is when (pf(Z) − 𝑍 𝑖 𝑃 𝑖 ,𝑧 ) >0-the business is preferred and if Ci,friction =0, the business is at the same time non-nuisance creating or there are no material NIMBY effect, leading to positive net home-owning impacts. Eq. (7) thus becomes, , () iz fZ pP D Z D         (8) where (pf(Z) − 𝑍 𝑖 𝑃 𝑖 ,𝑧 )>0 leads to (p 𝜕𝑓 𝜕𝑍 − 𝑃 𝑖 ,𝑧 )>0. 𝜕𝑍 𝜕𝐷 illustrates how labor stability and local amenity change with distance. From the discussion above, 𝜕𝑍 𝜕𝐷 <0, meaning that the positive spillover decreases with distance between the business and residence. Thus, in this case, we have positive net home-owning impacts in a decreasing trend. The third case happens when (pf(Z) − 𝑍 𝑖 𝑃 𝑖 ,𝑧 ) >0 and Ci,friction >0, we have a preferred and nuisance creating business. In this case, to graph the net home-owning impact (Fig 3-4), we set out four assumptions, which can be loosened later to include all possible outcomes. In the figure, the upper dashed line represents the net positive spillover, the lower dashed line the negative one and the solid line the net home-owning impact. 69 Figure 3-4: The prediction of the net impact of home-owning on business with respect to distance Assumptions: (i) At D=0, Ci,friction >(pf(Z) − 𝑍 𝑖 𝑃 𝑖 ,𝑧 ), (ii) 𝜕 𝐶 𝑓𝑟𝑖𝑐𝑡𝑖𝑜𝑛 𝜕𝐷 < 0, (iii) Da< Do, negative spillover at Da=0 and net positive spillover at Do=0, (iv) | 𝜕 𝐶 𝑓𝑟𝑖𝑐𝑡𝑖𝑜𝑛 𝜕𝐷 | > |(p 𝜕𝑓 𝜕𝑍 − 𝑃 𝑖 ,𝑧 ) 𝜕𝑍 𝜕𝐷 |. Item (i) assumes that for zero distance, the negative spillover is larger than the net positive spillovers, leading to a negative net home-owning impact at distance zero. Whether this is true depends on the neighborhoods and the business types. If for a certain business, the nuisance is not as large as the positive spillover then the net impact becomes positive. Item (ii) signifies that the negative spillover decreases as distance increases. Item (iii) implies that the distance for the negative impact to work, Da, is shorter than that of the positive impact, Do. This leads to the summit Distance Impact D 0 D a Pos Neg Net 70 of the solid line at Da. If this assumption does not hold that the negative spillover distance is longer than the positive one, with the first assumption, the net home-owing impact will be negative for all distances. Item (iv) specifies that between zero distance and Da, the negative impact decreases faster than that of the positive one, which leads to a positive slope on the solid line for this interval. If otherwise, the shape of the net home-owning impact differs but still the submit is at the distance Da, which is the optimal distance to identify empirically. We summarize our theoretical model below with three cases: Case One: If (pf(Z) − 𝑍 𝑖 𝑃 𝑖 ,𝑧 )>0 and Ci,friction >0 (preferred business and positive friction costs): (i) At D=0, π<0 (ii) At 0<D< Da, 𝜕 𝜋 𝜕𝐷 >0 (iii)At Da <= D <= D0, 𝜕𝜋 𝜕𝐷 <0 and π>=0; Case Two: If (pf(Z) − 𝑍 𝑖 𝑃 𝑖 ,𝑧 )>0 and Ci,friction =0 (preferred and zero friction costs), at 0<=D< =D0, 𝜕𝜋 𝜕𝐷 <0 and π>=0; Case Three: If (pf(Z) − 𝑍 𝑖 𝑃 𝑖 ,𝑧 )<0 (not preferred), π<0 For different income levels, the shape of the curve is shown in Fig. 3-5. There are three income levels, I0, I1 and I2. I0 is the income level where there is no friction cost for preferred 71 business since to exert material NIMBY effects, the neighborhoods need to invest their time, money and resources into the stringent zoning and protesting process. I1 is the income level where the NIMBY impact starts to work. I2 is the income level higher than I1 with a larger net negative impact at distance zero and a larger positive impact at distance Da with the assumption that higher income generates more positive spillovers and at the same larger negative spillovers. We finally develop our hypotheses: Figure 3-5: The prediction of the net impact of home-owning on business with respect to distance for different income levels H1: home-owning increases the job counts of preferred business for low income neighborhoods for all distances. H2: home-owning reduces the job counts of preferred but nuisance creating business located at adjacent distance for high income neighborhoods. H3: home-owning increases the job counts of preferred business located at close but not adjacent distance for high income neighborhoods. Distance Net impact D 0 D a I 2 I 0 I 1 72 H4: home-owning decreases the job counts of not preferred business at all distances for both income groups. 3. Methodology The former theoretical discussion translates into the empirical model specified below: , , , , , , , , log( ) j t j l t j CBSA t j l t j t JobCount Own FE Controls         (9) Where log (JobCount)j,t represents the job count for cluster j in year t, Ownj,l,t is the home- ownership rate of the donut ring l={1,2,3,4,5,6} surrounding the job cluster j in period t. A negative coefficient of Ownj,l,t suggests the existence of NIMBY effects while positive coefficient indicates that there are positive spillovers from home-owning. Besides home-owning, there are other factors influencing job counts, including transportation access, agglomeration effects, local features (eg. historical development, education, weather, and tax policies) as well as common macro-economic factors (Krugman, 1991; Forkenbrock and Foster, 1996; Ellison and Glaeser, 1997; Rosenthal and Strange, 2001; Hanson and Rohlin, 2010). We account for these with fixed effects of CBSA, individual job cluster j and time periods t in FEj,CBSA,t. The first two locational fixed effects control for compounding factors not varying across time. The time fixed effects are then able to remove time-varying un- observables that equally affect the observations. Aside for the fixed effects, we further include Controlsj,l,t comprising of the lagged job counts controlling for business development inertia, residential population, income and education for the consumer effects, employee education and population density for the agglomeration effects, home-ownership rates of other rings for other home-owning impacts, distance to CBD and the job 73 counts of job clusters adjacent to the job cluster under study for the impacts from other job centers. Aside from the FE model, I also conduct the OLS regression to check the direction of the relationship but the standard errors are correlated in a panel dataset. Albeit with the FE, Eq. (9) potentially suffers from 1) omitted variable bias which can’t be removed by FE, like the ones varying across time and at the same time not equally affecting the observations; and 2) the reverse causation. Thus, if we found significant FE results, we also implement the IV method to correct for the endogeneity by creating exogenous variation. Two instruments are employed for the adjacent and farther donut rings respectively. The first instrument is for adjacent donut rings: the Federal Housing Administration (FHA) loan limit divided by the median house price. FHA announces its loan limit by county every year, which is exogenous to cluster job counts. I divide this with the donut ring average median house price. The ratio represents the housing purchasing power and is therefore supposed to be positively related to home-ownership rates. A potential concern with this instrument is that within the same county or for counties with the same loan limit, the major variation in the instrument is caused by the median house price which might be correlated with the job counts. However, this correlation varies for donut rings of different radii. For adjacent distance, the house price depends heavily on factors besides the job counts, like the central city problems-poverty, crime and school quality. While for farther distances, the instrument depends more on job counts. The second instrument I use is the ratio of families with children under 18, which should be positively related to home-ownership rates. Coulson and Fisher (2009) used the instrument that whether the first two children are of the same sex for the tenure choice 17 . The primary concern 17 Their instrument is not dependent on whether the sex of the children is the same or different. They actually use the sex of children to equate the existence of children, which is related to home-owning. 74 with this instrument is that families with children might choose to live far from business. However, firstly, people’s preferences are heterogenous that some might prefer farther locations while others might not. Secondly, for communities with creative and pleasant design, families will not be bothered by proximity to business. Thirdly, the concern is for families who can choose their locations at their will. There are people forced to locate close to businesses either for work, family or monetary reasons. Fourthly, for farther distances, like 3-5 miles, this concern is of less worry because even for families who want to live far from businesses, 2 or 3 miles and beyond are far enough. Then whether the families live at 3 miles, 3.5 miles or 4 miles is irrelevant to the job counts in the center. For consistency check, I also perform the analysis with lagged home-ownership rates as the independent variable since firstly, the home-owning impact possibly needs time to accumulate till to exert a material impact on business. Secondly, the lagged specification suffers less from the reverse causation. For example, for job counts in 2010, the 2009 homeownership rates are the average of the years of 2005, 2006, 2007, 2008, and 2009. Thus, the impact from the 2010 job counts on the home-ownership rates averaging from 2005 to 2009 stays limited. One additional robustness check is to use an alternative measure of home-owning, the home-owner density to examine the relationship. 4. Data We assemble together multiple data sources for this analysis. The dependent variable is the log of job counts from LEHD Origin-Destination Employment Statistics (LODES) WAC. The LEHD LODES provides job counts of people, categorized by their age, sex, education, income and business industry types at the block level from 2002 to 2014. The LODES data includes the 75 WAC (Workplace Area Characteristics) and RAC (Residence Area Characteristics) data. The WAC is for the people who work in the block while the RAC is for the people who reside in the block. This data is the lowest geographical level data we could find for measuring business development 18 . Since the homeownership rate is from the American Community Survey (ACS) five-year estimate at the block group level from 2009 to 2014, we aggregate the WAC data to the block group level. And the WAC lacks data in Massachusetts in 2009 and 2010, DC in 2009 and Wyoming in 2014. We drop all the data in the three states/areas and finalize the data at the block group level from 2009 to 2014. For the independent variables, the variable of interest is the home-ownership rate. The controls include the residential population, education and income from the ACS. Other controls, like employee population density and employee education, are drawn from the WAC. Because we need to calculate the distances among the block groups. We obtain the geographic coordinates, the latitudes and longitudes, together with the land area and the CBSA codes, from the geographic files of the 2010 Decennial Census. To conduct the spatial analysis, I need to form the business clusters and draw residential donut rings of different radii around them. The first is to identify the business-oriented block groups. A block group is business-oriented if the number of job counts is larger than that of housing units. This standard differs from literatures studying job clusters. The commonly used standard is the employment density and total employment (Giuliano and Small, 1991; Waddell and Shukla, 1993). We employ a different standard since the major objective for this paper is to study the effect of home-owning on business thus differentiating between business and residence is the major 18 There are two establishment level data, Longitudinal Business Database (LBD) and National Establishment time series dataset (NETS). But they are not publicly available. 76 requirement. For example, even if a block group is of low density (lower than the standard) and low total employment, it could still be a job oriented block group in our analysis since some small neighborhood business might fall into this category but should be included in the analysis of the impact of home-owning on business. In addition, the standard of high employment density would bias our result since certain kinds of people would select to locate near or far from the high job density areas. Also, a concern might be that there are residents in our job oriented block groups. The job oriented is not resident exclusive and even for the density standard, it does not specify that there should be no residents in the block group. For different years, the business oriented block groups might vary when applying this standard, we thus pick 2011 as the base year to study the job count variation for the same geographical area across time 19 . Of the 217,778 observations/block groups in 2011, 50,138 are characterized as business-oriented block groups. Only the observations in the metropolitan and micropolitan areas are included in the analysis. These block groups are matched with other years to check for their area change and job growth over the years to delete the outliers. The block groups with extreme job count percentage change (>100 and <-.7) are deleted and the panel is kept balanced and is finalized with 49,810 business-oriented block groups. We then cluster block groups with a K-means clustering method 20 , which groups the geographically adjacent block groups by calculating their distances. The detailed explanation of this method is in Appendix 3-1: K means-clustering method. This method is commonly used in cluster analysis, like grouping the observations with respective to their similarities of certain 19 The reason to pick 2011 is that 2009 uses 2000 decennial census geo codes and the results of using 2010 and 2012 as the base year are similar with that of 2011. 20 This method might not perform well on deciding the cluster boundaries associated with different densities. But since the job clusters are separated from each other to avoid overlapping (Fujita and Mori, 2005), we are not suffering from this drawback. The mapping of it on actual job clusters in Appendix 3-2 also confirms the validity of this method. 77 attributes, like age, employment density, and industry types. For my paper, the similarity attribute is the location and the K-means calculates the distances of the location for identifying their location similarity/adjacency. In the literature, there are clustering methods called Hot Spot, density-based clustering and etc.. The difference of K-means with Hot Spot is that K-means is an algorithm while Hot Spot is a methodology used for clustering. Hot Spot method firstly identifies the hot spot areas and secondly groups the geographically close ones together. K-means is commonly used for hot spot detection or adjacency detection in clustering (Lawson, 2010; Fisher, 1958; Bailey and Gatrell, 1995; Levine, 1999a). Therefore, my method could be thought of as a “Hot Spot” though the name is not quite appropriate because hot spot means they are hot in some sense like most affected crime or disease area so it is more commonly related to density. But methodologically, the only difference between my method and the Hot Spot is the standard for identifying the block groups if Hot Spot uses K-means to identify the adjacent hot spots. In ArcGIS, the method used in this article is called Grouping. More detailed explanation of the K-means is provided in Appendix 3-1. All together, we set the business cluster number as 10,000 21 for the whole nation’s metropolitan and micropolitan areas. The number of block groups for each cluster varies from 1 to 38 and averages at 5.0138. The job cluster’s average land area is 31.79 square miles with an average radius of 2.008 miles. Eventually, from 2009 to 2014, we have 59,772 observations. We map the business clusters in Orange County and two business clusters around University of Southern California in Appendix 3-2 to check this method: K-mean clustering mapping. With the business clusters in the center, we then draw block groups of residents around them to form residential donut rings with a K-nearest neighborhood method. We follow the 21 This 10,000 is an arbitrary number set for this K-means clustering method. I also tried the 11000, 12,000, 8,000 and 9,000. The performance regarding grouping consistencies and accuracies is not as good as 10,000. 78 literatures using centroids of block groups to calculate the distances (Ramsey and Bell, 2014; Niedzielski, O'Kelly and Boschmann, 2015; Meltzer and Ghorbani, 2017). A block group is identified as residence oriented if its housing unit number is larger than its job count number. We select the residence-oriented block groups that are within 10 miles of the job cluster and they fit into 6 donut rings with different radii to the cluster (Fig. 3-6). Figure 3-6: The illustration for the geographical relationship of the donut rings of different radii Tables 3-1 and 3-2 show the descriptive statistics for the residential and job features for levels of block groups, job clusters and donut rings. Table 3-1 shows that the home-ownership rate increases gradually as the distance from the cluster center increases, indicating that owner concentrated areas tend to locate farther away from business. For the change, the home-ownership .3 .3-1 1-2 Job 79 rate decreases for all geographical levels from 2009 to 2014, with the largest decrease in the inner donut rings from within .3 miles. This creates the major variation in our regression models. The median income variation indicates that the highest income people cluster around the job center for close distances from .3 to 3 miles. For education, the bachelor and graduate ratios are just a little bit higher also from .3 to 3 miles. People from 25 to 35 seem to cluster near the job centers (from 0 to 3 miles) as well as the people from 65 and above. This indicates that people of higher income, higher education and younger or older age would like to live close to the business cluster. Table 3-1: Summary statistics for resident characteristics Variable (1)Block group (2) .3 (3) .3-1 (4) 1-2 (5) 2-3 (6) 3-5 (7) 5-10 (8) US Own .6536 .6323 .6529 .6666 .6735 .6845 .6916 .659 (.2600) (.1928) (.1847) (.1689) (.1623) (.1458) (.1258) - Own Change -.0059 -.0066 -.0067 -.0063 -.0058 -.0054 -.0048 -0.006 (.0700) (.0459) (.0464) (.0398) (.0392) (.0304) (.0225) - Median income 58526 58710 61836 62067 62302 61632 60837 53889 (31946) (27145) (26718) (24378) (22837) (20291) (16951) - High .4306 .4139 .4076 .4098 .4112 .4173 .4216 .431 (.1400) (.1127) (.1088) (.1005) (.0943) (.0853) (.0733) - Bachelor .1704 .1731 .1793 .1782 .1780 .1748 .1728 .1775 (.0900) (.0759) (.0720) (.0653) (.0602) (.0533) (.0456) - Graduate .0723 .0753 .0794 .0785 .0776 .0739 .0707 7.51% (.0700) (.0637) (.0605) (.0549) (.0500) (.0426) (.0349) - Age_25 .3263 .3377 .3338 .3335 .3332 .3339 .3348 .334 (.1100) (.0764) (.0708) (.0625) (.0588) (.0507) (.0428) - Age2535 .1329 .1380 .1354 .1336 .1324 .1301 .1296 .134 (.0800) (.0514) (.0479) (.0421) (.0399) (.0331) (.0266) - Age3545 .132 .1330 .1346 .1351 .1357 .1360 .1365 .13 (.0500) (.0314) (.0302) (.0265) (.0255) (.0215) (.0168) - Age4555 .1464 .1423 .1456 .1462 .1476 .1487 .1491 .141 (.0600) (.0334) (.0333) (.0290) (.0288) (.0242) (.0188) - Age5565 .1221 .1160 .1184 .1201 .1201 .1207 .1208 .123 (.0600) (.0349) (.0343) (.0318) (.0307) (.0269) (.0226) - Age65_ .1403 .1330 .1323 .1314 .1311 .1305 .1292 .137 (.1000) (.0616) (.0591) (.0527) (.0513) (.0441) (.0378) - Residential Population 1423 11974 15706 30790 40257 97588 320870 306058480 (844) (15388) (20614) (47019) (65525) (168442) (550810) - Res_pop_density .00253 .0023 .0021 .0018 .0017 .0014 .0010 (.0064) (.0053) (.0048) (.0039) (.0033) (.0028) (.0020) Obs 1,069,231 54,643 51,790 53,991 53,993 57,168 58,896 - Note: Column (1) is for the block group level. Columns (2) through (7) are for the donut rings of different radii to the business cluster center. Standard errors of the means are in the parenthesis. The unit for the radius length is mile. The data is from ACS from 2009 to 2014. The national average in column (8) is for 2014 from American Fact Finder. Education is for 25 years and older capita count divided by total population. 80 Table 3-2 presents the job characteristic statistics. Table 3-2.1 and 3-2.2 show the job counts and job count change respectively for both the block group level and cluster level, categorized by downtown, suburbs, higher income and lower income groups. The high income is categorized as the block groups with higher median income than the average median income of the block groups. Downtown is categorized as the block groups adjacent to the CBD with the population not exceeding 5% of the CBSA. Column (4) in Table 3-2.1 shows that for the cluster level, job concentrates in the suburbs and high-income neighborhoods. In Column (4) of Table 3- 2.2, it shows that jobs are still suburbanizing in the country with the suburbs and high-income neighborhoods exhibiting the highest level of job count growth. In addition, since the block group level descriptive statistics includes all the block groups while the job cluster only includes the business-oriented block groups in the micropolitan areas, Table 3-2.2 indicates that jobs are disproportionately concentrating in more populated areas with positive job growth in Column (4) but negative job count growth in Column (2). 81 Table 3-2: Summary statistics for job characteristics Table 3-2.1: Job counts Job count (1) Block group (2) Ratio (3) Cluster (4) Ratio Pooled 661 8859 Downtown 2685 8.90% 17255 25.50% Suburbs 616 91.10% 7592 74.50% High 604 39.80% 11394 59.40% Low 698 60.20% 6656 40.60% Downtown high 3435 2.25% 20353 15.82% Downtown low 2442 6.54% 13814 9.76% Suburban high 575 37.53% 9826 43.59% Suburban low 643 53.68% 5721 30.83% Obs 1,070,442 59,658 Note: Column (1) is the average block group level job counts. Column (2) is the ratio of job counts for downtown, suburbs, and etc. of all the job counts for 2014. Column (3) is the average job counts for the business clusters. Column (4) is the ratio of job counts for downtown, suburbs, and etc. of all the job counts for 2014. The data is from WAC from 2009 to 2014. The high income is categorized as the block groups with higher median income than the average median income of the block groups. Downtown is categorized as the block groups adjacent to the CBD, the total population of which are no more than 5% of the CBSA population. Table 3-2.2: Job count change Job count change (1) Block group (2) Growth (3) Cluster (4) Growth (5)2014 US Pooled -35 -5.30% 569 6.42% 1.62% Downtown -253 -9.42% 991 5.74% Suburbs -30 -4.87% 505 6.65% High -28 -4.64% 754 6.62% Low -40 -5.73% 398 5.98% Downtown high -244 -7.10% 1178 5.79% Downtown low -243 -9.95% 773 5.60% Suburban high -26 -4.52% 681 6.93% Suburban low -34 -5.29% 348 6.08% Obs 1,070,442 59,658 Note: Column (1) is the block group level job count change at levels. Column (2) is the percentage change of job counts. Column (3) is the change for the business clusters. Column (4) is its percentage change. Column (5) is the national average one year percentage change. The data from (1) to (4) is from WAC from 2009 to 2014. The data for Column (5) is from the Bureau of Labor Statistics (All employees, thousands, total nonfarm, seasonally adjusted). The high income is categorized as the block groups with higher median income than the average median income of the block groups. Downtown is categorized as the block groups adjacent to the CBD, the total population of which are no more than 5% of the CBSA population. Table 3-2.3 presents the job characteristics for control variables. The age and education both are similar with national averages. The higher income category with more than 39,996 annual 82 income is over represented than the national average in the job clusters. This might be because we only include job concentrated areas in the cluster analysis. The whites are a little over represented and the other race 22 subgroup is under represented in the analysis. Table 3-2.3: Other job characteristics Variable (1)Block group (2)Cluster (3)2014 US Age Age_29 23.55% 23.18% 25.70% (.11) (.06) Age3054 54.92% 56.13% 50.26% (.10) (.05) Age55_ 21.53% 20.69% 24.04% (.09) (.04) Income -15,000 33.30% 25.74% 33.32% (.16) (.09) 15,012-39,996 38.68% 37.11% 38.88% (.13) (.08) 39,996 28.02% 37.15% 27.79% (.17) (.13) White 81.14% 81.46% 72.41% (.18) (.13) Black 11.47% 11.73% 12.61% (.15) (.11) Other 7.39% 6.80% 14.98% (.05) (.04) Lesshigh 14.39% 12.61% 13.67% (.09) (.06) High 29.00% 28.05% 27.95% (.10) (.06) Lessba 31.56% 32.12% 29.09% (.08) (.03) Ba 25.05% 27.22% 29.28% (.11) (.08) Obs 1,070,442 59,658 Note: Column (1) is for the block group level. Column (2) is for the business clusters. Standard errors of the means are in the parenthesis. The data is from WAC from 2009 to 2014. The national average is from American Fact Finder except the income ratio, which is from RAC because AFF does not have the income categorized as WAC. 22 Other race includes American Indian or Alaska Native, Asian, and Native Hawaiian or Other Pacific Islander. 83 5. Result 5.1. The pooled sample results Our regression results are based on Eq. (9). We first present the result for within .3 miles in Table 3-3. Column (1) represents the simple relationship between home-ownership rates and job counts. The following columns add the control variables. Across all the columns, we can see that there is a negative relationship between home-owning and business job counts. For other variables, job counts in adjacent centers negatively affect job counts of the business center under study, which indicates that there are competitions among adjacent job clusters. Residential population and income positively affect job counts, which is consistent with the consumer effect. More educated 23 employees and residents also predict more job counts. 23 The omitted category is less than high school. 84 Table 3-3: Results on the cluster level of within .3 miles Variable (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) Own -1.209*** -1.227*** -.016*** -.019*** -.019*** -.012** -.019** -.019** -.026*** -.032*** -.129** Std (.058) (.059) (.004) (.005) (.005) (.006) (.008) (.008) (.008) (.009) (.052) L.Job .974*** .974*** .974*** .957*** .957*** .957*** .959*** .951*** .254*** (.002) (.002) (.002) (.003) (.002) (.002) (.002) (.003) (.023) OtherJob .000* .000* -.001*** -.002*** -.002*** -.002*** -.002*** -.086** (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.018) ResPop .021*** .021*** .021*** .020*** .024*** .142*** (.001) (.001) (.001) (.001) (.002) (.029) EmployeeEdu High -.211*** -.206*** -.214*** -.200*** -.142 .633 (.047) (.048) (.049) (.048) (.107) (.424) LessBa .254*** .257*** .235*** .234*** .315*** 2.052** (.078) (.078) (.080) (.079) (.095) (.454) Ba .186*** .179*** .175*** .153*** .349*** 2.591** (.026) (.029) (.030) (.030) (.048) (.394) ResIncome .005 .002 .004 .014** .014 (.005) (.006) (.005) (.006) (.020) ResEdu High .009 .014 .039** .199** (.016) (.016) (.019) (.078) Bachelor .069*** .077*** .049** .184** (.022) (.022) (.024) (.089) Graduate -.043 -.032 -.034 .265** (.029) (.029) (.033) (.121) Other ring own Y Y Y Y Y Y Y Y Y Y Distance to CBD Y Y Y Y Y Y N EmployeePopDen Y Y Y Y Y Y Year FE Y Y Y CBSA FE Y Y Cluster FE Y r2 .049 .049 .958 .958 .958 .958 .958 .958 .960 .961 .394 N 52157 52136 44370 44370 44370 44370 44370 44370 44370 44370 44370 Note: The dependent variable is the log of job counts. Column (1) represents the simple relationship between home- ownership rates and job count growth. Column (2) adds the other ring home-ownership rates. Column (3) adds the lagged job counts. Column (4) adds the job counts of adjacent job clusters. Column (5) adds the distance to the CBD. Column (6) adds the employee population density and education. Column (7) adds the residential population income. Column (8) adds the residential education. Column (9) adds the year fixed effects. Column (10) adds the CBSA fixed effects. Column (11) adds the cluster fixed effects. The interpretation of the Column (11) coefficient is that if ownership rate decrease by 1%, then job counts increase by 12.9%. So if ownership rate decreases from 66% to 65%, then job counts might increase from 1000 to 1012.9. Standard errors are clustered at the cluster level and are in parentheses. *** p<0.01, ** p<0.05, * p<0.1. Table 3-4 presents the results for all donut rings with specifications of OLS, FE, and IV. The OLS specification indicates the general direction of the relationship but due to panel data features, the standard errors of OLS coefficients are correlated. The FE model with CBSA and job 85 cluster fixed effects corrects this problem and could control for both regional and individual locational features not varying across time. Also, time fixed effects can remove the varying factors affecting observations equally across time. But the FE can’t correct for reverse causation. Thus, if we find significant results for FE, we will further test the IV result, which corrects for the endogeneity by creating exogenous variations thus being able to control for reverse causations. In conclusion, if both results of FE and IV are significant, it indicates there is a home-owning impact. 86 Table 3-4: Results for different radii (1) OLS (2)FE (3)IV_FHA (4)OLS_lag (5)FE_lag (6)IV_Lag_FHA 0.3 Own -.0319 -.1291 -.1086 -.0281 -.0592 -.1297 Std (.0097) (.0521) (.0524) (.0097) (.0284) (.0566) P_value .0011 .0132 .0380 .0036 .0375 .0220 Overall R 2 .9607 .3942 .9643 .9614 .3918 .9643 Obs 46498 44370 37211 45303 43253 37215 .3-1 Own -.0253 .0755 -.0327 -.0170 Std (.0113) (.0552) (.0117) (.0230) P_value .0255 .1714 .0053 .4603 Overall R2 .9613 .3352 .9618 .3218 Obs 44013 44012 42971 42970 1-2 Own -.0256 -.0266 -.0253 .0102 Std (.0128) (.0641) (.0132) (.0304) P_value .0454 .6780 .0547 .7373 Overall R 2 .9603 .3581 .9610 .3611 Obs 45666 45665 44835 44834 (1) OLS (2)FE (3)IV_Child (4)OLS_lag (5)FE_lag (6)IV_lag_Child 2-3 Own -.0166 -.1722 .0495 -.0086 .0001 Std (.0148) (.0674) (.0470) (.0146) (.0288) P_value .2640 .0106 .2918 .5563 .9984 Overall R2 .9612 .3665 .9640 .9616 .3552 Obs 45614 45613 34944 44842 44841 3-5 Own -.0053 -.2247 .0832 .0035 .1409 .1462 Std (.0172) (.1002) (.0660) (.0174) (.0450) (.0780) P_value .7588 .0249 .2078 .8387 .0018 .0608 Overall R2 .9602 .3673 .9636 .9607 .3541 .9636 Obs 48103 48102 36716 47505 45285 36716 5-10 Own .0303 .0966 .0260 -.0374 Std (.0215) (.1855) (.0218) (.0709) P_value .1590 .6023 .2321 .5977 Overall R2 .9596 .3505 .9601 .3508 Obs 49275 49273 48940 48939 Note: The dependent variable is the log of job counts. Column (1) is the OLS with controls of lagged job counts, residential population, employee education, employee population density, residential median income, educational levels, ownership rates of other rings, distance to CBD, job counts for job clusters adjacent to the job cluster under study, cbsa and year fixed effects. Column (2) controls for the individual (cluster) fixed effects. Column (3) is the IV result. The instrument for the within .3 miles is the FHA loan limit divided by the median house price. The instrument for the 2-3 and 3-5 miles is the ratio of families with children under 18. Column (4) to column (6) use the lagged homeownership rates as the independent variables. The interpretation of the coefficient (.1291) for within .3 miles for Column (2) for within .3 miles is that if the home-ownership rate decreases by 1%, then job counts increase by 12.91%. So if ownership rate decreases from 66% to 65%, job counts increase from 1000 to1012.91 Standard errors are clustered at the cluster level and are in parentheses. *** p<0.01, ** p<0.05, * p<0.1. 87 Table 3-4 shows that for the distances of within .3 miles, 2-3 miles and 3-5 miles, the results of FE are significant. I subsequently conduct the IV estimation to further test the relationship. The first stage regression in Table 3-5 shows significant relevance between home- owning and the instruments with F statistics of 1422.26, 1541.56 and 1548.25 24 for the three distances. For the adjacent distance of within .3 miles, we use the instrument of FHA loan limit divided by the median house price and for the distances of 2-3 and 3-5 miles, we use the instrument of the ratio of families with children under 18. Table 3-5: The instrument variable results (1) 0.3_FHA loan (2) 2-3_Child (3) 3-5_Child IV .0745 .2621 .2618 Std (.0047) (.0130) (.0145) P_value .0000 .0000 .0000 Overall R 2 .7241 .7808 .8199 Obs 35519 34944 36716 Note: The dependent variable is the home-ownership rates of the donut ring. The independent variable is the instruments. Column (1) tests the significance between home-ownership rates and FHA loan limit divided by the median house price for within .3 miles. Columns (2) and (3) test the relevance between home-ownership rates and child ratio for the distances of 2-3 and 3-5 respectively. Controls include the job counts controlling for all other explanatory variable including home-ownership rates of the donut rings under study, lagged job counts, residential population, employee education, employee population density, residential median income, educational levels, ownership rates of other rings, distance to CBD, cbsa and year fixed effects. Standard errors are clustered at the cluster level and are in parentheses. *** p<0.01, ** p<0.05, * p<0.1. 24 Large F-statistics could alleviate the bias of the IV coefficient with existence of correlations between instruments and dependent variables. 88 For within .3 miles, the IV result of (-.1086) in Column (3) of Table 3-4 is significant and is a little smaller in magnitude than that of FE. This suggests that the FE might suffer from reverse causation since the FE is biased downward. For the lagged impact of within .3 miles, the IV result of (-.1297) is also significant with a larger impact compared to the FE in columns (5) but is similar in magnitude with that of FE in Column (2). For the distance of 2-3 miles, although there is negative significance for the FE, the IV result is not significant, implying that the significance of the FE is possibly caused by omitted variable bias or reverse causation. It is also the case with the 3-5 distance result for the contemporary effect. While for lagged impacts in column (6) for 3-5 miles, we find significant coefficient of .1462 similar with that (.1409) of FE in column (5). The results show that the NIMBY effect exists for the adjacent distance of within .3 miles. While positive spillover effects are identified for the distance of 3-5 miles for lagged effects. In addition, the existence of negative significance for both the contemporary and lagged results suggests that the negative home-owning spillover is immediate and long lasting. While the fact that the positive significance only exists in lagged effect indicates that positive home-owning spillovers need time to accumulate. 5.2. Results for neighborhoods of different income levels To further test the hypothesis and cross check the results, we present the results for different income groups in Tables 3-6 to 3-9. Tables 3-6 and 3-7 show the FE results for the lower income and higher income groups respectively. Tables 3-8 and 3-9 are the results for lagged FE for the two income groups 25 . We also perform the regressions for OLS and IV besides the FE and 25 The OLS, FD and IV results will be provided upon request for article length reasons. 89 only the results consistent 26 across all specifications and have more than two significant results across different income groups for each distance are marked bold. From Table 3-6, we can see that there are no effects from lower income groups. While Table 3-7 shows that negative impacts happen for .3-1 miles. This result is consistent across all the specifications (OLS, FE and IV). The probability that this negative significance is a coincidence is .000457 for FE alone. In conclusion, there are no impacts from lower income groups and higher income groups exert negative impacts on the job counts for .3-1 miles, suggesting the existence of NIMBY effect for high income groups. 26 With more than two significant results across all the specifications and have no contradictory results from other specifications. 90 Table 3-6: The fixed effect results for the lower income group .3 <5% <10% <15% <20% <25% <30% <35% <40% <45% <50% Own - 0.6285 - 0.1480 - 0.1128 - 0.0692 - 0.0435 - 0.0550 - 0.0992 - 0.0759 - 0.0448 - 0.1045 Std Error 0.5567 0.2635 0.1754 0.1274 0.1093 0.0973 0.0858 0.0777 0.0740 0.0702 P-value 0.2596 0.5746 0.5202 0.5870 0.6907 0.5722 0.2476 0.3291 0.5446 0.1369 Overall R 2 0.4115 0.2747 0.3056 0.3207 0.2686 0.2719 0.2791 0.2736 0.2601 0.3140 Obs 1300 3204 5150 7464 9720 11596 14083 16320 18116 21623 .3-1 Own 0.1556 - 0.0049 - 0.1647 - 0.0286 - 0.0604 - 0.0037 0.0134 - 0.0086 - 0.0170 - 0.0349 Std Error 0.5230 0.2630 0.1939 0.1468 0.1164 0.1131 0.1103 0.0979 0.0919 0.0814 P-value 0.7663 0.9852 0.3960 0.8455 0.6036 0.9739 0.9033 0.9298 0.8533 0.6684 Overall R 2 0.4157 0.3690 0.3715 0.3662 0.3293 0.3295 0.3538 0.1983 0.2146 0.2155 Obs 1296 3163 5009 7249 9322 11160 13459 15563 17342 20560 1-2 Own 0.2313 - 0.1799 - 0.2410 - 0.2787 - 0.1606 - 0.1189 - 0.0646 - 0.0643 - 0.0224 - 0.0945 Std Error 0.5248 0.2684 0.2643 0.1850 0.1558 0.1418 0.1215 0.1097 0.1053 0.0897 P-value 0.6596 0.5027 0.3620 0.1321 0.3026 0.4018 0.5950 0.5577 0.8317 0.2922 Overall R 2 0.3214 0.4248 0.3280 0.3163 0.3042 0.1183 0.1542 0.2138 0.2517 0.2629 Obs 1324 3212 5115 7380 9569 11413 13798 15937 17755 21092 2-3 Own - 0.1728 0.2467 0.1809 0.0923 - 0.0729 - 0.1461 - 0.0844 - 0.0739 - 0.0597 - 0.0697 Std Error 0.5415 0.3716 0.2651 0.2127 0.1860 0.1665 0.1444 0.1327 0.1247 0.1061 P-value 0.7498 0.5069 0.4950 0.6643 0.6953 0.3804 0.5591 0.5775 0.6320 0.5112 Overall R 2 0.2484 0.2371 0.2918 0.1611 0.2067 0.2249 0.2348 0.2226 0.2434 0.2830 Obs 1278 3104 5004 7205 9344 11129 13506 15643 17457 20591 3-5 Own - 0.3790 - 0.5826 - 0.3750 - 0.2900 - 0.3199 - 0.2846 - 0.2521 - 0.2646 - 0.2842 - 0.2296 Std Error 0.6832 0.4047 0.3533 0.2935 0.2365 0.2307 0.2099 0.1856 0.1867 0.1469 P-value 0.5794 0.1504 0.2886 0.3232 0.1763 0.2175 0.2298 0.1540 0.1280 0.1181 Overall R 2 0.4457 0.1789 0.3382 0.3060 0.3324 0.2789 0.2999 0.2992 0.3048 0.3160 Obs 1260 3054 4908 7045 9147 10945 13301 15379 17152 20467 5-10 Own - 0.9936 - 1.2054 0.4067 0.2966 0.6564 0.2659 0.0536 - 0.0075 - 0.1735 - 0.0759 Std Error 1.3128 1.0678 0.6267 0.5004 0.4188 0.3713 0.3110 0.3016 0.3155 0.2705 P-value 0.4498 0.2594 0.5165 0.5535 0.1172 0.4740 0.8633 0.9800 0.5823 0.7791 Overall R 2 0.5099 0.3961 0.4977 0.3530 0.4662 0.4371 0.4155 0.4219 0.3853 0.3999 Obs 1045 2495 4028 5666 7409 8869 10723 12344 13789 16492 Note: These are FE with the dependent variable as the log of the job counts and the independent variable as the home- ownership rates of the donut ring under study. Analysis are performed for different income percentiles. Controls include homeownership rates of other circles, residential population, income, education, employee’s population density, employee’s education, whether the location is in downtown or suburbs, year and cbsa fixed effects. Standard errors are clustered at the job cluster level and are in parentheses. *** p<0.01, ** p<0.05, * p<0.1. 91 Table 3-7: The fixed effect results for the higher income group 0.3 >55% >60% >65% >70% >75% >80% >85% >90% >95% Own -0.0896 -0.1023 -0.0943 -0.0799 -0.1293 -0.1102 -0.1700 -0.3433 -0.3143 Std Error 0.0727 0.0738 0.0794 0.0828 0.0912 0.0952 0.1183 0.1321 0.2241 P-value 0.2182 0.1657 0.2351 0.3345 0.1564 0.2469 0.1508 0.0095 0.1617 Overall R 2 0.4361 0.4568 0.4453 0.4293 0.4230 0.3140 0.2568 0.1271 0.6643 Obs 18109 16309 14074 11590 9710 7464 5167 3208 1305 .3-1 Own -0.1191 -0.1244 -0.1373 -0.1045 -0.0807 -0.1792 -0.2125 -0.5416 -0.6165 Std Error 0.0737 0.0740 0.0823 0.0926 0.1011 0.1164 0.1430 0.1767 0.2725 P-value 0.1062 0.0930 0.0952 0.2590 0.4248 0.1239 0.1376 0.0022 0.0242 Overall R 2 0.3759 0.3361 0.3539 0.3506 0.3076 0.3537 0.2633 0.3893 0.3524 Obs 17358 15578 13474 11165 9327 7242 5018 3174 1300 1-2 Own -0.1435 -0.1480 -0.1602 -0.2012 -0.2463 -0.2527 -0.2753 -0.1913 0.0979 Std Error 0.1073 0.1148 0.1268 0.1238 0.1357 0.1338 0.1745 0.2296 0.3011 P-value 0.1811 0.1976 0.2067 0.1041 0.0695 0.0591 0.1148 0.4049 0.7452 Overall R 2 0.3743 0.3600 0.3851 0.2953 0.1874 0.4189 0.4485 0.3921 0.4317 Obs 17771 15951 13810 11417 9575 7387 5118 3214 1325 2-3 Own -0.1154 -0.1069 -0.1522 -0.0943 -0.1588 -0.1879 -0.1934 -0.1924 -0.4085 Std Error 0.0979 0.0954 0.0966 0.1148 0.1325 0.1381 0.1503 0.1806 0.2290 P-value 0.2388 0.2627 0.1153 0.4115 0.2308 0.1736 0.1985 0.2870 0.0752 Overall R 2 0.4339 0.3783 0.3448 0.3195 0.3075 0.2587 0.3551 0.2844 0.2581 Obs 17477 15664 13521 11137 9349 7211 5016 3104 1279 3-5 Own 0.0030 -0.0389 -0.0761 -0.0123 -0.0399 -0.3584 -0.4621 -0.2631 0.7004 Std Error 0.1721 0.1718 0.1719 0.1944 0.2055 0.1909 0.2565 0.3515 0.4859 P-value 0.9862 0.8210 0.6581 0.9497 0.8462 0.0607 0.0719 0.4544 0.1504 Overall R 2 0.3644 0.3476 0.3327 0.3910 0.3872 0.3943 0.4506 0.3954 0.2875 Obs 17148 15375 13299 10941 9142 7040 4908 3054 1257 5-10 Own -0.6987 -0.8271 -0.6692 -0.6461 -0.5522 -0.8845 -0.5208 -0.1082 0.1722 Std Error 0.3227 0.3302 0.2863 0.3159 0.3461 0.4496 0.5303 0.6190 0.8987 P-value 0.0304 0.0123 0.0195 0.0410 0.1107 0.0493 0.3263 0.8613 0.8482 Overall R 2 0.4327 0.4598 0.3870 0.4251 0.4609 0.4078 0.5567 0.4028 0.5417 Obs 13779 12331 10700 8872 7412 5658 4021 2495 1045 Note: These are FE with the dependent variable as the log of the job counts and the independent variable as the home- ownership rates of the donut ring under study. Analysis are performed for different income percentiles. Controls include homeownership rates of other circles, residential population, income, education, employee’s population density, employee’s education, whether the location is in downtown or suburbs, year and cbsa fixed effects. Standard errors are clustered at the job cluster level and are in parentheses. *** p<0.01, ** p<0.05, * p<0.1. 92 We next present the FE results with lagged home-ownership rates as independent variables in Tables 3-8 and 3-9. From Table 3-8, we find that for income groups of less than 10 and 15 percentile income, the impact is positive for the within .3 miles. Another positive significance is found for the 3-5 miles in the income groups of less than 20, 25, 30, 40, and 50 percentile incomes. Thus, the positive lag impact in pooled sample result seems to be from the lower income groups for the 3-5 miles and even for adjacent distance of within .3 miles, which suggests no NIMBY effect for the lower income group or the positive spillovers outweigh the negative ones. This agrees with H1. From Table 3-9, we didn’t find significance across all the specifications. 93 Table 3-8: The fixed effect results with lagged homeownership rates as the independent variable for the lower income group 0.3 <5% <10% <15% <20% <25% <30% <35% <40% <45% <50% L.Own 0.2060 0.1962 0.1241 0.0547 0.0516 0.0149 - 0.0171 - 0.0182 - 0.0131 - 0.0293 Std Error 0.2045 0.0870 0.0725 0.0611 0.0561 0.0517 0.0469 0.0416 0.0410 0.0371 P-value 0.3143 0.0244 0.0870 0.3707 0.3580 0.7736 0.7157 0.6616 0.7485 0.4292 Overall R 2 0.4186 0.2707 0.3045 0.3193 0.2695 0.2675 0.2724 0.2723 0.2552 0.3216 Obs 1284 3162 5083 7345 9564 11414 13849 16043 17805 21212 .3-1 L.Own 0.0437 - 0.0526 0.0470 0.0303 0.0217 - 0.0156 - 0.0020 - 0.0284 - 0.0171 - 0.0106 Std Error 0.2570 0.1246 0.1038 0.0870 0.0636 0.0547 0.0485 0.0441 0.0420 0.0376 P-value 0.8651 0.6727 0.6509 0.7274 0.7325 0.7755 0.9669 0.5202 0.6848 0.7786 Overall R 2 0.4069 0.3606 0.3701 0.3559 0.3190 0.3123 0.3347 0.1856 0.1997 0.1951 Obs 1285 3137 4967 7178 9219 11035 13278 15353 17102 20230 1-2 L.Own - 0.3621 - 0.1110 - 0.0798 - 0.0586 0.0152 0.0451 0.0728 0.0539 0.0803 0.0520 Std Error 0.3067 0.1586 0.1140 0.0971 0.0713 0.0647 0.0600 0.0542 0.0495 0.0459 P-value 0.2385 0.4841 0.4841 0.5462 0.8313 0.4855 0.2248 0.3201 0.1046 0.2572 Overall R 2 0.3673 0.4201 0.3337 0.3155 0.3052 0.1172 0.1569 0.2140 0.2512 0.2612 Obs 1313 3186 5074 7317 9463 11287 13624 15738 17540 20802 2-3 L.Own - 0.0677 - 0.1028 - 0.0170 - 0.0558 - 0.0263 - 0.0413 - 0.0278 0.0065 0.0054 - 0.0087 Std Error 0.3324 0.1507 0.1212 0.0985 0.0886 0.0779 0.0674 0.0585 0.0534 0.0488 P-value 0.8387 0.4953 0.8883 0.5715 0.7664 0.5959 0.6800 0.9116 0.9195 0.8582 Overall R 2 0.2764 0.1987 0.2559 0.1484 0.1941 0.2136 0.2220 0.2116 0.2334 0.2775 Obs 1268 3065 4945 7120 9236 11005 13334 15442 17239 20312 3-5 L.Own 0.2852 0.1821 0.1576 0.3042 0.2443 0.2430 0.2296 0.2086 0.1938 0.1622 Std Error 0.3543 0.1539 0.1228 0.1134 0.0990 0.0992 0.0889 0.0832 0.0757 0.0719 P-value 0.4215 0.2371 0.1996 0.0074 0.0137 0.0143 0.0099 0.0122 0.0105 0.0241 Overall R 2 0.3964 0.1605 0.2794 0.2842 0.3292 0.2664 0.3077 0.3124 0.3090 0.3160 Obs 1244 3023 4860 6975 9048 10832 13153 15210 16974 20224 5-10 L.Own - 0.4239 - 0.3858 0.0441 - 0.0110 0.1727 0.1066 - 0.0019 - 0.0304 - 0.0754 - 0.0613 Std Error 0.5397 0.3957 0.2444 0.1763 0.1656 0.1573 0.1362 0.1213 0.1196 0.0966 P-value 0.4329 0.3300 0.8567 0.9504 0.2972 0.4983 0.9888 0.8020 0.5281 0.5256 Overall R 2 0.4476 0.4598 0.5037 0.3446 0.4321 0.4115 0.3935 0.4081 0.3748 0.3852 Obs 1033 2477 4005 5632 7357 8811 10645 12260 13701 16366 Note: These are FE with the dependent variable as the log of the job counts and the independent variable as the LAGGED home-ownership rates of the donut ring under study. Analysis are performed for different income percentiles. Controls include homeownership rates of other circles, residential population, income, education, employee’s population density, employee’s education, whether the location is in downtown or suburbs, year and cbsa fixed effects. Standard errors are clustered at the job cluster level and are in parentheses. *** p<0.01, ** p<0.05, * p<0.1. 94 Table 3-9: The fixed effect results with lagged homeownership rates as the independent variable for the higher income group 0.3 >55% >60% >65% >70% >75% >80% >85% >90% >95% L.Own -0.0530 -0.0456 -0.0379 -0.0492 -0.0586 -0.0470 -0.0577 -0.1702 -0.4134 Std Error 0.0410 0.0436 0.0476 0.0546 0.0648 0.0787 0.1151 0.1736 0.3662 P-value 0.1965 0.2952 0.4258 0.3669 0.3655 0.5506 0.6162 0.3271 0.2596 Overall R 2 0.4222 0.4413 0.4333 0.4377 0.4107 0.2912 0.2327 0.0952 0.6409 Obs 17536 15781 13609 11199 9367 7205 4985 3096 1255 .3-1 L.Own -0.0571 -0.0613 -0.0879 -0.0757 -0.0364 -0.0092 0.0299 -0.0325 0.0020 Std Error 0.0336 0.0359 0.0410 0.0461 0.0495 0.0534 0.0632 0.0721 0.1028 P-value 0.0898 0.0877 0.0323 0.1004 0.4623 0.8634 0.6357 0.6523 0.9844 Overall R 2 0.3658 0.3238 0.3430 0.3414 0.2979 0.3547 0.2458 0.4180 0.5101 Obs 16859 15115 13055 10820 9022 7001 4839 3061 1260 1-2 L.Own -0.0352 -0.0452 -0.0266 -0.0336 0.0164 -0.0328 0.0078 0.0686 -0.1300 Std Error 0.0423 0.0435 0.0470 0.0507 0.0558 0.0664 0.0718 0.0947 0.1613 P-value 0.4054 0.2987 0.5712 0.5071 0.7694 0.6213 0.9130 0.4687 0.4207 Overall R 2 0.3904 0.3789 0.4072 0.3025 0.1810 0.4180 0.4381 0.3722 0.3777 Obs 17416 15621 13517 11176 9359 7215 4987 3124 1291 2-3 L.Own -0.0142 -0.0107 -0.0076 0.0070 0.0200 0.0524 0.1451 0.0477 -0.0914 Std Error 0.0387 0.0396 0.0432 0.0493 0.0552 0.0635 0.0859 0.1019 0.1171 P-value 0.7146 0.7875 0.8596 0.8872 0.7168 0.4094 0.0915 0.6401 0.4356 Overall R 2 0.4210 0.3653 0.3321 0.3017 0.2913 0.2463 0.3363 0.2574 0.2375 Obs 17166 15368 13257 10930 9160 7054 4899 3030 1243 3-5 L.Own 0.0507 0.0016 -0.0059 0.0108 0.0167 -0.0558 -0.0576 -0.0953 0.1109 Std Error 0.0651 0.0607 0.0640 0.0677 0.0741 0.0755 0.0869 0.1280 0.1709 P-value 0.4363 0.9795 0.9260 0.8733 0.8221 0.4601 0.5079 0.4569 0.5168 Overall R 2 0.3647 0.3518 0.3459 0.4076 0.4106 0.3974 0.4550 0.3947 0.2911 Obs 16971 15209 13149 10823 9041 6960 4854 3016 1240 5-10 L.Own 0.0046 -0.0655 -0.0827 0.0020 -0.0054 -0.1069 -0.2368 0.0011 -0.1441 Std Error 0.1296 0.1360 0.1256 0.1216 0.1380 0.1653 0.1943 0.2086 0.3053 P-value 0.9719 0.6299 0.5100 0.9868 0.9688 0.5178 0.2232 0.9960 0.6374 Overall R 2 0.4131 0.4398 0.3782 0.4181 0.4483 0.3891 0.5370 0.3998 0.4815 Obs 13716 12271 10648 8830 7374 5636 4002 2485 1039 Note: These are FE with the dependent variable as the log of job counts and the independent variable as the LAGGED home-ownership rates of the donut ring under study. Analysis are performed for different income percentiles. Controls include homeownership rates of other circles, residential population, income, education, employee’s population density, employee’s education, whether the location is in downtown or suburbs, year and cbsa fixed effects. Standard errors are clustered at the job cluster level and are in parentheses. *** p<0.01, ** p<0.05, * p<0.1. 95 We summarize the results in Table 3-10, where we show that negative impacts are from the adjacent distances like .3 while the positive is from 3-5. For high and low income subgroups, negative impacts are for the .3-1 from high income groups and positive ones are for the .3 and 3-5 from low income groups. It suggests that low income people might care more about job opportunities while higher income groups might worry more about the nuisance from business. Table 3-10: Result summary for the pooled sample and samples of different incomes Pooled 0.3 .3-1 1-2 2-3 3-5 5-10 Contemporary - Lag - + Income 0.3 .3-1 1-2 2-3 3-5 5-10 Contemporary - high Lag + low + low Note: this table summarizes the result. It shows that for the pooled sample, the home-owning impact is negative for the within .3 miles both for the contemporary (with home-ownership rates in period t as the independent variable) and lagged (with home-ownership rates in period t-1 as the independent variable) periods. It also presents the results for high and low income subgroups: for the contemporary effect, home-owning has negative influences on business for the .3-1 miles; for the lagged effect, home-owning has positive influences on business for the within .3 miles and 3-5 miles. The blank part means no significance is found. 5.3. The results for different industries In this section, we intend to find the optimal mix between different income neighborhoods and business types by evaluating the home-owning impact on varying business types. The LEHD data contains job counts for 20 industry sectors. Table 3-11 shows the summary statistics for the job counts of different industries at the job cluster level from 2009 to 2014. Table 3-12 shows the street view of the maximum job count cluster. The detailed description of each industry is in Appendix 3-3. 96 Table 3-11: Descriptive statistics for the industry job counts at the cluster level (1)NAICS (2)Obs (3)Obs Ratio (4)Mean (5)Total job counts (6)Total job ratio (7)Max (8)Max Location Total 58,193 1 8866 93479011 1 467929 - NAICS sector 11 (Agriculture, Forestry, Fishing and Hunting) 29,683 0.5101 96 515923 0.0055 11322 North CA NAICS sector 21 (Mining, Quarrying, and Oil and Gas Extraction) 21,249 0.3651 113 453794 0.0049 17320 Houston downtown NAICS sector 22 (Utilities) 33,870 0.582 93 546710 0.0058 11712 LA downtown NAICS sector 23 (Construction) 57,126 0.9817 336 3675240 0.0393 12023 Vegas NAICS sector 31-33 (Manufacturing) 55,091 0.9467 907 8882819 0.095 33260 Seattle (Boeing) NAICS sector 42 (Wholesale Trade) 56,825 0.9765 420 4291942 0.0459 26227 NYC Manhattan NAICS sector 44-45 (Retail Trade) 57,573 0.9893 969 10092999 0.108 18221 New York NAICS sector 48-49 (Transportation and Warehousing) 54,570 0.9377 323 3232054 0.0346 52392 NYC (JFK airport) NAICS sector 51 (Information) 52,465 0.9016 258 2415339 0.0258 47206 Seattle (Microsoft headquarters) NAICS sector 52 (Finance and Insurance) 55,854 0.9598 446 4454237 0.0476 75889 NYC Manhattan NAICS sector 53 (Real Estate and Rental and Leasing) 54,652 0.9392 135 1331323 0.0142 15293 NYC Manhattan NAICS sector 54 (Professional, Scientific, and Technical Services) 57,096 0.9811 594 6320236 0.0676 73292 Chicago NAICS sector 55 (Management of Companies and Enterprises) 45,423 0.7806 225 1979462 0.0212 13630 Arkansas (Walmart home office) NAICS sector 56 (Administrative and Support and Waste Management and Remediation Services) 57,129 0.9817 576 6292557 0.0673 26435 Chicago NAICS sector 61 (Educational Services) 54,794 0.9416 886 8318011 0.089 173587 NYC Manhattan (NYU) NAICS sector 62 (Health Care and Social Assistance) 57,056 0.9805 1314 13601190 0.1455 54194 Houston (Texas Medical Center) NAICS sector 71 (Arts, Entertainment, and Recreation) 50,039 0.8599 162 1496857 0.016 41671 Orlando Disney NAICS sector 72 (Accommodation and Food Services) 57,018 0.9798 763 8207916 0.0878 91530 Vegas NAICS sector 81 (Other Services [except Public Administration]) 57,422 0.9868 257 2545756 0.0272 23551 Seattle downtown NAICS sector 92 (Public Administration) 54,042 0.9287 526 4824638 0.0516 90563 NYC Lower Manhattan Note: Column (1) is the NAICS codes for each industry. Column (2) provides the number of block groups with positive job counts for each industry. Column (3) shows the observation ratios of each category. A higher ratio means that more block groups contain jobs of this industry. Except for a few industries, like the agriculture, mining and utilities, most of the categories appear on almost every cluster. Column (4) provides the average number of job counts of each industry per job cluster. Column (5) is the total job counts and Column (6) is the job count ratio of each industry. The industries of manufacturing, retail trade, educational services, health care and accommodation and food services are with ratios above the average share. Column (7) shows the maximum job count of each industry. Column (8) shows the address of the maximum job count of each industry. 97 Table 3-12: The block group photo with maximum job counts of each industry 11 Agricultural North CA 21: Oil Houston 22: Utility LA 23: Construction Vegas 31-33: Manufacturing Seattle 11 Wholesale Trade NYC 44-45: Retail Trade New York 48-49: Transportation JFK 51: Information Seattle 52: Finance & Insurance NYC 53: Real Estate NYC 54: Professional service Chicago 98 Table 3-12 continued 55: Management of companies Arkansas 56: Admin and support Chicago 61: Educational service NYC 62: Health care Houston 71: Arts and entertain Orlando 72: Accommodation and food Vegas 81: Other service Seattle 92: Public admin NYC 99 For different industries, there is no way to predict ex ante for most industries in terms of whether an industry is nuisance creating or not. Even for manufacturing, some of them might be just like an office from the outside. Within each industry, there are smaller categories of the industries, the quality and ratio of which are uncertain. Thus, we can only draw inferences ex post about for which industries and which neighborhoods the relationship of the job counts and home- ownership rates are positive or negative. We perform the spatial analysis with distance for each industry for different income groups with FE specification. Table 3-13 and 3-14 present the results for the lower income groups. Table 3-13 shows the positive effect and Table 3-14 the negative impacts. Table 3-15 and 3-16 show the results of positive and negative impacts for higher income group respectively. We only select the significant results that have more than two significances across different income groups for reliability reasons. Additionally, we only present the results that don’t seem to suffer from the degree of freedom issues, like the ones with observation numbers less than 1,000 and with apparently inflated coefficients. 100 Table 3-13: The industry result for the lower income group: the positive effect part Real Estate and Rental and Leasing Income <5% <10% <15% <20% <25% <30% <35% <40% <45% <50% 1-2 1.2204 -0.9040 1.1882 0.7992 0.9096 0.8783 0.6407 0.5860 0.5271 0.1681 Std 0.7436 0.7684 0.5139 0.4041 0.3518 0.3157 0.2730 0.2470 0.2315 0.1906 P-value 0.1016 0.2404 0.0209 0.0481 0.0098 0.0054 0.0190 0.0177 0.0229 0.3779 Overall R 2 0.5513 0.2923 0.5588 0.3869 0.4143 0.3727 0.4056 0.5022 0.4664 0.5696 Obs 1205 1125 4774 6874 8916 10660 12874 14922 16656 19741 Manufacturing Income <5% <10% <15% <20% <25% <30% <35% <40% <45% <50% 1-2 1.2394 -3.6317 1.1167 0.6188 0.6862 0.6081 0.2622 0.2687 0.2196 0.2269 Std 0.7842 2.4079 0.5126 0.4073 0.3973 0.3606 0.3140 0.2859 0.2704 0.2313 P-value 0.1149 0.1325 0.0296 0.1288 0.0843 0.0918 0.4038 0.3474 0.4168 0.3267 Overall R 2 0.3117 0.0387 0.2928 0.5180 0.6186 0.2955 0.3383 0.5185 0.5129 0.4968 Obs 1186 1086 4749 6874 8916 10673 12911 14943 16662 19818 Information Income <5% <10% <15% <20% <25% <30% <35% <40% <45% <50% 0.3 0.3300 -0.3771 0.9917 1.0197 0.4546 0.3991 0.3242 0.1145 0.0277 0.0170 Std 0.7594 1.2636 0.4359 0.3529 0.3111 0.2845 0.2543 0.2312 0.2226 0.2009 P-value 0.6642 0.7655 0.0231 0.0039 0.1442 0.1608 0.2024 0.6206 0.9011 0.9324 Overall R 2 0.3528 0.1508 0.2969 0.3347 0.3701 0.3936 0.4133 0.3947 0.3938 0.4263 Obs 1142 1128 4606 6662 8674 10379 12569 14587 16214 19271 .3-1 1.1176 -0.7464 0.2796 0.3639 0.1844 0.3172 0.4978 0.3693 0.3827 0.4915 Std 0.6006 0.6156 0.4650 0.3533 0.2994 0.2791 0.2517 0.2246 0.2137 0.1907 P-value 0.0631 0.2257 0.5477 0.3031 0.5379 0.2558 0.0480 0.1002 0.0734 0.0100 Overall R 2 0.2904 0.2561 0.3025 0.4308 0.4212 0.4029 0.3529 0.2946 0.2857 0.2923 Obs 2851 2694 4439 6387 8266 9912 11919 13833 15436 18250 Accommodation and Food Services Income <5% <10% <15% <20% <25% <30% <35% <40% <45% <50% .3-1 0.0207 0.4451 0.3847 0.4581 0.4045 0.3619 0.2693 0.2532 0.2878 0.1588 Std 0.2873 0.4423 0.2541 0.2077 0.1680 0.1546 0.1443 0.1279 0.1237 0.1205 P-value 0.9427 0.3147 0.1304 0.0276 0.0161 0.0193 0.0621 0.0479 0.0200 0.1877 Overall R 2 0.5034 0.3545 0.6745 0.7328 0.7288 0.6782 0.6609 0.5636 0.6008 0.5600 Obs 3155 3029 4923 7124 9159 10968 13227 15310 17061 20206 3-5 0.6422 -4.8347 -0.1704 -0.2159 -0.4786 -0.5087 -0.4704 -0.3339 -0.2898 -0.1365 Std 0.6192 1.5644 0.3199 0.2808 0.2750 0.3039 0.2571 0.2064 0.2111 0.1847 P-value 0.3003 0.0022 0.5944 0.4421 0.0820 0.0942 0.0673 0.1059 0.1697 0.4597 Overall R 2 0.2392 0.4868 0.5643 0.6074 0.6382 0.6029 0.7094 0.6787 0.6657 0.6106 Obs 1287 1032 4801 6877 8924 10684 12946 14981 16733 19963 Note: These are FE with the dependent variable as the log of the job counts of different industries and the independent variable as the home-ownership rates of the donut ring under study. Analysis are performed for different income percentiles. Controls include homeownership rates of other circles, residential population, income, education, employee’s population density, employee’s education, whether the location is in downtown or suburbs, year and cbsa fixed effects. Standard errors are clustered at the job cluster level and are in parentheses. *** p<0.01, ** p<0.05, * p<0.1. 101 Table 3-14: The industry result for the lower income group: the negative effect part Retail Trade Income <5% <10% <15% <20% <25% <30% <35% <40% <45% <50% 2-3 0.6494 -0.2579 -0.3924 -0.2082 -0.4070 -0.4811 -0.3416 -0.3169 -0.2612 -0.1477 Std 0.2736 0.3872 0.4133 0.3317 0.2598 0.2428 0.1989 0.1781 0.1641 0.1380 P-value 0.0178 0.5055 0.3427 0.5302 0.1174 0.0477 0.0859 0.0752 0.1116 0.2846 Overall R 2 0.3511 0.8035 0.3106 0.3488 0.5161 0.5650 0.6373 0.6142 0.6125 0.7024 Obs 3133 2992 4925 7082 9192 10953 13305 15424 17223 20323 Wholesale Trade Income <5% <10% <15% <20% <25% <30% <35% <40% <45% <50% 0.3 -0.7869 1.2744 -0.5335 -0.4113 -0.5264 -0.5487 -0.4692 -0.3284 -0.2109 -0.1758 Std 0.5893 1.0292 0.3115 0.2411 0.2047 0.1791 0.1693 0.1589 0.1519 0.1373 P-value 0.1826 0.2164 0.0870 0.0882 0.0102 0.0022 0.0056 0.0388 0.1652 0.2005 Overall R 2 0.0902 0.2426 0.5700 0.6575 0.6749 0.6749 0.6319 0.5897 0.5823 0.6258 Obs 1257 1231 5022 7286 9481 11317 13744 15939 17694 21104 Other Services [except Public Administration] Income <5% <10% <15% <20% <25% <30% <35% <40% <45% <50% .3-1 -0.4892 0.5900 -1.0392 -0.8289 -0.6804 -0.5428 -0.3818 -0.3294 -0.2607 -0.1580 Std 0.5085 0.4474 0.3126 0.2518 0.2095 0.1854 0.1646 0.1483 0.1413 0.1255 P-value 0.3363 0.1876 0.0009 0.0010 0.0012 0.0035 0.0204 0.0264 0.0652 0.2080 Overall R 2 0.6429 0.6919 0.4056 0.4277 0.6119 0.6576 0.6685 0.5907 0.6315 0.5684 Obs 3163 3066 4948 7156 9210 11027 13302 15382 17150 20324 5-10 -0.0328 -1.0657 -1.1084 -0.1493 -0.3928 -0.3627 -0.4631 -0.7094 -0.6711 -0.5860 Std 0.4657 1.2532 0.8774 0.8148 0.5784 0.5131 0.4240 0.4011 0.3827 0.3207 P-value 0.9439 0.3955 0.2068 0.8546 0.4972 0.4797 0.2748 0.0771 0.0796 0.0677 Overall R 2 0.6939 0.3863 0.3614 0.4314 0.4772 0.4340 0.3941 0.3892 0.4088 0.3969 Obs 3153 2446 3947 5562 7279 8721 10558 12168 13606 16261 Management of Companies Income <5% <10% <15% <20% <25% <30% <35% <40% <45% <50% 3-5 -0.1760 -11.9211 0.4767 -0.5484 -1.7808 -2.0914 -1.5930 -1.4095 -1.4241 -1.2911 Std 2.7696 5.4911 1.1520 0.9375 0.9355 0.8918 0.7386 0.6886 0.6665 0.5494 P-value 0.9494 0.0310 0.6791 0.5586 0.0571 0.0191 0.0311 0.0408 0.0327 0.0188 Overall R 2 0.2550 0.0045 0.4519 0.3352 0.4359 0.4405 0.3777 0.3614 0.3983 0.3966 Obs 931 760 3424 4965 6416 7762 9384 10963 12357 14700 Arts, Entertainment Income <5% <10% <15% <20% <25% <30% <35% <40% <45% <50% 3-5 1.9625 1.3423 -3.7247 -2.1656 -1.5110 -1.3409 -0.9005 -0.6601 -0.8397 -0.8213 Std 1.9660 2.2796 1.1275 0.9165 0.7487 0.7083 0.6048 0.4993 0.5188 0.4017 P-value 0.3190 0.5566 0.0010 0.0183 0.0437 0.0585 0.1366 0.1863 0.1057 0.0409 Overall R 2 0.4261 0.1651 0.1835 0.2881 0.3501 0.2850 0.3411 0.3672 0.3212 0.2997 Obs 979 838 3781 5469 7129 8619 10506 12253 13790 16507 Public Administration Income <5% <10% <15% <20% <25% <30% <35% <40% <45% <50% 0.3 -0.4666 2.1519 -0.0552 0.3151 0.2049 -0.1369 -0.3999 -0.5125 -0.5350 -0.4047 Std 1.2816 1.4740 0.6328 0.4757 0.4189 0.3623 0.3173 0.2998 0.2832 0.2436 P-value 0.7160 0.1452 0.9305 0.5079 0.6248 0.7056 0.2075 0.0875 0.0589 0.0967 Overall R 2 0.3201 0.1113 0.2135 0.1945 0.1461 0.1436 0.1150 0.1079 0.1063 0.1304 Obs 1139 1151 4627 6754 8819 10521 12801 14880 16522 19791 1-2 -0.5872 2.2515 -0.7707 -0.2566 0.2248 0.5463 0.6352 0.6082 0.5716 0.5119 Std 1.1213 1.9391 0.6827 0.5788 0.4707 0.4871 0.4133 0.3718 0.3550 0.2915 P-value 0.6008 0.2465 0.2591 0.6576 0.6329 0.2621 0.1244 0.1019 0.1075 0.0791 Overall R 2 0.1850 0.1937 0.1406 0.1347 0.1414 0.0558 0.0712 0.0887 0.0947 0.1058 Obs 1139 1080 4495 6522 8484 10158 12294 14251 15940 19028 3-5 -2.2697 -0.6631 -0.1795 -0.5161 -0.4905 -0.7430 -1.1648 -1.0742 -1.2217 -0.7224 Std 1.5727 2.0853 0.9269 0.7484 0.6129 0.5723 0.5058 0.4765 0.4650 0.3932 P-value 0.1499 0.7508 0.8465 0.4905 0.4236 0.1943 0.0213 0.0242 0.0086 0.0662 Overall R 2 0.2103 0.0870 0.0965 0.0807 0.1006 0.1081 0.0924 0.0884 0.0842 0.0940 Obs 1128 955 4327 6229 8115 9707 11848 13756 15362 18375 5-10 -0.5286 -0.5406 -0.8891 -1.0546 -1.0184 -1.4632 -1.6502 -1.5054 -1.6403 -1.1563 Std 0.8415 1.6687 1.2247 1.1444 0.9393 0.8596 0.6874 0.6517 0.6259 0.5107 P-value 0.5301 0.7461 0.4680 0.3570 0.2785 0.0889 0.0164 0.0210 0.0088 0.0236 Overall R 2 0.1710 0.1170 0.0293 0.0298 0.0282 0.0223 0.0359 0.0319 0.0288 0.0310 Obs 2798 2247 3594 5048 6632 7932 9622 11092 12399 14850 Note: These are FE with the dependent variable as the log of the job counts of different industries and the independent variable as the home-ownership rates of the donut ring under study. Analysis are performed for different income percentiles. Controls include homeownership rates of other circles, residential population, income, education, employee’s population density, employee’s education, whether the location is in downtown or suburbs, year and cbsa fixed effects. Standard errors are clustered at the job cluster level and are in parentheses. *** p<0.01, ** p<0.05, * p<0.1. 102 Table 3-15: The industry result for the higher income group: the positive effect part Arts, Entertainment Income >55% >60% >65% >70% >75% >80% >85% >90% >95% 0.3 0.0035 -0.0215 0.1131 0.3133 0.3533 0.3196 0.7805 0.5787 -0.2355 Std 0.2182 0.2315 0.2521 0.2792 0.3114 0.3686 0.4223 0.5262 0.8587 P-value 0.9872 0.9261 0.6538 0.2619 0.2567 0.3861 0.0648 0.2718 0.7840 Overall R 2 0.4748 0.4875 0.5884 0.5157 0.4987 0.2809 0.1972 0.0971 0.0394 Obs 16041 14471 12513 10388 8703 6755 4751 2986 1236 Retail Trade Income >55% >60% >65% >70% >75% >80% >85% >90% >95% 0.3 -0.2952 -0.3604 -0.2367 -0.2325 -0.2677 -0.0743 -0.0415 -0.4366 0.0725 Std 0.1269 0.1356 0.1466 0.1637 0.1750 0.1884 0.2544 0.3103 0.5121 P-value 0.0200 0.0079 0.1065 0.1557 0.1263 0.6934 0.8703 0.1598 0.8875 Overall R 2 0.7781 0.7190 0.7704 0.8018 0.7076 0.6520 0.6148 0.2208 0.4077 Obs 17929 16143 13931 11489 9626 7396 5112 3173 1295 3-5 0.4844 0.5412 0.5887 0.4615 0.6692 0.4254 0.4867 0.7739 1.6528 Std 0.2459 0.2668 0.2888 0.3532 0.3835 0.4714 0.4266 0.5431 1.0379 P-value 0.0489 0.0426 0.0416 0.1915 0.0811 0.3670 0.2542 0.1546 0.1122 Overall R 2 0.6065 0.5489 0.5663 0.5502 0.5284 0.4919 0.5624 0.5591 0.4825 Obs 16948 15191 13147 10810 9027 6954 4851 3021 1242 Professional, Scientific, and Technical Services Income >55% >60% >65% >70% >75% >80% >85% >90% >95% 0.3 -0.3021 -0.2688 -0.3069 -0.2283 -0.2099 -0.3505 -0.5051 -0.6052 -0.3294 Std 0.1538 0.1502 0.1537 0.1646 0.1867 0.1978 0.2430 0.2927 0.4728 P-value 0.0496 0.0736 0.0459 0.1655 0.2610 0.0765 0.0378 0.0390 0.4865 Overall R 2 0.6737 0.6520 0.6562 0.5954 0.5854 0.4828 0.4098 0.2515 0.4520 Obs 17881 16103 13898 11460 9606 7398 5131 3190 1300 3-5 0.0824 0.2524 0.3152 0.6501 0.5702 0.1064 0.0611 0.5687 0.9495 Std 0.2588 0.2612 0.2869 0.3142 0.3473 0.3723 0.4187 0.5177 0.8629 P-value 0.7503 0.3339 0.2719 0.0386 0.1008 0.7751 0.8840 0.2723 0.2720 Overall R 2 0.5466 0.5718 0.5783 0.6328 0.5962 0.5850 0.6937 0.6495 0.5290 Obs 16897 15144 13116 10802 9014 6946 4859 3026 1248 Public Administration Income >55% >60% >65% >70% >75% >80% >85% >90% >95% 0.3 0.1426 0.2459 0.3470 0.3892 0.5922 1.0539 1.2567 1.9098 2.5184 Std 0.2592 0.2721 0.3030 0.3515 0.3913 0.4560 0.5942 0.7319 1.2446 P-value 0.5823 0.3661 0.2523 0.2683 0.1304 0.0209 0.0346 0.0092 0.0437 Overall R 2 0.0792 0.0810 0.1014 0.0798 0.0530 0.0269 0.0140 0.0296 0.1522 Obs 16566 14912 12844 10562 8809 6746 4649 2874 1172 5-10 -1.8143 -1.7007 -1.1631 -1.6393 -1.0353 -1.0320 -1.9044 -0.9634 1.8730 Std 0.7785 0.8270 0.8567 1.0175 1.0295 1.2533 1.5633 2.4628 3.4914 P-value 0.0198 0.0398 0.1747 0.1073 0.3147 0.4104 0.2235 0.6958 0.5921 Overall R 2 0.2668 0.2701 0.2563 0.2416 0.1966 0.1793 0.1702 0.0969 0.0299 Obs 12765 11427 9917 8230 6885 5245 3742 2330 982 Administrative and Support and Waste Income >55% >60% >65% >70% >75% >80% >85% >90% >95% .3-1 0.2315 0.3023 0.3741 0.3698 0.3499 0.3958 0.4064 0.0313 0.3556 Std 0.1931 0.2032 0.2272 0.2527 0.2722 0.3129 0.3795 0.4905 0.7200 P-value 0.2308 0.1369 0.0997 0.1435 0.1987 0.2061 0.2844 0.9491 0.6217 Overall R 2 0.4817 0.3923 0.3753 0.3467 0.3006 0.3489 0.2667 0.3864 0.1962 Obs 17175 15416 13342 11055 9232 7155 4955 3141 1284 2-3 -0.2796 -0.2668 -0.0920 -0.0981 -0.0804 -0.0182 -0.6731 -1.0021 -1.4998 Std 0.2742 0.2871 0.2774 0.2964 0.3183 0.3622 0.4320 0.5828 0.8345 P-value 0.3081 0.3529 0.7401 0.7408 0.8006 0.9600 0.1194 0.0859 0.0732 Overall R 2 0.5092 0.4852 0.4549 0.4147 0.4069 0.3247 0.4080 0.3380 0.2321 Obs 17230 15446 13339 10986 9222 7119 4952 3061 1258 Note: These are FE with the dependent variable as the log of the job counts of different industries and the independent variable as the home-ownership rates of the donut ring under study. Analysis are performed for different income percentiles. Controls include homeownership rates of other circles, residential population, income, education, employee’s population density, employee’s education, whether the location is in downtown or suburbs, year and cbsa fixed effects. Standard errors are clustered at the job cluster level and are in parentheses. *** p<0.01, ** p<0.05, * p<0.1. 103 Table 3-16: The industry result for the higher income group: the negative effect part Wholesale Trade Income >55% >60% >65% >70% >75% >80% >85% >90% >95% 0.3 -0.2526 -0.3072 -0.3199 -0.2824 -0.3293 -0.3071 -0.2403 -0.8375 -0.3117 Std 0.1768 0.1877 0.1964 0.2286 0.2659 0.2780 0.3347 0.4232 0.7005 P-value 0.1532 0.1018 0.1035 0.2168 0.2156 0.2694 0.4729 0.0481 0.6566 Overall R 2 0.8071 0.8310 0.7505 0.7957 0.7496 0.7127 0.5432 0.2151 0.4111 Obs 17828 16062 13867 11428 9582 7367 5104 3176 1286 Finance and Insurance Income >55% >60% >65% >70% >75% >80% >85% >90% >95% 0.3 -0.2314 -0.2597 -0.3130 -0.2672 -0.2697 -0.2322 -0.1962 -0.1522 -0.2282 Std 0.1476 0.1549 0.1660 0.1895 0.2139 0.2491 0.3186 0.4257 0.7722 P-value 0.1170 0.0937 0.0594 0.1586 0.2075 0.3512 0.5381 0.7209 0.7677 Overall R 2 0.6785 0.6277 0.5924 0.5314 0.4807 0.4425 0.3640 0.2129 0.6692 Obs 17462 15728 13558 11180 9366 7225 5040 3123 1290 1-2 -0.3670 -0.2851 -0.3122 -0.1418 -0.1608 0.0616 -0.2780 -0.5476 -0.6797 Std 0.1745 0.1753 0.1890 0.2165 0.2458 0.2886 0.3473 0.4339 0.6154 P-value 0.0355 0.1039 0.0986 0.5125 0.5131 0.8311 0.4236 0.2073 0.2701 Overall R 2 0.5164 0.4790 0.4886 0.4878 0.3854 0.5370 0.5323 0.6536 0.7534 Obs 17206 15453 13408 11097 9301 7166 4976 3130 1300 Manufacturing Income >55% >60% >65% >70% >75% >80% >85% >90% >95% .3-1 -0.3080 -0.4294 -0.4122 -0.5063 -0.3378 -0.4366 -0.5647 -0.7335 -0.9389 Std 0.1997 0.1863 0.2121 0.2439 0.2668 0.2970 0.3726 0.4641 0.7397 P-value 0.1230 0.0212 0.0520 0.0380 0.2055 0.1417 0.1299 0.1144 0.2052 Overall R 2 0.6530 0.5644 0.5567 0.5427 0.5448 0.3957 0.4149 0.2606 0.2586 Obs 16385 14712 12708 10508 8758 6797 4680 2947 1179 Management of Companies Income >55% >60% >65% >70% >75% >80% >85% >90% >95% 0.3 -0.1251 -0.1799 -0.2761 -0.0854 -0.2884 -0.3681 -1.4651 -1.5195 0.1483 Std 0.3400 0.3678 0.3922 0.4362 0.4813 0.5672 0.7013 0.7853 1.2061 P-value 0.7129 0.6248 0.4816 0.8448 0.5491 0.5164 0.0369 0.0534 0.9022 Overall R 2 0.7406 0.7486 0.7038 0.6685 0.6585 0.6359 0.6282 0.4216 0.5568 Obs 13933 12555 10853 8983 7459 5782 4069 2585 1051 .3-1 0.0542 0.0573 -0.3114 -0.5894 -0.7456 -1.0935 -1.7717 -2.5167 -3.0053 Std 0.3339 0.3473 0.3742 0.4133 0.4379 0.4753 0.6047 0.7308 1.0474 P-value 0.8710 0.8690 0.4054 0.1540 0.0888 0.0215 0.0035 0.0006 0.0044 Overall R 2 0.5424 0.5694 0.6087 0.5673 0.5586 0.4713 0.3267 0.3461 0.3721 Obs 13690 12325 10642 8874 7391 5773 4023 2582 1073 2-3 -0.6854 -0.6132 -0.8610 -1.1522 -0.9605 -0.8278 -1.1963 -1.4134 -3.3500 Std 0.4488 0.4817 0.4947 0.5690 0.6252 0.6919 0.7819 0.9102 1.4897 P-value 0.1268 0.2031 0.0819 0.0430 0.1246 0.2317 0.1263 0.1209 0.0252 Overall R 2 0.6395 0.6277 0.5802 0.5452 0.4699 0.4441 0.3637 0.2171 0.2010 Obs 13677 12240 10589 8716 7291 5633 3992 2476 1022 Real Estate and Rental and Leasing Income >55% >60% >65% >70% >75% >80% >85% >90% >95% 0.3 -0.2631 -0.3272 -0.3348 -0.4513 -0.3991 -0.4316 -0.6460 -0.7413 -0.5619 Std 0.1789 0.1867 0.2017 0.2276 0.2585 0.3041 0.3737 0.4651 0.7629 P-value 0.1414 0.0798 0.0971 0.0475 0.1227 0.1560 0.0841 0.1113 0.4618 Overall R 2 0.6272 0.6303 0.5998 0.6025 0.5689 0.5346 0.4141 0.0965 0.5382 Obs 17158 15442 13327 11018 9209 7083 4928 3070 1245 2-3 -0.4485 -0.4805 -0.4278 -0.4532 -0.5016 -0.4523 -0.9164 -0.3482 -0.6550 Std 0.2247 0.2427 0.2611 0.2762 0.3070 0.3585 0.4558 0.5532 0.7379 P-value 0.0460 0.0478 0.1014 0.1009 0.1025 0.2073 0.0446 0.5292 0.3753 Overall R 2 0.4789 0.4565 0.4434 0.4276 0.4022 0.3800 0.3761 0.3210 0.3884 Obs 16595 14856 12802 10555 8873 6862 4806 2990 1241 5-10 -0.5209 -0.6164 -0.5517 -0.8919 -1.2476 -1.5561 -1.1618 0.0781 0.8542 Std 0.5229 0.5505 0.5928 0.6789 0.6957 0.8412 0.9796 1.4506 2.0205 P-value 0.3193 0.2629 0.3521 0.1891 0.0731 0.0646 0.2359 0.9571 0.6728 Overall R 2 0.4089 0.4950 0.6347 0.5928 0.6793 0.6104 0.6451 0.3772 0.1976 Obs 13225 11827 10278 8533 7115 5440 3863 2396 1007 Educational Services Income >55% >60% >65% >70% >75% >80% >85% >90% >95% 0.3 -0.2476 -0.3360 -0.5403 -0.4458 -0.3491 -0.3131 -0.3874 -0.9858 -1.2292 Std 0.2260 0.2412 0.2610 0.2670 0.2895 0.3390 0.4305 0.4905 0.6685 P-value 0.2733 0.1638 0.0385 0.0951 0.2279 0.3558 0.3684 0.0448 0.0667 Overall R 2 0.2852 0.2652 0.2567 0.2898 0.2860 0.2573 0.2103 0.0875 0.2713 Obs 17114 15414 13299 10983 9190 7060 4901 3056 1244 104 5-10 -1.9208 -1.7253 -1.4807 -1.4596 -0.6327 -0.6401 -0.7740 -2.0279 -2.8050 Std 0.7868 0.7979 0.7807 0.8167 0.8200 0.9038 1.3454 2.0023 2.9720 P-value 0.0147 0.0307 0.0580 0.0740 0.4404 0.4790 0.5652 0.3116 0.3462 Overall R 2 0.2139 0.1933 0.2263 0.1456 0.1297 0.1069 0.1035 0.1463 0.0899 Obs 13095 11719 10171 8430 7041 5382 3835 2387 1007 Note: These are FE with the dependent variable as the log of the job counts of different industries and the independent variable as the home-ownership rates of the donut ring under study. Analysis are performed for different income percentiles. Controls include homeownership rates of other circles, residential population, income, education, employee’s population density, employee’s education, whether the location is in downtown or suburbs, year and cbsa fixed effects. Standard errors are clustered at the job cluster level and are in parentheses. *** p<0.01, ** p<0.05, * p<0.1. Table 3-17 shows the result summary. For the higher income group, the positive impact generating industries are mostly service oriented. While for lower income groups, the industries with positive impacts seem to be the ones generating jobs. Additionally, positive effects happen for adjacent and close distances for the lower income group. The Accommodation and Food Services exerts negative impacts for farther distance of 3-5. Thus, for lower income group, they might either not care much about the nuisance or they care about nuisance but care more about job opportunities such that the positive spillovers outweigh the oppositions. 105 Table 3-17: The result summary for the industry Positive Lower Income Higher Income Manufacturing (+:1-2) Retail Trade (-: .3/+: 3-5) Real Estate and Car Rental and Leasing (+:1-2) Arts, Entertainment (+:.3) Information (+:.3;+: .3-1) Public Administration (+: .3; -: 5-10) Accommodation and Food Services (+:.3-1;-: 3-5) Professional, Scientific, and Technical Services (-: .3/+: 3-5) Administrative and Support and Waste (+: .3-1; -: 2-3) Negative Lower Income Higher Income Management of Companies (3-5) Manufacturing (.3-1) Wholesale Trade (.3) Real Estate and Rental and Leasing (.3; 2-3; 5-10) Other Services [except Public Administration] (.3-1; 5-10) Wholesale Trade (.3) Retail Trade (2-3) Management of Companies (.3; .3-1; 2-3) Arts, Entertainment (3-5) Finance and Insurance (.3; 1-2) Public Administration (.3; 1-2: 3-10) Educational Services (.3; 5-10) Note: this table summarizes the results for different industries. Take the Manufacturing for an example, the result means that there is a positive effect from home-owning on Manufacturing job counts for the distance of 1-2 miles for the lower income group and there is a negative impact from home-owning on Manufacturing job counts for the distance of .3-1 miles for the higher income group. For the higher income group, the Arts and Entertainment generates a positive impact for the most adjacent distance (.3). While for Retail Trade, Professional, Scientific, and Technical Services, they both have a negative impact for the adjacent distance of within .3 miles but a positive impact for farther distances of 3-5 miles. This might be because that these two industries are liked by the high-income neighborhoods but creates nuisance for adjacent distances. They also like the Administrative and Support and Waste and Public Administration to be close but not far. For the negative impact part, we can see that the Manufacturing and Real Estate and Rental and Leasing are the positive impact generating industries for the lower income groups and at the same time the negative impact generating industries for the higher income group. On the other hand, three positive impact generating industries for the higher income group, Arts and 106 Entertainment, Retail Trade, and Public Administration, are the negative impact generating industries for the lower income groups. The lower and higher income groups also have similarities. Both groups have negative impacts for Wholesale Trade and Management of Companies. For higher income groups, negative impacts are all for adjacent distances of .3 or .3-1, again suggesting the NIMBY effect at adjacent distances. 6. Conclusion Business and residents are coexistent. One residential area is equipped with jobs, retails, hospitals and schools to service the area. Business also locates in places with employees and customers. While the distance between them might create different impacts. They tend to be near but being too close might create frictions. We perform a series of analyses on the spatial relationship between home-owning and business. For the pooled sample, we find negative significance at the distance of within .3 miles and positive significance at the distance of 3-5 miles. This partially agrees with our hypothesis that for adjacent distances, businesses are most likely to suffer from the NIMBY effect but for farther but close distances, the positive impact rules. The analysis for different income percentiles suggests that higher income groups exert the negative impacts for the .3-1 miles while the lagged positive effects are from the lower income groups for the distance of .3 and 3-5 miles. For different industries, it is interesting that some of the positive impact generating industries for the low-income groups are at the same time the negative impact generating ones for the high-income ones and vice versa. The preferred businesses for the higher income groups are mostly service oriented, like the Retail and Arts and Entertainment. While for the lower income 107 groups, the ones that generate jobs have positive impacts, like the Manufacturing, Real Estate and Car Rental and Leasing. In conclusion, the micro level analysis indeed finds the evidence of NIMBY effect for the high-income groups from the industries they don’t like at adjacent distances. At the same time, we also discover some YIMBY and labor market effect that the impacts are positive for lower income neighborhoods for the industries that will generate job opportunities. This might provide insights on the appropriate combinations of industries and neighborhoods to avoid unnecessary frictions but to achieve the optimal positive interactions between the business and residents. 108 7. References Blanchflower, D., and A. Oswald (2013). Does High Home-Ownership Impair the Labor Market? (Working Paper No. 19079). Retrieved from National Bureau of Economic Research website: http://www.nber.org/papers/w19079 Baum-Snow, Nathaniel. (2007). Did Highways Cause Suburbanization? Quarterly Journal of Economics, 122(2): 775-805. Campbell, John Y., Stefano Giglio and Parag Pathak. (2011). Forced Sales and House Prices. American Economic Review, 101(5), 2108-31. Coulson, N., and Lynn M. Fisher. (2002). Tenure Choice and Labour Market Outcomes. Housing Studies,17(1), 35-49. doi:10.1080/02673030120105875 Coulson, N., and Lynn M. Fisher. (2009). Housing tenure and labor market impacts: The search goes on. Journal of Urban Economics, 65(3), 252-264. doi:10.1016/j.jue.2008.12.003 Dinkelman, Taryn. (2011). The Effects of Rural Electrification on Employment: New Evidence from South Africa, American Economic Review, 101(7): 3078-3108. DiPasquale, Denise and Edward L. Glaeser (1999). Incentives and Social Capital: Are Homeowners Better Citizens?, Journal of Urban Economics, 45 (March): 354-84. Duranton, Gilles and Matthew A. Turner. (2011). The Fundamental Law of Road Congestion: Evidence from the US. American Economic Review, 101(6): 2616-2652. Duranton, Gilles and Matthew A. Turner. (2012). Urban Growth and Transportation. Review of Economic Studies, 79 (4): 1407-1440. Farber, H.S. (1999). Mobility and stability: the dynamics of job change in labor markets. In: Ashenfelter, O.C., Card, D. (Eds.). Handbook of Labor Economics, vol. 3B Elsevier, 109 Amsterdam, The Netherlands, pp. 2439– 2483. Ferreira, Fernando, Joseph Gyourko and Joseph Tracy. (2010). Housing Busts and Household Mobility. Journal of Urban Economics, 68(1), 34-45. Fischel, W. (2004). An Economic History of Zoning and a Cure for its Exclusionary Effects. Urban Studies, 41( 2), 317–340. Flatau, P., M. Forbes, P. Hendershott, and G. Wood. (2003). Homeownership and unemployment: The roles of leverage and public housing (Working Paper No. 10021). Retrieved from National Bureau of Economic Research website: http://www.nber.org/papers/w10021 Giuliano, Genevieve and Kenneth Small. (1991). Subcenters in the Los Angeles Region. Regional Science and Urban Economics, 21(2), 163-182. Glaeser, Ed and Joseph Gyourko. (2017). The Economic Implications of Housing Supply. (Working Paper No. 23833). Retrieved from National Bureau of Economic Research website: http://realestate.wharton.upenn.edu/wp-content/uploads/2017/03/802.pdf Goss, E., Phillips, J., Summer. ( 1997). The impact of homeownership on the duration of unemployment. The Review of Regional Studies, 27 (1), 9–27. Green, Richard K. and Bingbing Wang. (2015). Housing Tenure and Unemployment. Available at SSRN: https://ssrn.com/abstract=2628242 or http://dx.doi.org/10.2139/ssrn.2628242 Hoxby, Caroline. (2000). Does competition among public schools benefit students and taxpayers? American Economic Review, 90(5): 1209-1238. Liu, Cathy and Gary Painter. (2012). Immigrant Settlement and Employment Suburbanization: Is There a Spatial Mismatch?. Urban Studies, 49(5) 979–1002. Mumford, K., and K. Schultz (2009). The Effect of Underwater Mortgages on Unemployment. Retrieved from 110 http://www.krannert.purdue.edu/faculty/kjmumfor/papers/Underwater_and_Unemployed pdf Munch, J., M. Rosholm, and M. Svarer. (2006). Are Homeowners Really More Unemployed?. Economic Journal, 116(514), 991-1013. doi:10.1111/j.1468-0297.2006.01120 Schulhofer-Wohl, Sam. (2011). Negative Equity Does Not Reduce Homeowners’ Mobility (Working Paper No. 16701). Retrieved from National Bureau of Economic Research website: http://www.nber.org/papers/w16701 Stephens, Josh. (2017). Planning's New Rivalry: Housing Advocates vs. Radical Left. Waddell, Paul and Vibhootwe Shukla. (1993). Manufacturing Location in a Polycentric Urban Area: A Study in the Composition and Attractiveness of Employment Subcenters, Urban Geography, 14, 277–296. 111 8. Appendix 8.1. K- means clustering method Given a set of observations (X1, X2, ….., Xn), where each observation is a d-dimensional real vector, k-means clustering aims to partition the n observations into k(<=n) sets={S1, S2, 。。。。Sk} so as to minimize the within-cluster sum of squares (WCSS) (sum of distance functions of each point in the cluster to the K center). In other words, its objective is to find:      k i Si X i S X 1 2 || || min arg  ɥi is the mean of points in Si. So from the algorithms, the choice of k matters for the result. In the paper, we choose k to be 10,000. So altogether, we cluster the business block groups to form 10,000 business cluster in the US. k-means clustering is a commonly used method in cluster analysis, like grouping the observations with respective to their similarities of a certain aspect, like age, employment density, 112 and industry types. For my paper, my aspect for similarity is the location and it calculates the distances of the location for identifying their location similarity/adjacency. The relationship of K-means clustering to the names of cluster methods mentioned like hotspot is that K-means clustering is an algorithm for hot spot detection or adjacency detection in clustering (Lawson, 2010; Fisher, 1958; Bailey and Gatrell, 1995; Levine, 1999a). Hotspot is widely used in geography, ecology and spatial economics. It is a clustering method and clustering is an aggregation of unusual events (Andrew B., 2010). The application of the hotspot clustering in spatial economics is to firstly identify the hot spot areas and secondly to group the geographically close ones together. The geographically adjacency in the hotspot is commonly achieved by K-means. The difference between the hotspot and my method used for clustering is the identification standards, employment density versus the ratio of job counts over housing units. This standard is decided from the research objective. In ArcGIS, they belong to two categories, grouping and hotspot. But ArcGIS tends to name the clustering methods by the purpose as opposed to the algorithms used for clustering. Thus, my method used in the paper is just a clustering method and could even be seen as a hotspot (not quite appropriate by naming because hot spot means they are hot in some sense like most affected crime or disease area so more commonly related to density) method that groups the geographically adjacent block groups with more job counts than housing units. Thus, the alternative of K-means clustering is not the hot spot method or the Giuliano and Small method. Besides the k-means, there are other clustering methods. Before the discussion on other clustering methods, we first lay out the advantages and disadvantages of k-means clustering (Santini, 2016): Advantages: 113 1) Easy to perform 2) Time efficient 3) Produce tighter clusters Disadvantages: 1) Sensitive to outliers 2) Need to decide on k initially: The k is set to 10,000 in the paper. I checked by changing the 10,000 to 8,000, 9,000 and 12,000 and the 10,000 performs the best. The best means that I compare the results on the job clusters I am familiar with like Orange County and USC neighbors and I found the 10,000 matches better with realities or matches better if I used the Giuliano and Small (1991) method. 3) Sensitive to the initial seeds 4) This method might not perform well on deciding the cluster boundaries associated with different densities. But we are not concerned about density difference in the clusters but the difference between residence and business so this does not create a serious problem. The alternatives to k-means are: (from http://www.sthda.com/english/articles/25-cluster- analysis-in-r-practical-guide/111-types-of-clustering-methods-overview-and-quick-start-r-code/) 1) If we want to do partitioning clustering, we can try k-medians, k-medoids, or PAM (Partitioning Around Medoids, Kaufman & Rousseeuw, 1990), which is less sensitive to outliers compared to k-means. The k-medians minimized the absolute deviations as opposed to the within-cluster variance minimized in k-means. Thus, if the clustering does not suffer a lot from outliers, the accuracy improvement of 114 these three is limited. For my analysis, seen from the Orange County mapping, I did not discover an outlier problem for the grouping. A disadvantage of these methods is that they take a much longer time than k-means, which is why k-means is more commonly used both in industry and academia. It is like we do OLS as opposed to median regressing. 2) Hierarchical clustering: this does not need to pre-specify the number of clusters to be generated. 3) Fuzzy clustering: items could belong to more than one clusters. This does not seem to be appropriate for my study. 4) Model-based clustering: this is not suitable. 5) Density-based clustering: better for data with noise and outliers but the purpose of this method is to cluster the areas with similar densities. My research focuses on differentiation between jobs and residence as opposed to the density perspective. For the above clustering methods, some can be done in statistical software and some can’t. The k-means is more commonly used in both industry and academia and other methods take much longer time to perform. According to my knowledge, ArcGIS could do the k-means and density based although it has more names than this but their naming is more related to purposes as opposed to clustering algorithms. 115 8.2. K-mean clustering mapping 1. The mapping of Orange County The above graph shows the ratio of job counts to owners in Orange County, with higher ratios represented by deeper blues. And the below graph shows the job clusters in Orange County. The different color just shows that they are of different clusters. 116 117 2. The mapping of the neighborhood business cluster above the University of Southern California 1) Below is the geoid information for cluster 4228. Geoid is the block group number. c000 is the job counts for this block group. geoid c000 cbsa lat lon landarea cluster 60372098102 391 31100 34.04816 -118.283 70446 4228 60372211202 743 31100 34.04205 -118.286 191940 4228 60372212202 878 31100 34.04307 -118.297 443972 4228 60372216021 494 31100 34.03792 -118.297 277474 4228 60372216022 306 31100 34.03761 -118.288 147917 4228 60372216023 270 31100 34.03909 -118.288 97426 4228 60372242001 870 31100 34.04111 -118.276 191666 4228 60372242002 178 31100 34.03848 -118.278 176310 4228 60372243201 678 31100 34.04294 -118.28 150184 4228 60372244101 2464 31100 34.03525 -118.277 246890 4228 60372244102 718 31100 34.03573 -118.282 206817 4228 60372244202 745 31100 34.03003 -118.277 222900 4228 60372247002 963 31100 34.02955 -118.282 235489 4228 2) Below is the mapping of them on google map with blue polygons: 118 119 3) Below is the geoid information for cluster 9228. Geoid is the block group number. c000 is the job counts for this block group. geoid c000 cbsa lat lon landarea cluster 60372219002 488 31100 34.02633 -118.294 141337 9228 60372227001 18945 31100 34.01941 -118.284 817399 9228 60372246001 530 31100 34.02066 -118.269 228576 9228 60372246002 2170 31100 34.02265 -118.274 490348 9228 60372247001 708 31100 34.02357 -118.281 204166 9228 60372247003 809 31100 34.02514 -118.277 196165 9228 60372311001 1450 31100 34.01463 -118.28 363898 9228 60372311002 2364 31100 34.01509 -118.276 401096 9228 60372311003 323 31100 34.01429 -118.273 129064 9228 60372312202 2053 31100 34.01446 -118.291 785974 9228 60372316001 456 31100 34.00948 -118.295 179145 9228 60372317202 232 31100 34.00662 -118.281 133543 9228 4) The actual mapping on Google Map: 120 121 8.3. Detailed description of each industry NAICS Code Description NAICS sector 11 (Agriculture, Forestry, Fishing and Hunting) The Agriculture, Forestry, Fishing and Hunting sector comprises establishments primarily engaged in growing crops, raising animals, harvesting timber, and harvesting fish and other animals from a farm, ranch, or their natural habitats. The establishments in this sector are often described as farms, ranches, dairies, greenhouses, nurseries, orchards, or hatcheries. A farm may consist of a single tract of land or a number of separate tracts which may be held under different tenures. For example, one tract may be owned by the farm operator and another rented. It may be operated by the operator alone or with the assistance of members of the household or hired employees, or it may be operated by a partnership, corporation, or other type of organization. When a landowner has one or more tenants, renters, croppers, or managers, the land operated by each is considered a farm. NAICS sector 21 (Mining, Quarrying, and Oil and Gas Extraction) The Mining sector comprises establishments that extract naturally occurring mineral solids, such as coal and ores; liquid minerals, such as crude petroleum; and gases, such as natural gas. The term mining is used in the broad sense to include quarrying, well operations, beneficiating (e.g., crushing, screening, washing, and flotation), and other preparation customarily performed at the mine site, or as a part of mining activity. NAICS sector 22 (Utilities) The Utilities sector comprises establishments engaged in the provision of the following utility services: electric power, natural gas, steam supply, water supply, and sewage removal. Within this sector, the specific activities associated with the utility services provided vary by utility: electric power includes generation, transmission, and distribution; natural gas includes distribution; steam supply includes provision and/or distribution; water supply includes treatment and distribution; and sewage removal includes collection, treatment, and disposal of waste through sewer systems and sewage treatment facilities. NAICS sector 23 (Construction) The construction sector comprises establishments primarily engaged in the construction of buildings or engineering projects (e.g., highways and utility systems). Establishments primarily engaged in the preparation of sites for new construction and establishments primarily engaged in subdividing land for sale as building sites also are included in this sector. NAICS sector 31-33 (Manufacturing) The Manufacturing sector comprises establishments engaged in the mechanical, physical, or chemical transformation of materials, substances, or components into new products. Establishments in the Manufacturing sector are often described as plants, factories, or mills and characteristically use power-driven machines and materials-handling equipment. However, establishments that transform materials or substances into new products by hand or in the worker's home and those engaged in selling to the general public products made on the same premises from which they are sold, such as bakeries, candy stores, and custom tailors, may also be included in this sector. Manufacturing establishments may process materials or may contract with other establishments to process their materials for them. Both types of establishments are included in manufacturing. NAICS sector 42 (Wholesale Trade) The Wholesale Trade sector comprises establishments engaged in wholesaling merchandise, generally without transformation, and rendering services incidental to the sale of merchandise. The merchandise described in this sector includes the outputs of agriculture, mining, manufacturing, and certain information industries, such as publishing. The wholesaling process is an intermediate step in the distribution of merchandise. Wholesalers are organized to sell or arrange the purchase or sale of (a) goods for resale (i.e., goods sold to other wholesalers or retailers), (b) capital or durable nonconsumer goods, and (c) raw and intermediate materials and supplies used in production. Wholesalers sell merchandise to other businesses and normally operate from a warehouse or office. These warehouses and offices are characterized by having little or no display of merchandise. In addition, neither the design nor the location of the premises is intended to solicit walk-in traffic. Wholesalers do not normally use advertising directed to the general public. Customers are generally reached initially via telephone, in-person marketing, or by specialized advertising that may include Internet and other electronic means. Follow-up orders are either vendor-initiated or client-initiated, generally based on previous sales, and typically exhibit strong ties between sellers and buyers. In fact, transactions are often conducted between wholesalers and clients that have long-standing business relationships. NAICS sector 44-45 (Retail Trade) The Retail Trade sector comprises establishments engaged in retailing merchandise, generally without transformation, and rendering services incidental to the sale of merchandise. The retailing process is the final step in the distribution of merchandise; retailers are, therefore, organized to sell merchandise in small quantities to the general public. This sector comprises two main types of retailers: store and nonstore retailers. 1. Store retailers operate fixed point-of-sale locations, located and designed to attract a high volume of walk-in customers. In general, retail stores have extensive displays of merchandise and use mass-media advertising to attract customers. They typically sell merchandise to the general public for personal or household consumption, but some also serve business and institutional clients. In addition to retailing merchandise, some types of store retailers are also engaged in the provision of after-sales services, such as repair and installation. 122 2. Nonstore retailers, like store retailers, are organized to serve the general public, but their retailing methods differ. The establishments of this subsector reach customers and market merchandise with methods, such as the broadcasting of "infomercials," the broadcasting and publishing of direct-response advertising, the publishing of paper and electronic catalogs, door-to-door solicitation, in-home demonstration, selling from portable stalls (street vendors, except food), and distribution through vending machines. NAICS sector 48-49 (Transportation and Warehousing) The Transportation and Warehousing sector includes industries providing transportation of passengers and cargo, warehousing and storage for goods, scenic and sightseeing transportation, and support activities related to modes of transportation. Establishments in these industries use transportation equipment or transportation related facilities as a productive asset. The type of equipment depends on the mode of transportation. The modes of transportation are air, rail, water, road, and pipeline. NAICS sector 51 (Information) The Information sector comprises establishments engaged in the following processes: (a) producing and distributing information and cultural products, (b) providing the means to transmit or distribute these products as well as data or communications, and (c) processing data. The main components of this sector are the publishing industries, including software publishing, and both traditional publishing and publishing exclusively on the Internet; the motion picture and sound recording industries; the broadcasting industries, including traditional broadcasting and those broadcasting exclusively over the Internet; the telecommunications industries; Web search portals, data processing industries, and the information services industries. The Information sector groups three types of establishments: (1) those engaged in producing and distributing information and cultural products; (2) those that provide the means to transmit or distribute these products as well as data or communications; and (3) those that process data. NAICS sector 52 (Finance and Insurance) The Finance and Insurance sector comprises establishments primarily engaged in financial transactions (transactions involving the creation, liquidation, or change in ownership of financial assets) and/or in facilitating financial transactions. Three principal types of activities are identified: 1. Raising funds by taking deposits and/or issuing securities and, in the process, incurring liabilities. Establishments engaged in this activity use raised funds to acquire financial assets by making loans and/or purchasing securities. Putting themselves at risk, they channel funds from lenders to borrowers and transform or repackage the funds with respect to maturity, scale, and risk. This activity is known as financial intermediation. 2. Pooling of risk by underwriting insurance and annuities. Establishments engaged in this activity collect fees, insurance premiums, or annuity considerations; build up reserves; invest those reserves; and make contractual payments. Fees are based on the expected incidence of the insured risk and the expected return on investment. 3. Providing specialized services facilitating or supporting financial intermediation, insurance, and employee benefit programs. In addition, monetary authorities charged with monetary control are included in this sector. NAICS sector 53 (Real Estate and Rental and Leasing) The Real Estate and Rental and Leasing sector comprises establishments primarily engaged in renting, leasing, or otherwise allowing the use of tangible or intangible assets, and establishments providing related services. The major portion of this sector comprises establishments that rent, lease, or otherwise allow the use of their own assets by others. The assets may be tangible, as is the case of real estate and equipment, or intangible, as is the case with patents and trademarks. This sector also includes establishments primarily engaged in managing real estate for others, selling, renting and/or buying real estate for others, and appraising real estate. These activities are closely related to this sector's main activity, and it was felt that from a production basis they would best be included here. In addition, a substantial proportion of property management is self-performed by lessors. The main components of this sector are the real estate lessors industries (including equity real estate investment trusts (REITs)); equipment lessors industries (including motor vehicles, computers, and consumer goods); and lessors of nonfinancial intangible assets (except copyrighted works). NAICS sector 54 (Professional, Scientific, and Technical Services) The Professional, Scientific, and Technical Services sector comprises establishments that specialize in performing professional, scientific, and technical activities for others. These activities require a high degree of expertise and training. The establishments in this sector specialize according to expertise and provide these services to clients in a variety of industries and, in some cases, to households. Activities performed include: legal advice and representation; accounting, bookkeeping, and payroll services; architectural, engineering, and specialized design services; computer services; consulting services; research services; advertising services; photographic services; translation and interpretation services; veterinary services; and other professional, scientific, and technical services. NAICS sector 55 (Management of Companies and Enterprises) The Management of Companies and Enterprises sector comprises (1) establishments that hold the securities of (or other equity interests in) companies and enterprises for the purpose of owning a controlling interest or influencing management decisions or (2) establishments (except government establishments) that administer, oversee, and manage establishments of the company or enterprise and that normally undertake the strategic or organizational planning and decision making role of the company or enterprise. Establishments that administer, 123 oversee, and manage may hold the securities of the company or enterprise. Establishments in this sector perform essential activities that are often undertaken, in-house, by establishments in many sectors of the economy. By consolidating the performance of these activities of the enterprise at one establishment, economies of scale are achieved. NAICS sector 56 (Administrative and Support and Waste Management and Remediation Services) The Administrative and Support and Waste Management and Remediation Services sector comprises establishments performing routine support activities for the day-to-day operations of other organizations. These essential activities are often undertaken in-house by establishments in many sectors of the economy. The establishments in this sector specialize in one or more of these support activities and provide these services to clients in a variety of industries and, in some cases, to households. Activities performed include: office administration, hiring and placing of personnel, document preparation and similar clerical services, solicitation, collection, security and surveillance services, cleaning, and waste disposal services. NAICS sector 61 (Educational Services) The Educational Services sector comprises establishments that provide instruction and training in a wide variety of subjects. This instruction and training is provided by specialized establishments, such as schools, colleges, universities, and training centers. These establishments may be privately owned and operated for profit or not for profit, or they may be publicly owned and operated. They may also offer food and/or accommodation services to their students. Educational services are usually delivered by teachers or instructors that explain, tell, demonstrate, supervise, and direct learning. Instruction is imparted in diverse settings, such as educational institutions, the workplace, or the home, and through diverse means, such as correspondence, television, the Internet, or other electronic and distance-learning methods. The training provided by these establishments may include the use of simulators and simulation methods. It can be adapted to the particular needs of the students, for example sign language can replace verbal language for teaching students with hearing impairments. All industries in the sector share this commonality of process, namely, labor inputs of instructors with the requisite subject matter expertise and teaching ability. NAICS sector 62 (Health Care and Social Assistance) The Health Care and Social Assistance sector comprises establishments providing health care and social assistance for individuals. The sector includes both health care and social assistance because it is sometimes difficult to distinguish between the boundaries of these two activities. The industries in this sector are arranged on a continuum starting with those establishments providing medical care exclusively, continuing with those providing health care and social assistance, and finally finishing with those providing only social assistance. The services provided by establishments in this sector are delivered by trained professionals. All industries in the sector share this commonality of process, namely, labor inputs of health practitioners or social workers with the requisite expertise. Many of the industries in the sector are defined based on the educational degree held by the practitioners included in the industry. NAICS sector 71 (Arts, Entertainment, and Recreation) The Arts, Entertainment, and Recreation sector includes a wide range of establishments that operate facilities or provide services to meet varied cultural, entertainment, and recreational interests of their patrons. This sector comprises (1) establishments that are involved in producing, promoting, or participating in live performances, events, or exhibits intended for public viewing; (2) establishments that preserve and exhibit objects and sites of historical, cultural, or educational interest; and (3) establishments that operate facilities or provide services that enable patrons to participate in recreational activities or pursue amusement, hobby, and leisure-time interests. Some establishments that provide cultural, entertainment, or recreational facilities and services are classified in other sectors. NAICS sector 72 (Accommodation and Food Services) The Accommodation and Food Services sector comprises establishments providing customers with lodging and/or preparing meals, snacks, and beverages for immediate consumption. The sector includes both accommodation and food services establishments because the two activities are often combined at the same establishment. NAICS sector 81 (Other Services [except Public Administration]) The Other Services (except Public Administration) sector comprises establishments engaged in providing services not specifically provided for elsewhere in the classification system. Establishments in this sector are primarily engaged in activities, such as equipment and machinery repairing, promoting or administering religious activities, grantmaking, advocacy, and providing drycleaning and laundry services, personal care services, death care services, pet care services, photofinishing services, temporary parking services, and dating services. NAICS sector 92 (Public Administration) The Public Administration sector consists of establishments of federal, state, and local government agencies that administer, oversee, and manage public programs and have executive, legislative, or judicial authority over other institutions within a given area. These agencies also set policy, create laws, adjudicate civil and criminal legal cases, provide for public safety and for national defense. In general, government establishments in the Public Administration sector oversee governmental programs and activities that are not performed by private establishments. Establishments in this sector typically are engaged in the organization and financing of the production of public goods and services, most of which are provided for free or at prices that are not economically significant. Government establishments also engage in a wide range of productive activities covering not only public goods and services but also individual goods and services similar to those produced in sectors typically identified with private-sector establishments. In general, ownership is not a criterion for classification in NAICS. Therefore, government establishments engaged in the production of private-sector-like goods and services should be 124 classified in the same industry as private-sector establishments engaged in similar activities. As a practical matter, it is difficult to identify separate establishment detail for many government agencies. To the extent that separate establishment records are available, the administration of governmental programs is classified in Sector 92, Public Administration, while the operation of that same governmental program is classified elsewhere in NAICS based on the activities performed. For example, the governmental administrative authority for an airport is classified in Industry 92612, Regulation and Administration of Transportation Programs, while operating the airport is classified in Industry 48811, Airport Operations. When separate records are not available to distinguish between the administration of a governmental program and the operation of it, the establishment is classified in Sector 92, Public Administration. Examples of government-provided goods and services that are classified in sectors other than Public Administration include: schools, classified in Sector 61, Educational Services; hospitals, classified in Subsector 622, Hospitals; establishments operating transportation facilities, classified in Sector 48-49, Transportation and Warehousing; the operation of utilities, classified in Sector 22, Utilities; and the Government Printing Office, classified in Subsector 323, Printing and Related Support Activities. 125 Chapter 4. The Effect of Walmart and Whole Foods on Nearby Property Values 1. Introduction A large literature has explored the effect of Walmart on markets. This is in part because of the size of the chain, but also because Walmart has been alleged to have a number of negative spillover effects on obesity (Courtemanche and Carden, 2010), crime rates (Wolfe and Pyrooz, 2014), pollution and poverty (Goetz and Swaminathan, 2006), Mom-and-Pop business (Haltiwanger, Jarmin and Krizan, 2010), job losses (Merriman, Persky and Davis, 2012) and wages (Neumark et al, 2008). This paper revisits the question of how Walmart affects residential real estate markets, because previous work (Popes, 2015; Slade, 2018) suffers from methodological shortcomings that may bias the results and lead to incorrect inferences. In particular, the old method suffers from two problems: 1) it assumes the properties used as control and treatment are similar only except their distance to the store. This creates a problem that Walmart sorts into certain neighborhoods with (dis)amenities, which would affect the property values differently for the controls and treatments. The old method uses store FE (as opposed to property FE) of time and space to control for the compounding factors. But even individual property level FE can’t control for time-variant unobservables and the unobservables that don’t affect the observations equally across time; 2) it is based on the assumption that the house price trends are the same for the properties used as control (far) and treatments (close) for before and after had the store not been built, which is not true. We apply a revised difference-in-differences approach, introduced in Diamond and McQuade (2015), that improves these two problems by: 1) finding the controls and treatments in 126 a small local area to ensure their similarity in land prices and neighborhood features; 2) controlling for the house characteristics including whether it has cooling, heating, a pool and a new fireplace, living square footage, number of rooms, baths and stories, the age of the house, property condition, quality, garage types and the transfer dates. The results differ from the prior studies. In particular, we find the effects of Walmart are negative with the largest effect of 30% decrease for the distance of 1.5 kilometers (.94 miles) in the year right after Walmart is built and the smallest effect of 10% decrease for the adjacent distance (less then .5 kilometers (.3 miles)) in the tenth year after opening date. Further analysis suggests that biases from the unadjusted methodology are behind the difference in results. This study also explores the dynamic between residential real estate and commercial real estate markets. We look to establish whether such markets discriminate within a product class based on differences in underlying business models across companies. We compare market reactions to the introduction of Walmart and Whole Foods grocery stores in California to answer this question. Here are what we find, in contrast with the result of Walmart, Whole Foods exerts positive effects on house prices. The largest effect of 15% increase happens at the distance of 2 kilometers (1.25 miles) in the year when Whole Foods opens. The lowest effect of around 0% exists for the adjacent distance of year 3 to 10 after openning. The similarity with the two results are that the most hit area (with the large negative and positive impacts) are for the distance of 1.5 to 2.5 miles (.94-1.56 miles) for the first five years after the openning. The rest of the paper is as follows. In Section 2, we will introduce the methodolgy, the new dif-in-dif for estimating the effect of Walmart and Whole Foods on nearby property values. Section 3 describes the data we use. Section 4 provides the results. And Section 5 concludes. 127 2. Methodology Our aim is to quantify the effect of the two stores on house values varying with distance between them. Location choice of stores represents their choice sets of profit maximization, the preferences of proximity to their targeted customers, the local population density, education, income, access to transportation hubs, locational physical characteristics like parking space, whether it is a stand lone space, whether it is visible directly off the street, considerations for land prices and local (dis)amenities. Residential house values, on the other hand, are the solution to the utility maximization of house buyers subject to their budget constraint. The house value can be thought of measuring the local amenity as well as the house characteristics (Rosen, 1974). The presence of the two stores in proximity constitutes parts of the local (dis)amenities. The relationship of property values with the local (dis)amenities suffers from a common issue evaluating the effect of amenities on house values that the un-observables in the location choice of the two retailers might be correlated with property values. Eq. (1) describes the house price determinants including neighborhood characteristics and housing features. LPn,j,t represents the log of house prices for house j, in neighborhood n, at period t as a function of local characteristics f , housing characteristics X and the effect of the retailers, g, with distance r and years after the opening date δ. f is a mapping on local transportation, school qualities, and amenities. If there are unobservables in ԑn,j,t that is correlated with g, Eq. (1) could not identify the effect of proximity to retailers on housing prices. The retailer might choose the places with higher property values thus violating the exogeneity condition. , , , , , , , , , ( , ) n j t n j t j t n j t n j t LP f X g r     (1) 128 Due to the endogeneity of stores’ location choice, we could not compare the house price values of the neighborhoods with the store with those without. Former literatures (Linden and Rockoff, 2008; Campbell, Giglio and Pathak, 2011; Autor, Palmer and Pathak, 2014) perform difference in difference analysis by treating the properties closer to factories/projects as treatment while using properties farther as controls. The choice of the distance within which the properties are treatments is arbitrary however. Diamond and McQuade (2016) render the distance continuous hence the effect for each distance can be evaluated. We follow their method by creating multiple small subsets from the sample and within each subset we compare the house prices of close and far for both before and after the opening date of the stores. Then we use a non-parametric method to smooth the price differences to obtain a continuous effect with respect to distance. However, one assumption is that the house characteristics are similar once the properties are in a similar neighborhood with similar transaction time, which is untrue. We correct this by accounting for the house characteristic differences including transaction time. Specifically, we first restrict our analysis for the properties within 3.5 kilometers radius of the stores. We then identify a similarity neighborhood for each property with a transaction record. The similarity neighborhood contains all the properties within .5 kilometers of the property with a transaction record. Then within each similarity neighborhood, we identify the controls and treatments by comparing their distance from the retailers to the distance of the property with a transaction record. We thus rewrite Eq. (1) as below to construct the housing price as a function of the retailer’s effect, g, and its housing and neighborhood characteristics effect, λ, and a time trend, t: , , , , , , , , , ( , , , ) ( , ) ( , ) ( , ) n j t n j t j t n j t n j t LP r X f g r X f t n t         (2) 129 g is a function of the years that the retail has been open for, δ, and the distance of the property to the retail project, r. λ measures the effect of housing and neighborhood characteristics, X and f, t measures the time trend effect for neighborhood n for time t. With this setting, we could add the far and close, before and after dummy like previous literatures to estimate the equations for each distance to obtain a continuous mapping of housing prices with respect to distance, r, nonparametrically. However, this might suffer from the issue that the controls and treatments used to calculate the price difference are not identical. Thus, within each similarity subset, we take the differences of house values of treatments and controls. Before that, we account for their housing characteristic differences regarding whether it has cooling, heating, a pool and a new fireplace, living square footage, number of rooms, baths and stories, the age of the house, property condition, quality, garage types and the transfer dates to obtain the house price as mappings only on the effect of the retailers and neighborhood characteristics: , , , , , , ( , , , ) ( , ) ( ) n j t n j t n j t LP r X f g r f     (3) By taking differences of house prices according to Eq. (3), properties with similar locations and neighborhood features would yield house prices as a function of just distance and years after opening time. The similarity neighborhood is employed to select the properties with similar locations at each transaction point with distance r. Equation (4) shows the price difference. , , , , , , , , 1 1 (log log ) j j j j L n j t l r t close l r t far l LP p p L      (4) ΔLPn,j,t is the differences of close and far for distance rj and years δ. rj is the distance of house j, where we want to obtain the differences. j is the property that has a transaction record during the study period. tj is the transaction time for the far properties. δ is tj minus the opening time of the retailer. For every transaction within 3.5 miles, we will construct the differences. 130 pl,rj,tj,far is the price of the property for pair l, chosen as far which distance is beyond rj. pl,rj,tj,close is the price of the property chosen as close which is within rj. The far is the control group and the close is the treatment group. The far and close properties are paired according to their distance to the retail. The properties are sequenced as p1,rj,tj,far/close, p2,rj,tj,far/close,..., pL,rj,tj,far/close according to their distances to the retail. p1,rj,tj,far/close is closer to the retail than p2,rj,tj,far/close. L is the total number of the pair of properties. The pre-treatment and post-treatment differences are constructed respectively using the properties transacted before and after the opening date. And with the price differences, we smooth them around each retailer with the nonparametric method below: 1 ,, 1 1 1 (( , ) ( , )) ( , ) (( , ) ( , )) N N N H j j n j t j N H j j j N K r t r t LP rt N K r t r t          (5) ) , ( 1 )) , ( ) , (( , , , , n t j n r j n t n r j j H h t t h r r K h h t r t r K N     N is the total number of properties with transactions for each retail site. K(.,.) is the two- dimensional Epichanokov kernel with bandwidths hr,n, ht,n. The treatment effect is then provided below as the difference of Ѱ(r,t) of after and before the opening of the retail: ( , ) ( , 1) r T r T       (6) T is the year that the store is built. δ is the number of years since its opening. 3. Data 131 The Whole Foods and Walmart data of opening time and location are collected online. The statistics of the data is shown in Table 4-1. There are altogether 159 Walmart stores with opening time from 1990 to 2006 and 64 Whole Foods stores from 1988 to 2013 in California. The property data is from Data Quick where we use the transaction time, price, location and house characteristic data. Table 4-1: Sample statistics for the data Walmart Whole Foods # of stores 159 64 Opening time 1990-2006 1988-2013 Transaction Before 281,369 494,877 After 160,645 185,772 Pairs Before 23 30 After 17 13 Price Close before 181,253 496,950 Far before 149,557 381,389 Close after 363,161 1,024,953 Far after 404,662 947,928 Note: this tables shows the sample statistics for the data. Around each of the 159 Walmart stores with 3.5 km radius, there are on average 281 thousand properties with transaction records for before and 160 thousand for after. Around each of these properties, we draw similarity zones with an average of 23 pair of properties for before and 17 pair of properties for after inside each similarity zone. The average price for far before is 149,557 and close before is 181,253. While for after, the average prices for close and after are 363,161 and 404,662 respectively. Therefore, for Walmart, it seems that it chooses better locations (with close before price larger than far before price) but the price of close properties does not increase as much as that of far properties after the opening of the stores. 132 For Whole Foods, around each of the 64 stores, there are on average 494 thousand properties with transaction records for before and 185 thousand for after. Around each of these properties, we draw similarity zones with an average of 30 pair of properties for before and 13 pair of properties for after inside each zone. The average price for far before is 381,389 and close before is 496,950. While for after, the average prices for close and after are 1024,953 and 947,928. Whole Foods seems to choose better neighborhoods but not closest to the richest people in the neighborhood. 4. Results We present our result in Figure 4-1 for the effect of Walmart and Figure 4-2 for the effect of Whole Foods. Figure 4-1 illustrates that for before the opening, the price is increasing from close to far at a rate of between 10% to 20%. It means that there might be some un-observables making the Walmart location site more attractive, like existing projects. There are no clear variations of the house price differences across different distances. However, we see that for the one year before, the increasing rate drops a little for within 1.5 km compared to farther distance for this year. It signifies that there are not adequate driving forces to sustain the house price increase for the houses adjacent compared to these farther apart. 133 Figure 4-1: The effect of Walmart on nearby house prices: Note: the Wal_effect is the increasing rate (the differences of log of prices) of house prices. Dist is the distance for the effect in miles and yrs stands for the number of years after the opening of the store. For net impact, we take the difference of the after with the before average for the last five years. We did not use the one year before since there might be effect from expectations. The blue lines represent that the house price is decreasing at rates between 30% to 15%. The largest impact 134 happens in the first four years at the distance of 1.5 km. While after the first five years, the negative impact lessens and for ten years after, the effect is around 15%. Also, the biggest hit distance is between 1.5 and 2.5 km. It clearly signifies a negative externality on the house prices. Figure 4-2: The effect of Whole Foods on nearby house prices: Note: the Wf_effect is the increasing rate (the differences of log of prices) of house prices. Dist is the distance for the effect in miles and yrs stands for the number of years after the opening of the store. 135 For Whole Foods, the result is the opposite of Walmart. Before the opening, the house price change rate increases at a rate of 5% to 10% for adjacent distances and decreases as the distance grows. This differs from Walmart in that, Whole Foods seems to choose prime locations with price increase variations across the distances. For after, we net the after difference with the average of before for the last five years. We did not take the one year before since it seems to differ with the other five years before. The red and yellow lines represent the net effect after the opening of Whole Foods. The effect is positive and the house price increasing rate increases as the distance increases from 0 to 1.5 km and stays stable for the distance between 1.5 and 2.5 kms and then drops beyond. The effects are larger at 15% for more recent years, first 3 years and then revert to close to 5% in the after years, where the increase rate slows down. 4.1. The difference with the current literature In this section, we test the results with the difference-in-difference method used by Popes (2015) and Slade (2018) for Walmart effects in CA. Our objective is to find whether the result differences for Walmart are due to the new method improvement or the possibility that in CA Walmart’s impact on house values are different. The results are shown in Table 4-2, which shows that the impacts are positive. This indicates that the differences in the Walmart effect lie in the differences of the new methods with the ones used by current literatures. 136 Table 4-2: The effect of Walmart in CA using the method in current literatures Walmart CA Result Within .5 miles -.5791*** （.0317 ） Within .5 miles*post .8265*** （.0391 ） .5-1 miles -.5446*** （.0284 ） .5-1 miles*post .7625*** （.0310) 1-2 miles -.5527*** （.0231 ） 1-2 miles*post .7632*** （.0254) Store by year by month fixed effects Y Store-level clustering of std. errors Y Housing characteristics Y # Of Walmart openings 159 Observations 3,985,879 Note: We did a traditional difference in difference following the specification in Popes (2015) and Slade (2018). The omitted category is 2-4 miles. Standard errors are clustered at the store level and are in parentheses. *** p<0.01, ** p<0.05, * p<0.1. From the discussion above, the traditional method suffers from two problems: 1) heterogeneity in the controls and treatments and the unobservables in the neighborhood affecting the property values; 2) the same counterfactual assumption. For the first problem, we will try to sign the bias direction. The Walmart effect shown in Figure 4-1 shows that before the opening, the impact is positive, from 12% to 20% for the one year before. Thus, it might be possible that the bias is upward. For the second problem, it assumes that the trends of house values for before and after if there were no Walmart built are the same. But from 1990 to 2006, the house price was on a rise. Although they control for store opening time, the price difference of the increase for after between the store opening time to transaction time and the price decrease for before between the transaction time and opening time is counted as the Walmart effect, thus biasing the result upward. 137 Popes (2015) and Slade (2018) might suffer from these problems. While for Popes (2015)’ result, besides the two problems discussed above, their data is biased towards the high-income neighborhood Walmart stores, thus increasing the bias. Saengchote (2014) found positive impacts for Whole Food using the old difference in difference method. For the first problem, from Figure 4-2, before the opening, the effects of Whole Foods on house values is between (-5%) to 5% for the ones year before. Thus, it seems like the bias direction is uncertain. And for the second issue, the Whole Foods time frame is from 1988 to 2013, for which period it is also hard to sign the bias directions. 5. Conclusion In this article, we study the effect of Walmart and Whole Foods on nearby property values. The result suggests that Walmart exerts a negative externality on property values which differs from the existing literatures. The difference might be due to the improvement of the new difference in difference method in controlling for unobervables affecting house values. While exploring the difference of stores targeting at difference customer segments, we find that Whole Foods provides positive spillovers to the house values surrounding it. 138 6. References Courtemanche, Charles and Art Carden. (2010). Supersizing Supercenters? The Impact of Wal- Mart Supercenters on Body Mass Index and Obesity. Available at SSRN: https://ssrn.com/abstract=1263316 Diamond, Rebecca and Timothy McQuade. (2016). Who Wants Affordable Housing in their Backyard? An Equilibrium Analysis of Low Income Property Development (Working Paper No. 22204). Retrieved from National Bureau of Economic Research website: http://www.nber.org/papers/w22204 Goetz, S. and H. Swaminathan. (2006). Wal-Mart and County-wide Poverty. Social Science Quarterly, 87, 211–226. Haltiwanger, J., R. Jarmin and C.J. Krizan. (2010). Mom-and-Pop Meet Big-Box: Complements or Substitutes? Journal of Urban Economics, 67(2010), 116-134. Hausman, J. and E. Leibtag. (2007). Consumer Benefits from Increased Competition in Shopping Centers and House Values: An Empirical Investigation Effect of Wal-Mart. Journal of Applied Econometrics, 22 (7), 1157–1177. Johnston, Josee and Michelle Szabo. (2010). Reflexivity and the Whole Foods Market Consumer: The Lived Experience of Shopping for Change. Agriculture and Human Values, 28, 303– 319. Kinmonth, A.L., R.M. Angus, P.A. Jenkins, M.A. Smith and J.D. Baum. (1982). Whole Foods and Increased Dietary Fibre Improve Blood Glucose Control in Diabetic Children. Arch. Dis. Child. 57, 187–194. Liu, R. H. (2003). Health Benefits of Fruits and Vegetables Are from Additive and Synergistic Combination of Phytochemicals. Am. J. Clin. Nutr. 78, 517S–520S. 139 Merriman D., J. Persky, J. Davis and R. Baiman. (2012). The Impact of an Urban WalMart Store on Area Businesses. Economic Development Quarterly. 26(4), 321-333. Neumark, D., J. Zhang and S. Ciccarella. (2008). The Effects of Wal-Mart on Local Labor Markets. Journal of Urban Economics, 63, 405–430. Pope, G. Devin and Jaren C. Pope. (2015). When Walmart Comes to Town: Always Low Housing Prices? Always? Journal of Urban Economics, 87(2015), 1-13. Rosen, Sherwin. (1974). Hedonic Prices and Implicit Markets: Product Differentiation in Pure Competition. Journal of Political Economy, 82(1), 34-55. Saengchote, Kanis. (2014). Positive Externalities in the Real Estate Market. Available at SSRN: https://ssrn.com/abstract=2348834 Sharp, Rick L. (2013). Role of Whole Foods in Promoting Hydration after Exercise in Humans, Journal of the American College of Nutrition, 26, 592S-596S. Slade, Barrett. (2018). Big-Box Stores and Urban Land Prices: Friend or Foe?. Real Estate Economics, 46(1), 7-58. Wolfe, Scott and David Pyrooz. (2013). Rolling Back Prices and Raising Crime Rates? The Walmart Effect on Crime in the United States. British Journal of Criminology, 54(2), 161- 373, 2014. Available at SSRN: https://ssrn.com/abstract=2402238 140 Chapter 5. Conclusions My dissertation studies how housing market and business interact with each other. This contributes to the literature in the following three ways: 1) the first essay examines how home- owning characteristics, like presence of mortgages, negative equity, wealth accumulation and attachment to communities, affect home-owners’ personal job outcomes. Formerly, the literatures just focus on studying the differences of personal job outcomes between home-owners and renters but did not explain why the differences exist and what lead to these differences; 2) the second essay fills in the gap of the analysis of NIMBYism. Literatures point out that there are NIMBYism that the home-owners tend to oppose business projects built near their neighborhood through stringent zoning or protesting. And this impact is larger for higher income neighborhoods. But there are no articles studying whether this NIMBYism exert material impacts on business development; 3) the third essay uses an improved difference-in-difference method to study the impact of Walmart and Whole Foods on nearby property values. This method improves the bias caused by unobervables in the old difference-in-difference method. The dissertation finds that: 1) home-owning does not increase peoples’ unemployment probabilities or significantly increase people’s unemployment spells or decrease people’s wages. But home-owning does affect employment by lengthening employment durations and decreasing peoples’ mobility. By investigating some features of home-owning-presence of mortgages, negative equity, wealth accumulation and attachment to communities-we conclude that presence of mortgages elongates home-owners’ employment spell length. Wealth accumulation decreases owners’ unemployment probability but shortens peoples’ annual working time; 2) home-owning decreases the job counts of within .3 miles but benefits the business in 3-5 miles. Negative impacts are identified in higher income groups for adjacent distances while the lower income groups 141 benefit business for both adjacent and farther distances. Service industries like Retail, Art and Professional Services are found with positive home-owning impacts in higher income groups and job counts of Manufacturing, Real Estate and Car Rental and Leasing are positively correlated with home-owning in lower income groups; 3) Walmart exerts a net negative effect on house prices with a decrease of between 15% to 30%. The largest impact happens at the distance of around 2 kilometers (1.25 miles) for the first three years after its opening. Whole Foods produces a positive impact between 15% to 0% increase with the largest impact at the distance around 1.5 (.94 miles) kilometers for the first three years. The policy implications are that: 1) home-owning doesn’t harm employment thus the home-owning promoting policies, like the tax deduction of mortgages payments, are not harmful for the economic development; 2) higher income home-owners might exert negative impacts for business development. For planning arrangements, different industries match with neighborhoods of different income levels.

Abstract (if available)

Abstract The research examines the relationships between housing and business. It firstly studies whether home-owning hurts employment and identifies the home-owning factors affecting home-owners’ job outcomes. Secondly, it studies whether home-ownership rates affect business development nearby. Thirdly, it measures the effect of the proximity to Walmart and Whole Foods on property values. These relationship is important in how we manage the relationship between housing and business development regarding the match and distance among different kinds of neighborhoods and business types.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

The impact of mobility and government rental subsidies on the welfare of households and affordability of markets

PDF

Essays on the economics of cities

PDF

Three essays in United States real estate markets

PDF

Case studies for real estate valuation

PDF

Curating gastronomy: restaurants and social media in the cultural economy

PDF

Childhood obesity and prevention: nutrition, cooking, and gardening learning approach for an educational intervention program to increase fruit and vegetable knowledge, preference, and consumptio...

PDF

Household mobility and neighborhood impacts

PDF

The impact of demographic shifts on automobile travel in the United States: three empirical essays

PDF

University campus expansion: an analysis of Harvard University’s expansion into Boston, Massachusetts

PDF

Property and labor formalization in the age of the sharing economy: Airbnb, housing affordability, and entrepreneurship in Havana

PDF

Three essays on aging, wealth, and housing tenure transitions

PDF

Examining the federal credit union model in the 21st century

PDF

Workplace conflict and employment retaliation in law enforcement; an examination of the causes, effects and viable solutions

PDF

Reshaping Los Angeles: housing affordability and neighborhood change

PDF

Avoiding middle-class planning 2.0: media arts and the future of urban planning

PDF

The spatial economic impact of live music in Orange County, CA

PDF

Mitigating the energy efficiency gap through Property Assessed Clean Energy (PACE): an assessment of the HERO Program in Riverside County, CA

PDF

Environmental justice in real estate, public services, and policy

PDF

Who learns where: understanding the equity implications of charter school reform in the District of Columbia

PDF

Housing market crisis and underwater homeowners

Asset Metadata

Creator Wang, Bingbing (author)

Core Title The interactions between housing and business

School School of Policy, Planning and Development

Degree Doctor of Philosophy

Degree Program Urban Planning and Development

Publication Date 06/24/2020

Defense Date 06/24/2018

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag business development,home-ownership,house price,OAI-PMH Harvest,Unemployment

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Green, Richard K. (committee chair), Bostic, Raphael (committee member), Painter, Gary (committee member)

Creator Email bingbingwang123@gmail.com,bingbinw@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c40-505287

Unique identifier UC11268174

Identifier etd-WangBingbi-6343.pdf (filename),usctheses-c40-505287 (legacy record id)

Legacy Identifier etd-WangBingbi-6343.pdf

Dmrecord 505287

Document Type Dissertation

Format application/pdf (imt)

Rights Wang, Bingbing

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA