The McMansion Curse: Housing Size Inequality, Status Competition and House Valuation in American Suburbs

Transcription

1 The McMansion Curse: Housing Size Inequality, Status Competition and House Valuation in American Suburbs By Clement S. Bellet Despite a major upscaling of single-family houses since 1980, house satisfaction has remained steady in American suburbs. At any point in time, however, house satisfaction rises with house size. This Easterlin paradox in the realm of housing can be explained by upward-looking comparisons in the size of neighboring houses. Combining data from the American Housing Surveys with an original dataset of three million suburban houses built from 1920 to 2009, I find that the construction of so-called McMansions lowers the satisfaction that neighbors derive from a rise in their own house size. Upward-looking comparisons are stronger among people living in larger houses and decrease with the distance from McMansions. I provide further evidence that homeowners exposed to the construction of big houses in their neighborhood value their own home less, and are more likely to upscale to a bigger home and subscribe to a new mortgage. JEL: D01, D03, I30, R20 Keywords: subjective well-being, housing, reference effects, household debt Since 1940, the surburban population of America has grown more than metropolitan areas, so that by 2000 half the U.S. population lived in the suburbs. This period was characterized by an impressive upscaling of single-family houses: the median new-build house is now twice bigger than in Since 1980, however, house size inequality has widened, driven by the construction of ever bigger houses at the top of the distribution. In 1980, only 5% of new-built houses were larger than 3,000 square feet; on the eve of the 2008 financial crisis, McMansions represented 15% of new construction and the largest 10% of all homes were about four times bigger than below-median houses, compared with a factor of three 30 years before. 1. In his analysis of economic growth and competition, Hirsch (1976) argues that consumption choices are ultimately positional. In other words, the utility derived LSE, INSEAD, clement.bellet@insead.edu. I am extremely grateful to Paulo Albuquerque, Samuel Bowles, Pierre Chandon, Andrew Clark, Nicolas Coeurdacier, Frank Cowell, Pierre Cahuc, Stefano Dellavigna, Paul Frijters, Steve Gibbons, Sergei Guriev, Marco Manacorda, Florian Oswald, Thomas Piketty, Alexandra Roulet and Claudia Senik for very helpful discussions and comments. I also thank the seminar and conference participants at Sciences Po Department of Economics, INSEE-CREST, London School of Economics, Paris School of Economics, INSEAD, 2016 HEIRS Conference and 2017 ECINEQ conference for useful comments. I acknowledges financial support from CREST, Centre for Economic Performance (LSE) and INSEAD 1 The dataset from which these figures are derived is described later in the paper. 1

2 2 from the possession of material goods depends on relative rather than absolute levels of consumption. If needs are relative, consumption inequality may in turn generate dissatisfaction for those households who cannot keep up with the consumption level of richer households (Frank, 2013). Indeed, the notion of relative needs is as old as political economics itself. In Wage, Labor and Capital published in 1847, Marx argued that A house may be large or small; as long as the neighboring houses are likewise small, it satisfies all social requirement for a residence. But let there arise next to the little house a palace, and the little house shrinks to a hut. This article offers a first empirical estimate of social comparisons in the housing market using large-scale field data. Matching an original online dataset of more than three millions single-family houses to representative survey data on homeowners, I relate current house satisfaction to individual experiences in the construction of new houses built after a move. I show that an experienced rise in size at the top of the distribution lowers the satisfaction gained by others house size, and provide evidence that this McMansion effect contributed to the size upscaling and increase in mortgage debt observed in the decades preceding the Great Recession. The Easterlin paradox posits that increasing the income of all does not increase the happiness of all (Easterlin, 1995, 2001; Easterlin et al., 2010) 2. I document the existence of a Joneses paradox (from the expression Keeping up with the Joneses ), which echoes the Easterlin paradox in the housing realm. Within any given year, new movers to bigger houses systematically report higher levels of house satisfaction. Yet, since 1980, despite the widespread upscaling of American homes, the average house satisfaction of new movers has remained steady 3. One explanation for the Easterlin paradox is the role of income comparisons (Clark, Frijters and Shields, 2008) - that is, the capacity of individuals to observe others incomes (Card et al., 2012; Perez-Truglia, 2016), or infer their relative position from visible consumption choices (Heffetz, 2011). Housing ranks among the most positional and visible consumption categories in lab experiments (Alpizar, Carlsson and Johansson-Stenman, 2005; Solnick and Hemenway, 2005), and house size is arguably the most salient characteristic of status in American suburbs (Wright, 1983) 4. Hence social comparisons in housing size are likely to come into play. 2 A modified version of the Easterlin hypothesis claims that the Paradox only verifies beyond a certain income threshold. Stevenson and Wolfers (2008, 2013) challenge the Easterlin hypothesis showing some of the previous results were statistical artifacts. However, the critique tends to neglect a different definition of the Paradox, which results from the contradiction between a strong positive correlation in crosssectional data and an absence of (or much weaker) positive longitudinal correlation in the long-run. 3 The paradox is robust to the inclusion of household and house controls (excluding house size), for both old and recent movers. 4 In Building the Dream, Wright (1983) writes: Allotments were based on a position and wealth of a family within the community, so that the land divisions and size of houses varied considerably.

3 THE MCMANSION CURSE 3 Due to endogenous sorting of households across suburbs, the identification of social comparisons is no easy task. Indeed, households that are more negatively affected by social comparisons will self-sort in suburbs with smaller houses. Regressing the size of existing houses on current house satisfaction is hence likely to give an estimate biased towards zero. To circumvent this issue, I exploit crosssectional differences between households current house satisfaction and the size of new houses built in their suburb after they moved in 5. Let us suppose two similar households, in the same suburb, are surveyed in The only difference between household A and household B is that A moved in 1990 while B moved in McMansions were built between 1990 and 1995 but there has been no change in top-size housing since then. Hence when B moved, these large houses were already in place. Unless both households perfectly internalized future variations in top housing size when buying a house, household A, who experienced a rise, should feel less satisfied than household B who experienced no change at all 6. This strategy addresses Manski (1993) s classical reflexion problem as it exploits differences in the size of houses built after moving choices have been made. It also allows me to introduce county-year fixed effects to account for any time varying suburb characteristics affecting all households regardless of when they moved, from congestion costs to local amenities or land prices, and tenure length fixed effects to account for general time effects like hedonic adaptation. I find evidence of upward-looking comparison effects on housing size. The impact of own house size on house satisfaction is of similar magnitude to that of own income on life satisfaction documented in the subjective well-being literature (Stevenson and Wolfers, 2008; Deaton and Stone, 2013). House size also shows diminishing marginal gains in house satisfaction, with a negligible effect of size beyond 3000 square feet. A 1% increase in the size of newly built houses at the top of the distribution almost offsets the satisfaction gains from a 1% increase in own housing size. Top size - or McMansions - is defined as houses in the top decile of the local size distribution 7. I find smaller or non-significant impact for lower size decile of new-build houses. The McMansion externality is stronger for households living in larger houses. This echoes the trickle-down consumption (or expenditure cascade) hypothesis of Frank, Levine and Dijk (2014), as evidenced by Bertrand and Morse (2016) for total consumption by the rich and non-rich at the state level. Lastly, I find households adapt to the size of their house but do not adapt to the size of new-build McMansions. 5 This methodology echoes Malmendier and Nagel (2011, 2016) who look at personal experiences in inflation rates across cohorts. However, while they exploit differences in inflation histories over time, my measure of personal experience in new-build houses combines time effects and differences across locations. 6 Perfect foresight over the size and location of new houses built is very unlikely, especially over a period of several years (the median length of tenure in my dataset is 8 years). Still, if anything, any negative and significant effect should be interpreted as a conservative estimate of the true effect, as I argue later in the paper. 7 I also use the local share of new houses built larger than 3000 square feet, and find similar results. The two measures are highly collinear. However, when using both in the same regression, only the 90th percentile size remains significant.

4 4 When households become dissatisfied with their house, they may be willing to sell the house at a lower price. Assuming upward-looking comparisons are internalized in the potential asking price, one may infer a monetary cost of experienced relative downscaling. I therefore complement the house satisfaction approach by a hedonic regression on house market values. Without controlling for suburb-year fixed effects, the construction of big houses during households tenure period increases the market value of their house. However, this first-order effect is likely driven by a positive correlation between top housing size and suburban quality, leading to higher land prices. The inclusion of suburb-year fixed effects allows me to identify the impact of households own experience in the construction of McMansions in their suburb separately from the market price effect affecting all households in a similar way. In the presence of upward-looking comparisons, I expect the coefficient on top housing size to become negative. In line with my results on house satisfaction, I find that a 1% rise in the size of McMansions constructed after a household moved leads to a 0.35% fall in the market value homeowners ascribed to their house. The increase in size driven by lower parts of the size distribution are either not significantly or positively related to house market values. Lastly, I explore whether people react to the relative downscaling of their house by increasing its size, and subscribe to new mortgage loans. Indeed, if the effect is driven by social comparison in housing size, I should expect homeowners to upscale the size of their own house following the construction of McMansions in their suburbs. I find that relatively deprived households strive to keep up with the Joneses. Controlling for household and house fixed effects, a 1% rise in top housing size within a suburb is associated with a 0.16% rise in own housing size though home improvements, and a 0.35% increase in mortgage debt. Again, lower size decile are negatively or not significantly related to home size improvements. This relative size effect can explain up to 20% of the rise in the mortgage debt to income ratio since While these results are consistent with a relative size effect, alternative noncausal explanations may also lead to a negative correlation between house satisfaction and an experienced rise in top housing size. A first concern is that the negative effect found may capture a neighborhood externality associated with top housing size, which lowers neighborhood and house satisfaction together. An increase in top housing size could indeed be associated with higher population density and households could be negatively affected by neighborhood gentrification. I show that contrary to house satisfaction, neighborhood satisfaction is positively or not significantly associated with an increase experienced in the size of new houses. Controlling for the experienced variation in population density does not affect my results. The construction of big houses could simply block the view or be considered unaesthetic. Though I cannot directly exclude this possibility, these externalities should also affect neighborhood satisfaction, and there is no reason to expect these effects to play more negatively on householders

5 THE MCMANSION CURSE 5 living in larger houses. Another concern relates to the endogenous sorting of new houses within suburbs. In particular, richer movers may choose to build big houses far away from small ones. My own data confirms that prediction: since 1980, the McMansion effect is characterized by an increasing segregation between small and big houses. The negative coefficient found on the rise in top housing size may be driven by homeowners feelings with respect to the economic segregation experienced. Using multiple imputations, I estimate the distance of homeowners houses from newly built McMansions and interact it with the average size of these same houses. If the effect is driven by upward-looking comparisons in size, the McMansion effect should remain significant and be stronger for homeowners living closer to new McMansions. I find the estimated distance to McMansions reduces house satisfaction, but the interaction effect with size is positive. The coefficient on top housing size remains negative and significant for houses built less than 15 miles away from McMansions. There remains the possibility that households which are more sensitive to social comparison self-sort on the basis of the accuracy of their predictions in future housing size and how it may affect their house satisfaction. In particular, if households more negatively affected by social comparisons do better at predicting future increases in housing size, the coefficient will be biased towards zero. The negative effect on the size of new McMansions may therefore be a conservative estimate of the true effect. This seems very unlikely considering the evidence on the difficulty to predict future utility (Loewenstein, O Donoghue and Rabin, 2003). However, to account for this possibility, I run a household fixed effect model on a subsample of my dataset. The ability to predict future increases in top housing size should be captured by the household fixed effect. Results remains significant and the estimated coefficient on top housing size is even stronger. This paper contributes to various strand of the literature on consumer behaviors, relative income and well-being by documenting a strong link between experienced relative deprivation and housing consumption. Evidence on conspicuous consumption and reference effects can be found in Charles, Hurst and Roussanov (2009) who look at relative income within racial groups, and Bertrand and Morse (2016) who identify trickle-down effects in total consumption. These papers infer reference effects from revealed preferences using cross-sectional variation at the state level. They also do not explicitly tackle the reflection problem. My methodology addresses endogeneity issues, and I can also look at self-reported differences in subjective satisfaction. I show past experiences can have long-lasting effects on current wellbeing and house valuations. While the reference point is usually based on the mean value of an income variable at a given location, I identify which part of the housing size distribution acts as the reference group and provide evidence that space matters. This is also the first paper estimating positional externalities in housing using field data. The impact of neighbors incomes on life satisfaction has been studied by

6 6 Luttmer (2005), Dynan and Ravina (2007), or Brodeur and Fleche (2018) who find a negative effect of others mean income. Housing satisfaction being a major component of life satisfaction (Van Praag, Frijters and Ferrer-i Carbonell, 2003), my results suggest a strong channel behind the relative income effect and the Easterlin paradox, namely visible consumption choices. What matters for life satisfaction may not be relative income but relative visible income. The contrasting effects of reference size on house and neighborhood satisfaction also provide further evidence that I am able to identify housing size comparisons separately from general neighborhood effects. To a lesser extent, this paper also relates to the urban economics literature on neighborhood effects. Using a different methodology, Ioannides and Zabel (2003) provide evidence of social interaction effects on home improvements. However, the literature tends to emphasize the contribution of positive neighborhood externalities (Glaeser and Shapiro, 2002; Rossi-Hansberg, Sarte and Owens III, 2010; Guerrieri, Hartley and Hurst, 2013). The fact that I identify a negative relative size externality suggests a trade-off between the decision to live in a neighborhood with better amenities and the costs implied by status competition in housing size. It also suggests the benefits of moving to more expensive locations may be overestimated by homeowners who do not anticipate the status loss experienced by the construction of increasingly bigger houses. Regarding the link between inequality and household debt, the existing evidence is mixed. Using survey data, Carr and Jayadev (2014), Bertrand and Morse (2016), and Christen and Morgan (2005) provide evidence that inequality is related to financial distress and borrowing, while Coibion et al. (2014) finds the opposite result. Exploiting random variation in relative deprivation through lottery wins, Agarwal, Mikhed and Scholnick (2018) finds a positive effect of relative income on financial distress. However, none of these studies look at the housing market specifically, and cannot identify status preferences directly. My results suggest reference-dependent preferences in housing consumption can lead to a positive effect of top income inequality on mortgage debt. The rest of the paper proceeds as follows. Section 2 presents the two main datasets and discusses important stylized facts on housing size and house satisfaction. Section 3 presents the methodology. Section 4 presents the results on house satisfaction and housing prices, and discusses their behavioral impact on individual choices. Section 5 presents a series of robustness checks. Section 6 concludes. I. Data and Stylized Facts A. American Housing Survey (AHS) The American Housing Survey (AHS), a longitudinal survey of American houses conducted every two years by the US Census Bureau at the national and metropolitan level, collects extensive information on house quality, neighborhood quality,

7 THE MCMANSION CURSE 7 and household characteristics. This includes information on the historical purchase price and the current market value of the house as assessed by the owner. In addition to a house identifier, a common household identifier can be recovered from consistent answers across waves within the same house (e.g. age, sex, race) 8. Starting in 1984, the US Census Bureau added two questions about subjective satisfaction, as follows: The next two questions are about your home and your neighborhood and how you feel about each of them, considering everything that we have talked about during this interview. On a scale of 1 to 10, how would you rate the house as a place to live? 10 is best, 1 is worst. How would you rate the neighborhood on a scale of 1 to 10? 10 is best, 1 is worst 9. Both questions are closer to evaluative (or cognitive) measures of subjective wellbeing (SWB) like the life satisfaction question, as opposed to more hedonic (or affective) measures that do not require the cognitive effort necessary to answer evaluative questions (Diener et al., 1999) 10. When it comes to the evaluation of externalities or non-market goods, subjective satisfaction measures have been used as an alternative to revealed preferences methods like hedonic pricing (Van Praag and Baarsma, 2005; Luechinger, 2009; Frey, Luechinger and Stutzer, 2009), which relies heavily on market clearing conditions and assumes perfect information (Rosen, 1974; Roback, 1982). Such conditions are unlikely to be met given homeowners moving costs or difficulty predicting future reference points in long-term investment decisions. Hedonic pricing methods also require sufficient information on house and neighborhood characteristics at the time of purchase, whereas the AHS mostly provides information at the time of survey. There is, in fact, consistent evidence showing that subjective satisfaction measures capture the influence of past experiences or habits (Kimball and Willis, 2006). Similarly, homeowners asking prices, which better capture sellers preferences and personal experiences, have been shown to be more reactive to reference effects than transaction prices (Genesove and Mayer, 2001). Lastly, subjective satisfaction measures are good predictors of individual choices (Kahneman et al., 1993; Benjamin et al., 2012, 2014). From the national files of the AHS, I can compare cross-sectional and longitudinal correlations between house satisfaction and housing size. These have surveyed a nationally representative sample of households every two years between 1985 and On average, suburban households are satisfied with their house, with an average of 8.5 points out of 10. About 8% report a value below or equal to six, 11% report seven, 26% height, 18% nine and 36% say ten (Figure 1). Compared 8 The household panel structure of the data leads to a much smaller sample. For that reason, the main empirical specification of the paper relies on the full sample of houses. 9 The introductory sentence preceding the two questions was only introduced after There is no sign of discontinuity before and after 1995 as for the way new movers answered the question. In the empirical analysis, the inclusion of survey-year fixed effects for the entire sample should account for general phrasing bias. 10 Evaluative measures tend to increase more linearly with respect to log income than affective measures (Kahneman and Deaton, 2010; Stevenson and Wolfers, 2013). They also differ with respect to the importance of relative income (Deaton and Stone, 2013)

8 8 to the average American household, they are richer and somewhat less representative of racial minorities. In 2009, the median household income of suburban households was $70,000 while the national median at that time was $55,000. The proportion of racial minorities (African Americans and Hispanic Americans) was 16.1%, compared to a national average of 19.5% (see Table E1 in appendix). I first test whether there is any evidence of an Easterlin paradox in housing size. Despite a positive relation between income and SWB in cross-sectional data, the Easterlin paradox finds no positive correlation between the growth in annual income levels and long-run subjective wellbeing. Since housing is a durable good, and that the AHS includes both new and old movers, I restrict the analysis to the flow of new suburban movers over time 11. It also allows me to abstract from hedonic adaptation (Loewenstein and Ubel, 2008). A cross-sectional regression of house satisfaction on housing size systematically produces a positive correlation, as can be seen in figure 2a using the 2011 AHS survey 12. For instance, moving from a 1000 square feet house to a 2000 square feet house increases house satisfaction by a 0.5 point. Percent House satisfaction (1-10) Figure 1. : Distribution of Self-Reported House Satisfaction, AHS National Surveys ( ) I then plot the average housing size of new movers in suburban areas between 1985 and 2013 against their average house satisfaction over the same period (Figure 2b). Despite an increase in average house size of about 500 square feet between 11 From Table E1, the average length of tenure in the sample is 13 years. 12 Figure E2 in appendix shows the regression coefficient of log housing size on house satisfaction for all waves of new movers. The coefficient on own size is always positive and significant (between 0.5 and 1), though it shows some decline over time.

9 THE MCMANSION CURSE 9 Subjective House Satisfaction (1-10) Own housing size (sqft), new movers Average house satisfaction (1-10), new movers Average house satisfaction Average house size Average housz size (sqft), new movers (a) House Satisfaction and House Size, AHS Cross-Section 2011 (b) House Satisfaction and House Size, AHS Longitudinal Figure 2. : Joneses Paradox, New Movers Note: This Figure replicates Easterlin s Paradox in the housing realm. Figure 2a plots the result of a quadratic regression of subjective house satisfaction on own housing size for new movers in The vertical-left axis of figure 2b indicates the average house satisfaction of new movers, while the verticalright axis shows the average size of their house. The two measures are constructed using the national surveys of the AHS. Sampling weights included. Source: AHS national surveys 1985 and 2005, new movers report no positive rise in house satisfaction over the period. Since average household size fell from 3 to 2.8, this cannot be explained by a fall in average space per person. The paradox is robust for all households to the inclusion of controls related to the quality of the house (house prices per square feet and age of the house) or household characteristics like household size, length of tenure or income (see Figure E3 in appendix). As a possible explanation for the paradox, I explore whether households are negatively affected by the increase in size of their neighbors houses with which they compare their own. For the main empirical analysis, I combine 18 waves of the metropolitan files from 1984 to The metropolitan surveys are conducted every year from 1984 to 1996 and every two years hereafter. They are representative at the MSA level, and contrary to the national files, all houses have a county identifier. A different set of metropolitan areas (MSA) are surveyed each year, so that on average a given MSA gets surveyed four times over the period. Each MSA comprises an average of five counties and each county is divided between its central city area and its suburban area. In this article, a suburb therefore corresponds to the suburban area of a metropolitan county. I restrict the analysis to homeowners, who account for about 90% of suburban households and for these observations I have information on the purchase price and market value of their house. After removing observations with missing values, I end up with a sample of about 137,000 individual observations, corresponding to 88,000 individual houses distributed in 154 suburbs between

10 and Figure E1 in appendix maps the location of these suburbs. These represent about 54% of the total American population, and covers close to the entire suburban population of the country. Using the metropolitan files, a simple test for relative size can be performed. Following Deaton and Stone (2013), it consists in comparing the coefficient of own housing size on individual satisfaction in the micro data to the coefficient of average suburban housing size on average suburban satisfaction. If the coefficient on suburban averages is not significant or lower than the coefficient on householdlevel data, it is indicative of a negative relative size effect. A higher coefficient indicates a positive externality (e.g. public goods). I first regress individual house and neighborhood satisfaction on the log of own housing size and a suburb-year fixed effect 13. I then run similar regressions on average suburb-year level data and a year fixed effect. Results are shown in Table 1. Table 1 : Self-Reported Satisfaction and Relative Housing Size: A Simple Test House satisfaction Neighborhood satisfaction β on log housing size Observations β on log housing size Observations Household-level data , ,787 (0.0112) (0.0124) Averaged by suburb-year (0.0678) (0.0710) Note: I use house satisfaction or neighborhood satisfaction as the dependent variable, and I report the regression coefficient on log housing size (β). Following Deaton and Stone (2013), regressions include controls for age, sex, and race, either individual or averaged within suburb-year cells. Regressions on house (neighborhood) satisfaction also control for neighborhood (house) satisfaction. Household-level regressions add suburb-year dummies and regressions averaged by suburb-year control for year effects. Source: AHS Metropolitan Surveys ( ) Own housing size is positively related to both house and neighborhood satisfaction. However, the positive effects of own size on house satisfaction are entirely offset by the negative effects of average suburban housing size. Conversely, average housing size affects neighborhood satisfaction positively: the suburb-year coefficient is more than twice that of the micro-data coefficient. These preliminary findings suggest households may face a trade-off between a positive neighborhood externality and a negative housing size externality. However, these results cannot be interpreted as causal. They may simply reflect similar characteristics of households who endogeneously sort in specific locations. Other than this it is hard to say anything about the heterogeneity of the effect across households or 13 As in Deaton and Stone (2013), I control for age, sex, and race. The regression on house (neighborhood) satisfaction also controls for neighborhood (house) satisfaction.

11 THE MCMANSION CURSE 11 the reference group to whom people compare. Indeed, the AHS does not offer enough observations at the suburban level to capture the local distribution of suburban houses over time. Like most surveys, it does a poor job at capturing the top of the housing size distribution. B. Web Scraping of Real Estate Data (Zillow) Zillow.com, a leading online real estate compagny in the US, makes publicly available information on millions of houses for sale or rent. Using web scraping techniques, I gathered an original geo-localized sample of 3.2 millions single-family houses located in each of the 154 metropolitan counties surveyed in the AHS. This corresponds to an average of about 20,000 houses per county. I restricted the scraping to those built between 1920 and 2009 to account for all possible tenure periods of AHS households. I collected information on the location of the house (latitude and longitude), the year it was built, the size of the house and the lot size. From this dataset, I can look at the evolution and distribution of housing size, over time and across suburbs. Figure 3a plots the median and 90th percentile size of all houses by year of construction. Over the last 50 years, the median new-built house doubled in size. Considering that average household size has decreased by about 20% since 1960, the amount of private space per person has risen at an even higher rate. Furthermore, the rise in median housing size was associated with a strong rise in size at the top of the distribution. Starting in the 1980s, houses in the top decile grew at a higher rate than median housing size. Size of new houses (sqft) th percentile size 50th percentile size th percentile house to median size house ratio (a) General Upscaling, New Houses Built (b) The McMansion Effect, Housing Stock Figure 3. : General Upscaling and the McMansion Effect ( ) Note: Figure 3a plots the median (solid line) and 90th percentile size (dashed line) of houses by year of construction from 1920 to Figure 3b looks at the stock of houses each year and plots the ratio between the 90th percentile size and the median housing size Source: Author s own calculation from Zillow

12 12 The McMansion effect becomes salient when one looks at how the construction of new houses altered the distribution of the housing stock over the period. Figure 3b plots the gap in size between the average top decile house and the median house for the stock of existing units each year. 14 It reveals a U-shape pattern of top housing size inequality which echoes the pattern of top wealth inequality over the last century (Saez and Zucman, 2016). From 1940 to 1960, the 90th percentile size of the housing stock declined by 10 percentage points compared to the median house. The ratio remained stable until 1980, when the reverse trend occurred. The 90th percentile size house then increased by almost 15 percentage points relative to the median. On the eve of the 2008 financial crisis, it had recovered to the level of the 1930s. Looking at the entire distribution of houses by construction decade, it is clear that the increase in size inequality was driven by the top of the size distribution, which became increasingly fat-tailed after 1980 (Figure E4 in appendix). As explained in detail below, I identify upward-looking comparison effects from the size of big houses constructed in a household s suburb after the householder moved in. Importantly, the distribution of housing size varies greatly between and within suburbs, even for houses located within the same MSA. To illustrate, Figure 4a maps the location of about 70,000 single-family houses built in Baltimore metropolitan area. There are seven suburbs in the Baltimore MSA, and each grey dot represents a house web-scrapped from Zillow. In Figure 4b, I plot the 90th percentile size to median size ratio in four suburbs of the Baltimore MSA for the entire stock of houses since The latter goes from 2.25 times to 1.6 times the median house size. In 1960, the 90th percentile house was 70% bigger than the median size house in both Baltimore and Howard suburbs. In 1980, the size of McMansions was 60% bigger in Howard suburb, while in Baltimore suburb the figure was 80%. The use of Zillow to recover historical changes in housing size may be problematic if attrition or home remodeling effects alter the measurement of past housing size in a systematic way. For instance, assuming that small houses are more likely to be destroyed first, the average size of older houses will decrease. Conversely, assuming improvements are made to the size of older houses, their average size will increase. I address this concern in Annexe C by comparing the mean to median housing size ratio in Zillow to the equivalent measure from the Survey of Construction (SOC) across regions. Looking at the period from 1970 to 2009, there is no evidence of a diverging gap in mean to median size ratio over time between the two datasets. 14 Taking the gini coefficient of housing size gives a similar picture.

13 THE MCMANSION CURSE 13 90th percentile size relative to median Baltimore MSA, Maryland Anne Arundel Suburb Carroll Suburb Baltimore Suburb Howard Suburb (a) Mapping of houses, Baltimore MSA (b) P90/P50 housing size, housing stock Figure 4. : The McMansion Effect Within Baltimore MSA ( ) Note: Figure 4a maps the 70,000 single-family houses web-scrapped from Zillow within the seven suburbs of Baltimore metropolitan area. Figure 4b plots the size of the 90th percentile house relative to the median size house for the stock of houses each year between 1920 and 2009 in four suburbs of the Baltimore metropolitan area (Maryland). A coefficient of 1.5 means the 90th percentile house is 50% bigger than the median house in the suburb. Source: Author s own calculation from Zillow II. Empirical Strategy A. Preference for Relative Housing Size Assume a household i who lives in suburb s at time t in a house of size h ist. Following Abel (1990), relative size can be defined as h ist = h ist /Hst, σ where H st captures the (external) reference housing size at time t in suburb s. In the case of the McMansion effect, H st is defined by the 90th percentile size, so that coefficient σ captures households sensitivity to top housing size. I take a Cobb-Douglas specification for house satisfaction: U ist = (h ist /H σ st) α N j=1 z β j jist Use of Cobb-Douglas is common in the empirical analysis of durable consumption (such as housing) in the United States. 15 Coefficient α captures the importance of size for house utility and β j the effect of any other house or neighborhood characteristics z jist. If σ = 0, only absolute size matters. If σ > 0, house satisfaction is negatively impacted by top housing size. A negative σ would be indicative of a positive size externality for instance because the presence of big houses is associated with better amenities locally See in particular Davis and Ortalo-Magné (2011) or Fernandez-Villaverde and Krueger (2011). 16 The case where σ < 0 leads to the specification of utility chosen by Guerrieri, Hartley and Hurst (2013) to model positive neighborhood externalities.

14 14 Figure 5 plots non-parametrically the measure of house satisfaction as a function of own housing size within suburbs characterized by weak (vs strong) McMansion effects. The latter are defined, respectively, as the bottom and top quartile suburbs of my dataset in 90th percentile housing size. First, the link between house satisfaction and housing size echoes the concave relationship found between wellbeing and income (Stevenson and Wolfers, 2013). Figure 5 also suggests a positive σ coefficient: a householder from a top quartile suburb, living in 1250 square feet, must increase the size of the house by 500 square feet to experience the same level of satisfaction than a householder from a top quartile suburb, living in a similar size house. Interestingly, the drop in house satisfaction is less pronounced for bigger houses; households with less than 1250 square feet seem much less affected by top housing size. This supports the ratio specification of relative size as the difference specification h ist = h ist σh st (also used in the literature) would lead to the opposite prediction. 17 Own house satisfaction (1-10) Own housing size (sqft) (A) Bottom quartile suburbs kernel = epanechnikov, degree = 3, bandwidth = (B) Top quartile suburbs Figure 5. : Non-Parametric House Satisfaction Curves Note: Figure 5 plots nonparametrically house satisfaction as a function of own housing size for households living in bottom quartile (A) and top quartile (B) suburbs in terms of 90th percentile housing size. A household from suburb B living in a 1250 square feet house must increase the size of his house by 500 square feet to experience the same level of house satisfaction than a household from suburb A living in a house of similar size. Source: AHS metropolitan files and author s own calculation from Zillow Knowing U ist, the sensitivity to the size of other houses could be estimated directly from a log linear regression: 17 Since σ > 0 and since house satisfaction can be shown to be a concave function of relative size (0 < α 1), sign( U ist/ h ist ) ( H = sign α 2 σ ) < 0. This specification of relative size implies that the st interaction term between own and reference housing size should be negative.

15 THE MCMANSION CURSE 15 (1) ln(u ist ) = constant + αln(h ist ) ασln(h st ) + N β j ln(z jist ) + u ist j This specification is similar to the reduced-form regression estimated in the literature on relative income and relative consumption 18. Coefficient σ can also be interpreted as the ratio between two elasticities. The well-being literature usually reports level-log or ordered logit estimates but my results are robust to these specifications. However, regression (1) poses a number of identification challenges that need to be addressed. B. Identification of Social Comparisons in Size The identification of σ is challenging due to Manski (1993) s reflection effect and the endogenous sorting bias. Indeed, social comparison effects are particularly difficult to capture since the decision to buy a house may reflect unobservable characteristics shared between a household and their neighbors at the time they decided to move. In particular, homeowners living in suburbs with big houses may simply share stronger preferences for larger houses. This implies those households negatively affected by top housing size will sort into suburbs with lower levels of top housing size. Regressing house satisfaction on the size of all existing houses at time t, including houses already built when the household moved in, will result in a σ coefficient biased towards zero. In other words, σ cannot be identified from the stock of existing houses at time t. To address this concern, I exploit the fact that personal experiences in the construction of new houses differ across households within any suburb at time t. The identifying assumption relies on the difficulty individuals have predicting future changes in reference housing size and how these may affect their future house satisfaction. Following Loewenstein, O Donoghue and Rabin (2003), If a person buys a small house in a wealthy neighborhood in part because it has a certain status value in her apartment building, she may not fully appreciate that her frame of references may quickly become the larger houses and bigger cars that her new neighbors have. I depart from mobility concerns by looking at the size of houses constructed after the moving decision has been made. Upward-looking comparisons are identified within suburbs based on cross-sectional differences in households own experiences in the construction of McMansions 19. Appendix A illustrates this intuition with a model of reference-dependent preferences under projection bias. Consider two home owners interviewed in Both moved in 1982 in a median 18 See in particular Luttmer (2005), Charles, Hurst and Roussanov (2009) or Bertrand and Morse (2016). 19 This strategy is also used by Malmendier and Nagel (2011, 2016) to measure the impact of experienced inflation and stock market prices on risk aversion and investment decisions.

16 16 size house of their respective suburbs;:household A moved to Carroll suburb while household B moved to Howard suburb 20. All else being equal, this suggests B had a lower preference for top housing size than A. Indeed, when A moved, existing McMansions were 70% bigger than her house, while they were less than 58% bigger when B moved. Six years later, there has been very little change in top housing size relative to A s house, while B experienced a strong rise. We should expect household B, who experienced a fall in relative size, to be less satisfied with their house than household A, who experienced almost no change. Rather than taking the entire stock of McMansions at time t in a household s suburb s, i.e. H st in equation (1), I look at the construction of new McMansions experienced by each household i since moving in, H ist. For each i, I construct a measure called 90thP ercentilesize ist, which corresponds to the 90th percentile size of houses built in the household s suburb since he moved in. Under the assumption that households do not perfectly internalize the size and location of future houses in their suburbs (or how it may affect their satisfaction), 90thP ercentilesize ist should be orthogonal to the choice of moving. In line with equation (1), I run the following regression: log(housesatisfaction) ist = γ 0 + γ 1 log(ownsize) ist + γ 2 log(90thpercentilesize) ist +SuburbTrends st + TenureLength it + γ 3 Controls ist + u ist (2) where HouseSatisfaction ist is the house satisfaction evaluated from 1 to 10 at time t by household i who lives in suburb s. OwnSize ist is the size of the household i s house, SuburbT rends st is a set of about 500 dummies controlling for suburb specific time effects, and T enurelength it a set of 84 dummies for the households length of tenure (in years). A negative γ 2 will be indicative of relative downscaling, and households sensitivity to top housing size can be expressed as σ = γ 2 /γ 1. Coefficients γ 1 and γ 2 also corresponds to the elasticity of house satisfaction with respect to own and others housing size. In other words, a σ coefficient of unity implies a household experiencing the construction of x% bigger McMansions must increase the size of his own house at the same rate in order to maintain a constant house satisfaction level. Unless I control for suburb-specific time trends, γ 2 is likely to be biased upward as individuals who experienced the construction of bigger houses are also likely to live in richer suburbs. Adding suburb-year fixed effects allows me to control for any suburb characteristics at time t affecting all households similarly within suburbs, regardless of when they moved. This includes the current characteristics of the housing stock, land prices or local amenities correlated with the size of houses. The addition of dummies for length of tenure captures hedonic adaptation to the house and other general time effects correlated to the experienced change in top 20 Figure E5 in appendix illustrates this example graphically.

17 THE MCMANSION CURSE 17 housing size. Lastly, I include a list of controls for house characteristics, 21 household characteristics 22 and neighborhood satisfaction evaluated by the household at time t on a scale from 1 to 10. Besides accounting for changes in neighborhood characteristics not captured by the suburb-year effect, the latter also controls for possible measurement errors in the way households answer subjective questions. The next section presents the main results from equation (2). I also show alternative specifications for the house satisfaction regression, test for other outcome variables, and look at the heterogeneity of coefficient σ as a function of own housing size. I explore different explanations to the relative size effect and run a series of robustness checks, including whether lower percentiles of the housing size distribution also affect current house satisfaction. For a sub-sample of my dataset, I can also check whether my results holds within households after adding household fixed effects. III. Main Results A. House Satisfaction and Reference Size Effect Table 2 shows the result for the pooled OLS log-log regression (2) where I add more controls progressively. All regressions include robust standard errors clustered at the suburb-year level. In the first column, I regress log house satisfaction on own log housing size, the 90th percentile log size of new houses built, length of tenure and suburb-year fixed effects. The coefficient on own housing size is positive and significant. A 1% rise in own size leads to a 0.08% rise in house satisfaction. However, a 1% rise in the 90th percentile size of new houses leads to a 0.07% fall in house satisfaction. The estimated σ coefficient is slightly below unity, which means an owner must upscale his house at almost the same rate as top housing size to maintain his house satisfaction over time. Column (2) adds controls for house and neighborhood quality. Some of these correlate to housing size such as the current market value of the house (normalized per square feet), the number of extra bathrooms or the mortgage service ratio. The coefficients on housing size are therefore smaller. However, the σ coefficient remains significant and close to one. The estimated impact of own size on house satisfaction is of similar magnitude as estimates previously found on the relationship between log income and life satisfaction. For level-log OLS and ordered probit regressions, the coefficient on log housing size lies within the range of estimates found on log income for the Cantril ladder (Stevenson and Wolfers 21 House characteristics include the purchased price of the house per square feet, its current market value per square feet, the age of the house, whether the unit has a balcony, whether the heating equipment is functional, the presence of holes in the floor or roof, whether the unit has extra bathrooms, whether the unit experienced any water leak in the past twelve months, and whether there has been home remodeling since the house was bought. 22 Households control are the age of the household s head and its square, his race, sex and level of education, the log of the household s annual income, the mortgage debt service ratio, the number of persons in the household and the number of cars in the household.

18 18 Table 2 : Baseline Regression of Log House Satisfaction Dependent variable: (1) No Controls (2) House + Neighbor. (3) Full Control Log(HouseSatisfaction) Controls List Coeff. S.E. Coeff. S.E. Coeff. S.E. Log(OwnSize) ( ) ( ) ( ) Log(90thPercentileSize) (0.0181) (0.0174) (0.0174) House and neighborhood quality: Log(MarketValue/Sqft) ( ) ( ) Log(PurchasePrice/Sqft) ( ) ( ) Mortgage service ratio ( ) ( ) Upscaling last 2 years ( ) ( ) Number of full bathrooms ( ) ( ) Age of the house ( ) ( ) Water leaks in last 12 months ( ) ( ) Broken heating equipment ( ) ( ) Unit has porch/deck/balcony ( ) ( ) Log(NeighborhoodSatisfaction) ( ) ( ) Household characteristics: Ln household annual income ( ) Number of cars ( ) Household size ( ) Age of householder ( ) Education of householder ( ) Latino ( ) Black ( ) Sex of householder ( ) Observations Adjusted R County x Year FE Yes Yes Yes Length of tenure FE Yes Yes Yes Note: This table reports the OLS estimation of equation (2), which regresses the logged subjective house satisfaction index on the logged values of own housing size and the 90th percentile size of houses built in the household s suburb after he moved in. The first column includes suburb-year fixed effects and length of tenure fixed effects. Column (2) also controls for house and neighborhood quality. Column (3) adds controls for household characteristics. Sampling weights are included in all regressions. Robust standard errors clustered at the county-year level are reported in parentheses. p < 0.10, p < 0.05, p < 0.01 (2008); Deaton and Stone (2013), see Table E2 in appendix). Experiencing the construction of twice bigger McMansions has a similar negative impact on house satisfaction as a water leak in the last twelve months. Home improvements or neighborhood satisfaction are positively related to house satisfaction, while an aging house or a dysfunctional heating system reduce house satisfaction. The purchase price and current market value of the house are both positively associated with house satisfaction, but only the latter shows a significant coefficient.

19 THE MCMANSION CURSE 19 This suggests households care more about changes in what than levels 23. Adding household characteristics in the regression does not significantly affect the housing size coefficients, as can be seen in column (3). Since the identification relies on the unanticipated construction of McMansions, any construction that occurred in the years preceding the moving decision should not affect current house satisfaction as these houses were already in place when the decision to buy a house was made. Table E3 in the appendix reproduces the analysis by adding the 90th percentile size of houses built during the n years prior to the household s moving decision, where n corresponds to the household s length of tenure. This placebo measure has no impact on current house satisfaction once we control for household and other house characteristics, which further supports the identification strategy. Homeowners' sensitivity to size of newly built houses (σ coefficient) Suburb/time FE Suburb/time FE + Controls p th percentile size of newly built houses 99 Figure 6. : Homeowners Sensitivity to the Size of New Houses Note: The figure shows the sensitivity to the size of new houses σ = γ 2 /γ 1 estimated from regression (2), with or without the addition of controls for house and household characteristics. Each dot represents the estimated σ coefficient of a separate regression where the p th percentile size of newly built houses is used as reference housing size. In the relative income literature, results can vary based on how the reference group is defined. Data limitations explain why most empirical studies rely on a single measure like the mean attribute of a reference group. 24 My sample 23 Ratcliffe et al. (2010) finds a positive impact of house price inflation on life satisfaction similar for homeowners and renters. This suggests housing wealth has no strong impact on life satisfaction. 24 For a discussion on the relative income hypothesis and the definition of the reference group in happiness economics, see Clark, Frijters and Shields (2008).

20 20 of American houses is large enough to test different reference groups along the size distribution of new houses. A first hypothesis is the trickle-down effect (or expenditure cascade ), according to which any reference level can be traced back to the biggest houses built 25. Though Table 2 seems to support this hypothesis, the 90th percentile size could capture a general effect of other houses size on house satisfaction. To test whether households are upward looking, I estimate the relative size coefficient σ across ten size percentiles of newly built houses. I first run each regression separately taking a given size percentile as reference group to avoid issues related to correlated measurement errors. Figure 6 plots the estimated σ coefficients from the 10th to the 90th percentile, along with the 99th percentile. As discussed earlier, adding other house characteristics as controls does not alter significantly the value of the coefficients but generates less precise estimates of relative size 26. I hence show the estimated coefficients with and without the addition of controls. Results are consistent with upward-looking comparisons in housing size, or trickle-down reference effects. The sensitivity to the size of new houses built starts to be significant within the top quintile of the size distribution. It is higher for the 90th size percentile, with a σ coefficient close to unity. In other words, the house satisfaction benefits of a 1% increase in own housing size are almost offset by a 1% increase in the size new McMansions. Interestingly, the 99th percentile coefficient is lower and less significant. This could be due to the lower visibility of these superstar houses compared to relatively smaller McMansions. Indeed, these superstar houses tend to be isolated from other houses and located in very exclusive areas (Blakely and Snyder, 1997). Each coefficient in Figure 6 corresponds to a separate regression. However, the results hold if I control for other parts of the size distribution. Table E4 in the appendix shows the results on other housing size for the 10th, 30th, 70th and 90th size percentiles of newly built houses. Again, only the 90th percentile size of new houses built affects house satisfaction significantly. In the rest of the paper, I use the terms McMansions and top housing size interchangeably to characterize this measure of reference housing size. As discussed in the previous section, if the effect is indeed driven by relative size, we should expect a negative interaction between own housing size and households experience in the construction of new McMansions. First, social interactions are more frequent between similar individuals, for instance because they live close to each other. Second, people stop forming positive aspirations when the reference point is not reachable, or lies outside of their aspiration window (Genicot and Ray, 2014). Trickle-down comparisons therefore imply the σ coefficient on top housing size should be stronger for households living in big houses. The first column of Table 3 shows the coefficients on own size and 90th percentile size with- 25 See Schor (1999), Frank et al. (2010), or Bertrand and Morse (2016) for further discussion and evidence on trickle-down consumption. 26 Controls include variables partly collinear with own housing size such as the current market value of the house (which I normalize per square feet to reduce collinearity) or the number of bathrooms.

21 THE MCMANSION CURSE 21 Table 3 : Trickle-Down Reference Effects - Interaction with Own Housing Size Dependent variable: (1) (2) (3) Log(HouseSatisfaction) Coeff. S.E. Coeff. S.E. Coeff. S.E. Log(OwnSize) ( ) (0.0827) (0.0812) Log(90thPercentileSize) (0.0174) (0.0813) (0.0851) Log(OwnSize) Log(90thPercentileSize) (0.0102) (0.0108) Log(OwnSize) Log(OwnSize) ( ) Observations Adjusted R Suburb x Year FE Yes Yes Yes Length of tenure FE Yes Yes Yes Controls Yes Yes Yes Note: Regressions have the logged house market value as the dependent variable and I report coefficients on the logged values of own housing size and the 90th percentile size of houses built in the household s suburb after he moved in. Column (1) controls for suburb-year fixed effects, tenure length fixed effects, and the full list of house, neighborhood and household characteristics. The second column adds the interaction between own and 90th percentile housing size. Column (3) adds a quadratic term on own size. Sampling weights are included in all regressions. Robust standard errors clustered at the county-year level are reported in parentheses. p < 0.10, p < 0.05, p < 0.01 out the interaction. Column (2) adds the interaction term, which is negative and highly significant. It shows the negative sign on top housing size is entirely driven by its interaction with own house size. The negative interaction could capture non-linearities in the coefficient on log house size as bigger houses may be located in neighborhoods which also experienced the construction of bigger houses. However, adding a quadratic term on log own house size does not significantly alter the results (column (3)) 27. In order to provide a more intuitive interpretation of the results, Figure 7 plots the estimated σ coefficients of equation (2) resulting from the interaction of own and top housing size with the decile of own house size within suburbs. It shows a clear increasing sensitivity to new-build McMansions as one s own house becomes comparable in size to the 90th percentile houses. Within suburbs, the house satisfaction of households living in first decile houses is not affected by the construction of new McMansion. However, σ quickly becomes significant and close to unity for above-median houses. Pre-existing top decile houses are the most negatively impacted by the McMansion effect: the percent rise in own house size must be twice as large as the percent rise in newly built McMansions for them to maintain their level of house satisfaction (σ = 2). These results are consistent with the trickle-down consumption (or expenditure cascade) hypothesis which posits that households ranked well down the hierarchy are less affected by the higher ranked group than households just below this group (Frank, Levine and 27 The quadratic term is negative and significant when I remove the interaction with top housing size. This suggests satiation in absolute (and not relative) housing size, contrary to Stevenson and Wolfers (2013) who find no evidence for absolute income satiation using the life satisfaction ladder.

22 22 Sensitivity to top housing size (σ coefficient) Size decile of own house within suburb Figure 7. : Sensitivity to Top Housing Size by Own Size Decile Within Suburbs Note: The figure shows the sensitivity to top housing size σ estimated from regression (2) by decile of own housing size. Homeowners belonging to the 1st decile live in the bottom 10% of houses in their suburb in terms of size. Dijk, 2014; Bertrand and Morse, 2016). They also suggest social comparisons in housing size may amplify housing size inequality as status competition on relative size seems to be particularly strong at the top of the size distribution. B. Impact of the McMansion Effect on Self-Assessed Market Values Since house satisfaction is a strong predictor of house market values, 28 we should expect householders who experienced a rise in top housing size to value their house at a lower price. Assuming upward-looking comparisons are internalized in homeowners potential selling or asking price, one may infer a monetary cost of experienced relative downscaling. 29 The American Housing Survey provides information on the current market value of homes as assessed by the homeowner. This represents the discounted present value of the total services provided by the house. These services incorporate the structure services along with the service flows coming from neighborhood amenities or disamenities. Market values in the AHS are generally higher in levels than transaction prices, but have quite similar time-series patterns (DiPasquale and Somerville, 1995; Kiel and Zabel, 1999) The elasticity of house market values to house satisfaction is The underlying assumptions distinguishing the hedonic pricing method from the house satisfaction method are discussed in appendix B. 30 Kiel and Zabel (1999) also shows the gap between market values and transaction prices is not significantly related to household characteristics except the length of tenure, which I add as a control in

23 THE MCMANSION CURSE 23 Table 4 : Regression of Log Current House Market Values Dependent variable: (1) (2) (3) Log(MarketValue) Coeff. S.E. Coeff. S.E. Coeff. S.E. Log(OwnSize) (0.0175) (0.0107) (0.363) Log(90thPercentileSize) (0.361) (0.0857) (0.338) Log(70thPercentileSize) (0.630) (0.128) (0.128) Log(50thPercentileSize) (0.655) (0.125) (0.125) Log(30thPercentileSize) (0.427) (0.0992) (0.0990) Log(10thPercentileSize) (0.271) (0.0711) (0.0708) Log(OwnSize) Log(90thPercentileSize) (0.0429) Log(OwnSize) Log(OwnSize) (0.0122) Observations Adjusted R Year FE Yes Yes Yes Length of tenure FE Yes Yes Yes Suburb x Year FE No Yes Yes Controls No Yes Yes Note: Regressions have the logged house market value as the dependent variable and I report coefficients on the logged values of own housing size and the 90th, 70th, 50th, 30th and 10th percentile size of houses built in the household s suburb after he moved in. Column (1) control for tenure length fixed effects. Column (2) adds suburb-year dummies and the full list of house, neighborhood and household characteristics. The third column adds the interaction between own and 90th percentile housing size and a quadratic term on own size. Sampling weights are included in all regressions. Robust standard errors clustered at the county-year level are reported in parentheses. p < 0.10, p < 0.05, p < 0.01 The same identification issue previously discussed arises if one regresses current market values on the current size of neighboring houses: due to the endogenous sorting of householders who dislike big houses in areas with smaller houses, the relative size effect will be biased towards zero. In the case of hedonic pricing, another challenge relates to the first order effect of housing size on land prices. If bigger houses are more likely built in expensive areas (or lead to increasing land prices), the coefficient will rise. Conversely, if a negative link exists at the county level between top housing size and the general level of house prices (e.g. because price-per-square-foot is lower in economically deprived areas), the negative McMansion effect may simply capture a negative wealth shock and lower price expectations. The latter should affect all houses similarly regardless of relative size and homeowners length of tenure and will be captured by suburbyear fixed effects. 31 It seems reasonable to assume that the first order effect is positive: the construction of McMansions is associated with increasing housing prices, in part all regressions. 31 Note that this first order effect on land prices should be even less of a concern for the house satisfaction approach as we control for purchase price of houses and current market values.

24 24 because it corresponds to the entry of richer households. This will be especially true in metropolitan areas where housing supply is constrained (Gyourko, Mayer and Sinai, 2006; Mian and Sufi, 2009). However, it may also be that due to lower land prices, economically deprived areas lead to a rise in top housing size. Zillow provides a time series of county-level house prices for all counties in my dataset between 1997 and Controlling for year and county fixed effects, I find yearon-year variation in housing prices to be positively correlated to changes in 90th percentile housing size (see Table E5 in appendix). This remains true when controlling for other parts of the size distribution. To derive the monetary cost of the McMansion effect, I follow the log linear approach for estimating hedonic house price functions (Ioannides and Zabel, 2003; Zabel, 2004) replacing the subjective house satisfaction index from regression (2) with the current market value of the house: log(marketvalue) ist = γ 0 + γ 1 log(ownsize) ist + γ 2 log(pthpercentilesize) ist +SuburbTrends st + TenureLength it + γ 3 Controls ist + u ist (3) The positive first order effect caused by the construction of big houses should affect all units regardless of when households moved. Removing this effect via the inclusion of suburb-year fixed effects allows me to identify the impact of households own experience in the construction of McMansions. In the presence of upward-looking comparisons, the coefficient on top housing size should become negative. All else being equal, households who experienced the construction of McMansions after moving should value their house relatively less compared to similar households who moved at a time where such houses were already built. Results are shown in Table 4. Without controlling for county-year effects, the construction of big houses in one s suburb is associated with a major rise in the current market value of a house, as can be seen in column (1). In column (2), I introduce county-year fixed effects. The coefficient on 90th percentile housing size becomes negative and significant. Controlling for suburb-year fixed effects, households who experienced a 1% higher increase in top housing size value their house 0.35% less. Interestingly, this effect offsets the residual impact of own house size, which echoes the result found on house satisfaction. There is also evidence that this effect is stronger for houses comparable in size to McMansions, as can be seen in column (3). C. Impact of McMansion Effect on Housing Choices If homeowners care about how their house compares to the largest houses in their suburbs, we should expect them to react following the construction of new McMansions. They may move to an even bigger house or increase the size of their current home. The later is likely to happen if moving costs are high. Moreover, if households fund additions to their house in part by taking up more debt, I should

25 THE MCMANSION CURSE 25 expect some positive effect of top housing size on mortgage debt. Between 1945 and 2009, mortgage debt went from 20% to 90% of households annual income. Figure 8 plots the mortgage debt-to-income ratio against the 90th percentile size of new houses built each year. These two measures follow very similar trends, with a correlation coefficient of Mortgage debt to income ratio Mortgage debt to income ratio 90th percentile housing size, new houses th percentile housing size (sqft) Figure 8. : American mortgage debt to income ratio vs. 90th percentile house size ( ) Note: The left axis shows the variation in average mortgage debt to annual income ratio. The right axis shows the variation in the 90th percentile size of houses built each year over the same period. Source: Lustig and Van Nieuwerburgh (2005) and author s own calculation from Zillow First, I want to check whether households who recently upscaled the size of their house are less sensitive to the construction of new-build McMansions experienced since they first moved. The AHS asks respondents whether there has been a change in the amount of living space in the housing unit because of putting on an addition since two years. Considering the average tenure length in my dataset is twelve years, this upscaling in size can be considered as quite recent. Table 5 shows the house satisfaction gains of a recent improvement in size is entirely driven by its interaction with households experience in the construction of McMansions. Householders who recently added extra space to their house do not experience the negative effect on house satisfaction. Keeping up with the Joneses can be considered as a reasonable strategy to restore house satisfaction. Table 5 only looks at recent size improvements. It may also be that households who decide to upscale are less sensible to the size of other houses. To estimate the effect of McMansions on house size improvements, I focus on a panel sub-sample of my dataset. This reduces my smaller sample size, but the panel specification

26 26 Table 5 : Impact of Recent Home Size Improvements on McMansion Effect Dependent variable: (1) (2) Log(HouseSatisfaction) Coeff. S.E. Coeff. S.E. Log(OwnSize) ( ) ( ) Any upscaling in last 2 years ( ) (0.236) Log(90thPercentileSize) (0.0174) Log(90thPercentileSize) Upscaling in last 2 years (0.0334) Log(90thPercentileSize) No upscaling in last 2 years (0.0174) Observations Adjusted R County x Year FE Yes Yes Length of tenure FE Yes Yes Controls Yes Yes Note: Regressions have the logged house satisfaction as the dependent variable. All regressions control for suburb year fixed effects, length of tenure fixed effects and the full list of controls for household, house and neighborhood characteristics. Column (1) reports the coefficients on own size, top housing size and whether the householder improved the size of his house in the last 2 years. Column (2) adds the interaction between upscaling and the 90th percentile housing size. Sampling weights are included in all regressions. Robust standard errors clustered at the county-year level are reported in parentheses. p < 0.10, p < 0.05, p < 0.01 allows me to look at within-household changes in own housing size accounting for any household-specific time invariant characteristics. I recover household identifiers from consistent answers to four basic questions across waves: whether within a given house, householders have the same sex, race, moved in the same year, and have the age we should expect based on the number of years separating each survey. After removing missing values on the main control variables, I end up with an unbalanced panel of 32,634 homeowners surveyed two to four times between 1984 and I then run the following household fixed effects model: log(ownsize) ist = γ 1 log(pthpercentilesize) st + γ 2 Controls ist +Household i + Year t + u ist (4) where log(ownsize) it and log(p thp ercentilesize) st are the logged values of household i s house size at time t and the p th percentile size of other houses in suburb s at time t. The panel specification allows me to look at whether withinsuburb changes in McMansions size between two surveys affect within-household changes in the size of their own houses (i.e. home size improvements). Household fixed effects Household i account for any unobserved house and household specific characteristics that do not vary over time. It eliminates Household i by 32 More precisely, 78% of householders are only surveyed twice, 19% three times and 2% four times. Five years separate each survey on average.

27 THE MCMANSION CURSE 27 demeaning the variables between survey years using the within transformation. Robust standard errors are clustered at the household level. I also control for house and household characteristics that may change over time. A positive γ 1 coefficient could be explained by a general upward trend in preferences for bigger houses. Year fixed effects partly account for that possibility, but I also control for changes in the size of new-build houses coming from lower percentiles of the size distribution. Results are shown in Table 6. Table 6 : Panel Regressions - Own Size Improvements Dependent variable: (1) FE Estimator (2) FE Estimator Log(OwnSize) Coeff. S.E. Coeff. S.E. Log(90thPercentileSize) (0.0348) (0.0667) Log(70thPercentileSize) (0.0937) Log(30thPercentileSize) (0.0885) Log(10thPercentileSize) (0.0985) House and neighborhood quality Log(CurrentMarketValue/Sqft) ( ) ( ) Number of full bathrooms in unit ( ) ( ) Age of the house ( ) ( ) Any inside water leaks in last 12 months ( ) ( ) Main heating equipment broke down ( ) ( ) Unit has porch/deck/balcony/patio ( ) ( ) Log(NeighborhoodSatisfaction) ( ) ( ) Household characteristics Ln household annual income ( ) ( ) Number of cars ( ) ( ) Household size ( ) ( ) Age of householder ( ) ( ) Education of householder ( ) ( ) Observations 72,796 72,796 Adjusted Within R Adjusted Between R Year FE Yes Yes Household/House FE Yes Yes Note: Panel fixed effect regressions have the logged value of own housing size in year t as the dependent variable. All regressions control for household fixed effects, year fixed effects and a list of time varying household, house and neighborhood characteristics. Column (1) takes as a reference group the 90th percentile house size of the stock of houses in suburb s at time t. Column (2) adds the 70th, 30th and 10th percentile house size. Panel robust standard errors clustered at the household level are reported in parentheses. p < 0.10, p < 0.05, p < 0.01 A 1% rise in the 90th percentile house size of a given suburb is associated to a

28 % improvement in the size of existing houses. The elasticity of own house size to the construction of new McMansions goes to 0.2 once I control for lower percentiles of the size distribution. Again, only the 90th percentile remains positive and significant. These results are consistent with the trickle-down consumption hypothesis. Households react to relative downscaling by increasing the size of their own house. They may do so by subscribing to new mortgages or by emptying their saving accounts. They may also use home equity loans, which became a popular source of credit in the decade preceding the 2008 financial crisis. However, information on the later only started being collected in last five waves of the AHS, from 1998 onwards. I therefore use the total value of acquired mortgages as a measure of mortgage debt. To test whether an increase in top housing size leads existing homeowners to take up new mortgages, I run a similar regression as specification (4) where I replace log(ownsize) ist by the total value of the acquired mortgages at the time a household is being surveyed, OwnMortgage ist. However, a strong fraction of my panel dataset is left-censored at zero. 33 I estimate a Poisson fixed effect regression as a way to deal with zero values of the dependent variable (Silva and Tenreyro, 2006). Results are shown in Table 7. The coefficient on top housing size in weakly significant but positive: a 1% rise in the 90th percentile house size leads to a 0.35% estimated increase in mortgage debt. Adding other percentile increases the value of the coefficient on top housing size but lowers its significance. This is partly due to the strong collinearity between these measures of reference size. Running each measure separately, only top housing size stands out. On average, the 90th percentile size of the housing stock went from 2650 square feet to 3350 square feet between 1980 and 2007, which corresponds to a 26% rise. Comparatively, below median houses went from an average 1150 square feet to 1300 square feet, a 13% rise. Had McMansions grown at the same rate as below median houses, the mortgage debt to income ratio would have been 4.5 percentage points lower at the eve of the 2008 financial crisis. Since the mortgage debt to income ratio increases by 40 percentage points from 1980 to 2008, the McMansion effect can explain up to 20% of the rise in the mortgage debt-to-income ratio over the period. IV. Alternative Hypotheses A. Relative Size vs. Neighborhood Externalities? The negative effect of top housing size on house satisfaction and house prices may hide a more general negative externality correlated with the experienced construction of McMansions. For instance, the construction of McMansions may be associated with experienced gentrification altering the identity of homeowners neighborhood, or with a rise in segregation within suburbs. The opposite effect could also arise if gentrification is perceived as a positive externality (i.e. a negative σ coefficient). The inclusion of neighborhood satisfaction in all regressions 33 Only 62% of homeowners in the panel reported a mortgage value for at least one period.

29 THE MCMANSION CURSE 29 Table 7 : Poisson FE Regressions - Mortgage Debt Dependent variable: (1) Poisson FE (2) Poisson FE MortgageDebt Coeff. S.E. Coeff. S.E. Log(90thPercentileSize) (0.191) (0.396) Log(70thPercentileSize) (0.533) Log(30thPercentileSize) (0.567) Log(10thPercentileSize) (0.551) House and neighborhood quality Log(CurrentMarketValue/Sqft) ( ) ( ) Number of full bathrooms in unit (0.0127) (0.0127) Age of the house ( ) ( ) Any inside water leaks in last 12 months (0.0103) (0.0103) Main heating equipment broke down (0.0210) (0.0210) Unit has porch/deck/balcony/patio (0.0142) (0.0142) Log(NeighborhoodSatisfaction) (0.0179) (0.0179) Household characteristics Ln household annual income ( ) ( ) Number of cars ( ) ( ) Household size ( ) ( ) Age of householder ( ) ( ) Education of householder ( ) ( ) Observations Wald Chi-Square Year FE Yes Yes Household/House FE Yes Yes Note: Panel fixed effect regressions have the logged value of all mortgages reported by the household at time t as the dependent variable. All regressions control for household fixed effects, year fixed effects and a list of time varying household, house and neighborhood characteristics. Column (1) takes as a reference group the 90th percentile house size of the stock of houses in suburb s at time t. Column (2) adds the 70th, 30th and 10th percentile house size. Panel robust standard errors clustered at the household level are reported in parentheses. p < 0.10, p < 0.05, p < 0.01 should capture these effects. However, a more direct test can be performed taking neighborhood satisfaction as the main outcome variable in equation (2). Comparing the γ 2 coefficient on top housing size for house and neighborhood satisfaction should be indicative of whether top housing size generates a positive or negative neighborhood externality. The results show top housing size is not significantly related to neighborhood satisfaction (Table E6 in appendix). If anything, others housing size generate a positive (though not significant) externality on neighborhood satisfaction when it comes from the bottom of the size distribution. The construction of bigger McMansions may also be correlated to population density within counties, as it captures the construction of additional houses. The

30 30 fact that only the size of big houses is driving my results lowers the concern, but it may still bias the coefficient. The impact of population density on house satisfaction is theoretically ambiguous. Higher density increases the price of land for existing home owners, which can lead to higher house satisfaction, but may also be associated with higher congestion costs not necessarily captured in prices, which is likely to reduce house satisfaction. To address this concern, I compute county-specific trends in population density between 1920 and 2009 for each AHS county from US Census population data and NHGIS and control for the average growth rate in population density experienced over the length of tenure. Adding experienced growth in population density to the regressions do not affect any of the main results found on top housing size. B. Geographic Distance to McMansions I found evidence of a stronger sensitivity to McMansions for homeowners living in bigger houses. This trickle-down effect evidenced in Figure 7 is consistent with a stronger preference for status at the top of the distribution. However, it could also be explained by the higher visibility of McMansions for homeowners already living in big houses, assuming McMansions get built closer to other big houses. In other words, rather than a heterogeneity in social status concerns, the decreasing sensitivity to McMansions as a function of own housing size may simply capture the endogenous segregation of new McMansions. To illustrate this concern, I compute for each year and within each county the geodetic distance separating the average biggest ten percent houses from below median size houses, using latitude and longitude information from Zillow. This measure of housing segregation averaged over all suburbs is correlated to the McMansions effect, at least since 1960 (see Figure E6 in appendix). The estimation of the social comparison effect within suburbs must therefore account for this critical fact. Unfortunately, the AHS does not provide the exact location of a house within suburbs. However, the overlapping variables between the American Housing Surveys and Zillow allow me to predict the location of each AHS house. I run multiple imputations within each suburb using the size of the house, the size of the lot, and their interaction with the year the house was built. To avoid outliers in predicted longitude and latitude that would lie outside of the suburb s boundaries, I use a truncated regression that defines the minimum and maximum latitude and longitude of the suburb as the upper and lower bounds. 34 I can then construct for each homeowner the average predicted distance Distance ist separating his own house from McMansions built in his suburb since he moved in. To see whether the elasticity of house satisfaction to top housing size varies with the predicted distance to newly built McMansions, I look at the interaction coefficient between the size of the 90th percentile houses built during the household s tenure period and the 34 See appendix D for the details of the imputation method. The use of non-parametric imputation methods like the random forest algorithm proposed by Stekhoven and Bühlmann (2011) did not improve my predictions significantly.

31 THE MCMANSION CURSE 31 estimated distance between his own house and these newly built McMansions. Table 8 : Impact of Geographic Distance to New McMansions Dependent variable: (1) (2) Log(HouseSatisfaction) Coeff. S.E. Coeff. S.E. Log(OwnSize) (0.0933) (0.0937) Log(90thPercentileSize) (0.0949) (0.0954) Log(90thPercentileSize) Log(OwnSize) (0.0115) (0.0116) Distance from McMansions (miles) ( ) Log(90thPercentileSize) Distance from McMansions (miles) ( ) Observations 94,207 94,207 Adjusted R Suburb x Year FE Yes Yes Length of tenure FE Yes Yes Controls Yes Yes Note: Regressions have the logged house market value as the dependent variable. All regressions controls for suburb-year fixed effects, tenure length fixed effects, and the full list of house, neighborhood and household characteristics described in section II. Column (1) report coefficients on the logged values of own housing size, the 90th percentile size of new-build houses and their interaction. Column (2) adds the average distance separating a household s house from new-build McMansions and its interaction with top housing size. Sampling weights are included in all regressions. Robust standard errors clustered at the county-year level are reported in parentheses. p < 0.10, p < 0.05, p < 0.01 Table 8 shows the interaction coefficient between own and top housing size with and without controlling for distance. As can be seen in column (2), the interaction coefficient between top housing size and distance is positive and significant. In other words, the negative impact of McMansions is smaller when these new houses are constructed further away from the households house. However, the effect of predicted distance is small and does not seem to be driving the interaction with own housing size. Indeed, the interaction term between top and own housing size does not change significantly between column (1) and column (2). The fact that most suburban households commute by car also implies within suburb distance might play a smaller role compared to areas where individuals commute on foot. To provide more intuitive estimates and to account for possible non-linearities, Figure E7 in appendix plots the σ coefficient of relative housing size obtained from the interaction between log(90thpercentilesize) ist and dummies of estimated distance to newly constructed McMansions. As before, an increase in top housing size during tenure period reduces house satisfaction. However, this effect tends to decrease with distance, so that beyond 15 miles the effect of top housing size is not significant. This corresponds to a 20 minute commute by car, which is slightly below the average commuting time in American suburbs The average predicted distance between a given household and the McMansions built during his tenure period is 10 miles. See Figure D2 in appendix for the distribution of average distances.

32 32 C. Unobserved Household Heterogeneity The regressions of house satisfaction and house values presented in section III did not controlled for household specific fixed effects, which may bias the results if an individual trait is linearly related to the construction of new McMansions. This should not be a concern if the construction and size of new houses is exogenous to the household s initial moving decision. However, households more sensitive to social comparison may be better at forecasting the construction of bigger houses, and so be less likely to experience unpredicted future changes in relative house size. This would bias the coefficient on top housing size towards zero. The fact that I still find significant estimates indicates that the influence of this effect may be limited. Still, to account for household-specific time invariant characteristics like unobserved forecasting abilities, I run the house satisfaction regression on the smaller panel of households described in section III.C. I reproduce specification (4) except that I replace log(ownsize) it by the logged value of house satisfaction log(housesatisfaction) it. The within-household specification gives even stronger results, as shown in Table 9. Table 9 : Fixed Effect Estimator of Relative Housing Size on House Satisfaction Dependent variable: (1) Pooled OLS (2) FE Estimator (3) FE Estimator Log(HouseSatisfaction) Coeff. S.E. Coeff. S.E. Coeff. S.E. Log(OwnSize) ( ) ( ) ( ) Log(90thPercentileSize) ( ) (0.0480) (0.102) Log(70thPercentileSize) (0.146) Log(30thPercentileSize) (0.148) Log(10thPercentileSize) (0.142) Observations 64,063 64,063 64,063 Adjusted R Adjusted Within R Adjusted Between R Household/House FE No Yes Yes Year FE Yes Yes Yes Controls Yes Yes Yes Note: Pooled OLS and fixed effects regressions have the logged value of house satisfaction in year t as the dependent variable. Column (1) controls for year fixed effects and the list of time varying household, house and neighborhood characteristics shown in Table 6. It takes as a reference group the 90th percentile house size of the stock of houses in suburb s at time t. Column (2) adds household fixed effects. Column (3) adds the 70th, 30th and 10th percentile house size as reference groups. Panel robust standard errors clustered at the household level are reported in parentheses. p < 0.10, p < 0.05, p < 0.01 Column (1) runs a pooled OLS regression with year fixed effects. It also controls for the list of time varying household, house and neighborhood characteristics

33 THE MCMANSION CURSE 33 shown in Table 6. The coefficients are similar in magnitude as the ones found on the full sample using specification (2) in section III.A. It gives a σ coefficient close to unity. In column (2), I add the household fixed effects. The coefficient on own house size now captures the impact of own size improvements, and the coefficient on top housing changes in the 90th percentile size of the housing stock between two survey years. The latter is now twice the size of the coefficient on own house size, though less significant. This suggests some degree of endogenous sorting based on future reference housing size. When I control for lower percentiles of the size distribution, only the 90th percentile is significant. Overall, these results confirm my previous findings on the full sample (section III.A). If anything, the latter should be interpreted as upper bound estimates of the true effect. D. Relative Novelty and Hedonic Adaptation Bigger houses also tend to be newer. Therefore, the construction of McMansions may correlate with their unobserved quality, capturing better design, more efficient heating technologies, or the mere value of novelty. Controlling for the age of a household s house like I do in all regressions partly addresses this issue. But variations in top housing size may still capture a relatively higher proportion of newer houses, which in the presence of a relative novelty effect would bias the coefficient on reference housing size upward. A more convincing test is to look at the interaction between a house s construction year and a household s experience in the size of new-build houses. If relative size is also capturing a novelty effect, the impact of new-build McMansions should be more negative on older houses. Column (1) in Table 10 shows that the interaction term between top housing size and the age of the house is positive for both house satisfaction and the market value of the house, dismissing the relative novelty explanation. However, the positive sign is consistent with the impact of distance previously discussed, since newer houses tend to be built further away from older ones. An alternative explanation related to time may also explain the negative effect on house valuation. Households who experienced bigger increases in top housing size also stayed in the same house for a longer period. They may therefore adapt to the size of their house. Hedonic adaptation to the house in general is controlled for by the inclusion of tenure length fixed effects. However, adaptation to size in particular is better captured by the interaction between house size and length of tenure. Households may also adapt to the size of other houses. Conversely, they may only remember their most recent experiences in the construction of new houses. Indeed, retrospective evaluations of past experiences tend to follow a weighted average where lower weights are placed on older experiences (Kahneman et al., 1993; Malmendier and Nagel, 2016). However, in the case of house constructions, households are constantly reminded of older experiences since newbuild houses rarely get destroyed. To test whether there is evidence of hedonic adaptation in own and other s house size, I interact own and reference house size with a continuous measure of

34 34 Table 10 : Testing for Relative Novelty and Hedonic Adaptation in Size Dependent variable: (1) (2) Log(HouseSatisfaction) Coeff. S.E. Coeff. S.E. Log(OwnSize) ( ) ( ) Log(90thPercentileSize) (0.0184) (0.0173) Log(90thPercentileSize) Age of the house ( ) - Log(OwnSize) Length of tenure ( ) Log(90thPercentileSize) Length of tenure ( ) Observations Adjusted R County x Year FE Yes Yes Length of tenure FE Yes Yes Note: Regressions have the logged house satisfaction as the dependent variable. All regressions controls for suburb-year fixed effects, tenure length fixed effects, and the full list of house, neighborhood and household characteristics described in section II. Column (1) report coefficients on the logged values of own housing size, the 90th percentile size of new-build houses and the the latter s interaction with the house s age. Column (2)report coefficients on the logged values of own housing size, the 90th percentile size of new-build houses and their interaction with the length of tenure. Sampling weights are included in all regressions. Robust standard errors clustered at the county-year level are reported in parentheses. p < 0.10, p < 0.05, p < 0.01 homeowners length of tenure. I keep the same full sample specification described in equation (2) of section II. Column (2) in Table 10 shows households adapt to the size of their house, but this does not explain the negative sign found on top housing size. Conversely, there is no adaptation to the size of other houses. The sign is positive but not significant. This dichotomy in hedonic adaptation when it comes to oneself versus others may be specific to the case of durable consumption. However, it suggests that unless individuals react to relative deprivation, the psychic cost associated with the feeling of falling behind may only exacerbate over time. V. Conclusion This article showed that despite a major upscaling in size of single-family houses in US suburbs, households did not experience a significant increase in house satisfaction since the 1980s. However, cross-sectional analysis suggests households living in bigger homes tend to be more satisfied with their home. This result echoes the Easterlin paradox, which is usually explained by adaptation and rising aspirations due to the presence of income comparisons. I identify comparison effects in the size of neighboring houses. My methodology addresses endogenous sorting as I look at personal experiences in the size of new-build houses within a household s suburb after his purchase decision has been made. It also allows me to control for time effects and any suburb-specific

35 THE MCMANSION CURSE 35 characteristics affecting all households similarly over time. I show householders who experienced a high increase in top housing size report lower house satisfaction than similar households who experienced a smaller rise. The comparison effect is upward-looking: social comparison are driven by the size of McMansions, defined as houses belonging to the top decile of the size distribution. These results provide a strong support to prior research on trickle-down consumption. The utility gains from a bigger house are almost offset by a rise in size new-build houses at the top of the distribution, and the effect is stronger for larger homes. This McMansion externality is robust to alternative specifications. Taking the variation of top housing size experienced by the same household between two survey years give coefficients of similar magnitude, or even stronger after controlling for household fixed effects. Using the current market value of the house instead of subjective house satisfaction, I also show households ascribe a lower market value to their house if they experienced a higher increase in top housing size during their tenure period. This means real estate agents should account for the overall size distribution in a suburb and the personal histories of owners when trying to assess the current asking price for a house. The McMansion effect also affects the decision to upscale the size of one s own house. Households react to relative deprivation by increasing the size of their house at the cost of higher levels of mortgage debt. Controlling for household fixed effects, a 1% rise in size of McMansions leads to a 0.16% rise in size through home improvements and a 0.35% rise in the level of outstanding mortgage debt. In the absence of keeping up with the Joneses on top housing size, the level of mortgage debt to income ratio would have been 9 percentage points lower on the eve of the 2008 financial crisis. This accounts for 20% of its rise since More generally, this paper showed inequality can have major implications for the dynamics of consumer markets through its impact on status competition. In terms of house satisfaction, competing for relative size becomes a negative sum game for those living in above median size houses. Unfortunately, I cannot evaluate the general welfare implications of relative size competition due to the absence of a general life satisfaction question. However, I show McMansions do not improve neighborhood satisfaction; only when size improvements come from the bottom of the distribution some positive effects can be identified. Furthermore, once the satisfaction gains of living in a relatively bigger house vanish, households are left with the negative long-run impact of their decisions in terms of extra borrowing costs. These results are supportive of regulations (zoning laws, housing permits) aimed at restricting the maximum size allowed for a house to be built within local areas 36. Some counties put these regulations in place already, such as in DeKalb County, Georgia or Arlington County, Virginia. In this regard, the extensive use of minimum lot size requirements in suburban communities may have amplified 36 In most cases, these regulations take the form of a maximum lot size ratios that can be used to build a house.

36 36 upward comparison effects and increased financial distress, with no long-term improvements in house satisfaction overall. Lastly, though I did not find evidence for this in the paper, reference effects in consumption can also be downwardlooking and generate avoidance of snob effects, as evidenced by the literature on fashion cycles. More research should therefore be given to empirical studies aimed at uncovering which type of comparison forces drives consumer behaviors at one point in time, and the consequences these effects can have on various measures of social welfare. REFERENCES Abel, Andrew B Asset prices under habit formation and catching up with the Joneses. The American Economic Review, Agarwal, Sumit, Vyacheslav Mikhed, and Barry Scholnick Does inequality cause financial distress? Evidence from lottery winners and neighboring bankruptcies. Working Paper. Alpizar, Francisco, Fredrik Carlsson, and Olof Johansson-Stenman How much do we care about absolute versus relative income and consumption? Journal of Economic Behavior & Organization, 56(3): Benjamin, Daniel J, Ori Heffetz, Miles S Kimball, Alex Rees-Jones, et al What do you think would make you happier? What do you think you would choose? American Economic Review, 102(5): Benjamin, Daniel J, Ori Heffetz, Miles S Kimball, Alex Rees-Jones, et al Can marginal rates of substitution be inferred from happiness data? evidence from residency choices. American Economic Review, 104(11): Bertrand, Marianne, and Adair Morse Trickle-down consumption. Review of Economics and Statistics, 98(5): Blakely, Edward J, and Mary Gail Snyder Fortress America: gated communities in the United States. Brookings Institution Press. Brodeur, Abel, and Sarah Fleche Neighbors Income, Public Goods, and WellBeing. Review of Income and Wealth, forthcoming. Card, David, Alexandre Mas, Enrico Moretti, and Emmanuel Saez Inequality at Work: The Effect of Peer Salaries on Job Satisfaction. American Economic Review, 102(6): Carr, Michael D., and Arjun Jayadev Relative Income and Indebtedness: Evidence from Panel Data. Review of Income and Wealth.

37 THE MCMANSION CURSE 37 Charles, Kerwin Kofi, Erik Hurst, and Nikolai Roussanov Conspicuous Consumption and Race. The Quarterly Journal of Economics, 124(2): Christen, Markus, and Ruskin M Morgan Keeping up with the Joneses: Analyzing the effect of income inequality on consumer borrowing. Quantitative Marketing and Economics, 3(2): Clark, Andrew E., Paul Frijters, and Michael A. Shields Relative Income, Happiness, and Utility: An Explanation for the Easterlin Paradox and Other Puzzles. Journal of Economic Literature, 46(1): Coibion, Olivier, Yuriy Gorodnichenko, Marianna Kudlyak, and John Mondragon Does Greater Inequality Lead to More Household Borrowing? New Evidence from Household Data. National Bureau of Economic Research. Davis, Morris A, and François Ortalo-Magné Household expenditures, wages, rents. Review of Economic Dynamics, 14(2): Deaton, Angus, and Arthur A Stone Two happiness puzzles. The American economic review, 103(3): 591. Diener, Ed, E Suh, Richard E Lucas, and Heidi L Smith Subjective well-being: Three decades of progress. Psychological Bulletin, 125(2): DiPasquale, Denise, and C Tsuriel Somerville Do house price indices based on transacting units represent the entire stock? Evidence from the American housing survey. Journal of Housing Economics, 4(3): Dynan, Karen E, and Enrichetta Ravina Increasing income inequality, external habits, and self-reported happiness. The American Economic Review, Easterlin, Richard A Will raising the incomes of all increase the happiness of all? Journal of Economic Behavior & Organization, 27(1): Easterlin, Richard A Income and happiness: Towards a unified theory. The economic journal, 111(473): Easterlin, Richard A, Laura Angelescu McVey, Malgorzata Switek, Onnicha Sawangfa, and Jacqueline Smith Zweig The happiness income paradox revisited. Proceedings of the National Academy of Sciences, 107(52): Fernandez-Villaverde, Jesus, and Dirk Krueger Consumption and saving over the life cycle: How important are consumer durables? Macroeconomic dynamics, 15(05):

38 38 Ferreira, Susana, and Mirko Moro On the use of subjective well-being data for environmental valuation. Environmental and Resource Economics, 46(3): Frank, Robert Falling behind: How rising inequality harms the middle class. Vol. 4, Univ of California Press. Frank, Robert H., Adam Seth Levine, and Oege Dijk Expenditure Cascades. Review of Behavioral Economics, 1(12): Frank, Robert H, Adam Seth Levine, Oege Dijk, et al Expenditure cascades. Social science. Frey, Bruno S, Simon Luechinger, and Alois Stutzer The life satisfaction approach to valuing public goods: The case of terrorism. Public Choice, 138(3-4): Genesove, David, and Christopher Mayer Loss aversion and seller behavior: Evidence from the housing market. National bureau of economic research. Genicot, Garance, and Debraj Ray Aspirations and Inequality. National Bureau of Economic Research, Inc NBER Working Papers Glaeser, Edward L, and Jesse M Shapiro The benefits of the home mortgage interest deduction. National Bureau of Economic Research. Goodman, John L, and John B Ittner The accuracy of home owners estimates of house value. Journal of Housing Economics, 2(4): Guerrieri, Veronica, Daniel Hartley, and Erik Hurst Endogenous gentrification and housing price dynamics. Journal of Public Economics, 100: Gyourko, Joseph, Christopher Mayer, and Todd Sinai Superstar cities. National Bureau of Economic Research. Heffetz, Ori A test of conspicuous consumption: Visibility and income elasticities. Review of Economics and Statistics, 93(4): Hirsch, Fred Social limits to growth. Cambridge, Massachusetts: Havard University Press. Ioannides, Yannis M, and Jeffrey E Zabel Neighbourhood effects and housing demand. Journal of applied Econometrics, 18(5): Kahneman, Daniel, and Angus Deaton High income improves evaluation of life but not emotional well-being. Proceedings of the national academy of sciences, 107(38):

39 THE MCMANSION CURSE 39 Kahneman, Daniel, Barbara L Fredrickson, Charles A Schreiber, and Donald A Redelmeier When more pain is preferred to less: Adding a better end. Psychological science, 4(6): Kiel, Katherine A, and Jeffrey E Zabel The accuracy of ownerprovided house values: The American Housing Survey. Real Estate Economics, 27(2): Kimball, Miles, and Robert Willis Utility and happiness. University of Michigan, Loewenstein, George, and Peter A Ubel Hedonic adaptation and the role of decision and experience utility in public policy. Journal of Public Economics, 92(8): Loewenstein, George, Ted O Donoghue, and Matthew Rabin Projection bias in predicting future utility. The Quarterly Journal of Economics, 118(4): Luechinger, Simon Valuing Air Quality Using the Life Satisfaction Approach*. The Economic Journal, 119(536): Luechinger, Simon, and Paul A Raschky Valuing flood disasters using the life satisfaction approach. Journal of Public Economics, 93(3): Lustig, Hanno N, and Stijn G Van Nieuwerburgh Housing collateral, consumption insurance, and risk premia: An empirical perspective. The Journal of Finance, 60(3): Luttmer, Erzo FP Neighbors as negatives: relative earnings and wellbeing. The Quarterly journal of economics, 120(3): Malmendier, Ulrike, and Stefan Nagel Depression Babies: Do Macroeconomic Experiences Affect Risk Taking?*. The Quarterly Journal of Economics, 126(1): Malmendier, Ulrike, and Stefan Nagel Learning from inflation experiences. The Quarterly Journal of Economics, 131(1): Manski, Charles F Identification of endogenous social effects: reflection problem. The review of economic studies, 60(3): Mian, Atif R, and Amir Sufi House prices, home equity-based borrowing, and the US household leverage crisis. National Bureau of Economic Research. Perez-Truglia, Ricardo The effects of income transparency on wellbeing: evidence from a natural experiment. Browser Download This Paper. The

40 40 Ratcliffe, Anita, et al Housing wealth or economic climate: Why do house prices matter for well-being? Centre for Market and Public Organisation, University of Bristol. Roback, Jennifer Wages, rents, and the quality of life. The Journal of Political Economy, Rosen, Sherwin Hedonic prices and implicit markets: product differentiation in pure competition. Journal of political economy, 82(1): Rossi-Hansberg, Esteban, Pierre-Daniel Sarte, and Raymond Owens III Housing externalities. Journal of political Economy, 118(3): Saez, Emmanuel, and Gabriel Zucman Wealth inequality in the United States since 1913: Evidence from capitalized income tax data. The Quarterly Journal of Economics, 131(2): Schor, Juliet B The Overspent American: Why We Want What We Don t Need. Harper Perennial, New York. Silva, JMC Santos, and Silvana Tenreyro The log of gravity. The Review of Economics and statistics, 88(4): Solnick, Sara J, and David Hemenway Are positional concerns stronger in some domains than in others? The American Economic Review, 95(2): Stekhoven, Daniel J, and Peter Bühlmann MissForestnonparametric missing value imputation for mixed-type data. Bioinformatics, 28(1): Stevenson, Betsey, and Justin Wolfers Economic growth and subjective well-being: Reassessing the Easterlin paradox. National Bureau of Economic Research. Stevenson, Betsey, and Justin Wolfers Subjective well-being and income: Is there any evidence of satiation? National Bureau of Economic Research. Van Praag, Bernard, and Barbara E Baarsma Using happiness surveys to value intangibles: The case of airport noise*. The Economic Journal, 115(500): Van Praag, Bernard MS, Paul Frijters, and Ada Ferrer-i Carbonell The anatomy of subjective well-being. Journal of Economic Behavior & Organization, 51(1):

41 THE MCMANSION CURSE 41 Wright, Gwendolyn Building the dream: A social history of housing in America. MIT press. Zabel, Jeffrey E The demand for housing services. Journal of Housing Economics, 13(1): Appendix A: Projection bias in relative housing size Suppose a person decides to buy a suburban house at time τ. The opportunity cost of buying the house is P, which includes any other goods that could have been bought had the house not been purchased. The person has just one opportunity to purchase the house. Assume her valuation of the house depends on its size compared to the size of other houses in the area. The latter can be considered as a consumption externality. Typically, a person may experience lower house satisfaction if her house looks comparatively smaller than neighboring houses, but the externality may also be positive, for instance if bigger houses are associated with aesthetic amenities. A house is a durable good which can last for several periods. The satisfaction the person will experience is therefore likely to change. First, the person may adapt to the house so that her absolute valuation decreases over time 37. Second, the housing stock may look very different after new houses get built. Formally, the satisfaction u τ corresponding to a house bought in period τ is { hτ νh u τ τ at the time τ the house is purchased γ k τ h τ νh k if the house has been purchased k > τ periods ago where h τ is the size of the house at time of purchase and H τ is the average size of houses in the suburb at that time, or the initial value of the average housing size. Coefficient ν characterizes the housing externality, which can be positive or negative and the term γ k τ captures the rate at which the person adapts to his house, with γ [0, 1] a constant. I assume the average size of houses follows an autoregressive process of order one, so that H k = φh k 1 + ɛ k for all k > τ where φ > 0 captures the growth rate of the average housing size between two periods, and ɛ t is a random, independent and identically distributed term with zero mean and constant variance σɛ 2. Define T = τ τ the length of tenure between the purchase date and some later period τ. The person does not discount future levels of house satisfaction, which does not affect the intuition of 37 Assuming physical depreciation would have a similar effect, which is why we identify separately the two in the empirical analysis. Evidence on hedonic adaptation is surveyed by Loewenstein and Ubel (2008).

42 42 the model. Her true expected inter-temporal house satisfaction between period τ and τ corresponds to [ E U τ τ ] = E [ ] τ [ k=τ γ k τ ] h τ νh k P This formulation assumes the person predicts her future instantaneous utility correctly. She fully accounts for adaptation and has perfect beliefs regarding how the suburb in which she decides to live may change. In reality, both are hard to anticipate. In particular, one may overestimate the long-term satisfaction of moving in an area facing changes in comparison groups. Typically, as argued by Loewenstein, O Donoghue and Rabin (2003), a person buying a big house in a wealthy suburb may not fully appreciate the reaction of future movers to her own decision to move, and the resulting change in the housing stock. A classical example of imperfect beliefs is projection bias, where a person s evaluation of the future depends on the state of the world at the time the decision is made. Theoretically, a person exhibiting simple projection bias will behave as if she was maximizing E E[ U τ τ [ τ [ k=τ (1 α)(γ k τ h τ νh k ) + α(h τ νh τ ) ] P ] = ] with 0 α 1 ] When α = 0, the person experiences no projection bias so that E[ U τ τ = [ ] E U τ τ. When α = 1, the person exhibits full projection bias towards her house: she perceives her future valuation as identical to her present valuation. The cumulative dissatisfaction D τ τ measured in period τ of a person who chose a house in period τ, then exactly equals the difference between her perceived intertemporal utility and her true intertemporal utility, which after some computations equal D τ τ [ E U τ τ ] [ E U τ τ ] = { [ α T 1 γ ] T 1 γ hτ if φ = 1 α [ ] T 1 γt 1 γ hτ + αν [ 1 φ T 1 φ T ] H τ otherwise This expression is a function of two terms. The first term reflects the person s misperception of her future adaptation to living in a house of size h τ. Since T > 1 γt 1 γ, the person will systematically overvalue a given house at the time it is bought, leading to investments she may regret in the future. In the presence of adaptation, the effect of own housing size h τ on house satisfaction measured in period τ will be a decreasing function of the length of tenure T. This is in line

43 THE MCMANSION CURSE 43 with evidence on how owners evaluate the current market value of their house 38. The second term captures the cumulative impact of the housing externality due to misperceived variations in the size of the housing stock after the date of purchase. In the case of a negative externality, it predicts that a misperceived increase in future housing size should imply a lower valuation of the house by the household in time τ. This corresponds to the cost of experienced relative downscaling. Typically, the person imperfectly accounts for future increase in housing size at the date of purchase and buys a house that ends up being too small. The second term disappears in the absence of any change in the size of the housing stock (φ = 1), is positive when the size of the housing stock is growing over time (φ > 1), but negative in the case of a declining size of the housing stock (φ < 1). Now, suppose two households, A and B, interviewed in time τ + 1 who moved in the same suburb. A bought his house at time τ while B bought his house one year later, at time τ + 1. Both houses are comparable in size h A τ = h B τ+1 = h. For T > 1, the difference in relative dissatisfaction of household A compared to household B is D τ D τ+1 = α(1 γ T 1 )h + αν(t 1)(φ 1)H τ First, household A will be less satisfied than household B simply because of the additional year of adaptation. This is captured by the first term, and the difference is a decreasing function of the length of tenure. The second term also shows household A will be less satisfied than household B in a suburb with growing housing size (φ > 1), but this time the difference is an increasing function of the length of tenure. This result is due to the interaction between projection bias and reference-dependent preferences. Because the late mover has a higher reference point than the early mover, the gap between his perceived and his true intertemporal utility is relatively lower. This simple set-up makes it clear that in the presence of projection bias, one should expect variations in construction histories between households to affect their subjective well-being, even controlling for the housing stock at the time of survey. It also shows that without controlling for households length of tenure, any cross-sectional estimation of relative size effects may simply capture adaptation to the house, or any other general time trends 39. Appendix B: The house satisfaction approach vs. hedonic pricing Assume a household with income y has the choice between two similar houses in suburbs s 1 and s 2 at time τ. The only difference between the two suburbs is the size of the other houses at that time H 1 τ > H 2 τ (hereafter called H 1 and H 2 ). The household chooses h to maximize 38 Goodman and Ittner (1992) find that owners over-estimate its value by 5% on average but Kiel and Zabel (1999) show that this over-valuation is greater for new owners and declines with the length of tenure. 39 Note that the model also generates different predictions regarding the sign of the interaction term between length of tenure and each of the two effects.

44 44 max U(x, h, H s ) such that y = x + ph with x a composite commodity, h the size of the house, H s the housing size externality in suburb s and p the housing price per square feet. The marginal utility is positive in own housing size U h > 0 and negative in reference housing size U H s < 0. In a perfectly competitive economy, the housing market internalizes the externality so p and y adjust to variations in H s. In equilibrium, utility is equalized across the two suburbs so that the household is equally happy in both places, with no incentive to move. The problem can be rephrased from the indirect utility function V as (B1) V ( y(h s ), p(h s ), H s) = k s where k is a constant. This market equilibrium condition is the starting point of the hedonic pricing (HP) approach introduced by Rosen (1974) or Roback (1982). The indirect utility of housing is an increasing function of income (V y > 0) and a decreasing function of housing prices for new movers (V p < 0) 40. The marginal impact of a change in the housing size stock depends on whether the externality is positive (V H s > 0) or negative (V H s < 0). The implicit cost of relative downscaling C experienced by an existing home owner can be defined as the increase in income required to make new movers indifferent net of the variation in the market value of houses: (B2) C = dy/dh s h(dp/dh s ) with h = V p /V y (Roy s identity) Taking the total derivative of equation (B1) gives (B3) dv/dh s = V y (dy/dh s ) + V p (dp/dh s ) + V H s = 0 And combining equation (B3) and (B2), the implicit hedonic cost of the housing externality equals (B4) C = dy/dh s ( V p /V y ) (dp/dh s ) = V H s/v y > 0 When the labor and housing markets are in equilibrium, the implicit cost of relative deprivation exactly equals the marginal willingness to pay (MWTP) to avoid feeling relatively deprived. Therefore, by regressing housing prices and 40 The fact that higher income allows for better house quality logically leads to a positive marginal utility of income. The estimation of the later is therefore very sensitive to the inclusion of dwelling specific controls for quality, an issue I address later in the paper.

45 THE MCMANSION CURSE 45 households income on the experienced variation in reference housing size, one can recover the MWTP of relative deprivation. However, if a direct proxy of house utility is available, the right hand side of equation (B4) can be estimated directly. This method is known as the life satisfaction (LS) approach 41. Typically, it consists in regressing a subjective measure of house satisfaction on income and the externality, holding house prices and income constant, to recover respectively V y and V H s. In the case presented above, it requires that the subjective measure of house satisfaction at time τ be a function of the cumulative instantaneous utility flows over the T periods since the person moved in 42. If the two methods give similar estimates, one can claim the market perfectly internalizes the externality through higher price differentials between relatively small and relatively big houses. There exist various reasons why the market equilibrium condition is unlikely to hold. A classical issue is the presence of moving costs. This generates a downward bias in the cost of the relative size externality, as households who would like to move to a relatively bigger house must also pay an extra moving cost. A similar bias may arise in the presence of loss aversion, which is typically associated with reference dependent preferences (Genesove and Mayer, 2001). Loss aversion can be experienced by existing home owners but not by potential buyers. Formally, if condition (B1) does not hold, house satisfaction is not equalized across all counties, so that dv/dh s < 0. It follows that the new implicit cost of relative deprivation estimated through the HP approach C is in fact lower than the true MWTP as estimated by the LS approach: (B5) C = dy/dh s + ( V p /V y ) (dp/dh s ) = V H s/v y + (dv/dh s )/V y < V H s/v y The hedonic cost of relative deprivation computed from the wage and price gradients would therefore give a downward biased estimate of the true cost, as it neglects the residual effect (dv/dh s )/V y not capitalized in private markets. Appendix C: Zillow and the measurement of historical variation in housing size One way to test whether Zillow does well at measuring past housing size is to compare my measures to the US Census Survey of Construction (SOC). The Survey of Construction (SOC) provides measures for the mean and median size of new single-family housing units constructed each year since Figure C1 plots the mean housing size of newly built houses from Zillow and SOC datasets over the period The trend correlation between both datasets is very close to one over the forty years period. This is reassuring as the empirical analysis 41 For a discussion of the LS approach, see Van Praag and Baarsma (2005); Luechinger and Raschky (2009); Luechinger (2009); Frey, Luechinger and Stutzer (2009) or Ferreira and Moro (2010). 42 Evidence that happiness differs from flow utility is reviewed by Kimball and Willis (2006).

46 46 exploits differences in personal experiences over time, controlling for county-year levels. The figure also shows that Zillow seems to capture bigger houses than the SOC on average. Mean size of newly built houses year Zillow sample SOC sample Zillow sample, top-coded SOC sample - MSA only Figure C1. : Average Size of New-Build Single Family Houses ( ) Note: Figure C1 plots the mean size of new-build single family houses from 1971 to 2009 using data from Zillow (dark lines) and from the Survey of Construction (grey lines). Source: Survey of Construction (SOC) and author s own calculations from Zillow. Two important reasons can explain that discrepancy in levels. First, the SOC estimates regroup both urban and rural single-family houses, while my Zillow sample is restricted to urban suburbs, where houses are on average bigger. A better comparison is to restrict the SOC to houses built within MSA. This only partly addresses the problem as suburban and central city houses cannot be distinguished in the SOC. However, the grey dashed line shows it already reduces part of the gap 43. Second, the SOC is top-coded for the biggest 1% houses, which means Zillow does a better job at measuring the true size of the biggest houses built. If I truncate the Zillow sample to exclude the top percentile, the gap is also reduced. To further check for the presence of an attrition bias affecting the distribution of houses over time, I take the ratio of mean to median size in each census region for each year t as a first approximation of the size distribution for both datasets. I then compute the difference between these two measures and see whether the gap varies over time in a systematic way. The right hand side variable used to test for attrition is therefore: 43 The Census Bureau does not compute averages at the MSA level for the period , and access to the micro data of the SOC is restricted to the period.

47 THE MCMANSION CURSE 47 Table C1 : Testing for attrition over time, SOC vs. Zillow (1) (2) AttritionIndex AttritionIndex Coeff. S.E. Coeff. S.E. Time since the house was built (years) ( ) ( ) North East region ( ) (0.0128) South region ( ) ( ) West region ( ) ( ) North East region Time since the house was built (years) ( ) South region Time since the house was built (years) ( ) West region Time since the house was built (years) ( ) Observations Adjusted R Note: In column (1), I regress the measure of attrition defined in equation (1) against the number of years since the house was built, controlling for Census region dummies. In column (2), I interact the number of years since the house was built with the Census region dummies. Source: Survey of Construction (SOC) and author s own calculations from Zillow. Attrition measure t = ( Mean Median ) Zillow,t ( Mean Median ) SOC,t (1) Table C1 first shows the results of a regression where I regress this measure on the number of years since houses were built and region fixed effects. The coefficient on time is not significant. I also interact time with the region fixed effect, as attrition and remodeling could play differently across regions, but except in the South, I find no significant differences across regions. This further reduces the attrition concern. Appendix D: Imputation of longitude and latitude Besides the suburb indentifier, three other variables can be used to predict the location of an AHS house from Zillow: house size, lot size and the house s year of construction. The imputation method relies on within-suburbs truncated regressions of longitude and latitude on these three variables and their interactions. The truncation restricts the range of possible values for longitude and latitude to their maximum and minimum value within any suburb. I take the mean values of 50 imputations as the final imputed measures for longitude and latitude. To test the validity of the imputation method, I simulate random missing values for each suburbs in Zillow in the same proportions as the corresponding missing values in the AHS. I can then look at the correlation coefficient between the imputed and true values within suburbs. Figure D1 plots the distribution of the correlation coefficients for all suburbs. Figure D2 shows the distribution of the

48 48 Cumulative distribution Longitude Latitude Correlation coefficient between true and simulated values (within suburbs) Figure D1. : Cumulated Distribution of Correlation Coefficients Between True and Simulated Latitudes/Longitudes Density Own house average distance to 90th percentile newly built houses kernel = epanechnikov, bandwidth = Figure D2. : Distribution of Average Distance Between AHS House and New McMansions (Sources: AHS and Zillow) estimated average distance between a household s house and all new McMansions built during his tenure period. The average house is located about 10 miles away from newly built McMansions.

49 THE MCMANSION CURSE 49 Appendix E: Other Figures and Tables Figure E1. : Mapping of Metropolitan Counties Note: The figure maps all 154 counties and pseudo-counties located within the Metropolitan Statistics Areas (MSA) surveyed in the American Housing Survey. These locations cover 55% of the total US population and almost the entire suburban population. Impact of log housing size on house satisfaction (1-10) Figure E2. : Regression Coefficient of Log House Size on House Satisfaction. Note: I use house satisfaction as the dependent variable, and I plot the regression coefficient on log housing size. Separate regressions are estimates for each year on new movers using the national files of the American Housing Surveys ( ). Sampling weights are included.