Portal de Periódicos da CAPES

Place‐based subsidies and employment growth in rural America: Evidence from the broadband initiatives programme

2023; Elsevier BV; Volume: 102; Issue: 4 Linguagem: Inglês

10.1111/pirs.12740

1435-5957

Anil Rupasingha, John Pender, Ryan Williams, Joshua Goldstein, Devika Nair,

Education Systems and Policy

This paper studies the labour market effects of the Broadband Initiatives Program (BIP), a programme authorized by the American Recovery and Reinvestment Act of 2009 to promote broadband deployment, mainly in rural areas. The BIP is one of the largest USDA broadband programmes implemented to date, providing more than $3.4 billion in grants and loans in FY 2010. We investigate the impacts of BIP investments on employment outcomes in BIP-recipient Census tracts compared to similar tracts outside of BIP project service areas between the inception of the programme in 2010 and 2019. We use a quasi-experimental research design that combines difference-in-difference regression with propensity score matching estimation to identify the causal effect of the BIP investments on employment outcomes. We find that the BIP investments had a positive overall effect on employment growth that increased over time. The subsidized investments had a greater effect on employment in startups than in incumbent businesses, in the goods-producing sector and the information and communications technology sector than other sectors, and a greater effect in micropolitan census tracts than tracts located in metropolitan areas or in small town and remote rural locations. Este artículo estudia los efectos en el mercado laboral del Programa de Iniciativas de Banda Ancha (BIP, por sus siglas en inglés), un programa autorizado por la Ley de Recuperación y Reinversión de Estados Unidos de 2009 para promover el despliegue de la banda ancha, principalmente en zonas rurales. El BIP es uno de los mayores programas de banda ancha del USDA ejecutados hasta la fecha, y en el 2010 concedió más de 3.400 millones de dólares en subvenciones y préstamos. Se investigaron los impactos de las inversiones del BIP en los resultados de empleo en los tramos censales receptores de BIP en comparación con tramos similares fuera de las áreas cubiertas por el proyecto BIP entre el inicio del programa en 2010 y 2019. Se utilizó un diseño de investigación cuasi-experimental que combina la regresión de diferencias en diferencias con una estimación basada en pareamiento por puntaje de propensión para identificar el efecto causal de las inversiones del BIP en los resultados de empleo. Las inversiones del BIP tuvieron un efecto global positivo en el crecimiento del empleo, que aumentó con el tiempo. Las inversiones subvencionadas tuvieron un mayor efecto sobre el empleo en las empresas de nueva creación que en las ya existentes, en el sector de la producción de bienes y en el sector de las tecnologías de la información y las comunicaciones que en otros sectores, y un mayor efecto en los tramos censales micropolitanos que en los situados en áreas metropolitanas o en localidades rurales pequeñas y remotas. 本稿では、2009年アメリカ復興・再投資法によって認可された、主に農村地域におけるブロードバンドの導入を促進するためのプログラムであるブロードバンド・イニシアティブ・プログラム(Broadband Initiatives Program:BIP)の労働市場への影響を検討する。BIPは、2010年度には34億ドル以上の補助金と融資を提供しており、これまでに実施された米国農務省のブロードバンド・プログラムの中でも最大規模である。プログラムが開始された2010年から2019年までの、国勢統計区におけるBIP受給者の雇用成果に対するBIP投資の影響を、BIPプロジェクトのサービス対象地域以外の同様の地域と比較して調査した。差分の差分法と傾向スコアマッチング推定を組み合わせた疑似実験デザインを用いて、BIP投資が雇用成果に及ぼす因果効果の解明を試みた。BIP投資は全体的に雇用増加に全体的なプラスの効果をもたらし、さらにその効果は時間の経過とともに増大した。補助金による投資は、既存の企業の雇用よりもスタートアップの雇用に大きな影響を与え、他の部門よりも財生産部門や情報通信技術部門に大きな影響を与え、小都市の国勢統計区では大都市圏や小さな町あるいは僻地の国勢統計区よりも大きな影響を与えていた。 The deployment of high-speed broadband in underserved rural areas has become an important rural development policy over the last two decades. Benefits of access to broadband in rural areas are manifold, including business development through reduced input costs and e-commerce, job growth through business creation and expansion, access to healthcare and telemedicine, and other benefits (Abrardi & Cambini, 2019; Atasoy, 2013; Benda et al., 2020; Duvivier, 2019; Goldfarb & Tucker, 2019; Stenberg et al., 2009). However, broadband deployment in rural underserved areas is expensive due to remoteness and sparse population in these places, requiring very large up-front capital expenditures by service providers (Kandilov & Renkow, 2010). As a result, 17% of rural Americans did not have access to fixed terrestrial broadband service anywhere in their census block in December 2019 (Federal Communications Commission (FCC), 2021), while 41% of rural households did not have wired high-speed Internet service in their homes in November 2021 (National Telecommunications and Information Administration (NTIA), 2022). Federal investments in the United States promoting access to broadband in rural communities have been increasing over the last two decades. During fiscal years 2015 to 2020, the Federal Government spent more than $50 billion to promote broadband deployment and use through 12 programmes, focused mainly on rural areas, and Congress has appropriated nearly $75 billion for 13 new broadband programmes established since the beginning of the COVID-19 pandemic (U.S. Government Accountability Office (GAO), 2022). The general objective of many of these programmes is to provide broadband infrastructure in unserved and underserved areas, which is expected to stimulate improvements in social and economic well-being in these areas. This paper explores the impacts of one of these federal investments—the USDA Broadband Initiatives Program (BIP)—on employment growth. Several USDA programmes have supported broadband development in rural areas in the United States for the past two decades. The BIP is one of the largest of these programmes, created by the American Recovery and Reinvestment Act (ARRA) of 2009, which appropriated a total of $2.5 billion to USDA's Rural Utilities Service (RUS) to promote broadband deployment, mainly in rural areas of the country. Despite the large size of the BIP and length of time since the programme was completed, few published studies have investigated impacts of this programme, except two recent papers by Bai et al. (2022) and Pender et al. (2022). 1 This paper contributes to this small literature by investigating the impacts of BIP on employment, which was not addressed in either of those studies. The impacts of provision of broadband infrastructure on employment growth and other socio-economic outcomes depend on whether such infrastructure results in increased adoption and use of broadband, or use of broadband at higher speeds than previously, by businesses, households, and other end users. 2 Unfortunately, data on adoption and use of broadband by businesses are not generally available. Existing empirical evidence on effects of broadband availability on broadband adoption by households supports the notion that increased availability leads to more adoption, especially in rural areas (Pender et al., 2022; Silva et al., 2018; Whitacre et al., 2013). In addition to finding that initial broadband availability contributes to subsequent household broadband adoption, Pender et al. (2022) found that BIP increased the average share of households adopting broadband by about 1.1–3.0 percentage points by 2016. Beyond the effect of BIP on broadband adoption, there is the question of how the programme affected socio-economic outcomes such as employment growth. The socio-economic benefits of increasing access to broadband in rural communities are widely documented as described in the literature review below. However, no published studies have investigated impacts of the BIP on employment growth and few studies have investigated employment impacts of any Federal broadband programme. In this study, we investigate the impacts of the BIP on employment growth in BIP-recipient census tracts compared to similar census tracts outside of BIP project service areas between the inception of the programme in 2010 and 2019. The expectation from a policy perspective was that the BIP would lead to increased access to broadband in unserved and underserved areas, promoting economic growth by saving existing jobs and creating more. However, theoretically, the employment impacts of such investment incentives are ambiguous. Although broadband investments may increase job opportunities due to the direct employment and multiplier effects of broadband deployment and maintenance, promotion of new business formation, retention and expansion of existing businesses, attraction of workers, reduction in search costs in labour markets, and other factors, they may also reduce employment in some regions or for some types of workers due to increased competition or substitution of local labour by capital expenditures or outsourcing of work to other firms and locations (Akerman et al., 2015; Atasoy, 2013; Duvivier, 2019; Goldfarb & Tucker, 2019; Kolko, 2012; Kuttner, 2016; Stephens et al., 2022). In addition to investigating the impact of BIP on overall employment changes, we examine whether employment effects are heterogeneous between incumbent establishments and startups, across broad industry sectors, and across the urban/rural spectrum. Although the BIP was intended to be mainly focused on rural areas, it was not strictly limited to such areas, and the programme service areas included areas located within metropolitan and micropolitan areas. 3 Earlier research has found that broadband provision has resulted in positive net establishment entry in the United States (Chen et al., 2023), which is the largest source of employment growth impacts of broadband provision, though broadband also has contributed to employment growth in incumbent businesses (Kolko, 2012). The literature has also shown that employment impacts of broadband provision vary across industry sectors (Atasoy, 2013; Kandilov & Renkow, 2010; Kolko, 2012; Shideler et al., 2007) and between urban and rural markets (Atasoy, 2013; Kandilov & Renkow, 2010; Kolko, 2012; Lobo et al., 2020; Whitacre et al., 2014b). Thus, we expect that the impacts of BIP may differ between incumbent businesses and startups and across industry sectors and geographic contexts. Our analysis uses a reduced form approach, investigating the effects of BIP on employment outcomes without investigating the mediating mechanisms by which such outcomes may have occurred due to data limitations—such as the lack of publicly available data on broadband adoption by businesses. It assumes that broadband deployment through BIP investments translates into adoption by consumers and businesses, influencing employment outcomes in programme recipient areas. We find that BIP investments had a positive effect on employment growth overall, with the impacts generally increasing over time. BIP had a greater effect on employment in startups than incumbent businesses, in the goods-producing sector and information and communication technology (ICT) subsector than in other sectors, and in micropolitan census tracts than in metropolitan and remote rural locations. Broadband can affect regional employment through many mechanisms. Broadband deployment can increase the demand for labour since labour is required to manufacture, deploy, and maintain the necessary infrastructure and parts (Atasoy, 2013), some of which may be provided by workers within a given region. Such direct employment effects may have indirect and induced multiplier effects on local labour demand (Kuttner, 2016). Broadband can affect demand for the goods and services produced in a region by increasing access of local firms to more distant markets and competition with more distant firms (Atasoy, 2013). Broadband may affect the productivity of firms, which can have mixed effects on labour demand depending on the substitution or complementarity relationships between the technologies that broadband enables and different types/skill levels of labour (Akerman et al., 2015; Atasoy, 2013). By increasing access to information and the productivity of firms, broadband availability may attract firms (especially those in knowledge-intensive industries) to startup or relocate to a region with better broadband service and a sufficiently skilled workforce (Atasoy, 2013; Chen et al., 2023; Conroy & Low, 2022; Deller et al., 2022; Duvivier, 2019; Kim & Orazem, 2017; Kolko, 2012; Mack, 2014, 2015; Mack & Rey, 2014; Mack et al., 2011; Mack, 2014; Mack & Rey, 2014; Mack & Wentz, 2017; Shideler & Badasyan, 2012; Stephens et al., 2022; Whitacre et al., 2014a, 2014b). Broadband access may help existing firms survive and grow (Kolko, 2012), though it may also undermine survival of local businesses by increasing competition with distant providers of goods and services (Stephens et al., 2022). All these mechanisms can affect the demand for labour in a region (Kolko, 2012). Broadband can also affect the supply of labour in a region. As a consumer amenity providing improved access to goods and services that affect quality of life, broadband can affect migration (Mahasuweerachai et al., 2010) and home buying decisions (Deller & Whitacre, 2019; Molnar et al., 2019) and thus the supply of labour in a region. Broadband access enables telework (Carvalho et al., 2022; Pender et al., 2022), which may attract people to live and telework in places that were previously too distant to commute from. Broadband also can reduce search costs in the labour market for job applicants and employers, helping to reduce unemployment and vacancies (Goldfarb & Tucker, 2019) and improve the worker-firm matching process (Atasoy, 2013). Broadband access (especially mobile broadband) may also facilitate peer-to-peer markets for employment opportunities ("gig jobs") such as Uber/Lyft drivers and domestic services (Abraham et al., 2019; Goldfarb & Tucker, 2019). A large and growing literature on broadband finds positive effects of high-speed internet on economic outcomes at a local level (usually the county or zip code level) in the United States, particularly labour market outcomes. Numerous studies have found a positive association of broadband availability with employment and/or a negative association with unemployment (Atasoy, 2013; Bai, 2017; Guidry et al., 2012; Isley & Low, 2022; Jayakar & Park, 2013; Kolko, 2012; Lobo et al., 2020; Shideler et al., 2007; Stenberg et al., 2009; Whitacre et al., 2014b). Not all studies have found positive effects of broadband availability on employment, however. Several found statistically insignificant effects, particularly when examining impacts of broadband deployment on changes in employment over time (Ford, 2018; Jayakar & Park, 2013; Whitacre et al., 2014a, 2014b, 2018). The impacts of broadband availability on employment have been found in several studies to be context dependent. For example, Kolko (2012) found that the employment impacts of broadband provision were greater in less densely populated Zip Code Tabulation Areas (ZCTAs). Similarly, Atasoy (2013) found that the positive effect of broadband availability on employment rate was greater in more rural areas, especially those with more skilled workers. By contrast, Whitacre et al. (2014b) found, using cross-sectional spatial error model (SEM) regressions, that high broadband availability was positively associated with greater total employment in metro counties but not in nonmetro counties, while they found no statistically significant effect of changes in broadband availability on employment change in either metro or nonmetro counties using first difference ordinary least squares (OLS) regressions. In a study of impacts of broadband availability on county-level unemployment rates in Tennessee, Lobo et al. (2020) found a significant negative association of high-speed broadband with unemployment rates, but only in rural counties. Thus several, but not all, studies that investigated the effects of rurality found greater impacts of improvements in broadband availability on employment in more rural areas. Several studies have found a positive association of broadband adoption with employment or a negative association with unemployment (Carvalho et al., 2022; Isley & Low, 2022; Whitacre et al., 2014a, 2014b). Although Whitacre et al. (2014b) found a positive association between high broadband adoption and total employment in nonmetro counties in their cross-sectional SEM regressions, they found no statistically significant impact of broadband adoption on employment in their first difference OLS regressions, so this result for impacts of broadband adoption was not robust to the method used. Carvalho et al. (2022) investigated the effects of broadband adoption combined with the ability of workers to telework, on county unemployment rates in the early months of the COVID-19 pandemic, finding that counties in the southeastern United States with greater initial adoption of broadband and a greater share of the workforce able to telework had less growth in unemployment. One study by Forman et al. (2012) investigated the impacts of commercial businesses' use of advanced Internet applications—which depend on broadband availability and adoption by businesses—on county employment growth in the late 1990s. That study found a weakly statistically significant (at 10% level) positive effect of use of advanced Internet applications on county employment growth, with this effect concentrated in counties with high levels of household income, education, information technology intensity, and population. Fewer studies have investigated impacts of programmes designed to increase broadband availability or adoption on employment outcomes. Several studies have investigated impacts of the USDA Rural Broadband Access Loan and Loan Guarantee Program (RBLP) and its predecessor, the pilot Broadband Loan Program (BBLP) (Dinterman & Renkow, 2017; Kandilov & Renkow, 2010, 2020; Kandilov et al., 2017; U.S. Government Accountability Office (GAO), 2014). Of these studies, only Kandilov and Renkow (2010) and U.S. Government Accountability Office (GAO) (2014) investigated impacts of these programmes on employment. 4 Kandilov and Renkow (2010) investigated impacts of the pilot BBLP and RBLP at both zip code and county level, focusing on zip code areas or counties that had a population of 20,000 or less in the year 2000. They found a positive impact of the pilot BBLP on employment in their baseline regression at the zip code level and a small and statistically insignificant impact of the RBLP on employment. They argued that the more limited effects of the RBLP could have been due to the limited amount of time between when loans began to be administered under that programme in 2004 and the end of their study period in 2007. They found that the pilot BBLP had a positive association with employment only in metro counties, while in nonmetro counties not adjacent to a metro area, both the pilot BBLP and the RBLP had a negative association with employment. They found no statistically significant impact of either programme on total employment in county-level regressions. Across industries, they found a negative impact of the pilot BBLP on employment in information services, negative impacts of the RBLP on employment in utilities and retail trade, and a positive impact of the RBLP on employment in the real estate and rental industry. They argued that the negative effects of these broadband programmes on employment in some sectors could be due to competition for local businesses facilitated by broadband access. In general, Kandilov and Renkow's results demonstrate that a broadband programme can have heterogeneous effects on employment in different geographic contexts and industries, and likely also depend on how much time has elapsed since the inception of the programme. The U.S. Government Accountability Office (GAO) (2014) also investigated impacts of the RBLP on employment, using county-level regressions and focusing on a longer period than Kandilov and Renkow (2010) (2003–2011). Unlike Kandilov and Renkow, U.S. Government Accountability Office (GAO) (2014) found robust positive impacts of the RBLP on employment across several analyses with alternative control groups, using its baseline sample of counties with at least 90% of the county population residing in a rural area or an urban cluster of less than 50,000 people. Using instead a more restrictive definition of rural counties—excluding people in urban clusters—U.S. Government Accountability Office (GAO) (2014) found statistically insignificant impacts of the RBLP on employment. The lack of significant impacts with the more rural sample of counties reflects a much smaller sample of counties classified as "rural" having received RBLP loans, resulting in larger standard errors of the estimated impacts, and does not necessarily reflect actual heterogeneity of impacts between more and less rural counties. Two peer-reviewed studies have investigated impacts of the Broadband Technology Opportunities Program (BTOP)—the sister programme to the BIP also funded by ARRA—on broadband adoption (Chang, 2021; Hauge & Prieger, 2015). Both these studies found statistically insignificant impacts of BTOP grants on broadband adoption. Two recent peer-reviewed studies have investigated impacts of BIP (Bai et al., 2022; Pender et al., 2022). Bai et al. (2022) investigated the impact of BIP on farm productivity (defined as farm sales per farm employment), finding a significant positive but short-term impact on farm productivity. Pender et al. (2022) investigated the impact of BIP on broadband adoption and telework, finding positive impacts on both outcomes. No published peer-reviewed study that we are aware of has investigated impacts of either BTOP or BIP on employment outcomes. The available literature suggests that a programme such as BIP, which financed deployment of broadband infrastructure in mostly (but not only) rural areas, could have affected employment in recipient areas through multiple mechanisms, with possibly different impacts on employment in business startups than on employment in incumbent businesses. The literature also suggests the impacts may have been heterogeneous across geographic contexts and industries. The present study contributes to the literature by investigating the impacts of BIP on employment, which has not been reported in any published studies. We investigate the heterogeneity of BIP impacts on employment across several dimensions, including geography, industry, and whether employment change occurred in startup or incumbent business at the time BIP was initiated. The Broadband Initiatives Program (BIP) was authorized as a part of the American Recovery and Reinvestment Act (ARRA) of 2009, which appropriated $2.5 billion to U.S. Department of Agriculture's Rural Utilities Service (RUS) to promote broadband deployment in rural areas of the country. ARRA also expanded the existing authority of RUS to make loans and provided new authority to make grants for the purpose of facilitating broadband deployment in rural communities. RUS leveraged the budget authority provided by ARRA to make grants, loans and loan/grant combination awards. In total, over $2.23 billion in grants and $1.19 billion in loans were made to 299 terrestrial broadband infrastructure projects throughout the country. 5 All BIP awards were made in two rounds in FY 2010. ARRA required the project service areas to be at least 75% rural (in terms of area) and without broadband availability at 5 megabits per second (mbps) (upstream plus downstream) in at least 50% of each project service area. Rural areas were defined as: "Any area, as confirmed by the latest decennial census of the Bureau of the Census, which is not located within: (1) a city, town, or incorporated area that has a population of greater than 20,000 inhabitants; or (2) an urbanized area contiguous and adjacent to a city or town that has a population of greater than 50,000 inhabitants." Based on these criteria, project service areas are mostly at the subcounty level, and some of the qualified rural areas are located within metro counties. Infrastructure projects supported by this programme were almost entirely (95%) last mile infrastructure projects and nearly two thirds of the projects (65%) provided fibre optic service to the household or business. Grants in the first round could only support projects serving exclusively remote unserved rural areas but this requirement was dropped in the second round. However, a counterfactual for the employment of BIP-recipient tracts, if they had not received BIP investments, can be estimated using the employment of tracts that never received BIP investments. This assumes that the mean outcome observed for the tracts that never received BIP investments is the same as the mean outcome that would have been observed for the treated group if they had not received BIP. However, approximating the counterfactual level for BIP-recipient tracts by using non-recipient tracts, our estimates of the ATET are likely to be affected by selection bias due to the presence of observable or unobservable differences between the treated and never treated tracts that are associated with differences in outcomes. One drawback of PSM estimation alone is that it controls only for differences in observable characteristics of treated and control groups and not for unobservable differences. Selection bias can be further reduced by combining PSM with difference-in-difference (DiD) estimation, which estimates impacts of treatment on changes in the outcome variable between pre- and post-treatment periods, thus subtracting out the additive effects of fixed characteristics, whether observed or not. 7 Thus we combine PSM with DiD estimation. The primary data sources for the study are BIP data from the USDA Rural Utilities Service (RUS), American Community Survey (ACS) data from the Census Bureau, Federal Communication Commission (FCC) data, and National Establishment Time-Series (NETS) data from Walls & Associates. All the data used in the analysis are at the census tract level using Census 2010 tract boundaries. Digitized maps of BIP areas and other RUS broadband programmes were provided by RUS and used to identify the share of tracts that were in a BIP project service area or service areas of other broadband programmes. Census tract-level demographic, housing, and industry employment share information were obtained from the ACS. Tract-level broadband availability information was obtained from the FCC. All tract-level employment data were compiled from the National Establishment Time-Series (NETS) data. The NETS is a longitudinal establishment-level database constructed by Walls & Associates using business-level data from Dun & Bradstreet (D&B) Market Identifier (DMI) files. Walls & Associates compiles repeated cross sections of the underlying D&B data on employment, sales and other variables into a longitudinal series using the unique D&B identification number (the DUNS number). They claim to cover nearly every US business unit that has operated in the United States over the past three decades, including sole proprietors, small privately owned firms, farms, nonprofit organizations and public sector establishments, such as post offices and public schools. Business data based on D&B DMI data have been criticized as not suitable for analysis (Davis et al., 1996). These criticisms claim that births and younger/smaller businesses are underreported and that there are discrepancies in the total U.S. employment figures in DMI files and data published by the Bureau of Labor Statistics. However, several more recent studies (Choi et al., 2013; Kunkle, 2011; Neumark et al., 2007, 2011) that investigated the accuracy of NETS data found fault with some of the above criticisms. Neumark et al. (2011) found that the claim of the underreporting of births in the NETS is without merit. They also claimed that the NETS data has better coverage of small business owners compared to those reported under the Statistics of Business (SOB) data. The main reason for the greater representation of smaller businesses in the NETS than in official business data sets is due to the inclusion of nonemployers in the NETS and imputing a value of 1 for the number of employees for nonemployer establishments. Most recently, Barnatchez et al. (2017) conducted a critical review of NETS data and concluded that the main discrepancies between NETS and official sources are largely driven by differences among small establishments and that NETS data greatly overrepresent establishments with less than 10 employees. They also highlighted issues with industry labeling that affect the comparability of results between the NETS and official business statistics and suggested that while the NETS is generally suitable for static analyses of establishments and employment across geographies or industries, it is less useful for studying business dynamics, though it is unclear how they arrived at this conclusion. RUS shape files of BIP project service areas 8 were overlaid on 2010 census tract boundaries using GIS software, 9 and the share of area of each census tract within BIP project service areas was computed. In our base analyses, we classified a census tract as "treated" by BIP if at least 90% of the tract area was in a BIP project service area. In robustness checks of our analysis, we investigate the robustness of our results to alternative thresholds of the minimum share of BIP area in a tract used to classify a tract as treated. All control tracts used in the analysis had no BIP project service areas intersecting the tract, and any tracts that included project service areas of any other RUS broadband programmes were excluded from the analysis. To implement the DiD/PSM design, treated tracts were defined in our base models as tracts with at least 90% of their area in a BIP project service area. 10 We selected a comparison group of tracts from among non-BIP areas. We excluded census tracts served by other USDA telecommunications programmes before matching and limited the control census tracts to those within 50 miles of the boundary of a treated tract. To account for county and state-level