Housing and Marcellus Shale Development

Transcription

1 Housing and Marcellus Shale Development The Marcellus Shale Impacts Study Wave 2: Chronicling Social and Economic Change in Northern and Southwestern Pennsylvania March 2017 Executive Summary The rapid increase in Marcellus Shale development in Pennsylvania over the past decade has raised questions about the potential impacts of drilling and related development activities on housing availability and affordability for industry workers and residents in the counties experiencing substantial development as well as neighboring counties that may have no wells but may house industry workers and/or residents relocating from counties with substantial development activity. This research studied trends in several housing indicators since 2000, including housing supply, age of housing, occupancy, value, rental costs, and household income inequality. The analyses featured in the report focus on four specific Pennsylvania counties with substantial drilling activity, including two in the northern tier (Bradford and Lycoming) and two in the southwestern region (Greene and Washington) of Pennsylvania. The research also compared housing outcomes in the four study counties to outcomes in neighboring counties located in the same regions, many of which also experienced significant well development over the past 8 years. Finally, the research used bivariate and multivariate statistical analyses to determine whether well development was statistically related to housing outcomes and/or changes in those outcomes over the recent period of widespread unconventional natural gas development in Pennsylvania. Summary of Results Availability and Occupancy Among the study counties, there was substantial variation in housing stock and changes in housing stock between 2000 and Bradford and Washington counties experienced increases in the total number of housing units, Lycoming County experienced almost no change in its housing stock, and Greene County experienced a slight decline in its total number of housing units between 2000 and Results suggest that well development is not related to changes in total housing stock. The significantly larger growth in housing stock experienced in counties with no wells was explained by larger increases in population density in those same counties. However, well development occurred over a period of widespread economic and housing market distress. Therefore, it is possible that the housing supply in counties with substantial well development increased more than it otherwise might have without any development at all in those same counties. Comparisons of counties along three groupings of well development (counties with no wells, counties in the top 25 th percentile of well count, and counties in the bottom 75 th percentile of well count) revealed that all three groups of counties experienced average declines in owner-occupied housing units between 2005/2009 and 2009/2013, and only counties with no wells had more owner-occupied housing units in 2009/2013 than in 2000, on average. All three groups of counties also experienced increases in vacant housing units over the study period, and counties with the most wells experienced the largest average increase in vacant housing (though the difference between them and the other two groups of counties was not statistically significant). Despite increases in vacant units, however, the vacant rental supply (the The Center for Rural Pennsylvania 1

2 stock of vacant homes available for rent) declined in all three well development categories between 2000 and 2009/2013, but it decreased the most in the counties with the most wells (i.e., those in the top 25 th percentile of well count). Moreover, counties with well development (both high and low-to-mid tier development) had a lower supply of available rental units in 2009/2013 versus counties with no wells (both in terms of overall stock and available rental units as a percentage of all housing units), suggesting increased rental demand and/or reduced supply of rental units in counties with the most well development. Lower rental availability in counties with drilling activity remains a challenge for residents of these counties, particularly among many counties in the northern tier that already had a small vacant rental supply in 2000 before well development began (e.g., Potter, Sullivan, Susquehanna, Tioga). Importantly, some high drilling counties were more strongly impacted by declines in rental availability than others. For instance, whereas Bradford County experienced a 16 percent increase in its stock of vacant rental units between 2000 and 2009/2013, Washington, Greene, and Lycoming counties experienced declines in vacant rental stock of 37 percent, 42 percent, and 42 percent, respectively, between 2000 and 2009/2013. These results suggest that local leaders must be mindful of rental shortages in counties with natural gas extraction activities. Though temporary housing for workers (e.g., man-camps, RVs, trailers) may be one possible solution, these strategies can be expensive and often bring challenges when extraction activities end and workers leave the area, including unpaid debt, abandoned and unkempt temporary units, and rodents (Oldham, 2015). Therefore, local leaders should consider enacting reasonable zoning laws and permitting policies that place fiscal responsibility on natural gas companies to properly dispose of temporary housing developed for their workers and clean the area on which that housing was located. The development of more permanent housing is also an option to combat rental shortages, but new development should include a sustainability plan. Local leaders could require housing developers to provide such a plan prior to granting permits for building. Longer-term structures could be adapted for use for tourism or safe housing for elders (Lycoming County Planning Department, 2012). New structures built to accommodate industry workers could also be modified into domestic violence shelters, youth services facilities, mental health and developmental disability rehabilitation facilities, office space for county social service workers, and rent-to-own or cooperative ownership housing for low-income residents and/or those with inadequate credit to attain conventional mortgages. Affordability Results of comparisons of home values across the three categories of well-development demonstrate that counties with the most wells experienced significantly larger increases in the median value of owner-occupied housing units between 2000 and 2009/2013 compared to counties with low-to-mid tier development. However, counties with no wells experienced significantly larger increases in the median value of owner-occupied housing units between 2000 and 2009/2013 compared to counties in both the low-to-mid tier development range and high well counts. These differences existed net of controls for county population and economic characteristics, including population density, percentage of working age (18-64) adults in poverty, and unemployment rate. This finding suggests that, at least within Pennsylvania, there is no evidence of home value decline in areas with significant extraction activity. However, residents in these areas should not bank of the value of their homes increasing during energy boom periods. All three county types experienced increases in the percentage of renters income spent on rent and in the percentage of renters spending more than 30 percent of their income on rent (a standard measure of housing affordability) between 2000 and 2009/2013. However, those increases were significantly The Center for Rural Pennsylvania 2

3 smaller for counties with wells (both low-to-mid tier drilling and high drilling) compared to counties without wells. Despite this, counties with the highest well counts experienced the largest average increase in median rent over the study period. However, these counties also had the lowest percentage of renters spending 30 percent or more of their income on rent in 2009/2013, and these counties experienced the smallest increases in the percentages of renters spending 30 percent or more of their incomes on rent between 2000 and 2009/2013. This remained true even after accounting for pre- Marcellus county characteristics and changes in population and economic characteristics over the same period. Together, these findings suggest that although the median cost of rent increased in high drilling counties, income also increased enough to offset the rising rental costs and/or renters in high drilling counties shared rental units (e.g., doubled- or tripled-up) in an effort to reduce individual expenses. Inequality There is substantial variability in household income inequality across the state. This includes the global measure of inequality (the Gini coefficient), as well as differences in household income between renters and homeowners. Household income inequality was lower among the four study counties than for Pennsylvania as a whole, and the study counties had comparable levels of household income inequality as the adjacent counties in the same regions, on average. However, in the northern tier, Clinton, Montour, and Wyoming counties had relatively low income inequality in 2005/2009 but experienced especially substantial increases in income inequality between 2005/2009 and 2009/2013, suggesting that high-income residents moved into these counties over this period and/or that a small percentage of residents experienced large increases in income while the majority of residents experienced stagnant or declining wages in these counties over this period. Ultimately, findings suggest that well development was not statistically related to overall household income inequality or increases or decreases in household income inequality over the study period. Local leaders should continue to monitor these trends because rising income inequality could lead to increasing debt burdens, deprivation of opportunity, and intergenerational transmission of disadvantage among families at the low end of the income distribution. Despite no relationship between well development and income inequality, there were changes in inflation-adjusted median income among renters and owners between 2000 and 2009/2013. Whereas counties with no wells and counties with low-to-mid-level drilling experienced declines in median household income among homeowners between 2000 and 2009/2013, counties with the most wells experienced a slight increase in median household income among homeowners over the same period. This may suggest that homeowners experienced income benefits from drilling, on average. If this is the case, it could be due either to increased income from leasing royalties or to increased wages related to natural gas sector employment or employment in related industries. These data and analyses were not able to tease out the merit of these potential explanations. Moreover, whereas median household income among renters declined in all three categories of well development over the study period, the decline was the smallest in counties with the most wells. Regression analyses revealed that changes in economic and demographic characteristics between 2000 and 2009/2013 explain the finding that renters in counties with the most wells had less severe reductions in median household income among renters over the same period compared to renters in counties with no wells. That is, counties with the most wells experienced smaller increases in poverty and unemployment over this period than did counties with no wells, and these factors were somewhat protective against declines in household income among renters. The Center for Rural Pennsylvania 3

4 Finally, the gap in median household income between renters and owners increased substantially in Pennsylvania between 2000 and 2009/2013. Whereas homeowners earned $23,010 more than renters in 2000 (in 2013 adjusted dollars), this gap had increased to $35,826 by 2009/2013. This is a trend that should be of concern to state and local leaders. A rising income gap between renters and owners may result in increasing debt among renters, disparities in educational opportunities for children of renters versus owners, and vastly different neighborhood quality between the two groups. The income gap between owners and renters increased more in Bradford County than in Lycoming County and more in Washington County than in Greene County. Whereas all adjacent counties in the southwestern region experienced increases in the owner/renter income gap between 2000 and 2009/2013, in the northern tier, both Tioga and Wyoming counties experienced decreases in the owner/renter income gap. These declines were mostly attributable to declines in income among homeowners rather than increases in income among renters. The gap in median household income between homeowners and renters was significantly lower in low-to-mid tier drilling counties versus counties with no wells, but pre-marcellus population and economic characteristics explained this difference. There were no significant differences in the owner/renter income gap between counties with the most wells versus counties with no wells, and the number of wells was not related to changes in the owner/renter income gap between 2000 and 2009/2013. This project was sponsored by a grant from the Center for Rural Pennsylvania, a legislative agency of the Pennsylvania General Assembly. The Center for Rural Pennsylvania is a bipartisan, bicameral legislative agency that serves as a resource for rural policy within the Pennsylvania General Assembly. It was created in 1987 under Act 16, the Rural Revitalization Act, to promote and sustain the vitality of Pennsylvania s rural and small communities. Information contained in this report does not necessarily reflect the views of individual board members or the Center for Rural Pennsylvania. For more information, contact the Center for Rural Pennsylvania, 625 Forster St., Room 902, Harrisburg, PA 17120, telephone (717) , info@rural.palegislature.us, The Center for Rural Pennsylvania 4

5 Table of Contents Executive Summary... 1 Summary of Results... 1 About this Project... 6 Study Counties... 6 Marcellus Shale Activity... 7 Potential Impacts of Marcellus Shale Activity on Housing... 8 Methods and Analysis Data Analysis Results Housing Availability and Occupancy Housing Affordability Household Income Inequality Summary and Implications Availability and Occupancy Affordability Inequality Further Considerations Report Authors Acknowledgements Funding References Appendix A. List of Tables in Report Appendix B. Tables with Coefficients for Well Categories from Unadjusted and Adjusted Regression Models Appendix C. Tables with all Coefficients from Adjusted Regression Models The Center for Rural Pennsylvania 5

6 About this Project The Marcellus Shale Impacts Study chronicles the effects of shale-based energy development in Pennsylvania by focusing on the experiences of four counties with significant natural gas extraction and production activity Bradford, Lycoming, Greene, and Washington counties. Wave 1 of the project was completed in 2013 and Wave 2 began in early Wave 1 focused predominantly on data collection and the use of descriptive statistics to present changes in various outcomes over time. Wave 2 used statistical analysis methods to describe relationships between Marcellus Shale development and a set of social and economic indicators, identify changes in social and economic outcomes that are associated with unconventional natural gas development, and identify the compositional and contextual characteristics of counties associated with the magnitude and types of impact experienced. A particular focus of Wave 2 was to explore the heterogeneity in unconventional natural gas development impact on different population groups. The purpose of this research on housing was to identify and describe changes in housing availability, affordability and inequality in counties with significant drilling, compare those trends with trends in neighboring counties, and identify whether unconventional natural gas development in Pennsylvania is related to variation in housing outcomes, above and beyond the demographic and economic characteristics of these counties prior to the commencement of drilling activities. The analyses also helped determine whether any changes in the demographic and economic characteristics of counties that occurred since 2000 explain changes in housing outcomes that also occurred during that time period. Study Counties This study focused on the same four counties examined in Wave 1 of the Marcellus Shale Impacts Study: Bradford, Lycoming, Greene, and Washington counties. These counties experienced among the highest levels of Marcellus Shale development in Pennsylvania over the past 8 years, and they have diverse populations, histories, economic bases, and geographic locations. These differences allow comparisons that facilitate understanding of the potential associations between Marcellus Shale development and various social and economic outcomes. Regional comparisons are also made based on adjacency to the study counties. The northern tier counties include Bradford, Lycoming, Clinton, Columbia, Montour, Northumberland, Potter, Sullivan, Susquehanna, Tioga, Union, and Wyoming. The southwestern counties include Greene, Washington, Allegheny, Beaver, Fayette, and Westmoreland. All four study counties are classified as rural by the Center for Rural Pennsylvania with population densities of less than 284 people per square mile. However, the U.S. Department of Agriculture s (USDA) Economic Research Service (ERS) classifies Lycoming and Washington counties as being located inside metropolitan areas. Lycoming County encompasses much of the Williamsport metropolitan area, and Washington County is part of the Pittsburgh metropolitan area. Bradford and Greene counties are classified by USDA ERS as being located outside of metropolitan areas. Bradford and Greene counties have small urban populations of less than 20,000 people. However, both are adjacent to metropolitan areas. The Center for Rural Pennsylvania 6

7 Marcellus Shale Activity Figure 1. Cumulative Number of Unconventional Gas Wells Drilled, Jan 1, 2005 June 1, 2015 Source: PA Department of Environmental Protection, Office of Oil and Gas Management Figure 1 presents the distribution of the cumulative number of unconventional natural gas wells 1 drilled in each county in Pennsylvania from Jan 1, 2005 to June 1, Counties are classified into five categories: counties with no wells and the four quartiles (25 th percentiles) of well development. Well development is concentrated in the northeast, north central and southwestern regions of Pennsylvania. In the northern tier, Bradford, Lycoming, Tioga, and Susquehanna counties all experienced similarly high levels of development. This suggests that comparisons of study outcomes among these counties are particularly useful because these counties experienced similarly high levels of drilling. Likewise, the most useful comparisons in the southwestern region are between Greene, Washington, Fayette, Allegheny, Beaver, and Westmoreland counties, although it is important to note that Greene and Washington counties experienced significantly more well development than the other counties in the southwestern region. The cumulative number of unconventional gas wells drilled in the four study counties since Jan 1, 2005 is shown in Figure 2. The rate of development of the Marcellus Shale reflects that of the traditional natural resource development boomtown model. That is, the rate of growth increased quickly over time with a particularly dramatic spike in Bradford County between 2009 and Bradford County has consistently maintained the highest number of wells in the state since widespread development began in Conventional wells are not included. The Center for Rural Pennsylvania 7

8 Figure 2. Cumulative number of wells drilled in four study counties, June 1, Bradford Lycoming Greene Washington Source: PA Dept. of Environmental Protection, Office of Oil and Gas Management Potential Impacts of Marcellus Shale Activity on Housing Rural boomtowns (communities undergoing rapid increases in natural resource extraction) may experience challenges with housing availability and affordability, including limits to housing stocks and rapid increases in housing costs. An influx of workers employed by the natural gas industry may quickly lead to depletion of available rental units and temporary housing (such as hotels and mobile homes) as the development of new housing can, at the very least, take many months if not several years. Moreover, local leaders and community members are often reluctant to approve large-scale development projects that may not be needed following the boom period. As noted by previous research, housing shortages can lead to rapid escalation of home purchase prices and rental rates, pushing long-term residents out of the housing market (Gilmore and Duff, 1975), but as noted by Meuhlenbachs et al. (2014), the positive association between well development and property values diminishes over time, potentially resulting in long-term negative economic implications for new homeowners who may end up underwater on their homes. In a study of residents in several Pennsylvania counties, Brasier et al. (2011) found that residents of Lycoming County were concerned with the potential for housing shortages and price inflation. On the other hand, a study on the housing impacts of gas and oil development in Central Alberta, Canada, found that property values were actually inversely associated with natural gas development (Boxall et al., 2005). The effects of development-related population and economic changes on housing are likely to vary with overall drilling activity but may also depend heavily on the status of the housing market and other economic and demographic characteristics of counties prior to any widespread development. Housing for new residents or workers may have been plentiful and affordable in some high drilling counties and less available and more expensive in others prior to well development. Moreover, some counties in The Center for Rural Pennsylvania 8

9 Pennsylvania housed greater proportions of economically disadvantaged residents than others prior to any well development, suggesting that larger populations of residents in some counties may have been dealing with housing insecurity before any well development activities began. These conditions have implications for the potential impacts of well development on housing outcomes. An influx of similar numbers of new residents may have very different effects in areas with small populations, high rates of poverty, limited vacant housing available for rent, and high rental costs compared to areas with ubiquitous housing supply, low rental costs, and larger populations that can absorb an influx of new workers and their families. For instance, Brasier et al. (2011) found that housing shortages created problems for social service agencies trying to place low-income and homeless residents in temporary housing in Bradford County, PA, a county with a small population and relatively large proportion of economically disadvantaged residents. Rapid natural gas development also coincided with the most severe recession the U.S. has experienced since the Great Depression. Significant heterogeneity in recession impacts on unemployment and foreclosure has implications for changes to housing characteristics and outcomes counties experienced over this period. Complicating issues related to housing and the influx of new workers is determining the length of time newcomers intend to stay in the area. A local housing market may not have the mix of housing to meet the preferences of different types of workers. More transient workers rig workers and workers constructing drill-pads or other infrastructure (pipelines, pumping stations), for example may be more likely to seek temporary housing, making it easier for them to move on in a few weeks or months. Higher turnover of renters may increase the wear-and-tear on existing properties or greater demand for rentals could cause low-income individuals and families to be priced out of rental markets. There is already evidence of overbuilt housing in North Dakota, where civic leaders and developers struggled to house a massive influx of workers during the oil boom. These same communities now face challenges with too few local residents who can afford the high rent on these housing units that now sit vacant as workers fled the area as oil production declined with global reductions in oil prices 2. Other workers may stay for several years or permanently relocate to the area. Gas-related workers in management or in maintenance of drilled wells may be more likely to purchase, rather than rent, housing. Those seeking to buy may have difficulty finding suitable housing in areas where the housing stock is older and has not been updated. Moreover, a rapid increase in demand for housing may lead to or exacerbate inequality between homeowners and renters. Individuals and families who already own homes may benefit from an influx of new residents if it leads to increases in housing values whereas renters may suffer displacement or economic hardship if the cost of rent increases dramatically with the arrival of new residents and increased housing demand. Questions that face local leaders and developers in active drilling areas include whether they should encourage building of new housing or rental units to meet the housing needs of temporary workers; and if so, what kind of housing? Who will pay for this new or renovated housing and how much will developers and renters be willing to pay? How can the impact and trade-offs associated with these decisions be identified, especially given uncertainty about where the next drilling boom will occur and how long activity will last? If decisions are made to build new housing, it is essential for local leaders to consider the long-term housing needs of the existing population. For example, if the population is aging, as is occurring in much of rural Pennsylvania, would it make sense to repurpose housing built during the gas development boom period to safely house elders when drilling activity ends and workers move on? 2 The Real Estate Crisis in North Dakota s Man Camps. Bloomberg Business. September 29, The Center for Rural Pennsylvania 9

10 If homeownership became more restrictive over this period due to increases in housing values and growing income inequality between renters and owners, can state and local leaders take a role in mitigating any increase in inequality that may be driven by natural gas development? Though the results from this report cannot answer all of these questions, they can help to inform leaders and developers about the current housing context within which these decisions must be made. Methods and Analysis Data Comparative information on housing over time is available from the U.S. Census Bureau through the 2000 Decennial Census and the American Community Survey (ACS). This research used county-level housing and population data from the 2000 Census and ACS 5-year estimates for 2005/2009 and 2009/2013. The American Community Survey is a mandatory, ongoing statistical survey that samples a small (but representative) sample of the population annually. The 5-year estimates combine 5 years of data collection into a 5-year rolling average. The Census first began the ACS in 2005, so the 2005/2009 estimates are the earliest 5 years (since the 2000 Decennial Census) for which housing data are available for all counties in Pennsylvania. Though the ACS also provides 1-year estimates beginning in 2005 and 3- year estimates beginning in 2007, 1-year estimates are available only for areas with populations of 65,000 or more, and 3-year estimates are available only for areas with populations of 20,000 or more. As of the 2010 Decennial Census, there were 27 counties in Pennsylvania with fewer than 65,000 residents and are therefore excluded from the 1-year estimates. These include two of the study counties (Bradford and Greene). Therefore, using the 1-year estimates was not an option for this report. There are challenges with using 5-year estimates in lieu of the more-timely 3-year estimates. Chief among these is that many of the housing changes that may have occurred as a result of Marcellus likely occurred very quickly upon the onset of major gas development. These sometimes rapid and pronounced changes may not be captured in the 5-year estimates that combine periods of rapid increase and decline. Moreover, as with any data collected from samples, there is some variability (margin of error) in the ACS estimates due to sampling error. Readers can visit the American Factfinder website ( to see specific margins of error for housing outcomes of interest. The larger the margins of error, the lower the accuracy of the estimate, and the less confidence one should have that the estimate is close to the true value. Because they combine the most data points, the 5-year estimates are the most precise (compared to the 1-year or 3-year estimates) and have the smallest margins of error. For instance, even among the smallest counties in Pennsylvania, the typical margin of error for total housing units is only around 0.8 percent total (0.4 percent +/-). Margins of error get larger (e.g. 5 percent) with more precise variables like renter-occupied or vacant housing stock. Despite these limitations, there are at least three reasons to use the 5-year estimates versus the 3-year estimates for these analyses. First, there are six counties in Pennsylvania with fewer than 20,000 residents. These are Montour, Potter, and Sullivan (which neighbor high drilling counties), Fulton, Forest, and Cameron. These counties are excluded from the 3-year estimates. These are the six most rural counties in Pennsylvania. Comparing these six counties to the state s other 61 counties reveals many important differences that would bias the results if these counties were excluded from the analyses. Compared to the rest of the counties, these six small counties have significantly fewer gas wells, fewer housing units, lower median household income, a smaller income gap between homeowners and renters, lower median home values The Center for Rural Pennsylvania 10

11 and rents, lower percentages of owner-occupied and rental housing, greater housing vacancy but a lower percentage of vacant housing units available for rent (likely because of the large proportion of seasonal and recreational housing units), lower unemployment rates, and greater employment in production and transportation. Using the 5-year estimates in lieu of the 3-year estimates allows for the inclusion of all 67 counties in Pennsylvania, thereby reducing the bias (risk of presenting incorrect findings) that would result from regression analyses that excluded these six counties that vary in important ways from the rest of Pennsylvania s counties. Second, just as important as the risk of biasing the results by excluding the most rural counties, losing six observations from the analysis would result in a substantial decline in statistical power from an already small sample of counties. Conducting proper regression analyses, which includes controlling for the widest possible range of county-level factors that might explain both well development and housing outcomes, requires as much statistical power as possible. The 5-year estimates, while not ideal for measuring rapid change, are the best estimates available for a study with a small sample size. Third, as a result of tight budgetary considerations, the U.S. Census Bureau plans to discontinue use of the 3-year estimates beginning in fiscal year 2016, meaning that after 2015, 3-year estimates will likely not be available to enable ongoing comparisons of change in housing and other social, economic, and demographic characteristics. The 2000 Census data provide information on pre-marcellus activity, the 2005/2009 ACS 5-year estimates capture the early period of Marcellus activity, and the 2009/2013 ACS 5-year estimates were the most recently available data at the time of analysis and capture the latest period of Marcellus activity, during which time most wells have been drilled (83 percent of all wells drilled between 2005 and 2013 were drilled between 2010 and 2013) 3. These Census data were merged with information about the cumulative number of natural gas wells drilled from 2005 through the end of Although well counts are available up through the present month, using 2013 well counts maintains temporal consistency with the ACS data used in these analyses. Analysis In the results sections below, descriptive statistics are presented on overall housing supply, age of housing, occupancy, affordability, household income inequality, and percentage change in each measure over time for Pennsylvania overall, the four study counties, and the counties adjacent to the four study counties. Following this description of trends over time, results of bivariate statistical tests comparing means in housing outcomes across three levels of well development are reported. These three categories of well development are: (1) no wells (28 counties); 2) high wells (counties in the top quartile of drilling with at least 178 wells drilled between 2004 and 2013), which include the four study counties plus Butler, Fayette, Susquehanna, Tioga, and Westmoreland counties; and 3) all other counties with some wells (N=30). After the presentation of results from difference of means tests, results from unadjusted and multivariate linear regression models for each outcome are presented. These models identify statistical associations between levels of well development (the independent/predictor variable) and housing outcomes (the dependent variables). All of the housing outcomes are quantitative variables 4, so linear 3 Readers should note that there is a 1-year overlap between the 2005/2009 and 2009/2013 data. 4 Quantitative variables are those that are measured on a numerical scale. In addition to being quantitative, well spud count is considered an interval variable because each interval has the same interpretation throughout (i.e., the difference in value between 1 and 2 is the same as the difference in value between 5 and 6. The Center for Rural Pennsylvania 11

12 models are most appropriate for an analysis of these outcomes. Unadjusted models are models that include only the primary predictor variable 5. The primary predictor variable of interest was natural gas production, measured with three categories of well spud count. Well spud count is highly skewed (not evenly distributed across all Pennsylvania counties). Nine counties contain over 80 percent of all wells, and 28 counties contain no wells. Therefore, spud count could not be entered into regression models in its original count format because the skewed (highly uneven) distribution of wells would increase the risk of biased (incorrect) regression estimates 6. Outcomes are assessed across the three well development categories described above. Outcomes for high well counties and counties with some wells are compared to outcomes from counties with no wells (reference category). Multivariate (adjusted) regression models are used to determine whether well spud count is related to housing outcomes above and beyond the impact of other demographic and socioeconomic factors that may influence both Marcellus Shale development and housing outcomes. For instance, Marcellus Shale development occurs almost entirely within rural Pennsylvania. Housing prices are also usually lower in rural counties. Therefore, population density is related to both Marcellus Shale development and housing prices. It is important to control for this reality by holding population density constant in any models that predict relationships between Marcellus Shale development and housing outcomes. Two sets of control variables are used. The first set of control variables represent pre-marcellus (year 2000) county-level demographic and economic characteristics that are likely to be related to housing outcomes: population density, percentage of working-age adults (aged 18-64) in poverty 7, unemployment rate, percentage of the civilian labor force aged 16 and older employed in construction, extraction, and maintenance, and percentage of civilian labor force aged 16 and older employed in production, transportation, and material moving occupations. This set of covariates enables readers to examine associations between well development and housing outcomes while accounting for the characteristics of the population prior to any substantial well development in the county. The second set of control variables represent changes in the demographic and economic characteristics of the population over the period of well development: percentage change between 2000 and 2009/2013 in each of the covariates listed above. These models enable readers to assess associations between well development and housing outcomes while accounting for the possibility that the housing market changed over the high well development period as a result of changes to the demographic and economic characteristics of the populations in counties with high drilling. This analysis assesses the question: Were any changes in housing outcomes observed between 2000 and 2009/2013 explained by other changes to economic and population characteristics across counties? Ideally, models would adjust 5 Predictor variables are variables used in regression models to predict the value of another variable. In this case, well development is being used to predict housing outcomes. It is sometimes referred to as an independent variable or an explanatory variable. 6 Entering the highly skewed well spud count into a regression model in its original form would violate linear regression assumptions of normality of residuals and homoscedasticity. These violations increase the risk of biased (incorrect) findings. 7 The percentage of county residents aged 25 and older with less than a high school diploma and the percentage of county residents aged 25 and older with at least a 4-year college degree were both strongly correlated with percentage of working age adults in poverty. Likewise, total population size in 2000 and 2010 were strongly correlated (r=.79) with population density in 2000 and 2010, and percent change in population size 2000 to 2010 was almost perfectly correlated with percent change in population density over the same time period (r=.998), leading to violations of the regression assumption that one predictor variable is not a linear predictor of another (i.e., the assumption of no multicollinearity). Therefore, the educational attainment variables and population size variables were excluded from the regression models presented in this report. Results were similar when substituting educational attainment variables for percentage of working age adults in poverty and when substituting population size variables in lieu of population density. The Center for Rural Pennsylvania 12

13 for both pre-marcellus characteristics and changes in those characteristics over the study period. However, the small sample of 67 counties does not enable the inclusion of this many control variables in one model due to loss of statistical power. At least 10 observations per predictor variable are needed in order to have enough statistical power to detect reasonable effect sizes (Harrell Jr., 2001). This means including no more than six covariates in any one regression model for the analysis of 67 counties in Pennsylvania. Accordingly, results from both sets of models are presented so readers can assess differences between models that control for pre-marcellus county characteristics versus models that control for changes in county characteristics over the study period. Differences in means of each of the control variables across the three categories of drilling are presented in Table 1. Data for all variables for 2000 came from the 2000 U.S. Decennial Census, and data for all variables for 2009/2013 came from the American Community Survey. Given that drilling occurs outside of urban centers, it is no surprise that population density is much lower in counties with well development. Perhaps less well known is that whereas population density increased between 2000 and 2009/2013 in counties with no wells, counties with high well development experienced slight declines in population density over that time period. These population declines are consistent with a long-term pattern of rural to urban population redistribution (urbanization) over the past 20 years (Brown, 2014). In addition to population decline, counties with well development (both some and high) had higher poverty rates in 2000 and 2009/2013 compared to counties with no wells, but poverty increased more between 2000 and 2009/2013 in counties with no wells. Whereas counties with no wells had a lower average unemployment rate in 2000 compared with counties with wells, the unemployment rate increased dramatically in counties with no wells between 2000 and 2009/2013, such that by the 2009/2013 measurement period, counties with no wells had a significantly higher unemployment rate than counties with some or high well counts. These trends may be a function of positive employment impacts of unconventional well development, or they may reflect the fact that the economic recession of disproportionately affected urban areas, particularly those that were reliant on manufacturing (Thiede and Monnat, 2014). There are two other differences in employment characteristics between counties with versus without wells. First, counties with a high well count had a significantly larger average percentage of their civilian labor force employed in construction, extraction, and maintenance occupations in 2009/2013 compared to counties with no wells. Second, counties with some wells had a significantly larger average percentage of their employment in production, transportation, and material-moving occupations compared to counties with no wells. If workers in these occupations are transient, temporary, or have different housing needs than workers in other occupations, these differences in workforce composition may have implications for housing outcomes in these counties. For instance, rather than an increase in homeownership, there may be an increase in renter-occupied housing and an associated increase in rental costs in counties with high drilling activity. The Center for Rural Pennsylvania 13

14 Table 1. Differences between Three Levels of Well Development in Means of County Characteristics included as Control Variables in Regression Models Control Variable No Wells Some Wells High Wells (N=28) (N=30) (N=9) Population Density (per square mile), (2122.0) (320.8) (111.3) Population Density (per square mile), 2009/ (2162.1) (307.5) (111.3) Percent change in population density, / (6.59) (11.91) a (4.45) b Percentage of adults (18-64) in poverty, (3.40) (2.44) a (3.12) b Percentage of adults (18-64) in poverty, 2009/ (3.93) (2.19) a (2.85) Percent change in percentage of adults (age 18-64) in poverty, / (11.54) (11.87) a 8.58 (9.43) b,c Unemployment rate, (0.87) 3.62 (0.51) a 3.51 (0.72) Unemployment rate, 2009/ (1.28) 4.57 (0.75) a 4.63 (0.72) b Percent change in unemployment rate, / (35.22) (25.94) a (34.35) b Percentage of employed civilian labor force age 16+ in construction, extraction, and maintenance occupations, (2.20) (1.98) (2.89) Percentage of employed civilian labor force age 16+ in construction, extraction, and maintenance occupations, 2009/ (2.60) (2.41) (2.18) b Percent change in employment in construction, extraction, and maintenance occupations, / (9.98) (14.18) (10.08) Percentage of employed civilian labor force age 16+ in production, transportation, and material moving occupations, (6.05) (5.94) (3.96) Percentage of employed civilian labor force age 16+ in production, transportation, and material moving occupations, 2009/ (4.51) (4.69) a (2.94) Percent change in employment in production, transportation, and material moving occupations, / (6.63) (6.68) (4.47) Source: Author calculations using data from U.S. Census and ACS. Notes: two-tailed difference of means t-tests. a Significant difference between no wells and some wells. b Significant difference between no wells and high wells. c Significant difference between some wells and high wells. The Center for Rural Pennsylvania 14

15 Results Housing Availability and Occupancy Tables 2 to 13 present (1) descriptive information about the housing supply and availability (number of housing units, housing age, percentage of units that are owner-occupied, renter-occupied, vacant, and percentage of vacant units that are available for rent) for the years 2000, 2005/2009, and 2009/2013) and (2) the percentage change in each of these outcomes from /2013, /2009, and 2005/ /2013 for Pennsylvania, the four study counties, and the counties adjacent to the four study counties. By presenting percentage change measures broken down across these three time periods, readers can see whether most of the change that occurred between 2000 and 2013 occurred prior to (2005/2009), or during (2009/2013), the height of unconventional natural gas development in Pennsylvania. Housing Units Table 2 presents information on the total number of housing units 8 from 2000 to 2014 for the northern tier study and adjacent counties and percentage change in total number of housing units. The Census began releasing annual estimates of housing units in 2011 via the data product: Annual Estimates of Housing Units. This enables annual comparisons from 2010 onward 9. The total number of housing units in Pennsylvania increased by nearly 6.5 percent between 2000 and 2014, with almost all of that growth occurring between 2000 and Between the two northern tier study counties, Bradford County experienced more growth in housing stock between 2000 and 2014 (5 percent) than Lycoming County (0.26 percent). It is noteworthy that whereas Lycoming County experienced a decline in housing stock between 2005/2009 and 2014, Bradford County experienced continuous increases. The average growth among northern tier adjacent counties between 2000 and 2014 was higher than the growth for the two study counties, but four adjacent counties (Clinton, Northumberland, Potter, and Sullivan) experienced slight declines in housing stock between 2010 and A housing unit may be a house, an apartment, a mobile home, a group of rooms or a single room that is occupied (or if vacant, intended for occupancy) as separate living quarters. Separate living quarters are those in which the occupants live separately from any other individuals in the building and which have direct access from outside the building or through a common hall. For vacant units, the criteria of separateness and direct access are applied to the intended occupants whenever possible. If that information cannot be obtained, the criteria are applied to the previous occupant. Both occupied and vacant housing units are included in the housing unit inventory. Boats, recreational vehicles (RVs), vans, tents, railroad cars, and the like are included only if they are occupied as someone s current place of residence. Excluded from the housing inventory are quarters being used entirely for nonresidential purposes, such as a store or an office, or quarters used for the storage of business supplies or inventory, machinery, or agricultural products. (U.S. Census Bureau American Community Survey yr Summary File: Technical Documentation, available at: 9 Note that this is the only housing outcome for which annual estimates are available for all PA counties. Therefore the ACS 5- year estimates are used for all other outcomes. The Center for Rural Pennsylvania 15

16 Table 2. Housing Supply: Number of Housing Units and Percentage Change in Northern Tier Study and Adjacent Counties, Sources: Census 2000 Summary File 3 (SF-3); Census 2010 Summary File 3 (SF-3); American Community Survey 5- year estimates; U.S. Census Annual Estimates of Housing Units. Note: 5-year estimates are used for because 3-year estimates are not available for Montour, Potter and Sullivan counties. Non-decadal annual estimates for all counties were not available until Information on the total number of housing units from 2000 to 2014 for the southwestern study and adjacent counties and percentage change in total number of housing units is presented in Table 3. Whereas Greene County experienced a decline in housing stock between 2000 and 2014, with a short period of growth between 2000 and 2005/2009, Washington County experienced continuous increases in housing stock over the 2000 to 2014 period. Indeed, Washington County experienced a greater increase than any of the adjacent counties, and Fayette County actually experienced a decline in housing stock between 2000 and Table 3. Housing Supply: Number of Housing Units and Percentage Change in Southwest Region Study and Adjacent Counties, Sources: Census 2000 Summary File 3 (SF-3); Census 2010 Summary File 3 (SF-3); American Community Survey 5- year estimates; U.S. Census Annual Estimates of Housing Units. The Center for Rural Pennsylvania 16

17 Housing Age Information on the age of housing stock in the northern tier study and adjacent counties is presented in Table 4. From 2000 to 2009/2013 the median age of housing units in Bradford County decreased by 4 years, whereas the median housing age for Lycoming County decreased by only 2 years; the Bradford County housing stock is younger than the Lycoming County housing stock. Northumberland County has the oldest median housing age of all northern tier counties. Potter and Sullivan counties experienced the largest declines in housing age over the study period, indicating significant recent building of new housing units and/or destruction of old housing units. Table 4. Housing Age (in years) in Northern Tier Study and Adjacent Counties Change Change Change / / / /2013 Pennsylvania Study Counties Bradford Lycoming Adjacent Counties Clinton Columbia Montour* Northumberland Potter Sullivan* Susquehanna Tioga Union Wyoming Sources: Census 2000 Summary File 3 (SF-3); American Community Survey 5-year estimates; American Community Survey 5-year estimates. Notes: The median age of housing is calculated by subtracting the median year of housing built from the year 2013 (the final year of the 2009/2013 estimates). *With very small populations, Montour and Sullivan counties have large margins of error. Estimates should be considered with caution. Information on the age of housing stock in the southwestern study and adjacent counties is presented in Table 5. Both Greene and Washington counties experienced declines in median housing age between 2000 and 2009/2013 of 6 and 5 years, respectively. Of the adjacent counties in the southwestern region, Fayette County experienced the greatest decline (i.e., the greatest move toward newer housing units), with 5 years over the study period. All other adjacent counties experienced declines in housing age of 3 years. The Center for Rural Pennsylvania 17

18 Table 5. Median Year Housing Units Built in Southwest Region Study and Adjacent Counties Change /2013 Change /2009 Change 2005/ / / /2013 Pennsylvania Study Counties Greene Washington Adjacent Counties Allegheny Beaver Fayette Westmoreland Sources: Census 2000 Summary File 3 (SF-3); American Community Survey 5-year estimates; American Community Survey 5-year estimates. Notes: The median age of housing is calculated by subtracting the median year of housing built from the year 2013 (the final year of the 2009/2013 estimates). Housing Occupancy: Owner-Occupied, Rented, and Vacant Information on the percentage of housing units that are owner-occupied in the northern tier study and adjacent counties is presented in Table 6. Like Pennsylvania as a whole, Bradford and Lycoming Counties experienced declines in owner-occupied housing units from 2000 to 2009/2013, but these declines were restricted to the 2005 to 2009/2013 period. From 2000 to 2009/2013, all adjacent counties in the northern tier, with the exceptions of Northumberland and Sullivan counties, experienced increases in the number of housing units that are owner occupied, ranging from a 14.2 percent increase in Union County to a 0.3 percent increase in Clinton County. The Center for Rural Pennsylvania 18

19 Table 6. Number and Percentage of all Housing Units that are Owner-Occupied and Percentage Change in Number of Owner-Occupied Housing Units in Northern Tier Study and Adjacent Counties, Pennsylvania ,406,337 3,496,696 (64.9) (63.8) / / / /2013 3,462,512 (62.2) Study Counties Bradford 18,455 (64.4) 18,464 (62.7) 18,142 (60.5) Lycoming 32,636 (62.2) 33,169 (61.4) 32,229 (61.4) Adjacent Counties 11,354 (59.2) 11,529 (58.0) 11,555 (56.9) Clinton 10,775 (59.3) 10,962 (58.0) 10,803 (56.8) Columbia 18,030 (65.0) 18,311 (63.2) 18,263 (62.0) Montour 5,171 (67.8) 5,533 (69.1) 5,339 (66.8) Northumberland 28,561 (66.2) 28,160 (64.1) 28,040 (62.3) Potter 5,418 (44.6) 5,275 (41.5) 5,436 (42.1) Sullivan 2,149 (35.7) 2,068 (33.0) 1,996 (31.7) Susquehanna 13,145 (60.2) 13,560 (60.0) 13,425 (58.5) Tioga 12,133 (61.0) 12,133 (60.4) 12,680 (59.4) Union 9,665 (65.8) 10,190 (65.5) 11,034 (65.0) Wyoming 8,492 (66.8) 8,675 (65.1) 8,529 (64.4) Sources: Census 2000 Summary File 3 (SF-1); American Community Survey 5-year estimates; American Community Survey 5-year estimates. Information on the number and percentage of housing units that are renter occupied in the northern tier study and adjacent counties is presented in Table 7. Both Bradford and Lycoming counties experienced declines in the stock of renter-occupied housing units between 2000 and 2009/2013 of 3.9 percent and 6.7 percent, respectively. Both counties experienced increases in the renter-occupied housing stock from 2000 to 2005/2009, but then experienced large decreases from 2005/2009 to 2009/2013. Changes among adjacent counties in the northern tier varied greatly. Sullivan County experienced the largest decline of almost 21 percent, whereas Columbia County experienced an increase of nearly 16 percent. The Center for Rural Pennsylvania 19

20 Table 7. Number and Percentage of all Housing Units that are Renter Occupied and Percentage Change in Number Renter Occupied Housing Units in Northern Tier Study and Adjacent Counties, Pennsylvania ,370,666 1,396,341 (26.1) (25.5) / / / /2013 1,495,915 (26.9) Study Counties Bradford 5,998 (20.9) 6,319 (21.4) 6,071 (20.3) Lycoming 14,367 (27.4) 14,813 (27.4) 13,817 (26.3) Adjacent Counties 3,813 (19.4) 3,991 (19.3) 4,208 (19.65) Clinton 3,998 (22.0) 4,149 (21.9) 4,264 (22.4) Columbia 6,885 (24.8) 7,215 (24.9) 7,962 (27.0) Montour 1,914 (25.1) 1,872 (23.4) 1,894 (23.7) Northumberland 10,274 (23.8) 10,487 (23.9) 11,308 (25.1) Potter 1,587 (13.1) 1,725 (13.6) 1,641 (12.7) Sullivan 511 (8.5) 432 (6.9) 406 (6.4) Susquehanna 3,384 (15.5) 3,766 (16.6) 3,738 (16.3) Tioga 3,792 (19.1) 4,166 (20.0) 4,378 (20.5) Union 3,513 (23.9) 3,558 (22.9) 4,028 (23.7) Wyoming 2,270 (17.9) 2,542 (19.1) 2,463 (18.6) Sources: Census 2000 Summary File 3 (SF-3); American Community Survey 5-year estimates; American Community Survey 5-year estimates. Table 8 presents trends for the number and percentage of housing units that are vacant in the northern tier study and adjacent counties. Like Pennsylvania as a whole, both Bradford and Lycoming Counties experienced increases in the number of vacant housing units from 2000 to 2009/2013. Bradford County experienced an increase of nearly 37 percent, whereas Lycoming County had an increase of nearly 18 percent. All northern tier adjacent counties experienced increases in their vacant housing stocks. The increases ranged from a high of 41 percent in Montour County to a low of 7.9 percent in Tioga County. The Center for Rural Pennsylvania 20

21 Table 8. Number and Percentage of all Housing Units that are Vacant and Percentage Change in Vacant Housing Units in Northern Tier Study and Adjacent Counties, / / / / Pennsylvania 472,747 (9.0) 588,549 (10.7) 607,226 (10.9) Study Counties Bradford 4,211 (14.7) 4,682 (15.9) 5,759 (19.2) Lycoming 5,461 (10.4) 6,005 (11.1) 6,424 (12.2) Adjacent Counties 3,232 (21.4) 3,584 (22.7) 3,757 (23.4) Clinton (18.7) 3,794 (20.1) 3,951 (20.8) Columbia 2,818 (10.2) 3,427 (11.8) 3,243 (11.0) Montour 542 (7.1) 606 (7.6) 765 (9.6) Northumberland 4,329 (10.0) 5,252 (12.0) 5,665 (12.6) Potter 5,154 (42.4) 5,698 (44.8) 5,826 (45.2) Sullivan 3,357 (55.8) 3,768 (60.1) 3,894 (61.8) Susquehanna 5,300 (24.3) 5,321 (23.5) 5,779 (25.2) Tioga 3,968 (19.9) 4,062 (19.5) 4,281 (20.1) Union 1,506 (10.3) 1,802 (11.6) 1,920 (11.3) Wyoming 1,951 (15.3) 2,109 (15.8) 2,244 (17.0) Sources: Census 2000 Summary File 3 (SF-3); American Community Survey 5-year estimates; American Community Survey 5-year estimates. Trends in the number and percentage of vacant housing units available for rent in the northern tier study and adjacent counties are presented in Table 9. Units that are vacant but not available for rent may be seasonal/vacation homes not occupied at the time of the census survey or may be vacant for some other reason. Like Pennsylvania as a whole, Lycoming County experienced a decline in the number of vacant housing units available for rent from 2000 to 2009/2013. Both counties experienced declines in vacant rental stock from /2009, but Bradford County experienced an increase from 2005/2009 to 2009/2013, whereas Lycoming County continued to decline over that time period. All adjacent counties except for Potter, Sullivan, and Union counties also experienced declines in the vacant rental stock from 2000 to 2009/2013. The largest declines were in Wyoming (91 percent) and Montour (84 percent) counties. Overall, these trends suggest a tightening of the rental market in northern tier counties between 2000 and 2009/2013. The Center for Rural Pennsylvania 21

22 Table 9. Number and Percentage of all Vacant Housing Units for Rent and Percentage Change in Vacant Housing Units for Rent in Northern Tier Study and Adjacent Counties, / / / / Pennsylvania 105,585 (22.3) 111,155 (18.9) 97,911 (16.1) Study Counties Bradford 501 (11.9) 392 (8.4) 582 (10.1) Lycoming 1,180 (21.6) 793 (13.2) 687 (11.0) Adjacent Counties 298 (11.6) 217 (6.7) 186 (5.1) Clinton 243 (7.2) 184 (4.8) 177 (4.5) Columbia 511 (18.1) 409 (11.9) 172 (5.3) Montour 166 (30.6) 95 (15.7) 26 (3.4) Northumberland 1,008 (23.3) 888 (16.9) 601 (10.6) Potter 112 (2.2) 106 (1.9) 135 (2.3) Sullivan 58 (1.7) 76 (2.0) 93 (2.4) Susquehanna 255 (4.8) 104 (2.0) 170 (2.9) Tioga 268 (6.8) 160 (3.9) 197 (4.6) Union 227 (15.1) 123 (6.8) 273 (14.2) Wyoming 130 (6.7) 26 (1.2) 12 (0.5) Sources: Census 2000 Summary File 3 (SF-3); American Community Survey 5-year estimates; American Community Survey 5-year estimates. Trends in housing occupancy for the southwestern region counties are presented in Tables As shown in Table 10, from 2000 to 2009/2013, Washington County experienced an increase in the number of housing units that were owner-occupied, as did Pennsylvania, whereas Greene County experienced a decline in owner-occupied housing units. However, unlike Washington County, which consistently declined over the period, Greene County experienced a slight increase in the number of owner-occupied housing from 2005/2009 to 2009/2013. All adjacent counties experienced declines in the number of housing units that were owner-occupied between 2000 and 2009/2013. Fayette County experienced the largest decline (9.8 percent), while Westmoreland County experienced the smallest decline (0.7 percent). The Center for Rural Pennsylvania 22

23 Table 10. Number and Percentage of all Housing Units that are Owner-Occupied and Percentage Change in Owner-Occupied Housing Units in Southwest Region Study and Adjacent Counties, ,406,337 (64.9) 3,496,696 (63.8) / /2009 The Center for Rural Pennsylvania / /2013 3,462,512 (62.2) Pennsylvania Study Counties Greene 11,159 (66.9) 11,058 (64.0) 10,526 (64.1) Washington 62,561 (71.7) 64,931 (70.5) 63,973 (68.7) Adjacent Counties 114,782 (67.5) 141,449 (65.2) 138,049 (64.3) Allegheny 360,036 (61.7) 351,807 (59.4) 344,618 (58.5) Beaver 54,367 (69.9) 54,391 (68.3) 52,018 (66.5) Fayette 43,876 (66.0) 42,263 (62.7) 52,018 (63.1) Westmoreland 116,849 (72.6) 117,335 (70.5) 116,000 (69.0) Sources: Census 2000 Summary File 3 (SF-3); American Community Survey 5-year estimates; American Community Survey 5-year estimates. Trends in the renter occupied housing stock in the southwestern region study and adjacent counties are presented in Table 11. Washington County experienced an overall increase in the number of renter occupied housing units (8.4 percent) from 2000 to 2009/2013, similarly to Pennsylvania. However, Greene County had a decline in its renter occupied housing stock from 2000 to 2009/2013. Allegheny and Beaver counties experienced decreases from 2000 to 2005/2009 and increases from 2005/2009 to 2009/2013. Westmoreland County experienced an increase between 2000 and 2005/2009 and between 2005/2009 and 2009/2013, whereas Fayette County experienced an increase between 2000 and 2005/2009 but a decrease from 2009/2013. The overall percent change in the number of renter occupied housing units ranged from an increase of about 9.5 percent in Westmoreland County to a decrease of about 8 percent in Fayette County. Table 11. Number and Percentage of all Housing Units that are Renter Occupied and Percentage Change in Renter Occupied Housing Units in Southwest Region Study and Adjacent Counties, ,370,666 1,396,341 (26.1) (25.5) / / / /2013 1,495,915 (26.9) Pennsylvania Study Counties Greene 3,901 (23.4) 3,472 (20.1) 3,891 (23.7) Washington 18,569 (21.3) 18,677 (20.3) 20,125 (21.6) Adjacent 61,095 (24.6) 60,307 (24.1) (25.0) Counties Allegheny 177,114 (30.3) 172,777 (29.2) 181,386 (30.8) Beaver 18,209 (23.4) 17,241 (21.7) 18,849 (24.1) Fayette 16,093 (24.2) 16,608 (24.6) 14,803 (23.6) Westmoreland 32,964 (20.5) 34,601 (20.8) 36,109 (21.5) Sources: Census 2000 Summary File 3 (SF-3); American Community Survey 5-year estimates; American Community Survey 5-year estimates.

24 Trends in the number and percentage of all housing units that were vacant in the Southwestern region study and adjacent counties are presented in Table 12. Like Pennsylvania, Greene and Washington counties experienced increases in the number of housing units that were vacant from 2000 to 2009/2013 of just over 24 percent and nearly 47 percent, respectively. Washington County experienced increases between 2000 and 2005/2009 and between 2005/2009 and 2009/2013, whereas Greene County experienced an increase between 2000 and 2005/2009 and a decrease between 2005/2009 and 2009/2013. All adjacent counties experienced increases in the number of housing units that were vacant from 2000 to 2009/2013. Table 12. Number and Percentage of all Housing Units that are Vacant and Percentage Change in Vacant Housing Units in Southwest Region Study and Adjacent Counties, / / / / Pennsylvania 472,747 (9.0) 588,549 (10.7) 607,226 (10.9) Study Counties Greene 1,618 (9.7) 2,736 (15.8) 2,010 (12.2) Washington 6,137 (7.0) 8,557 (9.3) 9,014 (9.7) Adjacent Counties 17,363 (7.9) 24,590 (10.7) 23,572 (10.7) Allegheny 46,496 (8.0) 67,422 (11.4) 62,640 (10.6) Beaver 5,189 (6.7) 7,979 (10.0) 7,332 (9.4) Fayette 6,521 (9.8) 8,556 (12.7) 8,339 (13.3) Westmoreland 11,245 (7.0) 14,402 (8.7) 15,975 (9.5) Sources: Census 2000 Summary File 3 (SF-3); American Community Survey 5-year estimates; American Community Survey 5-year estimates. Table 13 presents trends in rental availability in the Southwestern region study and adjacent counties. Greene and Washington counties experienced declines in the number of vacant housing units available for rent from 2000 to 2009/2013. Between 2000 and 2009/2013, the decline was nearly 42 percent in Greene County and just over 37 percent in Washington County. However, Greene County experienced an increase from 2000 to 2005/2009, while Washington County experienced declines in both time periods. All adjacent counties experienced decreases in vacant housing stock for rent during the 2000 to 2009/2013 period. Allegheny County experienced the largest decline (45.2 percent), followed by Greene County. The Center for Rural Pennsylvania 24

25 Table 13. Number and Percentage of all Vacant Housing Units for Rent and Percentage Change in Vacant Housing Units for Rent in Southwest Region Study and Adjacent Counties, / / / /2013 Pennsylvania 105,585 (22.3) 111,155 (18.9) 97,911 (16.1) Study Counties Greene 327 (20.2) 395 (14.4) 191 (9.5) Washington 1,946 (31.7) 1,924 (22.5) 1,220 (13.5) Adjacent Counties 5,844 (29.5) 5,667 (19.6) 3,313 (12.9) Allegheny 17,250 (37.1) 17,394 (25.8) 9,446 (15.1) Beaver 1,495 (28.8) 1,474 (18.5) 914 (12.5) Fayette 1,661 (25.5) 1,632 (19.1) 1,043 (12.5) Westmoreland 2,970 (26.4) 2,166 (15.0) 1,848 (11.6) Sources: Census 2000 Summary File 3 (SF-3); American Community Survey 5-year estimates; American Community Survey 5-year estimates. Statistical Relationships between Marcellus Activity and Housing Availability and Occupancy Descriptive statistics (means and standard deviations) for housing supply and availability outcomes across the three categories of well development (no wells, some wells, and high wells [top quartile of well count]) are presented in Table 14. Subscripts indicate when significant differences in outcomes exist between the three county types at the p<0.05 (95 percent confidence) level or better. Readers should be aware that even when there are substantial differences in means across the categories, those differences may not be statistically significant because there may also be substantial variability in that outcome among the counties in that category (i.e., large standard deviations). This is especially the case for the category for high well count, as this category contains only nine counties and large differences between any two counties can lead to very large standard deviations. Moreover, though the table includes the absolute values in each outcome for 2000, 2005/2009, and 2009/2013, readers may wish to focus on the variables representing percentage change in each outcome over the study period, as this report is intended to assess relationships between unconventional natural gas development and changes in outcomes. As shown in Table 14, counties with no wells have substantially more housing units, on average, than counties with some wells or high well counts (top quartile of well counts). This is largely because counties with no wells are located in urban areas and have large populations, whereas counties with natural gas wells are predominantly on the periphery of urban areas or are located in rural areas with much smaller populations. Results of statistical tests comparing these means also indicate that counties with no wells experienced a larger rate of increase in total number of housing units between 2000 and This is preliminary evidence that natural gas well development is not associated with increases in housing stock. However, these analyses cannot rule out the possibility that the housing supply in counties with wells increased more than it otherwise might have without any well development at all. Turning to the housing occupancy variables, counties with some wells have the lowest average number and average percentage of housing units that were owner-occupied in 2000, 2005/2009 and 2009/2013, although only the differences between counties with some wells and counties with no wells are The Center for Rural Pennsylvania 25

26 statistically significant. Counties with no wells experienced an average increase in owner-occupied housing units between 2000 and 2009/2013, whereas counties with some wells experienced an average decline, and counties with the most wells experienced no change, on average. All three groups of counties experienced declines in the percentage of housing units that were owner-occupied over the study period, with the largest average decline in counties with some wells and the smallest average decline in counties with no wells. This is likely largely a result of the Great Recession that led to above average foreclosure rates across the country. All three groups of counties experienced increases in the stock of renter-occupied housing between 2000 and 2009/2013, but at 12.1 percent, the rate of increase was significantly larger in counties with no wells than in counties with some wells and counties with the most wells. The overall share (percentage) of housing units that were renter-occupied remained fairly stable for the three groups of counties over the study period. There are not significant differences in the change in the share of housing units that were renter-occupied over the study period. However, all three groups of counties experienced increases in total number of vacant housing units and the percentage of housing units that were vacant between 2000 and 2009/2013. Counties with some wells have the highest average share of vacant housing units in all three periods, but the rate of increase in the number of vacant housing units and the rate of increase in the share of housing units that were vacant between 2000 and 2009/2013 were statistically similar across the three groups of counties. Finally, all three groups of counties experienced average declines in the stock of vacant housing units available for rent and in the share (percentage) of housing units that were vacant and available for rent between 2000 and 2009/2013, but the declines were significantly greater in counties with the most wells than in counties with no wells. This suggests increased demand and/or reduced supply of rental unit availability in counties with high unconventional gas development. The Center for Rural Pennsylvania 26

27 Table 14. Means for Housing Supply and Availability Outcomes by Level of Well Development No Drilling Some Drilling High Drilling HOUSING STOCK (N=28) (N=30) (N=9) Total number of housing units, ,892 (132,634) 54,952 (104,025) 58,246 (46,164) Total number of housing units, 2005/ ,901 (134,272) 56,313 (105,482) 60,782 (47,969) Total number of housing units, ,778 (137,092) 56,251 (105,116) 61,007 (48,707) Percent change in total number of housing units, (5.58) 2.75 (4.57)a 4.33 (5.99)b Number of owner-occupied housing units, ,634 (77,392) 34,614 (64,406) 40,233 (34,287) Number of owner-occupied housing units, 2005/ ,936 (75,687) 34,317 (62,976)a 40,977 (34,720) Number of owner-occupied housing units, 2009/ ,594 (74,468) 33,703 (61,613)a 40,310 (34,412) Percent change in number of owner-occupied housing units, / (7.60) (4.95)a (5.68)b Percent housing units owner-occupied, (6.70) (11.75)a (6.70) Percent housing units owner-occupied, 2005/ (6.55) (11.76)a (4.81) Percent housing units owner-occupied, 2009/ (6.37) (11.51)a (4.70) Percent change in percentage of housing units owner-occupied, / (3.71) (3.07)a (1.48) Number of renter-occupied housing units, ,884 (45,912) 14,056 (31,945) 12,627 (9,691) Number of renter-occupied housing units, 2005/ ,553 (46,666) 14,085 (31,179) 13,155 (10,140) Number of renter-occupied housing units, 2009/ ,437 (51,551) 14,660 (32,782) 13,321 (10,575) Percent change in number of renter-occupied housing units, / (6.52) 3.76 (7.41)a 5.48 (8.57)b Percent housing units renter-occupied, (5.49) (6.58) (3.34) Percent housing units renter-occupied, 2005/ (5.34) (6.48) (3.07) Percent housing units renter-occupied, 2009/ (5.98) (6.84) (2.78) Percent change in percentage of renter-occupied housing units, / (5.40) 1.06 (6.32) 1.63 (4.08) Number of vacant housing units, ,422 (13,214) 6,282 (8,076) 5,385 (2,638) Number of vacant housing units, 2005/ ,412 (17,302) 7,911 (11,637) 6,629 (3,485) Number of vacant housing units, 2009/ ,073 (16,291) 7,812 (10,843) 6,980 (3,955) Percent change in number of vacant housing units, / (25.21) (18.80) (13.77) Percent housing units vacant, (9.55) (16.62)a (6.35)c Percent housing units vacant, 2005/ (8.43) (16.67)a (5.45)c Percent housing units vacant, 2009/ (8.36) (16.68)a (6.00)c Percent change in percentage of vacant housing units, / (23.45) (16.63) (13.97) The Center for Rural Pennsylvania 27

28 Number of vacant housing units for rent, ,070 (3,386) 1,247 (3,126) 1,137 (928) Number of vacant housing units for rent, 2005/2009 2,411 (5,212) 1,168 (3,125) 958 (782) Number of vacant housing units for rent, 2009/2013 2,446 (4,335) 754 (1,686) 757 (561) Percent change in number of vacant housing units for rent, / (49.12) (38.14) (18.18)b Percent vacant housing units for rent, (12.03) (10.72)a (12.03) Percent vacant housing units for rent, 2005/ (11.03) (7.53)a (7.10) Percent vacant housing units for rent, 2009/ (11.50) 7.90 (5.27)a (4.25)b Percent change in percentage of vacant housing units for rent, / (29.47) (33.93) (13.78)b Notes: two-tailed difference of means t-tests a Significant difference between no wells and some wells b Significant difference between no wells and high wells c Significant difference between some wells and high wells The Center for Rural Pennsylvania 28

29 This section summarizes results from regression models predicting differences in the housing outcomes reported above between counties with a high well count (top quartile of well count) vs. counties with no wells and between counties with some wells versus counties with no wells. These regression models did not control for any other characteristics of counties. A table with regression coefficients is presented in Appendix B (Table B1) for readers interested in specific values. The table also includes the R 2 values from the regression models. R 2 values indicate the proportion of variation in each outcome explained by differences in well count. There were significant differences between counties with a high number of wells and counties with no wells on four outcomes. First, counties with the most wells experienced a significantly smaller average increase in the total number of housing units between 2000 and 2014 than did counties with no wells. Second, counties with the most wells experienced a significantly smaller average increase than counties with no wells in the stock (total number) of owner-occupied housing units between 2000 and 2009/2013. Third, counties with the most wells experienced a significantly smaller average increase than counties with no wells in the stock (total number) of renter-occupied housing units between 2000 and 2009/2013. Finally, counties with the most wells had a significantly lower percentage of vacant housing units available for rent in 2009/2013 compared with counties with no wells. Counties with the most wells also experienced a substantially larger reduction than counties with no wells in the stock of vacant housing units available for rent between 2000 and 2009/2013, but this difference was not statistically significant due to substantial variation in rental availability among the nine high well counties. There were also significant differences between counties with some wells versus counties with no wells for most outcomes. Compared to counties with no wells, counties with some wells had a significantly smaller average increase in total number of housing units between 2000 and 2014, a significantly lower average percentage of housing units that were owner-occupied in 2000 and 2009/2013, a significantly larger reduction in the share (percentage) of owner-occupied housing units between 2000 and 2009/2013, significantly lower average shares of renter-occupied housing units in 2000 and 2009/2013,and significantly higher average shares of vacant housing units in 2000 and 2009/2013, but significantly lower average shares of vacant housing units available for rent in 2000 and 2009/2013. All three groups of counties experienced average increases in the stock of vacant housing units between 2000 and 2009/2013, but the increase in the stock of vacant housing units was significantly smaller in counties with some wells than in counties with no wells. However, there was not a significant difference between the two groups of counties in the change in the stock of vacant housing units available for rent. Finally, counties with some wells experienced an average decrease in the total number of owneroccupied housing units between 2000 and 2009/2013, whereas counties with no wells experienced an increase. This difference was statistically significant. The results summarized above did not account for pre-marcellus, county-level characteristics or changes in county characteristics during the period of high development that may explain the differences in housing outcomes observed between the three groups of counties. The next two sets of regression models predict differences in the same housing outcomes discussed above between the three categories of well development while 1) controlling for pre-marcellus county characteristics (Appendix B, Table B2), and 2) controlling for changes in county-level characteristics between 2000 and 2009/2013 (Appendix B, Table B3). These county-level characteristics were described in the Analysis section and shown in Table 1 near the beginning of this report. These characteristics include year 2000 values for county population density, percent poverty, unemployment rate, and shares of the county workforce employed in construction, extraction or maintenance and production, transportation and material moving and changes in each of those values between 2000 and 2009/2013. The Center for Rural Pennsylvania 29

30 Coefficients that were significant in the unadjusted models but that lost statistical significance in the adjusted models are italicized in Tables B2 and B3 in Appendix B. Loss of statistical significance in the models that control for pre-development (year 2000) county-level characteristics indicate that the differences in these pre-marcellus characteristics in the three different groups of counties explain the differences in housing outcomes between the categories of well development found earlier in the simple bivariate analyses. Loss of statistical significance in the models that control for changes ( /2013) in county-level characteristics indicate that differences in those changes between the three groups of counties explain the differences in housing outcomes observed in the simple bivariate analyses. Some important findings emerge. First, there was only one housing outcome that varied significantly between the categories of well development after controlling for pre-marcellus county characteristics and changes in Marcellus characteristics. Compared to counties with no wells, counties with some wells had a significantly smaller share of housing units that were renter-occupied in 2009/2013. All other differences observed in the simple unadjusted models lost statistical significance in one or both of the adjusted models. Counties with the most wells and counties with some wells experienced significantly smaller average growth in total housing stock between 2000 and 2014 and in the renter-occupied housing stock between 2000 and 2009/2013 compared with counties with no wells, but those differences were explained by changes in other county-level characteristics between 2000 and 2009/2013. Moreover, counties with no wells experienced an average increase in the stock of owner-occupied housing units between 2000 and 2009/2013, but counties with the most wells experienced virtually no change, and counties with some wells experienced a decline in the stock of owner-occupied housing units. This difference was also explained by changes in other county-level characteristics between 2000 and 2009/2013. Specifically, these differences were explained by differences in population density changes between the three groups of counties. An increase in population density is related to increases in overall housing stock, owner-occupied housing stock, and renter-occupied housing stock. That is, counties that experienced increases in population density also experienced increases in housing stock, on average, between 2000 and 2009/2013. Counties with no wells experienced an average increase in population density between 2000 and 2009/2013, whereas counties with some wells and counties with the most wells experienced average declines in population density over the same period. These differences in population density changes explain the differences observed in total, owner-occupied, and renteroccupied housing stock growth between the three groups of counties. Third, pre-marcellus characteristics explained the significantly smaller increase in the stock of vacant housing units between 2000 and 2009/2013 in counties with some wells versus counties with no wells. Specifically, the difference observed in the simple unadjusted model was accounted for by larger shares of employment in 2000 in construction, extraction, and maintenance occupations and in production, transportation, and material moving occupations in counties with some wells compared to counties with no wells. Counties with some wells have a pre-marcellus history of transient laborers that need rental housing. Whereas all three groups of counties experienced average increases in their vacant housing stock, counties with larger shares of these kinds of workers experienced smaller increases in vacant housing stock. Finally, although the simple unadjusted model showed that the average share of vacant housing units available for rent in 2009/2013 was significantly lower in counties with the most wells and in counties The Center for Rural Pennsylvania 30

31 with some wells compared to counties with no wells, the magnitude of those differences declined substantially in the model that accounted for pre-marcellus county-level differences, and the statistical significance of those differences disappeared. Again, a history of higher shares of employment in construction, extraction, and maintenance occupations and in production, transportation, and material moving occupations in counties with wells explains the smaller supply of vacant and available rental units in counties with drilling activity in 2009/2013. However, this does not imply that counties with drilling activity are not at a rental disadvantage. Even though pre-marcellus industry characteristics explain the difference in rental availability between counties with wells versus those without, the reality of lower rental availability in counties with drilling activity remains a problem for residents of these counties, especially as counties with well activity already had lower vacant rental unit availability in 2000 before well development began. It is important to note that though the larger reduction in vacant rental unit availability between 2000 and 2009/2013 in high well counties versus counties with no wells is not statistically significant, the difference is substantively large (12.6 percentage points), even when controlling for pre-drilling characteristics. There was substantial variability in changes in vacant rental availability between the high drilling counties over this time period. That is, some high drilling counties experienced much larger declines in vacant units available for rent over this period than did others. For example, whereas Bradford County experienced an increase of 16.2 percent in vacant units available for rent, all other high drilling counties experienced reductions of more than 20 percent over the study period, as shown in Table 15. Declines in vacant rental availability were especially pronounced in Greene, Lycoming, Fayette, Washington, and Westmoreland counties. Therefore, some high drilling counties were more strongly impacted by declines in vacant rental availability than others. Table 15. Number and Percentage of Vacant Housing Units Available for Rent in High Drilling Counties, 2000 and 2009/2013 County N(%) vacant units for rent, 2000 N(%) vacant units for rent, 2009/2013 in stock of vacant housing units for rent, /2013 Bradford 501 (11.90) 582 (10.11) Butler 1,123 (28.03) 872 (16.63) Fayette 1,661 (25.47) 1,043 (12.51) Greene 327 (20.21) 191 (9.50) Lycoming 1,180 (21.61) 687 (10.69) Susquehanna 255 (4.81) 170 (2.94) Tioga 268 (6.75) 197 (4.60) Washington 1,946 (31.71) 1,220 (13.53) Westmoreland 2,970 (26.41) 1,848 (11.57) Sources: Census 2000 Summary File 3 (SF-3); American Community Survey 5-year estimates; American Community Survey 5-year estimates. The Center for Rural Pennsylvania 31

32 Housing Affordability Trends in housing affordability, including owner-occupied home values, median rental costs, and the percentage of the county population paying affordable rental prices (less than 30 percent of their incomes on rent) are presented in this section. Home Values Median owner-occupied home values for northern tier study and adjacent counties from 2000 to 2009/2013 are shown in Table 16. Both Bradford and Lycoming counties experienced increases in median owner-occupied housing values from 2000 to 2009/2013. Lycoming County experienced greater increases from 2000 to 2005/2009, whereas Bradford County experienced more growth from 2005 to 2009/2013. All adjacent counties also experienced increases in median housing values from 2000 to 2009/2013 with Wyoming County having the largest increase of just over 25 percent and Potter County experiencing the smallest increase of just over 1 percent. Table 16. Median House Value (2013 dollars) for all Owner-Occupied Housing Units and Percentage Change in Northern Tier Study and Adjacent Counties, / / / /2013 Pennsylvania $128,237 $165,403 $164, Study Counties Bradford $100,100 $105,997 $119, Lycoming $114,574 $123,916 $131, Adjacent Counties $110,855 $122,320 $129, Clinton $101,994 $104,042 $110, Columbia $116,062 $124,133 $135, Montour $123,773 $147,375 $153, Northumberland $94,690 $97,851 $98, Potter $95,366 $94,811 $96, Sullivan $109,299 $129,781 $137, Susquehanna $114,304 $127,609 $139, Tioga $97,666 $107,191 $116, Union $131,483 $149,004 $154, Wyoming $123,908 $141,401 $155, Sources: Census 2000 Summary File 3 (SF-3); American Community Survey 5-year estimates; American Community Survey 5-year estimates. Note: Inflation adjusted to 2013 dollars. Information on the median values of owner-occupied housing units (in 2013 constant dollars) in the southwestern region study and adjacent counties from 2000 to 2009/2013 is shown in Table 17. Both Greene and Washington counties experienced increases in median home values from 2000 to 2009/2013, with Washington County having the larger increase of just over 24 percent compared to just over 18 percent in Greene County. All adjacent counties experienced growth during this time period as The Center for Rural Pennsylvania 32

33 well. Allegheny and Westmoreland counties experienced growth from 2000 to 2005/2009 and from 2005/2009 to 2009/2013, whereas Beaver and Fayette counties experienced growth during the first period and decreases during the latter. Table 17. Median House Value (2013 dollars) for all Owner-Occupied Housing Units and Percentage Change in Southwest Region Study and Adjacent Counties, / / / /2013 Pennsylvania $128,237 $165,403 $164, Study Counties Greene $75,481 $85,905 $89, Washington $115,521 $135,428 $143, Adjacent Counties $106,492 $114,766 $113, Allegheny $112,951 $119,789 $122, Beaver $112,545 $119,138 $115, Fayette $81,974 $88,620 $84, Westmoreland $118,497 $131,518 $133, Sources: Census 2000 Summary File 3 (SF-3); American Community Survey 5-year estimates; American Community Survey 5-year estimates. Note: Inflation adjusted to 2013 dollars. Rental Costs Table 18 presents trends in median gross rent 10 in northern tier study and adjacent counties in 2005/2009 and 2009/2013. Estimates from 2000 are not included because Census 2000 tables were not released for total renter-occupied units. The population universe in 2000 was specified renteroccupied housing units whereas the universe in the ACS is renter occupied housing units. The Census indicates that comparisons cannot be made between the ACS and the 2000 Census. Median gross rent is lower in all northern tier counties than the Pennsylvania average of $813. The median gross rent in Pennsylvania increased 4.5 percent between 2000 and 2009/2013. The median rent in Bradford County increased more at nearly 11 percent, whereas in Lycoming County, the median rent increased by just over six percent. All but two adjacent counties also experienced increases during this time frame, with the largest increases in Sullivan, Tioga and Montour counties. 10 Gross rent is the contract rent plus the estimate average monthly cost of utilities (electricity, gas, and water and sewer) and fuels (oil, coal, kerosene, wood, etc.) if these are paid by the renter (or paid for the rent by someone else). Gross rent is intended to eliminate differentials that result from varying practices with respect to the inclusion of utilities and fuels as part of the rental payment. The estimated costs of water, sewer and fuels are reported on a 12-month basis but are converted to monthly figures for the tabulations. Gross rent provides information on the monthly housing cost expenses for renters. When the data are used in conjunction with income data, the information offers an excellent measure of housing affordability and excessive shelter costs. (U.S. Census Bureau American Community Survey yr Summary File: Technical Documentation, available at: The Center for Rural Pennsylvania 33

34 Table 18. Median Gross Rent (2013 dollars 11 ) and Percentage Change in Northern Tier Study and Adjacent Counties, 2005/ / / /2013 Pennsylvania $778 $ Study Counties Bradford $578 $ Lycoming $635 $ Adjacent Counties $610 $ Clinton $646 $ Columbia $640 $ Montour $639 $ Northumberland $552 $ Potter $628 $ Sullivan $460 $ Susquehanna $621 $ Tioga $609 $ Union $649 $ Wyoming $652 $ Sources: American Community Survey 5-year estimates; American Community Survey 5-year estimates. Note: Inflation adjusted to 2013 dollars. Table 19 shows trends in median gross rent in the southwestern region study and adjacent counties for 2005/2009 and 2009/2013. Median rents in all southwestern counties are also lower than the state average. The median rent in Greene County increased by 9.5 percent, and median rent in Washington County increased by 6.3 percent. All adjacent counties also experienced increases, though not as large as those in the study counties. The smallest increase was in Beaver County (0.5 percent). Table 19. Median Gross Rent (2013 dollars) and Percentage Change in Southwest Region Study and Adjacent Counties, 2005/ / / /2013 Pennsylvania $778 $ Study Counties Greene $545 $ Washington $605 $ Adjacent Counties $626 $ Allegheny $734 $ Beaver $620 $ Fayette $546 $ Westmoreland $605 $ To inflate gross rent amounts for years prior to 2013, Census inflates the dollar value to 2013 dollar values by multiplying by a factor equal to the average annual Consumer Price Index (CPI-U-RS) factor for 2013, divided by the average annual CPI-U-RS factor for the earlier year. The Center for Rural Pennsylvania 34

35 Sources: Census 2000 Summary File 3 (SF-3); American Community Survey 5-year estimates; American Community Survey 5-year estimates. Note: Inflation adjusted to 2013 dollars. Trends in the percentage of household income spent on rent in the northern tier study and adjacent counties between 2000 and 2009/2013 are shown in Table 20. Like Pennsylvania, both Bradford and Lycoming counties experienced continuous increases in the average percentage of income spent on rent, with most of the increase occurring between 2000 and 2005/2009. Yet some adjacent counties experienced larger increases in this measure between 2000 and 2009/2013, including Sullivan County, where the average percentage of income spent on rent increased nearly 36 percent. Table 20. Median Gross Rent as a Percentage of Household Income and Percentage Change in Northern Tier Study and Adjacent Counties, / / / /2013 Pennsylvania Study Counties Bradford Lycoming Adjacent Counties Clinton Columbia Montour Northumberland Potter Sullivan Susquehanna Tioga Union Wyoming Sources: Census 2000 Summary File 3 (SF-3); American Community Survey 5-year estimates; American Community Survey 5-year estimates. Rent is considered affordable when a household spends no more than 30 percent of its income on rent. Table 21 shows trends in the percentage of renters spending more than 30 percent of household income on rent in northern tier study and adjacent counties from 2000 to 2009/2013. The percentage of renters spending over 30 percent of housing income on rent increased by nearly 26 percent in Bradford County and 18 percent in Lycoming County from 2000 to 2009/2013. Both counties experienced increases from 2000 to 2005/2009 but then decreases from 2005/2009 to 2009/2013. All adjacent counties experienced increases in the share of renters spending over 30 percent of household income on rent from 2000 to 2009/2013. Montour, Sullivan, and Wyoming counties experienced declines from 2000 to 2005/2009 but increases from 2005/2009 to 2009/2013, whereas Potter County experienced increases from 2000 to 2005/2009 and decreases from 2005/2009 to 2009/2013. The remaining adjacent counties experienced continuous increases between 2000 and 2009/2013. The Center for Rural Pennsylvania 35

36 Table 21. Percentage of Renters Spending more than 30 Percent of Household Income on Rent and Percentage Change in Northern Tier Study and Adjacent Counties, / / / /2013 Pennsylvania Study Counties Bradford Lycoming Adjacent Counties Clinton Columbia Montour Northumberland Potter Sullivan Susquehanna Tioga Union Wyoming Sources: Census 2000 Summary File 3 (SF-3); American Community Survey 5-year estimates; American Community Survey 5-year estimates. Trends in median gross rent as a percentage of household income in the southwestern region study and adjacent counties from 2000 to 2013 are presented in Table 22. Rent became less affordable in all southwestern region counties over the study period. Greene County experienced an increase of just over 29 percent in the percentage of household income spent on rent from 2000 to 2009/2013, whereas Washington County experienced an increase of 12 percent. Adjacent counties experienced increases ranging from 10 percent in Fayette County to just over 22 percent in Beaver County. The majority of these increases occurred between 2000 and 2005/2009, with Washington, Fayette, and Allegheny counties experiencing slight declines in the percentage of household income spent on rent between 2005/2009 and 2009/2013. The Center for Rural Pennsylvania 36

37 Table 22. Median Gross Rent as a Percentage of Household Income and Percentage Change in Southwest Region Study and Adjacent Counties, / / / /2013 Pennsylvania Study Counties Greene Washington Adjacent Counties Allegheny Beaver Fayette Westmoreland Sources: Census 2000 Summary File 3 (SF-3); American Community Survey 5-year estimates; American Community Survey 5-year estimates. Trends in renters spending more than 30 percent of their household incomes on rent in Southwestern region study and adjacent counties from 2000 to 2009/2013 are shown in Table 23. Greene and Washington Counties both experienced increases in the percentage of renters spending more than 30 percent of household income on rent between 2000 and 2009/2013 a nearly 39 percent increase in Greene County and a nearly 21 percent increase in Washington County. All adjacent counties also experienced increases in the percentage of residents paying unaffordable rent between 2000 and 2009/2013, with the largest increase in Beaver County of 31 percent. Table 23. Percentage of Renters Spending more than 30 Percent of Household Income on Rent and Percentage Change in Southwest Region Study and Adjacent Counties, / / / /2013 Pennsylvania Study Counties Greene Washington Adjacent Counties Allegheny Beaver Fayette Westmoreland Sources: Census 2000 Summary File 3 (SF-3); American Community Survey 5-year estimates; American Community Survey 5-year estimates. Statistical Relationships between Marcellus Activity and Housing Affordability Descriptive statistics (means and standard deviations) for housing affordability outcomes across the three categories of well development (no wells, some wells, and high number of wells [top quartile of well count]) are presented in Table 24. Subscripts indicate when significant differences in outcomes exist The Center for Rural Pennsylvania 37

38 between the three county types at the p<0.05 (95 percent confidence) level or better. Readers should be aware that even when there are substantial differences in means across the categories, those differences may not be statistically significant because there may also be substantial variability in that outcome among the counties in that category (i.e., large standard deviations). Moreover, though the table includes the absolute values in each outcome for 2000, 2005/2009, and 2009/2013, readers may wish to focus on the variables representing percentage change in each outcome over the study period, as this report is intended to assess relationships between unconventional natural gas development and changes in outcomes. As shown in Table 24, the median value of owner-occupied housing units is significantly higher in counties with no wells compared with counties with some or high well counts. Counties with the most wells experienced significantly larger increases in the median value of owner-occupied housing units between 2000 and 2009/2013 compared with counties with some wells, but counties with no wells experienced significantly greater increases in the median value of owner-occupied housing units between 2000 and 2009/2013 compared both with counties with some wells and counties with high well counts. Turning to variation in the cost of rent across the three county types, counties with no wells have a significantly higher median gross rent compared to the other two county types, but counties with some wells and counties with a high well count have similar median rents. In terms of increases in rent between 2000 and 2009/2013, counties with some wells experienced the smallest increases that were significantly smaller increases than both counties with no wells and counties with a high well count. It is noteworthy that counties with a high well count experienced the largest increase in median rent over the study period, but the difference between counties with a high well count and counties with no wells is not statistically significant. All three county types experienced increases in the percentage of renters income spent on rent between 2000 and 2009/2013 and increases in the percentage of renters spending more than 30 percent of their income on rent (a standard measure of housing affordability), but those increases were significantly smaller for counties with wells compared to counties without wells. On average, high drilling counties had the lowest percentage of renters spending 30 percent or more of their income on rent in 2009/2013 and experienced the smallest increases in the percentages of renters spending more than 30 percent of their incomes on rent between 2000 and 2009/2013. The Center for Rural Pennsylvania 38

39 Table 24. Means for Housing Affordability Outcomes by Level of Well Development Outcomes No Wells Some Wells High Wells (N=28) (N=30) (N=9) Median value of all owner-occupied housing units, 2000 (adjusted to 2013 $) $141,078 (38,462) $99,870 (18,273) a $106,729 (20,388) b Median value of all owner-occupied housing units, (adjusted to 2013 $) $183,819 (68,855) $108,737 (26,380) a $119,065 (25,012) b Median value of all owner-occupied housing units, (adjusted to 2013 $) $177,161 (57,853) $110,130 (29,139) a $125,156 (26,667) b Percent change in median value of owner-occupied housing units, / (15.12) 9.12 (10.34) a (6.34) b,c Median gross rent, 2000 (adjusted to 2013 $) $719.7 (156.8) $561.4 (66.2) a $569.0 (50.7) b Median gross rent, 2005/2009 (adjusted to 2013 $) $771.0 (171.2) $601.8 (75.1) a $607.3 (53.2) b Median gross rent, 2009/2013 (adjusted to 2013 $) $816.9 (171.8) $614.0 (79.9) a $650.8 (49.5) b Percent change in median gross rent, / (4.26) 9.33 (4.79) a (3.73) c Percentage of renters' income spent on rent, (2.19) (2.24) (0.52) Percentage of renters' income spent on rent, 2005/ (2.79) (2.97) (0.70) Percentage of renters' income spent on rent, 2009/ (3.12) (2.34) (1.33) Percent change in percentage of renters' income spent on rent, / (6.58) (6.78) a (5.56) b Percentage of renters spending 30% or more of household income on rent, (5.17) (5.43) (2.03) Percentage of renters spending 30% or more of household income on rent, 2005/ (6.61) (5.73) (2.38) Percentage of renters spending 30% or more of household income on rent, 2009/ (7.15) (4.78) (2.94) Percent change in percentage of renters spending 30% or more of household income on rent, / (14.67) (13.81) a (7.98) b Notes: two-tailed difference of means t-tests. a Significant difference between no wells and some wells. b Significant difference between no wells and high wells. c Significant difference between some wells and high wells. Table B4 in Appendix B presents results from unadjusted linear regression models predicting differences in housing affordability outcomes between counties with a high well count (top quartile of well count) vs. counties with no wells, and between counties with some wells versus counties with no wells. To reiterate, these are models that do not control for the other county-level characteristics that may influence both well development and housing outcomes. As discussed in the Methods section, these factors may include county population density, unemployment and poverty rates, and shares of the county workforce employed in construction, extraction or maintenance and production, transportation or material moving. The table also includes R 2 values indicating the proportion of variation in each outcome explained by differences in well count. The Center for Rural Pennsylvania 39

40 Housing values, rental costs, and percentage of income spent on rent are higher in counties with no wells compared to counties with any wells. The median value of owner-occupied housing units and median gross rent are significantly lower in counties with any wells (high well count and some wells) compared to counties with no wells. Counties with some wells experienced significantly smaller increases in the value of owner-occupied housing units and in the median gross rent between 2000 and 2009/2013 versus counties with no wells. Counties with any wells (high and some) also experienced significantly smaller increases in the percentage of renters income spent on rent and the percentage of renters spending at least 30 percent of their incomes on rent between 2000 and 2009/2013 compared to counties with no wells. To determine whether these differences are simply a function of urban living or employment characteristics, given that most counties without wells are located in urban areas with higher cost of living and different employment conditions, models were adjusted to account for year 2000 population density, percent poverty, unemployment rates, and shares of the county workforce employed in construction, extraction or maintenance and production, transportation or material moving and changes in each of those values between 2000 and 2009/2013. Results from these adjusted models are presented in Appendix B, Tables B5 and B6. Coefficients that are italicized indicate loss of statistical significance between the unadjusted and adjusted model. Only two differences remained statistically significant after accounting for both pre-drilling characteristics and changes in county characteristics between 2000 and 2009/ the median value of owner-occupied housing units in 2009/2013 and the percent change in median value of owneroccupied housing units between 2000 and 2009/2013. Despite differences in population density, poverty, and employment conditions, counties with some wells have significantly lower median owneroccupied housing values and experienced significantly smaller increases in owner-occupied housing values between 2000 and 2009/2013 compared to counties with no wells. In addition, controlling for changes in county-level characteristics between 2000 and 2009/2013 brought to light a significant difference in percent change in median owner-occupied housing values between counties with no wells and counties with the most wells. Counties with the most wells experienced a significantly smaller increase in poverty rates between 2000 and 2009/2013 than counties with no wells, and increases in the poverty rates are related to declines in median owner-occupied housing values. If counties with the most wells had experienced similarly larger poverty rate increases as counties with no wells, counties with the most wells would have experienced significantly larger declines in median owner-occupied housing values. The smaller increases in poverty rates in counties with the most wells were protective against larger median home value declines between 2000 and 2009/2013. All other significant differences observed in the simple unadjusted model were accounted for by either pre-marcellus county characteristics or changes in county characteristics between 2000 and 2009/2013. First, the lower median gross rent in 2009/2013 in counties with the most wells compared to counties with no wells was explained by differences in population density changes. Counties with no wells experienced larger increases in population density between 2000 and 2009/2013. An increase in population density is related to an increase in gross rent. Second, in the unadjusted models, counties with the most wells and counties with some wells experienced smaller increases than counties with no wells in average percentages of renters incomes spent on rent and smaller increases than counties with no wells in the percentages of renters spending 30 percent or more of their household income on rent. These differences were explained by differences in population density and differences in poverty rate increases. Counties with no wells experienced larger increases in average percentages of income spent The Center for Rural Pennsylvania 40

41 on rent because they experienced larger increases in population density and larger increases in poverty rates between 2000 and 2009/2013 compared to counties with wells. Household Income Inequality This section presents trends in measures of income inequality. This includes median household incomes for households that are owner- and renter-occupied (adjusted to 2013 dollars), the gap in median household income between these groups, and the Gini coefficient (a global measure of household income inequality). It is important to note that median household incomes do not account for the number of people living in a household. Households occupied by homeowners may have more residents contributing to cost of living than households occupied by renters, but owner-occupied households may also have more children on average, than renters, thereby increasing overall living expenses. Table 25 shows trends in median household income among homeowners in the northern tier study and adjacent counties from 2000 to Pennsylvania experienced increases in median household income during this time period, but both Bradford and Lycoming counties experienced declines in household income among homeowners. This decline was slight for Bradford County (0.2 percent), but was larger in Lycoming County (3.8 percent). Both counties experienced declines in homeowner median household income from 2000 to 2005/2009, but Lycoming County continued to decline from 2005/2009 to 2009/2013, whereas Bradford County recovered over this period. Adjacent counties experienced declines in median household income from 2000 to 2013, overall, but there was variation among these counties. Clinton, Northumberland, Potter, Tioga, Union, and Wyoming counties all experienced declines in median household income among homeowners, but Columbia, Montour, Sullivan, and Susquehanna counties experienced improvements. Table 25. Median Household Income for Owner-Occupied Housing Units and Percentage Change in Northern Tier Study and Adjacent Counties, / / / /2013 Pennsylvania $47,611 $54,016 $65, Study Counties Bradford $55,439 $50,786 $55, Lycoming $57,222 $56,364 $55, Adjacent Counties $54,436 $52,630 $53, Clinton $51,993 $50,975 $51, Columbia $55,843 $54,605 $56, Montour $59,313 $59,506 $59, Northumberland $50,993 $50,545 $50, Potter $50,977 $45,616 $47, Sullivan $48,529 $48,163 $50, Susquehanna $52,426 $52,000 $54, Tioga $51,567 $50,666 $50, Union $65,589 $57,921 $56, Wyoming $57,134 $56,304 $56, Sources: Census 2000 Summary File 3 (SF-3); American Community Survey 5-year estimates; American Community Survey 5-year estimates. Note: Inflation adjusted to 2013 dollars. The Center for Rural Pennsylvania 41

42 Trends in median household income among renters in the northern tier study and adjacent counties from 2000 to 2013 are shown in Table 26. Whereas Pennsylvania as a whole experienced increases in median household income among renters, Bradford and Lycoming counties both experienced decreases of 11 percent between 2000 and 2009/2013. However, whereas both counties experienced declines from 2000 to 2005/2009, they experienced increases between 2005/2009 to 2009/2013. It is noteworthy that this represents a period of substantial well development in these counties. However, these increases were not large enough to offset the declines experienced between 2000 and 2005/2009. All adjacent counties, except for Tioga, experienced declines in renter median household income from 2000 to 2009/2013, ranging from nearly 26 percent in Potter County to 3.5 percent in Sullivan County. However, both Sullivan and Tioga counties experienced increases in renter median household income between 2005/2009 and 2009/2013. Yet, as with Bradford and Lycoming counties, these increases were not large enough to offset the declines that occurred between 2000 and 2005/2009. Table 26. Median Household Income for Renters and Percentage Change in Northern Tier Study and Adjacent Counties, / / / /2013 Pennsylvania $24,601 $30,349 $29, Study Counties Bradford $30,741 $25,719 $27, Lycoming $29,845 $25,435 $26, Adjacent Counties $28,909 $26,298 $25, Clinton $24,270 $24,142 $22, Columbia $29,026 $26,407 $23, Montour $34,285 $35,167 $29, Northumberland $26,377 $22,698 $24, Potter $29,980 $23,470 $22, Sullivan $26,456 $23,159 $25, Susquehanna $29,030 $26,620 $25, Tioga $26,690 $22,293 $27, Union $30,426 $26,321 $26, Wyoming $32,548 $32,699 $30, Sources: Census 2000 Summary File 3 (SF-3); American Community Survey 5-year estimates; American Community Survey 5-year estimates. Note: Inflation adjusted to 2013 dollars. Trends in the household income gap between owners and renters in the northern tier study and adjacent counties from 2000 to 2009/2013 are shown in Table 27. The gap increased by nearly 56 percent in Pennsylvania between 2000 and 2009/2013. By 2009/2013, household income among homeowners in Pennsylvania was nearly $36,000 higher than among renters. Bradford and Lycoming counties experienced smaller increases in the owner-renter income gap of just over 13 percent and just over 4 percent, respectively. Adjacent counties as a whole experienced a just over 8 percent increase in this gap between 2000 and 2009/2013, but this masks significant heterogeneity. For instance, the homeowner advantage increased by over 20 percent in Columbia, Montour, Potter and Susquehanna counties but income among renters began to converge with income among homeowners in Tioga and Union counties. The Center for Rural Pennsylvania 42

43 Table 27. Gap in Median Household Income between Owners and Renters and Percentage Change in Northern Tier Study and Adjacent Counties, / / / /2013 Pennsylvania $23,010 $23,667 $35, Study Counties Bradford $24,698 $25,067 $27, Lycoming $27,377 $30,929 $28, Adjacent Counties $25,528 $26,333 $27, Clinton $27,723 $26,833 $29, Columbia $26,817 $28,198 $32, Montour $25,028 $24,339 $30, Northumberland $24,616 $27,847 $26, Potter $20,997 $22,146 $25, Sullivan $22,073 $25,004 $24, Susquehanna $23,396 $25,380 $28, Tioga $24,877 $28,373 $23, Union $35,163 $31,600 $30, Wyoming $24,586 $23,605 $25, Sources: Census 2000 Summary File 3 (SF-3); American Community Survey 5-year estimates; American Community Survey 5-year estimates. Note: Inflation adjusted to 2013 dollars. Table 28 shows trends in median household income among homeowners in the southwestern region study and adjacent counties from 2000 to 2009/2013. Greene and Washington counties experienced increases of 5 percent and nearly 4 percent, respectively, in median household income among homeowners between 2000 and 2009/2013. Among adjacent counties, Allegheny and Beaver counties experienced declines in homeowner median household income from 2000 to 2009/2013, whereas Fayette and Westmoreland counties both experienced increases. Table 28. Median Household Income for Owner-Occupied Housing Units and Percentage Change in Southwest Region Study and Adjacent Counties, / / / /2013 Pennsylvania $47,611 $54,016 $65, Study Counties Greene $50,977 $52,474 $53, Washington $61,270 $64,271 $63, Adjacent Counties $58,272 $58,069 $58, Allegheny $67,198 $66,105 $67, Beaver $59,970 $58,550 $59, Fayette $46,290 $47,063 $46, Westmoreland $59,628 $60,556 $60, Sources: Census 2000 Summary File 3 (SF-3); American Community Survey 5-year estimates; American Community Survey 5-year estimates. Note: Inflation adjusted to 2013 dollars. The Center for Rural Pennsylvania 43

44 Trends in median household income among renters in the southwestern region study and adjacent counties from 2000 to 2009/2013 are shown in Table 29. Renters in Greene County experienced an average increase in household income of just over 7 percent, whereas renters in Washington County experienced a decline of nearly 8 percent. Renters in adjacent counties experienced overall declines in median household income from 2000 to 2009/2013, but these declines were most severe for Beaver and Westmoreland counties. Table 29. Median Household Income for Renters and Percentage Change in Southwest Region Study and Adjacent Counties, / / / /2013 Pennsylvania $24,601 $30,349 $29, Study Counties Greene $22,652 $21,523 $24, Washington $28,593 $24,546 $26, Adjacent Counties $29,080 $25,686 $25, Allegheny $31,863 $28,083 $28, Beaver $31,208 $27,970 $26, Fayette $22,707 $21,346 $21, Westmoreland $30,543 $25,344 $25, Sources: Census 2000 Summary File 3 (SF-3); American Community Survey 5-year estimates; American Community Survey 5-year estimates. Note: Inflation adjusted to 2013 dollars. Table 30 shows trends in the household income gap between owners and renters in the southwestern study and adjacent counties from 2000 to 2009/2013. The gap increased in both Greene and Washington counties from 2000 and 2005/2009 but decreased between 2005/2009 and 2009/2013. All adjacent counties also experienced increases in the household income gap between owners and renters between 2000 and 2009/2013, ranging from nearly 6 percent in Fayette County to just over 20 percent in Westmoreland County. Table 30. Gap in Median Household Income between Owners and Renters and Percentage Change in Southwest Region Study and Adjacent Counties, / / / /2013 Pennsylvania $23,010 $23,667 $35, Study Counties Greene $28,325 $30,951 $29, Washington $32,677 $39,725 $37, Adjacent Counties $29,191 $32,383 $33, Allegheny $35,335 $38,022 $38, Beaver $28,762 $30,580 $33, Fayette $23,583 $25,717 $24, Westmoreland $29,085 $35,212 $34, Sources: American Community Survey 5-year estimates; American Community Survey 5-year estimates. Note: Inflation adjusted to 2013 dollars. The Center for Rural Pennsylvania 44

45 The Gini coefficient is a global measure of household income inequality that ranges from 0 to 1. A Gini coefficient of 0 expresses perfect equality (i.e., all households earn the same income), whereas a Gini coefficient of 1 expresses perfect inequality (i.e., one household controls all of the income, and the rest of the households have no income). For example, a case where the richest 20 percent of households have 80 percent of the income would lead to a Gini coefficient of at least.60. In 2009/2013, the United States had a Gini coefficient of.47, and Pennsylvania had a Gini coefficient of.46. Trends in the Gini coefficient in the northern tier study and adjacent counties are presented in Table 31. Note that the Gini coefficient was not available from the Census in The Gini coefficient (county income inequality) remained fairly stable for Bradford, Lycoming, Northumberland, Susquehanna, Tioga, and Unions counties between 2005/2009 and 2009/2013. However, there were rather large increases in income inequality in Clinton, Montour, and Wyoming counties. Table 31. Gini Coefficient of Household Income Inequality and Percentage Change in Study and Northern Tier Study and Adjacent Counties, / /2013 Pennsylvania Study Counties Bradford Lycoming Adjacent Counties Clinton Columbia Montour Northumberland Potter Sullivan Susquehanna Tioga Union Wyoming Sources: American Community Survey 5-year estimates; American Community Survey 5-year estimates. Note: Gini coefficient not available for Trends in income inequality in the southwestern region study and adjacent counties are presented in Table 32. Among all four study counties, Washington County experienced the largest increase in the Gini coefficient (income inequality) between 2005/2009 and 2009/2013 at 2.5 percent. The increase in income inequality in Bradford County was similar at 2.4 percent. Greene County actually experienced a decline in income inequality over the study period of 1.7 percent. Based on trends shown in the household income tables above, this is likely explained by increases in median household income among renters between 2005/2009 and 2009/2013. The Center for Rural Pennsylvania 45

46 Table 32. Gini Coefficient of Household Income Inequality and Percentage Change in Southwest Region Study and Adjacent Counties, / /2013 Pennsylvania Study Counties Greene Washington Adjacent Counties Allegheny Beaver Fayette Westmoreland Sources: American Community Survey 5-year estimates; American Community Survey 5-year estimates. Note: Gini coefficient not available for Statistical Relationships between Marcellus Activity and Household Income Inequality Means and standard deviations for income inequality outcomes across the three categories of well development are presented in Table 33. There were no significant differences in the global measure of household income inequality (the Gini coefficient) during either of the measurement time periods or in the change in the Gini coefficient between the two time periods. This suggests that well development was not related to increases or decreases in global household income inequality. The Center for Rural Pennsylvania 46

47 Table 33. Means for Inequality Outcomes by Level of Well Development Outcomes No Wells Some Wells High Wells (N=28) (N=30) (N=9) Gini coefficient of household income inequality, 2005/ (.028).423 (.022).416 (.028) Gini coefficient of household income inequality, 2009/ (.030).427 (.023).427 (.009) Percent change in Gini coefficient, 2005/ / (2.80) 0.87 (3.53) 0.81 (1.71) Median household income among homeowners, 2000 $68,312 (14,227) $53,953 (5,555) a $55,892 (6,555) b Median household income among homeowners, 2005/2009 $67,216 (14,536) $52,439 (6,057) a $56,110 (7,684) b Median household income among homeowners, 2009/2013 $66,002 (13,742) $52,748 (6,166) a $56,577 (6,950) b Percent change in median household income among homeowners, / (3.70) (4.88) 1.21 (2.86) b,c Median household income among renters, 2000 $37,049 (7,329) $27,716 (2,768) a $28,188 (3,549) b Median household income among renters, 2005/2009 $33,334 (6,135) $24,810 (2,800) a $24,736 (2,711) b Median household income among renters, 2009/2013 $31,657 (5,590) $23,918 (2,635) a $26,109 (2,272) b,c Percent change in household income among renters, / (5.88) (5.78) (7.24) b,c Gap in median household income between owners and renters, 2000 $31,263 (7,966) $26,237 (4,752) a $27,704 (4,149) Gap in median household income between owners and renters, 2005/2009 $33,882 (9,490) $27,629 (5,168) a $31,374 (6,067) Gap in median household income between owners and renters, 2009/2013 $34,345 (8,996) $28,830 (5,060) a $30,467 (5,501) Percent change in household income gap between owners and renters, / (11.00) (10.23) 9.79 (8.75) Notes: two-tailed difference of means t-tests. a Significant difference between no wells and some wells. b Significant difference between no wells and high wells. c Significant difference between some wells and high wells. However, there were significant differences between the county types in all other inequality outcomes. First, total median household income, median household income among homeowners, and median household income among renters are significantly higher in counties with no wells compared to counties with some wells and counties with a high well count. However, whereas counties with no wells and counties with some wells experienced declines in median household income among homeowners between 2000 and 2009/2013, counties with a high well count experienced a slight increase in median household income among homeowners. Moreover, the median household income among renters declined in all three county types over the study period, but the decline was the smallest in counties with a high well count. Finally, whereas the gap in median household income between owners and renters was not statistically different between counties with no wells and counties with a high well count, the gap between renters and owners was significantly smaller in counties with some wells compared to counties with no wells. There was no significant difference in changes in the household income gap between renters and homeowners between 2000 and 2009/2013. That is, well development The Center for Rural Pennsylvania 47

48 was not associated with either increasing or decreasing income inequality between households occupied by renters versus owners. Table B7 in Appendix B presents coefficients from unadjusted linear regression models predicting associations between level of well development and household income inequality. Again, these are models that do not control for any other county factors (described earlier) that may be related to both Marcellus Shale development and housing outcomes. These results mirror those of the difference of means tests above. There are no significant differences between counties with versus without wells in the Gini coefficient of household income inequality, changes in the Gini coefficient of income inequality, or changes in the income gap between renters and homeowners over the study period. Median household income overall and separately for homeowners and renters is significantly higher in counties with no wells versus counties with wells, but counties with a high well count experienced significantly larger increases in median household income among homeowners than did counties with no wells. The difference between counties with no wells and counties with a high well count in the percentage change in median household income among renters between 2000 and 2009/2013 was also statistically significant. The decline in median income among renters between 2000 and 2009/2013 was significantly smaller in counties with a high well count compared to counties with no wells. Results from linear regression models predicting associations between level of well development and household income inequality controlling for pre-drilling county characteristics are presented in Table B8 in Appendix B. As with earlier results, italics represent differences in housing outcomes that lost statistical significance with the inclusion of control variables (i.e., differences in housing outcomes that are explained by pre-drilling county characteristics). The inclusion of control variables explained the significantly lower median household income among homeowners and among renters in 2009/2013 in counties with high well counts versus no wells, but did not explain these differences between counties with some wells versus no wells. Net of pre-development conditions (e.g., population density, unemployment rate), counties with some wells have significantly lower median household income among both homeowners and renters compared to counties with no wells. Pre-drilling county characteristics did explain the smaller gap in median household income between owners and renters in counties with some wells versus no wells. The addition of control variables in the multivariate model eliminated the significant difference in this gap for both 2000 and 2009/2013. Finally, net of controls for pre-drilling characteristics, counties with a high well count continued to maintain a significant difference in percent change in median household income among renters between 2000 and 2009/2013 compared to counties with no wells. Keeping in mind that all county types experienced a decline in median household income among renters, the positive coefficient for high wells versus no wells indicates that the decline in median household income among renters in counties with a high well count was less severe compared to counties with no wells. Finally, Table B9 in Appendix B presents results from linear regression models predicting associations between well development and inequality outcomes while controlling for changes in county characteristics over the study period. Changes in county-level demographic and economic characteristics between 2000 and 2009/2013 explained several of the same differences in housing inequality outcomes between counties with versus without well development, as did controls for pre-drilling characteristics (as shown in Table B8 in Appendix B). However, in both sets of adjusted models, the coefficient for the difference in median household income among renters in 2009/2013 between counties with some wells versus counties with no wells remained negative and statistically significant; compared to counties with The Center for Rural Pennsylvania 48

49 no wells, counties with some wells have a lower average median household income among renters. One major difference between this model and the prior adjusted model that controlled for pre-marcellus characteristics is that the difference between counties with high wells versus counties with no wells in the change in household income among renters between 2000 and 2009/2013 is no longer significant in this model. This suggests that economic and demographic changes that occurred over the period of well development explain the fact that renters in counties with high well development has less severe reductions in household income compared to renters in counties with no wells. Summary and Implications Availability and Occupancy There was substantial variation in housing stock and changes in housing stock between 2000 and Among the study counties, Bradford and Washington counties experienced increases in the total number of housing units, Lycoming County experienced almost no change in its housing stock, and Greene County experienced a slight decline in the total number of housing units between 2000 and Comparisons of counties along three groupings of well development (counties with no wells, counties in the top 25 th percentile of well count, and counties in the bottom 75 th percentile of well count) revealed that counties with no wells experienced an average increase in owner-occupied housing units between 2000 and 2009/2013, whereas counties with wells experienced slight declines in owner-occupied housing units over this same period. Regression models indicated that this difference was explained by changes in county-level population density over the same period (i.e., counties with no wells experienced larger increases in population density versus counties with wells, and this large increase in population is what led to larger increases in owner-occupied housing units). All three groups of counties also experienced increases in vacant housing units over the study period, and counties with the most wells experienced the largest increase in the stock of vacant housing. This suggests a transition from owner-occupied housing to vacant housing in counties with substantial well development. The vacant rental supply decreased in all three well development categories between 2000 and 2009/2013, but it decreased the most in the counties with the most wells (i.e., those in the top 25 th percentile of well count), suggesting increased rental demand and/or reduced supply of rental units in counties with the most well development. Moreover, some high drilling counties were more strongly impacted by declines in rental availability than others and were impacted at different times. For instance, whereas Bradford County experienced a 22 percent decline in its stock of vacant rental units between 2000 and 2005/2009, Bradford experienced a 48 percent increase in vacant rental units between 2005/2009 and 2009/2013. Washington, Greene, and Lycoming counties, on the other hand, experienced declines in vacant rental stock of 37 percent, 42 percent, and 42 percent, respectively, between 2000 and 2009/2013. Overall, results show that, compared to counties with no wells, counties with wells experienced smaller increases in overall housing stock, declines in owner-occupied housing stock (compared to an average increase in counties with no wells), smaller increases in renter-occupied housing stock and vacant housing stock, and larger declines in vacant housing available for rent. However, all of these differences were explained by either pre-marcellus characteristics (e.g., population density) or changes in county characteristics between 2000 and 2009/2013. Importantly, the analyses conducted here do not enable ruling out the possibility that the housing supply in counties with substantial well development The Center for Rural Pennsylvania 49

50 increased more than it otherwise might have without any development at all in those same counties, particularly given that well development occurred over a period of widespread economic and housing market decline. Unfortunately, the aggregated American Community Survey data provide limited information on housing changes, especially changes that occurred rapidly (i.e., within a 1-2 year period). Moreover, the need to use 5-year estimates to capture all counties in PA and avoid biasing the results prohibits investigating changes occurring since Pennsylvania s economy started to recover from the recession until after the estimates become available. These secondary data also do not reflect the experiences of those who reside in temporary housing or who have been displaced because of increases in rent or loss of income. Temporary housing units (e.g., travel trailers, motels and hotels, man camps ) are not counted as housing units in the U.S. Census or the American Community Survey but present important challenges for local civic leaders and community members. In Wave 1 of this project, McLaughlin et al. (2014) found that use of temporary housing was a strategy employed by gas workers and low-income families. In that study and another (Williamson and Kolb, 2011), gas industry workers were more likely to opt for hotels and motels, especially if they provided housekeeping services and meals. As indicated in the Wave 1 focus groups, temporary housing for gas workers often takes the form of travel trailers or RVs located in rural areas some in longstanding campgrounds, others in people s yards, farmer s fields or converted golf driving ranges. From a policy perspective, local water and sewage regulations and enforcement of those regulations are essential for ensuring that water sources are safe and that the high density of RVs does not contaminate the ground or local water sources with sewage. A recent Bloomberg Business article details the use of man-camps in the Bakken (another area that experienced population growth related to oil extraction) as an interim solution for a rapid influx of workers until subdivisions and apartment complexes could be built. This resulted in the processing of hundreds of permits and borrowing of tens of millions of dollars to pay for water and sewer systems for these camps. Now that oil extraction has slowed in the Bakken, many of these camps sit empty, and much of the debt has yet to be repaid. Discarded RVs and trailers sometimes contain rotting food and pests. These problems have led cities and counties in the Bakken to change permitting policies and toughen zoning laws to outlaw or restrict temporary colonies (Oldham, 2015). State and local county officials may find these previous experiences and strategies from the Bakken useful when considering how to deal with housing issues that may arise when natural gas extraction subsides and workers leave the area. Affordability Previous research on the association between Marcellus Shale Development and housing found no significant relationship between wells drilled and median home values (Farren et al., 2013). The results of the present study differ from those of Farren et al. (2013) in some ways. Results of comparisons of home values across the three categories of well-development demonstrate that counties with the most wells experienced significantly larger increases in the median value of owner-occupied housing units between 2000 and 2009/2013 compared to counties with low-to-mid tier development. However, counties with no wells experienced significantly greater increases in the median value of owneroccupied housing units between 2000 and 2009/2013 compared to both counties in the low-to-mid tier development range and counties with high well counts. One of the most pressing housing concerns related to natural gas development has been the possibility that it could lead to increases in rental costs, thereby pricing local residents out of the rental market. All The Center for Rural Pennsylvania 50

51 three county types experienced increases in the percentage of renters income spent on rent and increases in the percentage of renters spending more than 30 percent of their income on rent (a standard measure of housing affordability) between 2000 and 2009/2013, but those increases were significantly smaller for counties with wells (both low-to-mid tier drilling and high drilling) compared to counties without wells. However, these differences were explained by pre-marcellus county-level economic and population characteristics. That is, counties without wells tend to be part of urban areas that have overall higher costs of living. Accordingly, well development itself does not appear to be related to differences in rental affordability across the three types of counties. However, it is important to note that, despite expected differences in cost of living (i.e., that it is generally lower in rural areas), there were not significant differences in median gross rent in 2009/2013 between counties with no wells that are located mostly in urban areas versus counties with the highest well counts which are usually more rural. These findings imply that renters in smaller counties with the highest drilling activity are paying similar rental costs as those in urban parts of the state. Moreover, counties with the highest well counts experienced the largest average increase in median rent over the study period, but these counties also had the lowest percentage of renters spending 30 percent or more of their income on rent in 2009/2013 and experienced the smallest increases in the percentages of renters spending 30 percent or more of their incomes on rent between 2000 and 2009/2013. Given the finding of significantly lower median household income among renters in high drilling counties versus counties with no wells, it may be that renters in high drilling counties are sharing rental units (e.g., doubling- or tripling-up) in an effort to reduce individual expenses. It may also be the case that displacement of low-income families and individuals from housing results in rental housing appearing to be more affordable than it actually is. Only those who can afford the rent and initial costs to rent housing are included in the data. In either case, the findings related to rental prices highlight an important area for monitoring after well development ends and temporary workers leave these counties. Whether rents will drop back down to pre-marcellus levels post-drilling or whether rents will remain high once demand for housing decreases remains to be seen, but it is unlikely that local residents will be able to maintain these significantly higher rental costs once these relatively high paying jobs and the economic opportunities that gas workers create in the service sector leave these counties. Inequality There is substantial variability in household income inequality across the state. This includes the global measure of inequality (the Gini coefficient), as well as differences in household income between renters and homeowners. Household income inequality was lower among the four study counties than for Pennsylvania as a whole, and the study counties had comparable levels of household income inequality as the adjacent counties in the same regions, on average. However, in the northern tier, Clinton, Montour, and Wyoming counties had relatively low income inequality in 2005/2009 but experienced especially substantial increases in income inequality between 2005/2009 and 2009/2013, suggesting that high-income residents moved into these counties over this period and/or that a small percentage of residents experienced large increases in income while the majority of residents experienced stagnant or declining wages in these counties over this period. These data are unable to identify the exact explanations for these changes. Ultimately, findings from regression analyses suggest that well development was not associated with overall household income inequality nor increases or decreases in household income inequality over the study period. There were also differences across counties in changes to median income among renters and owners between 2000 and 2009/2013. Whereas counties with no wells and counties with low-to-mid level The Center for Rural Pennsylvania 51

52 drilling experienced declines in median household income among homeowners between 2000 and 2009/2013, counties with the most wells experienced a slight increase in median household income among homeowners over the same period, suggesting the possibility that homeowners experienced income benefits from drilling, on average. If this is the case, it could be due either to increased income from leasing mineral rights or royalties from gas production or to increased wages related to natural gas sector employment or employment in related industries. These data and analyses are not able to tease out the merit of these potential explanations. Moreover, whereas median household income among renters declined in all three categories of well development over the study period, the decline was the smallest in counties with the most wells. Regression analyses revealed that changes in economic and demographic characteristics between 2000 and 2009/2013 explain the finding that renters in counties with the most wells had less severe reductions in renter s household income over the same period compared to renters in counties with no wells. Finally, the gap in median household income between renters and owners increased substantially in Pennsylvania between 2000 and 2009/2013. Whereas homeowners earned $23,010 more than renters in 2000 (in 2013 adjusted dollars), this gap had increased to $35,826 by 2009/2013. The income gap between owners and renters increased more in Bradford County than in Lycoming County and more in Washington County than in Greene County. Whereas all adjacent counties in the southwestern region experienced increases in the owner/renter income gap between 2000 and 2009/2013, in the northern tier, both Tioga and Wyoming counties experienced decreases in the owner/renter income gap. These declines were mostly attributable to declines in income among homeowners rather than increases in income among renters. Whereas the gap in median household income between homeowners and renters was significantly lower in low-to-mid tier drilling counties versus counties with no wells, pre- Marcellus population and economic characteristics explained this difference. There were no significant differences in the owner/renter income gap between counties with the most wells versus counties with no wells, and number of wells was not associated with changes in the owner/renter income gap between 2000 and 2009/2013. Further Considerations Results of this study should be considered in light of some limitations not already addressed above. In addition to failing to account for temporary housing structures, these data do not separately capture seasonal or recreational homes (they are included in the count of housing units that are vacant, but not available for rent). Some of the findings related to rental vacancy may be related to more seasonal housing in rural areas. However, supplementary analyses (not shown in the report) confirmed that there were no significant differences in changes in temporary/seasonal housing units over the period of substantial drilling between the three categories of well development. Second, secondary data cannot capture how the residents of counties with high drilling experience housing challenges. Experiences among low-income families and elders, for instance, may vary dramatically from those of middle- and higher-income residents with financial security and stability. Wave 1 of this study reported temporary bouts of homelessness and negative interactions between low-income residents and landlords in highly drilled counties (McLaughlin et al., 2014). As noted in the housing report from Wave 1, the displacement of low-income families from housing so that housing units could be rented to individuals at much higher rents is a major problem facing communities with substantial drilling activities, especially those with smaller populations where housing was already limited prior to widespread development activities. There are few displacement protections in place particularly for low-income and elderly residents. Local social and human service providers in the study counties in Wave 1 were well aware of the potential for and actual displacement of these families and the strategies the families and individuals used to find The Center for Rural Pennsylvania 52

53 housing, including couch-surfing, moving in with relatives, moving to substandard units or a homeless shelter, or, in some cases, living in tents or cars (McLaughlin et al., 2014). These types of transitions in housing and the associated costs to families and local governments and nonprofit organizations trying to assist families in resolving these housing crises are not captured in the secondary data used in this report. Other factors may have been associated with changes in housing in specific counties over the study period, including a casino built in Washington County and gas-company built worker housing in the northern tier of Pennsylvania (Williamson and Kolb, 2011). Without annual consistent longitudinal data that captures housing and multiple other (e.g., economic, demographic, policy change) measures before, during, and after industry development, it is not possible to identify housing transactions and characteristics directly related to Marcellus Shale versus those related to these other county-specific changes. Another consideration is that activities supporting the natural gas industry may not be located directly in those areas of heavy drilling. Regional headquarters, for instance, may employ several higherincome workers who may be located far outside of drilling areas (Williamson and Kolb, 2011). These workers may be more likely to stay in the area for longer periods of time or well after drilling has ended, and they may be more likely to purchase homes versus rent. The specific composition (educational, income, intentions to stay after drilling has ended) of natural gas and natural gas-related workers in a county and the timing of various development activities influence the demand for different types of housing. Both Bradford and Washington counties are home to regional offices of gas companies. Lycoming County has become a location for companies that support the gas industry. This dispersion of natural gas-related activity to counties other than those experiencing the largest volume of active drilling may explain some of the non-significant differences observed between counties with active drilling and those with no drilling activity. Finally, the period of substantial drilling coincided with the Great Recession that officially began at the end of 2007 and concluded in Despite the official end date, however, some research suggests that the effects of the recession impacted some rural areas later and over a longer period of time than urban areas and that the economic recovery has been slower in rural areas (Hertz et al., 2014; Thiede and Monnat, 2014). One of the major causes and consequences of the recession was high rates of home foreclosure and declining housing values. Therefore, it is not possible to determine whether any of the changes in housing outcomes observed in this study were due to the recession affecting some counties more than others. However, the regression analyses account for both pre-marcellus economic characteristics and changes in economic characteristics between 2000 and 2009/2013, limiting the possibility that observed changes and/or non-significant findings are driven by heterogeneous recession impacts. Ultimately, results of this and other similar studies suggest that there have been very few objective changes in housing outcomes associated with well development in the Marcellus Shale, and many of the observed differences across counties are largely accounted for by heterogeneity in county economic and population characteristics prior to the existence of widespread drilling. Yet, these aggregated data cannot get at heterogeneity in housing experiences across diverse groups of homeowners and renters. For subjective housing experiences among residents in high-drilling Pennsylvania counties, readers should read results from focus groups conducted in Wave 1 of this study (McLaughlin et al., 2014). It is likely that housing needs and characteristics of housing depend on the types of jobs workers hold and the phase of natural gas production (Williamson and Kolb, 2011; Farren et al., 2013). The Center for Rural Pennsylvania 53

54 Ideally, local officials and developers would have time to prepare for massive influxes of new residents and any potential impacts that may have on long-term residents. However, gas companies rarely alert local government officials about plans to move into an area to drill, the planned timeline for drilling, or the intensity with which drilling activities will occur. Accordingly, shifts in housing needs are unpredictable. Moreover, most industry-related workers moving to areas with natural gas development tend to be those working in the active drilling phases of gas extraction. These workers also are the most mobile and most likely to be relocated from one drilling site to another. Gas industry representatives claim that drilling in the Marcellus Shale (or related Shales) will continue over decades. However, local officials and service providers should be aware that this long-term natural gas extraction will be mobile and geographically dispersed across various Shales. Drilling activity moves from one area to another in response to changes in the price of natural gas, the type of gas produced, the availability of pipelines to carry gas to processing facilities or to markets, and the discovery of new, richer, gas fields elsewhere. There is already evidence of this movement in Pennsylvania, as some locations with previous rapid and significant drilling are now experiencing no activity at all. As noted in the Wave 1 housing report, Pennsylvania s northern tier counties already experienced a mass exodus of industry workers from the region. As natural gas prices dropped in the northern tier, gas extraction efforts shifted to regions where wet natural gas was readily available. A central Pennsylvania staging area filled with equipment, pipe, and drilling rig parts was emptied out in four days as the equipment was moved west to Ohio (McLaughlin et al., 2014). This suggests that efforts to meet the housing needs of temporary workers should focus on preparing temporary facilities that would serve the housing needs of gas workers in the short run. Alternatively, new longer-term structures could be adapted for use for tourism or safe housing for elders (Lycoming County Planning Department, 2012). New structures built to accommodate industry workers could also be modified into domestic violence shelters, youth services facilities, mental health and developmental disability rehabilitation facilities, office space for county social service workers, and rent-to-own or cooperative ownership housing for low-income residents and/or those with inadequate credit to attain conventional mortgages. Report Authors Shannon M. Monnat, PhD Assistant Professor of Rural Sociology, Demography, and Sociology Raeven Faye Chandler PhD Student in Rural Sociology and Demography Danielle Ely PhD Student in Rural Sociology and Demography Other Research Team Members Kathryn Brasier, PhD Associate Professor of Rural Sociology Leland Glenna, PhD Associate Professor of Rural Sociology Arielle Hesse PhD Student in Geography and Women s Studies Timothy Kelsey, PhD Professor of Agricultural Economics Joshua Perchinski MS Student in Rural Sociology Kai Schafft, PhD Associate Professor of Educational Leadership & Director of Center on Rural Education and Communities Mark Suchyta MS Student in Rural Sociology All report authors are located in the Department of Agricultural Economics, Sociology, and Education at Penn Pennsylvania State University. The Center for Rural Pennsylvania 54

55 Acknowledgements The authors gratefully acknowledge Diane McLaughlin, Jonathan Williamson, and William Blevins who read and provided helpful comments and suggestions on a previous version of this report. Funding This research was sponsored by a grant from the Center for Rural Pennsylvania, a legislative agency of the Pennsylvania General Assembly. The Center for Rural Pennsylvania is a bipartisan, bicameral legislative agency that serves as a resource for rural policy within the Pennsylvania General Assembly. It was created in 1987 under Act 16, the Rural Revitalization Act, to promote and sustain the vitality of Pennsylvania s rural and small communities. Information contained in this report does not necessarily reflect the views of individual board members of the Center for Rural Pennsylvania. For more information, contact the Center for Rural Pennsylvania, 625 Forster St., Room 902. Harrisburg, PA 17120, telephone (717) ; info@rural.palegislature.us, References Boxall, Peter C., Wing H. Chan, and Melville L. McMillan The Impact of Oil and Natural Gas Facilities on Rural Property Values: A Spatial Hedonic Analysis. Resource and Energy Economics 27(3): Brasier, Kathryn, Matthew R. Filteau, Diane K. McLaughlin, Jeffrey Jacquet, Richard C. Stedman, Timothy W. Kelsey, and Stephan J. Goetz Residents Perceptions of Community and Environmental Impacts from Development of Natural Gas in the Marcellus Shale: A Comparison of Pennsylvania and New York Cases. Journal of Rural Social Sciences 26(1): Brown, David L Rural Population Change in Social Context. Page in Rural America in a Globalizing World: Problems and Prospects for the 2010s, edited by Conner Bailey, Leif Jensen, and Elizabeth Ransom. Morgantown, WV: West VirGinia University Press. Farren, Michael, Amandan Weinstein, Mark Partridge and Michael Betz Too Many Heads and Not Enough Beds: Will Shale Development Cause a Housing Shortage? The Swank Program in Rural- Urban Policy, The Ohio State University, Columbus, Ohio. Retrieved from Gilmore, John S. and Mary K. Duff Boom Town Growth Management: A Case Study of Rock Springs-Green River, Wyoming. Boulder, CO: Westview Press. Harrell, Jr., Frank E Regression Modeling Strategies: With Applications to Linear Models, Logistic Regression, and Survival Analysis. Springer Series in Statistics. Hertz, Thomas, Lorin Kusmin, Alexander Marre, and Timothy Parker Rural Employment in Recession and Recovery. Amber Waves. Retrieved from october/rural-employment-in-recession-and-recovery.aspx#.vdi_2unhtuq. The Center for Rural Pennsylvania 55

56 McLaughlin, Diane, Anne Delesio-Parson, and Danielle Rhubart Housing and Marcellus Shale Development: The Marcellus Impacts Report #5. Harrisburg, PA: The Center for Rural Pennsylvania. Retrieved at: Muehlenbachs, Lucija, Elisheba Spiller, and Christopher Timmins The Housing Market Impacts of Shale Gas Development. NBER Working Paper No The National Bureau of Economic Research. Available at: Oldham, Jennifer The Real Estate Crisis in North Dakota s Man Camps. Bloomberg Business. September 25, Thiede, Brian C. and Shannon M. Monnat The Great Recession and America s Geography of Unemployment. Presented at the Annual Meeting of the American Sociological Association. Williamson, Jonathan and Bonita Kolb Marcellus Natural Gas Development s Effect on Housing in Pennsylvania. Center for the Study of Community and the Economy (CSCE). Retrieved from PHFA%20Marcellus_report.pdf. The Center for Rural Pennsylvania 56

57 Appendix A. List of Tables in Report Figure 1. Cumulative Number of Unconventional Gas Wells Drilled, 2005 June 1, 2015 Figure 2. Cumulative number of wells drilled in four study counties, 2005 June 1, 2015 Table 1. Differences between 3 Levels of Well Development in Means of County Characteristics included as Control Variables in Regression Models Table 2. Housing Supply: Number of Housing Units and Percentage Change in Northern Tier Study and Adjacent Counties, Table 3. Housing Supply: Number of Housing Units and Percentage Change in Southwest Region Study and Adjacent Counties, Table 4. Housing Age: Median Year Housing Units Built in Northern Tier Study and Adjacent Counties Table 5. Median Year Housing Units Built in Southwest Region Study and Adjacent Counties Table 6. Number and Percentage of all Housing Units that are Owner-Occupied and Percentage Change in Owner-Occupied Housing Units in Northern Tier Study and Adjacent Counties, Table 7. Number and Percentage of all Housing Units that are Renter Occupied and Percentage Change in Renter Occupied Housing Units in Northern Tier Study and Adjacent Counties, Table 8. Number and Percentage of all Housing Units that are Vacant and Percentage Change in Vacant Housing Units in Northern Tier Study and Adjacent Counties, Table 9. Number and Percentage of all Vacant Housing Units for Rent and Percentage Change in Vacant Housing Units for Rent in Northern Tier Study and Adjacent Counties, Table 10. Number and Percentage of all Housing Units that are Owner-Occupied and Percentage Change in Owner-Occupied Housing Units in Southwest Region Study and Adjacent Counties, Table 11. Number and Percentage of all Housing Units that are Renter Occupied and Percentage Change in Renter Occupied Housing Units in Southwest Region Study and Adjacent Counties, Table 12. Number and Percentage of all Housing Units that are Vacant and Percentage Change in Vacan Housing Units Southwest Region Study and Adjacent Counties, Table 13. Number and Percentage of all Vacant Housing Units for Rent and Percentage Change in Vacant Housing Units for Rent in Southwest Region Study and Adjacent Counties, Table 14. Means for Housing Supply and Availability Outcomes by Level of Well Development Table 15. Percent of Vacant Housing Units Available for Rent in High Drilling Counties, 2000 and 2009/2013 The Center for Rural Pennsylvania 57

58 Table 16. Median House Value (2013 dollars) for all Owner-Occupied Housing Units and Percentage Change in Northern Tier Study and Adjacent Counties, Table 17. Median House Value (2013 dollars) for all Owner-Occupied Housing Units and Percentage Change in Southwest Region Study and Adjacent Counties, Table 18. Median Gross Rent (2013 dollars) and Percentage Change in Northern Tier Study and Adjacent Counties, Table 19. Median Gross Rent (2013 dollars) and Percentage Change in Southwest Region Study and Adjacent Counties, Table 20. Median Gross Rent as a Percentage of Household Income and Percentage Change in Northern Tier Study and Adjacent Counties, Table 21. Percentage of Renters Spending more than 30% of Household Income on Rent and Percentage Change in Northern Tier Study and Adjacent Counties, Table 22. Median Gross Rent as a Percentage of Household Income and Percentage Change in Southwest Region Study and Adjacent Counties, Table 23. Percentage of Renters Spending more than 30% of Household Income on Rent and Percentage Change in Southwest Region Study and Adjacent Counties, Table 24. Means for Housing Affordability Outcomes by Level of Well Development Table 25. Median Household Income for Owner-Occupied Housing Units and Percentage Change in Northern Tier Study and Adjacent Counties, Table 26. Median Household Income for Renters and Percentage Change in Northern Tier Study and Adjacent Counties, Table 27. Gap in Median Household Income between Owners and Renters and Percentage Change in Northern Tier Study and Adjacent Counties, Table 28. Median Household Income for Owner-Occupied Housing Units and Percentage Change in Southwest Region Study and Adjacent Counties, Table 29. Median Household Income for Renters and Percentage Change in Southwest Region Study and Adjacent Counties, Table 30. Gap in Median Household Income between Owners and Renters and Percentage Change in Southwest Region Study and Adjacent Counties, Table 31. Gini Coefficient of Household Income Inequality and Percentage Change in Study and Northern Tier Study and Adjacent Counties, The Center for Rural Pennsylvania 58

59 Table 32. Gini Coefficient of Household Income Inequality and Percentage Change in Study and Southwest Region Study and Adjacent Counties, Table 33. Means for Inequality Outcomes by Level of Well Development The Center for Rural Pennsylvania 59

60 Appendix B. Tables with Coefficients for Well Categories from Unadjusted and Adjusted Regression Models Notes for all tables: *p<0.05; **p<0.01; ***p<0.001; two-tailed tests; standard errors in parentheses. Table B1. Results from Unadjusted Linear Regression Models Predicting Associations between Level of Well Development and Housing Supply Availability Outcomes Outcome High Drilling vs. None β (SE) Some Drilling vs. None R 2 β (SE) Percent change in total number of housing units, (1.994)** (1.367)*** Percent housing units owner-occupied, (3.526) (2.418)* Percent housing units owner-occupied, 2009/ (3.425) (2.349)** Percent change in owner-occupied housing stock, / (2.409)** (1.652)*** Percent change in percentage of housing units owneroccupied, / (1.232) (0.845)* Percent housing units renter-occupied, (2.226) (1.527)* Percent housing units renter-occupied, 2009/ (2.339) (1.604)* Percent change in renter-occupied housing stock, / (2.763)* (1.895)*** Percent change in percent housing units renter-occupied, / (2.184) (1.498) Percent housing units vacant, (4.977) (3.413)** Percent housing units vacant, 2009/ (4.846) (3.323)** Percent change in vacant housing stock, / (8.147) (5.587)* Percent change in percent housing units vacant, / (7.487) (5.135) Percent vacant housing units for rent, (4.281) (2.936)** Percent vacant housing units for rent, 2009/ (3.220)* (2.208)*** Percent change in vacant housing stock for rent, / (15.884) (10.893) Percent change in percent vacant housing units for rent, / (11.569) (7.933) The Center for Rural Pennsylvania 60

61 Table B2. Results from Linear Regression Models Predicting Associations between Level of Well Development and Housing Supply and Availability Outcomes, Controlling for Pre-Drilling County Characteristics Outcomes High Drilling vs. None Some Drilling vs. None R 2 β (SE) β (SE) Percent change in total number of housing units, (1.967)** (1.565)*** Percent housing units owner-occupied, (3.923) (3.121) Percent housing units owner-occupied, 2009/ (3.700) (2.944) Percent change in owner-occupied housing stock, / (2.242)** (1.784)*** Percent change in percentage of housing units owneroccupied, / (1.242) (0.988) Percent housing units renter-occupied, (1.844) (1.467)* Percent housing units renter-occupied, 2009/ (1.922) (1.529)* Percent change in renter-occupied housing stock, / (3.019) (2.402)* Percent change in percent housing units renter-occupied, / (2.525) (2.009) Percent housing units vacant, (5.385) (4.285) Percent housing units vacant, 2009/ (5.229) (4.161) Percent change in vacant housing stock, / (7.963) (6.336) Percent change in percent housing units vacant, / (7.786) (6.195) Percent vacant housing units for rent, (3.695) (2.940)* Percent vacant housing units for rent, 2009/ (2.786) (2.217) Percent change in vacant housing stock for rent, / (14.430) (14.664) Percent change in percent vacant housing units for rent, / (13.729) (10.924) The multivariate model controls for county population density, percent poverty, unemployment rate, percent of civilian labor force aged 16+ employed in construction, extraction, and maintenance, and percent of civilian labor force aged 16+ employed in production, transportation, and material moving occupations (all in year 2000). 2. Italicized values indicate that the difference was statistically significant in the unadjusted model (the model without control variables) but not significant in the multivariate model (the model with control variables). The Center for Rural Pennsylvania 61

62 Table B3. Results from Linear Regression Models Predicting Associations between Level of Well Development and Housing Supply and Availability Outcomes, Controlling for Change in County Characteristics, /2013 Outcomes High Drilling vs. None β (SE) Some Drilling vs. None R 2 β (SE) Percent change in total number of housing units, (2.356) (1.874) Percent housing units owner-occupied, 2009/ (3.814) (3.034) Percent change in owner-occupied housing stock, / (2.437) (1.938) Percent change in percent housing units owner-occupied, / (1.401) (1.115) Percent housing units renter-occupied, 2009/ (2.789) (2.218)* Percent change in renter-occupied housing stock, / (3.323) (2.643) Percent change in percent housing units renter-occupied, / (2.907) (2.312) Percent housing units vacant, 2009/ (5.199) (4.135)* Percent change in vacant housing stock, / (10.597) (8.429) Percent change in percent housing units vacant, / (9.205) (7.322) Percent vacant housing units for rent, 2009/ (3.849) (3.062)** Percent change in vacant housing stock for rent, / (21.929) (17.43) Percent change in percent vacant housing units for rent, / (15.181) (12.076) The multivariate model controls for percentage change from 2000 to 2009/2013 in county population density, percent poverty, unemployment rate, percent of civilian labor force aged 16+ employed in construction, extraction, and maintenance, and percent of civilian labor force aged 16+ employed in production, transportation, and material moving occupations. 2. Italicized values indicate that the difference was statistically significant in the unadjusted model (the model without control variables) but not significant in the multivariate model (the model with control variables). 3. Bolded values indicate that the difference remained significant in both models (the model without control variables and the model with control variables). The Center for Rural Pennsylvania 62

63 Table B4. Results from Unadjusted Linear Regression Models Predicting Associations between Level of Well Development and Housing Affordability Outcomes Outcomes High Wells vs. No Wells Some Wells vs. No Wells β (SE) β (SE) R 2 Median value of all owner-occupied housing units, 2000 ($1,000s) (11.021)** (7.558)*** Median value of all owner-occupied housing units, 2009/2013 ($1,000s) (16.639)** (11.410)*** Percent change in median value of owneroccupied housing units, / (4.692) (3.218)*** Median gross rent, 2000 (adjusted to 2013 $) (43.15)*** (29.59)*** Median gross rent, 2009/2013 (adjusted to 2013 $) (47.93)*** (32.87)*** Percent change in median gross rent, / (1.703) (1.168)*** Percentage of renters' income spent on rent, (.797) (.547) Percentage of renters' income spent on rent, 2009/ (1.000) (.686) Percent change in percentage of renters' income spent on rent, / (2.511)* (1.722)* Percentage of renters spending 30% or more of household income on rent, (1.923) (1.319) Percentage of renters spending 30% or more of household income on rent, 2009/ (2.201) (1.509) Percent change in percentage of renters spending 30% or more of household income on rent, / (5.214)* (3.576)* The Center for Rural Pennsylvania 63

64 Table B5. Results from Linear Regression Models Predicting Associations between Level of Well Development and Housing Affordability Outcomes, Controlling for Pre-Drilling County Characteristics Outcomes High Wells vs. No Wells Some Wells vs. No Wells β (SE) β (SE) R 2 Median value of all owner-occupied housing units, 2000 ($1,000s) (7.579) (6.031)* Median value of all owner-occupied housing units, ($1,000s) (12.438) (9.897)* Percent change in median value of owneroccupied housing units, / (3.827) (3.045)** Median gross rent, 2000 (adjusted to 2013 $) (29.594)* (23.548)* Median gross rent, 2009/2013 (adjusted to 2013 $) (34.102)** (27.134)*** Percent change in median gross rent, / (1.778) (1.415)*** Percentage of renters' income spent on rent, (0.677) (0.538) Percentage of renters' income spent on rent, 2009/ (0.966) (0.769) Percent change in percentage of renters' income spent on rent, / (2.797) (2.225) Percentage of renters spending 30% or more of household income on rent, (1.428) (1.137) Percentage of renters spending 30% or more of household income on rent, 2009/ (2.167) (1.724) Percent change in percentage of renters spending 30% or more of household income on rent, / (5.253) (4.180) The multivariate models control for county population density, percent poverty, unemployment rate, percent of civilian labor force aged 16+ employed in construction, extraction, and maintenance, and percent of civilian labor force aged 16+ employed in production, transportation, and material moving occupations (all in year 2000). 2. Italicized values indicate that the difference was statistically significant in the unadjusted model (the model without control variables) but not significant in the multivariate model (the model with control variables. The Center for Rural Pennsylvania 64

65 Table B6. Results from Linear Regression Models Predicting Associations between Level of Well Development and Housing Affordability Outcomes, Adjusted for Change in County Characteristics, /2013 Outcomes High Wells vs. No Wells Some Wells vs. No Wells β (SE) β (SE) R 2 Median value of all owner-occupied housing units, ($1,000s) (20.915) (16.637)* Percent change in median value of owneroccupied housing units, / (5.706)** (4.539)*** Median gross rent, 2009/2013 (adjusted to 2013 $) (59.301) (47.171)* Percent change in median gross rent, / (2.230) (1.774)** Percentage of renters' income spent on rent, 2009/ (1.320) (1.050) Percent change in percentage of renters' income spent on rent, / (3.357) (2.670) Percentage of renters spending 30% or more of household income on rent, 2009/ (2.881) (2.292) Percent change in percentage of renters spending 30% or more of household income on rent, / (6.256) (4.977) The multivariate model controls for percentage change from 2000 to 2009/2013 in county population density, percent poverty, unemployment rate, percent of civilian labor force aged 16+ employed in construction, extraction, and maintenance, and percent of civilian labor force aged 16+ employed in production, transportation, and material moving occupations. 2. Italicized values indicate that the difference was statistically significant in the unadjusted model (the model without control variables) but not significant in the multivariate model (the model with control variables. 3. Bolded values indicate that the difference remained significant in both models (the model without control variables and the model with control variables). The Center for Rural Pennsylvania 65

66 Table B7. Results from Unadjusted Linear Regression Models Predicting Associations between Level of Well Development and Inequality Outcomes Outcomes High Wells vs. No Wells Some Wells vs. No Wells R 2 β (SE) β (SE) Gini coefficient of household income inequality, 2005/ (.009) (.006) Gini coefficient of household income inequality, 2009/ (.010) (.006) Percent change in Gini coefficient, 2005/ / (1.170) (0.803) Median household income among homeowners, (3.922)** (2.689)*** Median household income among homeowners, 2009/ (3.888)* (2.666)*** Percent change in median household income among homeowners, / (1.607)** (1.101) Median household income among renters, (2.017)*** (1.383)*** Median household income among renters, 2009/ (1.579)*** (1.083)*** Percent change in household income among renters, / (2.308)** (1.583) Gap in median household income between owners and renters, (2.398) (1.644)** Gap in median household income between owners and renters, 2009/ (2.697) (1.849)*** Percent change in household income gap between owners and renters, / (3.983) (2.731) The Center for Rural Pennsylvania 66

67 Table B8. Results from Linear Regression Models Predicting Associations between Level of Well Development and Inequality Outcomes, Controlling for Pre-Drilling County Characteristics Outcomes High Wells vs. No Wells Some Wells vs. No Wells R 2 β (SE) β (SE) Gini coefficient of household income inequality, 2005/ (0.006) (0.006) Gini coefficient of household income inequality, 2009/ (0.006) (0.005) Percent change in Gini coefficient, 2005/ / (1.389) (1.105) Median household income among homeowners, (2.264) (1.801)* Median household income among homeowners, 2009/ (2.169) (1.726)* Percent change in median household income among homeowners, / (1.531) (1.219) Median household income among renters, (1.362)* (1.084)** Median household income among renters, 2009/ (1.117) (0.889)** Percent change in household income among renters, / (2.533)* (2.015) Gap in median household income between owners and renters, (1.597) (1.270) Gap in median household income between owners and renters, 2009/ (1.596) (1.270) Percent change in median household income gap between owners and renters, / (3.991) (3.175) The multivariate models control for county population density, percent poverty, unemployment rate, percent of civilian labor force aged 16+ employed in construction, extraction, and maintenance, and percent of civilian labor force aged 16+ employed in production, transportation, and material moving occupations (all in year 2000). 2. Italicized values indicate that the difference was statistically significant in the unadjusted model (the model without control variables) but not significant in the multivariate model (the model with control variables). The Center for Rural Pennsylvania 67

68 Table B9. Results from Linear Regression Models Predicting Associations between Level of Well Development and Inequality Outcomes, Controlling for Change in County Characteristics, /2013 Outcomes High Wells vs. No Wells Some Wells vs. No Wells R 2 β (SE) β (SE) Gini coefficient of household income inequality, 2009/ (0.012) (0.009) Percent change in Gini coefficient, 2005/ / (1.597)* (1.270)* Median household income among homeowners, 2009/ (4.922) (3.915) Percent change in median household income among homeowners, / (1.968) (1.565) Median household income among renters, 2009/ (2.068) (1.645)** Percent change in household income among renters, / (3.048) (2.425) Gap in median household income between owners and renters, 2009/ (3.396) (2.701) Percent change in household income gap between owners and renters, / (5.371) (4.272) The multivariate model controls for percentage change from 2000 to 2009/2013 in county population density, percent poverty, unemployment rate, percent of civilian labor force aged 16+ employed in construction, extraction, and maintenance, and percent of civilian labor force aged 16+ employed in production, transportation, and material moving occupations. 2. Italicized values indicate that the difference was statistically significant in the unadjusted model (the model without control variables) but not significant in the multivariate model (the model with control variables. 3. Bolded values indicate that the difference remained significant in both models (the model without control variables and the model with control variables). The Center for Rural Pennsylvania 68

69 Appendix C. Tables with all Coefficients from Adjusted Regression Models Notes for all tables: *p<0.05; **p<0.01; ***p<0.001; two-tailed tests Table C1. Results from Linear Regression Model Predicting Percent Change in Total Number of Housing Units, , adjusted for pre-marcellus Characteristics Intercept (3.872) *** High Drilling (1.967) ** Some Drilling (1.565) *** Population Density, 2000 (100s) (.055) * Percent Poverty, (.255) Unemployment Rate, (1.079) * Percent employed workforce in construction/extraction/maintenance, (.283) Percent employed workforce in production/transport/material moving, (.106) *** Table C2. Results from Linear Regression Model Predicting Percent Change in Total Number of Housing Units, , adjusted for Changes in County Characteristics β SE P Intercept (3.239) * High Drilling (2.356) Some Drilling (1.874) Percent change in population density, / (.076) *** Percent change in percent poverty, / (.053) Percent change in unemployment rate, / (.020) Percent change in percent of employed workforce in construction/extraction/maintenance, / (.060) Percent change in percent of employed workforce in production/transportation/material moving, / (.100) Table C3. Results from Linear Regression Model Predicting Percent of Housing Units that were Owner-Occupied, adjusted for County Characteristics, 2000 Intercept (7.721) *** High Drilling (3.923) * Some Drilling (3.121) Population Density, 2000 (100s) (.110) Percent Poverty, (.508) Unemployment Rate, (2.153) Percent of employed workforce in construction/extraction/maintenance, (.564) Percent of employed workforce in product/transport/material moving, (.211) The Center for Rural Pennsylvania 69

70 Table C4. Results from Linear Regression Model Predicting Percent of Housing Units that were Owner-Occupied, 2009/2013, adjusted for Pre-Marcellus County Characteristics Intercept (7.282) *** High Drilling (3.700) Some Drilling (2.944) Population Density, 2000 (100s) (.104) Percent Poverty, (.479) Unemployment Rate, (2.030) Percent of employed workforce in construction/extraction/maintenance, (.532) Percent of employed workforce in product/transport/material moving, (.199) Table C5. Results from Linear Regression Model Predicting Percent Change in Number of Housing Units that were Owner-Occupied, /2013, adjusted for Pre-Marcellus County Characteristics Intercept (4.413) *** High Drilling (2.242) ** Some Drilling (1.784) *** Population Density, 2000 (100s) (.063) ** Percent Poverty, (.290) Unemployment Rate, (1.230) * Percent of employed workforce in construction/extraction/maintenance, (.322) Percent of employed workforce in production/transportation/material moving, (.121) *** Table C6. Results from Linear Regression Model Predicting Change in Percent of Housing Units that were Owner- Occupied, /2013, adjusted for Pre-Marcellus County Characteristics Intercept (2.444) High Drilling (1.242) Some Drilling (.988) Population Density, 2000 (100s) (.035) * Percent Poverty, (.161) Unemployment Rate, (.682) Percent of employed workforce in construction/extraction/maintenance, (.179) Percent of employed workforce in product/transport/material moving, (.067) * The Center for Rural Pennsylvania 70

71 Table C7. Results from Linear Regression Model Predicting Percent Change in Number of Housing Units that were Owner-Occupied, /2013, adjusted for Change in County Characteristics Intercept (3.350) High Drilling (2.437) Some Drilling (1.938) Percent change in population density, / (.078) *** Percent change in percent poverty, / (.055) Percent change in unemployment rate, / (.020) ** Percent change in percent of employed workforce in construction/extraction/maintenance, / (.062) Percent change in percent of employed workforce in production/transportation/material moving, / (.103) Table C8. Results from Linear Regression Model Predicting Change in Percent of Housing Units that were Owner- Occupied, /2013, adjusted for Change in County Characteristics Intercept (1.926) *** High Drilling (1.401) Some Drilling (1.115) Percent change in population density, / (.045) Percent change in percent poverty, / (.031) Percent change in unemployment rate, / (.012) Percent change in percent of employed workforce in construction/extraction/maintenance, / (.036) Percent change in percent of employed workforce in production/transportation/material moving, / (.059) Table C9. Results from Linear Regression Model Predicting Percent of Housing Units that were Renter-Occupied, adjusted for County Characteristics, 2000 Intercept (3.628) *** High Drilling (1.844) Some Drilling (1.467) * Population Density, 2000 (100s) (.052) Percent Poverty, (.239) ** Unemployment Rate, (1.012) Percent of employed workforce in construction/extraction/maintenance, (.265) *** Percent of employed workforce in product/transport/material moving, (.099) ** The Center for Rural Pennsylvania 71

72 Table C8. Results from Linear Regression Model Predicting Percent of Housing Units that were Renter-Occupied, 2009/2013, adjusted for Pre-Marcellus County Characteristics Intercept (3.782) *** High Drilling (1.922) Some Drilling (1.529) * Population Density, 2000 (100s) (.054) Percent Poverty, (.249) ** Unemployment Rate, (1.054) Percent of employed workforce in construction/extraction/maintenance, (.276) *** Percent of employed workforce in product/transport/material moving, (.103) ** Table C10. Results from Linear Regression Model Predicting Percent of Housing Units that were Renter-Occupied, 2009/2013, adjusted for Changes in County Characteristics Intercept (3.997) High Drilling (2.907) Some Drilling (2.312) Percent change in population density, / (.093) Percent change in percent poverty, / (.065) Percent change in unemployment rate, / (.024) Percent change in percent of employed workforce in construction/extraction/maintenance, / (.074) Percent change in percent of employed workforce in production/transportation/material moving, / (.123) Table C11. Results from Linear Regression Model Predicting Percent Change in Number of Housing Units that were Renter-Occupied, /2013, adjusted for Pre-Marcellus County Characteristics Intercept (5.941) *** High Drilling (3.019) Some Drilling (2.402) * Population Density, 2000 (100s) (.085) Percent Poverty, (.391) Unemployment Rate, (1.656) ** Percent of employed workforce in construction/extraction/maintenance, (.434) Percent of employed workforce in production/transportation/material moving, (.162) The Center for Rural Pennsylvania 72

73 Table C12. Results from Linear Regression Model Predicting Change in Percent of Housing Units that were Renter- Occupied, /2013, adjusted for Pre-Marcellus County Characteristics Intercept (4.969) High Drilling (2.525) Some Drilling (2.009) Population Density, 2000 (100s) (.071) Percent Poverty, (.327) Unemployment Rate, (1.386) Percent of employed workforce in construction/extraction/maintenance, (.363) Percent of employed workforce in product/transport/material moving, (.136) C13. Results from Linear Regression Model Predicting Percent Change in Number of Housing Units that were Renter-Occupied, /2013, adjusted for Change in County Characteristics Intercept (4.568) High Drilling (3.323) Some Drilling (2.643) Percent change in population density, / (.107) ** Percent change in percent poverty, / (.074) Percent change in unemployment rate, / (.028) * Percent change in percent of employed workforce in construction/extraction/maintenance, / (.085) Percent change in percent of employed workforce in production/transportation/material moving, / (.141) Table C14. Results from Linear Regression Model Predicting Change in Percent of Housing Units that were Renter- Occupied, /2013, adjusted for Change in County Characteristics Intercept (3.997) High Drilling (2.907) Some Drilling (2.312) Percent change in population density, / (.093) Percent change in percent poverty, / (.065) Percent change in unemployment rate, / (.024) Percent change in percent of employed workforce in construction/extraction/maintenance, / (.074) Percent change in percent of employed workforce in production/transportation/material moving, / (.123) The Center for Rural Pennsylvania 73

74 Table C15. Results from Linear Regression Model Predicting Percent of Housing Units that were Vacant, adjusted for County Characteristics, 2000 Intercept (10.60) * High Drilling (5.385) Some Drilling (4.285) Population Density, 2000 (100s) (.152) Percent Poverty, (.698) Unemployment Rate, (2.955) Percent of employed workforce in construction/extraction/maintenance, (.775) * Percent of employed workforce in product/transport/material moving, (.290) Table C16. Results from Linear Regression Model Predicting Percent of Housing Units that were Vacant, 2009/2013, adjusted for Pre-Marcellus County Characteristics Intercept (10.29) * High Drilling (5.229) Some Drilling (4.161) Population Density, 2000 (100s) (.147) Percent Poverty, (.677) Unemployment Rate, (2.870) Percent of employed workforce in construction/extraction/maintenance, (.752) * Percent of employed workforce in product/transport/material moving, (.281) * Table C17. Results from Linear Regression Model Predicting Percent of Housing Units that were Vacant, 2009/2013, adjusted for Changes in County Characteristics Intercept (7.147) * High Drilling (5.199) Some Drilling (4.135) * Percent change in population density, / (.167) *** Percent change in percent poverty, / (.116) Percent change in unemployment rate, / (.043) Percent change in percent of employed workforce in construction/extraction/maintenance, / (.132) ** Percent change in percent of employed workforce in production/transportation/material moving, / (.220) The Center for Rural Pennsylvania 74

75 Table C18. Results from Linear Regression Model Predicting Change in Number of Housing Units that were Vacant, /2013, adjusted for Pre-Marcellus County Characteristics Intercept (15.673) *** High Drilling (7.963) Some Drilling (6.336) Population Density, 2000 (100s) (.224) Percent Poverty, (1.032) Unemployment Rate, (4.370) Percent of employed workforce in construction/extraction/maintenance, (1.145) * Percent of employed workforce in production/transportation/material moving, (.429) *** Table C19. Results from Linear Regression Model Predicting Change in Percent of Housing Units that were Vacant, /2013, adjusted for Pre-Marcellus County Characteristics Intercept (15.32) *** High Drilling (7.786) Some Drilling (6.195) Population Density, 2000 (100s) (.219) Percent Poverty, (1.009) Unemployment Rate, (4.273) Percent of employed workforce in construction/extraction/maintenance, (1.120) Percent of employed workforce in product/transport/material moving, (.419) ** Table C20. Results from Linear Regression Model Predicting Percent Change in Number of Housing Units that were Vacant, /2013, adjusted for Change in County Characteristics Intercept (14.569) ** High Drilling (10.597) Some Drilling (8.429) Percent change in population density, / (.341) Percent change in percent poverty, / (.237) Percent change in unemployment rate, / (.088) Percent change in percent of employed workforce in construction/extraction/maintenance, / (.270) * Percent change in percent of employed workforce in production/transportation/material moving, / (.449) The Center for Rural Pennsylvania 75

76 Table C21. Results from Linear Regression Model Predicting Change in Percent of Housing Units that were Vacant, /2013, adjusted for Change in County Characteristics Intercept (12.66) ** High Drilling (9.205) Some Drilling (7.322) Percent change in population density, / (.296) ** Percent change in percent poverty, / (.206) Percent change in unemployment rate, / (.076) Percent change in percent of employed workforce in construction/extraction/maintenance, / (.234) ** Percent change in percent of employed workforce in production/transportation/material moving, / (.390) Notes: *p<0.05; **p<0.01; ***p<0.001; two-tailed tests Table C22. Results from Linear Regression Model Predicting Percent of Vacant Housing Units for Rent, adjusted for County Characteristics, 2000 Intercept (7.27) *** High Drilling (3.695) Some Drilling (2.940) * Population Density, 2000 (100s) (.104) Percent Poverty, (.479) Unemployment Rate, (2.027) Percent of employed workforce in construction/extraction/maintenance, (.531) *** Table C23. Results from Linear Regression Model Predicting Percent of Vacant Housing Units for Rent, 2009/2013, adjusted for Pre-Marcellus County Characteristics Intercept (5.484) *** High Drilling (2.786) Some Drilling (2.217) Population Density, 2000 (100s) (.078) Percent Poverty, (.361) Unemployment Rate, (1.529) Percent of employed workforce in construction/extraction/maintenance, (.401) *** Percent of employed workforce in product/transport/material moving, (.150) *** The Center for Rural Pennsylvania 76

77 Table C24. Results from Linear Regression Model Predicting Percent of Vacant Housing Units for Rent, 2009/2013, adjusted Changes in County Characteristics Intercept (5.292) ** High Drilling (3.849) Some Drilling (3.062) ** Percent change in population density, / (.124) Percent change in percent poverty, / (.086) * Percent change in unemployment rate, / (.032) Percent change in percent of employed workforce in construction/extraction/maintenance, / (.098) ** Percent change in percent of employed workforce in production/transportation/material moving, / (.163) Table C25. Results from Linear Regression Model Predicting Percent Change in Number of Vacant Housing Units for Rent, /2013, adjusted for Pre-Marcellus County Characteristics Intercept (36.273) High Drilling (18.430) Some Drilling (14.664) Population Density, 2000 (100s) (.519) Percent Poverty, (2.387) Unemployment Rate, (10.113) Percent of employed workforce in construction/extraction/maintenance, (2.651) Percent of employed workforce in production/transportation/material moving, (.992) Table C26. Results from Linear Regression Model Predicting Change in Percent of Vacant Housing Units for Rent, /2013, adjusted for Pre-Marcellus County Characteristics Intercept (27.02) High Drilling (13.73) Some Drilling (10.92) Population Density, 2000 (100s) (.387) Percent Poverty, (1.778) Unemployment Rate, (7.534) Percent of employed workforce in construction/extraction/maintenance, (1.975) Percent of employed workforce in product/transport/material moving, (.739) The Center for Rural Pennsylvania 77

78 Table C27. Results from Linear Regression Model Predicting Percent Change in Number of Vacant Housing Units for Rent, /2013, adjusted for Change in County Characteristics Intercept (30.147) High Drilling (21.929) Some Drilling (17.443) Percent change in population density, / (.705) Percent change in percent poverty, / (.491) Percent change in unemployment rate, / (.182) Percent change in percent of employed workforce in construction/extraction/maintenance, / (.558) Percent change in percent of employed workforce in production/transportation/material moving, / (.929) Table C28. Results from Linear Regression Model Predicting Change in Percent of Vacant Housing Units for Rent, /2013, adjusted for Change in County Characteristics Intercept (20.87) High Drilling (15.18) Some Drilling (12.08) Percent change in population density, / (.488) ** Percent change in percent poverty, / (.340) Percent change in unemployment rate, / (.126) Percent change in percent of employed workforce in construction/extraction/maintenance, / (.386) Percent change in percent of employed workforce in production/transportation/material moving, / (.643) Table C22. Results from Linear Regression Model Predicting Median Value of Owner-Occupied Housing Units, adjusted for County Characteristics, 2000 Intercept (14.92) *** High Drilling (7.579) Some Drilling (6.031) * Population Density, 2000 (100s) (.213) Percent Poverty, (.982) *** Unemployment Rate, (4.159) Percent of employed workforce in construction/extraction/maintenance, (1.090) Percent of employed workforce in product/transport/material moving, (.408) *** The Center for Rural Pennsylvania 78

79 Table C29. Results from Linear Regression Model Predicting Median Value of Owner-Occupied Housing Units, 2009/2013, adjusted for Pre-Marcellus County Characteristics Intercept (24.48) *** High Drilling (12.44) Some Drilling (9.897) ** Population Density, 2000 (100s) (.350) Percent Poverty, (1.611) ** Unemployment Rate, (6.825) Percent of employed workforce in construction/extraction/maintenance, (1.789) Percent of employed workforce in product/transport/material moving, (.669) *** Table C30. Results from Linear Regression Model Predicting Median Value of Owner-Occupied Housing Units, 2009/2013, adjusted for Change in County Characteristics Intercept (28.75) *** High Drilling (20.92) Some Drilling (16.64) * Percent change in population density, / (.672) * Percent change in percent poverty, / (.468) Percent change in unemployment rate, / (.174) Percent change in percent of employed workforce in construction/extraction/maintenance, / (.532) Percent change in percent of employed workforce in production/transportation/material moving, / (.886) * Table C31. Results from Linear Regression Model Predicting Percent Change in Median Value of Owner-Occupied Housing Units, /2013, adjusted for Pre-Marcellus County Characteristics Intercept (7.533) *** High Drilling (3.827) Some Drilling (3.045) ** Population Density, 2000 (100s) (.108) *** Percent Poverty, (.496) Unemployment Rate, (2.100) Percent of employed workforce in construction/extraction/maintenance, (.551) *** Percent of employed workforce in product/transport/material moving, (.206) *** The Center for Rural Pennsylvania 79

80 Table C32. Results from Linear Regression Model Predicting Percent Change in Median Value of Owner-Occupied Housing Units, /2013, adjusted for Change in County Characteristics Intercept (7.84) *** High Drilling (5.71) ** Some Drilling (4.54) *** Percent change in population density, / (.183) Percent change in percent poverty, / (.128) ** Percent change in unemployment rate, / (.047) Percent change in percent of employed workforce in construction/extraction/maintenance, / (.145) Percent change in percent of employed workforce in production/transportation/material moving, / (.242) * Table C33. Results from Linear Regression Model Predicting Median Gross Rent, adjusted for County Characteristics, 2000 Intercept (58.25) *** High Drilling (29.59) * Some Drilling (23.55) * Population Density, 2000 (100s) (.833) Percent Poverty, (3.833) *** Unemployment Rate, (16.24) Percent of employed workforce in construct/extraction/maintenance, (4.257) * Percent of employed workforce in product/transport/material moving, (1.593) *** Table C34. Results from Linear Regression Model Predicting Median Gross Rent, 2009/2013, adjusted for Pre- Marcellus County Characteristics Intercept (67.12) *** High Drilling (34.10) ** Some Drilling (27.13) *** Population Density, 2000 (100s) (.960) Percent Poverty, (4.417) ** Unemployment Rate, (18.71) Percent of employed workforce in construct/extraction/maintenance, (4.905) Percent of employed workforce in product/transport/material moving, (1.835) *** The Center for Rural Pennsylvania 80

81 Table C35. Results from Linear Regression Model Predicting Median Gross Rent, 2009/2013, adjusted for Changes in County Characteristics Intercept (81.53) *** High Drilling (59.30) Some Drilling (47.17) * Percent change in population density, / (1.906) * Percent change in percent poverty, / (1.328) Percent change in unemployment rate, / (.492) Percent change in percent of employed workforce in construction/extraction/maintenance, / (1.509) Percent change in percent of employed workforce in production/transportation/material moving, / (2.512) * Table C36. Results from Linear Regression Model Predicting Percent Change in Median Gross Rent, /2013, adjusted for Pre-Marcellus County Characteristics Intercept (3.500) * High Drilling (1.778) Some Drilling (1.415) *** Population Density, 2000 (100s) (.050) Percent Poverty, (.230) ** Unemployment Rate, (.976) Percent of employed workforce in construction/extraction/maintenance, (.256) * Percent of employed workforce in production/transportation/material moving, (.096) * Table C37. Results from Linear Regression Model Predicting Percent Change in Median Gross Rent, /2013, adjusted for Changes in County Characteristics Intercept (3.066) *** High Drilling (2.230) Some Drilling (1.774) ** Percent change in population density, / (.072) Percent change in percent poverty, / (.050) Percent change in unemployment rate, / (.019) Percent change in percent of employed workforce in construction/extraction/maintenance, / (.057) Percent change in percent of employed workforce in production/transportation/material moving, / (.094) The Center for Rural Pennsylvania 81

82 Table C38. Results from Linear Regression Model Predicting Percentage of Renters Household Income Spent on Rent, adjusted for County Characteristics, 2000 Intercept (1.332) *** High Drilling (.677) Some Drilling (.538) Population Density, 2000 (100s) (.019) Percent Poverty, (.088) *** Unemployment Rate, (.371) Percent of employed workforce in construct/extraction/maintenance, (.097) * Percent of employed workforce in product/transport/material moving, (.036) *** Table C39. Results from Linear Regression Model Predicting Percentage of Renters Household Income Spent on Rent, 2009/2013, adjusted for Pre-Marcellus County Characteristics Intercept (1.902) *** High Drilling (.966) Some Drilling (.769) Population Density, 2000 (100s) (.027) Percent Poverty, (.125) Unemployment Rate, (.530) Percent of employed workforce in construct/extraction/maintenance, (.139) Percent of employed workforce in product/transport/material moving, (.052) *** Table C40. Results from Linear Regression Model Predicting Percentage of Renters Household Income Spent on Rent, 2009/2013, adjusted for Changes in County Characteristics Intercept (1.814) *** High Drilling (1.320) Some Drilling (1.050) Percent change in population density, / (.042) * Percent change in percent poverty, / (.030) Percent change in unemployment rate, / (.011) Percent change in percent of employed workforce in construction/extraction/maintenance, / (.034) Percent change in percent of employed workforce in production/transportation/material moving, / (.056) The Center for Rural Pennsylvania 82

83 Table C41. Results from Linear Regression Model Predicting Percent Change in Percentage of Renters Household Income Spent on Rent, /2013, adjusted for Pre-Marcellus County Characteristics Intercept (5.505) * High Drilling (2.797) Some Drilling (2.225) Population Density, 2000 (100s) (.079) Percent Poverty, (.362) * Unemployment Rate, (1.535) Percent of employed workforce in construct/extraction/maintenance, (.402) Percent of employed workforce in product/transport/material moving, (.151) Table C42. Results from Linear Regression Model Predicting Percent Change in Percentage of Renters Household Income Spent on Rent, /2013, adjusted for Changes in County Characteristics Intercept (4.615) *** High Drilling (3.357) Some Drilling (2.670) Percent change in population density, / (.108) Percent change in percent poverty, / (.075) Percent change in unemployment rate, / (.028) Percent change in percent of employed workforce in construction/extraction/maintenance, / (.085) Percent change in percent of employed workforce in production/transportation/material moving, / (.142) Table C43. Results from Linear Regression Model Predicting Percentage of Renters Spending 30% or More of Household Income on Rent, adjusted for County Characteristics, 2000 Intercept (2.811) *** High Drilling (1.428) Some Drilling (1.137) Population Density, 2000 (100s) (.040) Percent Poverty, (.185) *** Unemployment Rate, (.784) Percent of employed workforce in construction/extraction/maintenance, (.205) ** Percent of employed workforce in production/transportation/material moving, (.077) *** The Center for Rural Pennsylvania 83

84 Table C44. Results from Linear Regression Model Predicting Percentage of Renters Spending 30% or More of Household Income on Rent, 2009/2013, adjusted for Pre-Marcellus County Characteristics Intercept (4.264) *** High Drilling (2.167) Some Drilling (1.724) Population Density, 2000 (100s) (.061) Percent Poverty, (.281) Unemployment Rate, (1.189) Percent of employed workforce in construct/extraction/maintenance, (.312) Percent of employed workforce in product/transport/material moving, (.117) *** Table C39. Results from Linear Regression Model Predicting Percentage of Renters Spending 30% or More of Household Income on Rent, 2009/2013, adjusted for Changes in County Characteristics Intercept (3.961) *** High Drilling (2.881) Some Drilling (2.292) Percent change in population density, / (.093) Percent change in percent poverty, / (.065) Percent change in unemployment rate, / (.024) Percent change in percent of employed workforce in construction/extraction/maintenance, / (.073) Percent change in percent of employed workforce in production/transportation/material moving, / (.122) Table C45. Results from Linear Regression Model Predicting Percent Change in Percentage of Renters Spending 30% or More of Household Income on Rent, /2013, adjusted for Pre-Marcellus County Characteristics Intercept (10.34) ** High Drilling (5.253) Some Drilling (4.180) Population Density, 2000 (100s) (.148) Percent Poverty, (.680) *** Unemployment Rate, (2.883) Percent of employed workforce in construct/extraction/maintenance, (.756) Percent of employed workforce in product/transport/material moving, (.283) ** The Center for Rural Pennsylvania 84

85 Table C46. Results from Linear Regression Model Predicting Percent Change in Percentage of Renters Spending 30% or More of Household Income on Rent, /2013, adjusted for Changes in County Characteristics Intercept (8.601) *** High Drilling (6.256) Some Drilling (4.977) Percent change in population density, / (.201) Percent change in percent poverty, / (.140) ** Percent change in unemployment rate, / (.052) Percent change in percent of employed workforce in construction/extraction/maintenance, / (.159) * Percent change in percent of employed workforce in production/transportation/material moving, / (.265) Table C47. Results from Linear Regression Model Predicting Gini Coefficient, 2005/2009, adjusted for Pre- Marcellus Characteristics Intercept (.013) *** High Drilling (.006) Some Drilling (.005) Population Density, 2000 (100s) (.000) Percent Poverty, (.001) * Unemployment Rate, (.004) Percent of employed workforce in construct/extraction/maintenance, (.001) *** Percent of employed workforce in product/transport/material moving, (.000) *** Table C48. Results from Linear Regression Model Predicting Gini Coefficient, 2009/2013, adjusted for Pre- Marcellus Characteristics Intercept (.012) *** High Drilling (.006) Some Drilling (.005) Population Density, 2000 (100s) (.000) Percent Poverty, (.001) * Unemployment Rate, (.003) Percent of employed workforce in construct/extraction/maintenance, (.001) *** Percent of employed workforce in product/transport/material moving, (.000) *** The Center for Rural Pennsylvania 85

86 Table C44. Results from Linear Regression Model Predicting Gini Coefficient, 2009/2013, adjusted for Changes in County Characteristics Intercept (.016) *** High Drilling (.012) Some Drilling (.009) Percent change in population density, / (.000) * Percent change in percent poverty, / (.000) Percent change in unemployment rate, / (.000) * Percent change in percent of employed workforce in construction/extraction/maintenance, / (.000) * Percent change in percent of employed workforce in production/transportation/material moving, / (.001) Table C49. Results from Linear Regression Model Predicting Percent Change in Gini Coefficient, 2005/ /2013, adjusted for Pre-Marcellus Characteristics Intercept (2.734) High Drilling (1.389) Some Drilling (1.105) Population Density, 2000 (100s) (.039) Percent Poverty, (.180) Unemployment Rate, (.762) Percent of employed workforce in construct/extraction/maintenance, (.200) Percent of employed workforce in product/transport/material moving, (.075) Table C50. Results from Linear Regression Model Predicting Percent Change in Gini Coefficient, 2005/ /2013, adjusted for Changes in County Characteristics Intercept (2.196) * High Drilling (1.597) * Some Drilling (1.270) * Percent change in population density, / (.051) Percent change in percent poverty, / (.036) Percent change in unemployment rate, / (.013) Percent change in percent of employed workforce in construction/extraction/maintenance, / (.041) Percent change in percent of employed workforce in production/transportation/material moving, / (.068) The Center for Rural Pennsylvania 86

87 Table C51. Results from Linear Regression Model Predicting Median Household Income among Homeowners, adjusted for County Characteristics, 2000 Intercept (4.456) *** High Drilling (2.264) Some Drilling (1.801) * Population Density, 2000 (100s) (.064) Percent Poverty, (.293) *** Unemployment Rate, (1.242) Percent of employed workforce in construct/extraction/maintenance, (.326) *** Percent of employed workforce in product/transport/material moving, (.122) *** Tables C52. Results from Linear Regression Model Predicting Median Household Income among Homeowners, 2009/2013, adjusted for Pre-Marcellus Characteristics Intercept (4.268) *** High Drilling (2.169) Some Drilling (1.726) * Population Density, 2000 (100s) (.061) Percent Poverty, (.281) *** Unemployment Rate, (1.190) Percent of employed workforce in construct/extraction/maintenance, (.312) ** Percent of employed workforce in product/transport/material moving, (.117) *** Tables C53. Results from Linear Regression Model Predicting Median Household Income among Homeowners, 2009/2013, adjusted for Changes in County Characteristics Intercept (6.767) *** High Drilling (4.922) Some Drilling (3.915) Percent change in population density, / (.158) Percent change in percent poverty, / (.110) Percent change in unemployment rate, / (.041) Percent change in percent of employed workforce in construction/extraction/maintenance, / (.125) Percent change in percent of employed workforce in production/transportation/material moving, / (.208) The Center for Rural Pennsylvania 87

88 Tables C54. Results from Linear Regression Model Predicting Percent Change in Median Household Income among Homeowners, /2013, adjusted for Pre-Marcellus Characteristics Intercept (3.014) High Drilling (1.531) Some Drilling (1.219) Population Density, 2000 (100s) (.043) Percent Poverty, (.198) Unemployment Rate, (.840) Percent of employed workforce in construct/extraction/maintenance, (.220) *** Percent of employed workforce in product/transport/material moving, (.082) *** Tables C55. Results from Linear Regression Model Predicting Percent Change in Median Household Income among Homeowners, /2013, adjusted for Changes in County Characteristics Intercept (2.705) High Drilling (1.968) Some Drilling (1.565) Percent change in population density, / (.063) Percent change in percent poverty, / (.044) *** Percent change in unemployment rate, / (.016) Percent change in percent of employed workforce in construction/extraction/maintenance, / (.050) Percent change in percent of employed workforce in production/transportation/material moving, / (.083) Table C56. Results from Linear Regression Model Predicting Median Household Income among Renters, adjusted for County Characteristics, 2000 Intercept (2.681) *** High Drilling (1.362) * Some Drilling (1.084) ** Population Density, 2000 (100s) (.038) * Percent Poverty, (.176) *** Unemployment Rate, (.747) Percent of employed workforce in construct/extraction/maintenance, (.196) Percent of employed workforce in product/transport/material moving, (.073) *** The Center for Rural Pennsylvania 88

89 Table C57. Results from Linear Regression Model Predicting Median Household Income among Renters, 2009/2013, adjusted for Pre-Marcellus Characteristics Intercept (2.198) *** High Drilling (1.117) Some Drilling (.889) ** Population Density, 2000 (100s) (.031) * Percent Poverty, (.145) *** Unemployment Rate, (.613) Percent of employed workforce in construct/extraction/maintenance, (.161) Percent of employed workforce in product/transport/material moving, (.060) *** Table C58. Results from Linear Regression Model Predicting Median Household Income among Renters, 2009/2013, adjusted for Changes in County Characteristics Intercept (2.843) *** High Drilling (2.068) Some Drilling (1.645) ** Percent change in population density, / (.066) Percent change in percent poverty, / (.046) Percent change in unemployment rate, / (.017) Percent change in percent of employed workforce in construction/extraction/maintenance, / (.053) Percent change in percent of employed workforce in production/transportation/material moving, / (.088) Tables C59. Results from Linear Regression Model Predicting Percent Change in Median Household Income among Renters, /2013, adjusted for Pre-Marcellus Characteristics Intercept (4.985) *** High Drilling (2.533) * Some Drilling (2.015) Population Density, 2000 (100s) (.071) Percent Poverty, (.328) * Unemployment Rate, (1.390) * Percent of employed workforce in construct/extraction/maintenance, (.364) Percent of employed workforce in product/transport/material moving, (.136) The Center for Rural Pennsylvania 89

90 Tables C60. Results from Linear Regression Model Predicting Percent Change in Median Household Income among Renters, /2013, adjusted for Changes in County Characteristics Intercept (4.191) * High Drilling (3.048) Some Drilling (2.425) Percent change in population density, / (.098) Percent change in percent poverty, / (.068) * Percent change in unemployment rate, / (.025) Percent change in percent of employed workforce in construction/extraction/maintenance, / (.078) Percent change in percent of employed workforce in production/transportation/material moving, / (.129) Table C61. Results from Linear Regression Model Predicting Gap in Household Income between Owners and Renters, adjusted for County Characteristics, 2000 Intercept (3.142) *** High Drilling (1.597) Some Drilling (1.270) Population Density, 2000 (100s) (.045) Percent Poverty, (.207) Unemployment Rate, (.876) Percent of employed workforce in construct/extraction/maintenance, (.230) *** Percent of employed workforce in product/transport/material moving, (.086) *** Table C62. Results from Linear Regression Model Predicting Gap in Household Income between Owners and Renters, 2009/2013, adjusted for Pre-Marcellus Characteristics Intercept (3.141) *** High Drilling (1.596) Some Drilling (1.270) Population Density, 2000 (100s) (.045) * Percent Poverty, (.207) * Unemployment Rate, (.876) Percent of employed workforce in construct/extraction/maintenance, (.230) *** Percent of employed workforce in product/transport/material moving, (.086) *** The Center for Rural Pennsylvania 90

91 Table C63. Results from Linear Regression Model Predicting Gap in Household Income between Owners and Renters, 2009/2013, adjusted for Changes in County Characteristics Intercept (4.669) *** High Drilling (3.396) Some Drilling (2.701) Percent change in population density, / (.109) Percent change in percent poverty, / (.076) Percent change in unemployment rate, / (.028) Percent change in percent of employed workforce in construction/extraction/maintenance, / (.086) Percent change in percent of employed workforce in production/transportation/material moving, / (.144) Table C64. Results from Linear Regression Model Predicting Percent Change in Gap in Household Income between Owners and Renters, /2013, adjusted for Pre-Marcellus Characteristics Intercept (7.854) High Drilling (3.991) Some Drilling (3.175) Population Density, 2000 (100s) (.112) Percent Poverty, (.517) Unemployment Rate, (2.190) Percent of employed workforce in construct/extraction/maintenance, (.574) ** Percent of employed workforce in product/transport/material moving, (.215) ** Table C65. Results from Linear Regression Model Predicting Percent Change in Gap in Household Income between Owners and Renters, /2013, adjusted for Changes in County Characteristics Β SE p Intercept (7.384) ** High Drilling (5.371) Some Drilling (4.272) Percent change in population density, / (.173) Percent change in percent poverty, / (.120) Percent change in unemployment rate, / (.045) Percent change in percent of employed workforce in construction/extraction/maintenance, / (.137) Percent change in percent of employed workforce in production/transportation/material moving, / (.228) The Center for Rural Pennsylvania 91

92 The Center for Rural Pennsylvania Board of Directors Chairman Senator Gene Yaw Vice Chairman Representative Garth D. Everett Treasurer Representative Sid Michaels Kavulich Secretary Dr. Nancy Falvo Clarion University Dr. Livingston Alexander University of Pittsburgh Stephen M. Brame Governor s Representative Dr. Michael A. Driscoll Indiana University Dr. Stephan J. Goetz Northeast Regional Center for Rural Development Dr. Timothy Kelsey Pennsylvania State University The Center for Rural Pennsylvania 625 Forster St., Room 902 Harrisburg, PA Phone: (717) P0317 The Center for Rural Pennsylvania 92