Evaluating the impact of the American Recovery and Reinvestment Act’s BTOP on broadband adoption

Abstract

The American Recovery and Reinvestment Act’s Broadband Technology Opportunities Program (BTOP) spent $4.7 billion during 2009–2013 to, inter alia, increase broadband adoption in underserved communities. We characterize the BTOP grants and examine the impact of the awards on broadband adoption. Econometric specifications controlling for award endogeneity related to observed and unobserved county-level factors find that spending is apparently associated with increased broadband adoption. Further investigation, however, reveals that the impacts of spending are nonlinear and even nonmonotonic over the range of county-level BTOP spending in the data. Controlling for trends to reduce the potential for spurious correlation between spending and outcomes removes most of the significance of the results. We conclude with three lessons for policymakers derived from the uncertain outcomes of BTOP spending found in our exploration.

Keywords:

• American Recovery and Reinvestment Act
• Broadband Technology Opportunities Program
• broadband adoption
• stimulus spending

JEL Classification:

• H31
• H54
• L96
• L98

Notes

¹ Gillett et al. (2006) and Katz and Suter (2009) identified positive links between broadband deployment and economic prosperity. Scott (2012) similarly found that a 10% increase in broadband penetration is correlated with a 1.35% increase in GDP for developing countries, and a 1.19% increase for developed countries. See also Koutroumpis (2009) and Qiang et al. (2009).

² NTIA is an executive branch agency under the US Department of Commerce that advises the executive branch of the federal government on the telecommunications industry. The remaining $2.5 billion the ARRA allocated to broadband was to be distributed through the Rural Utilities Service (RUS).

³ ASR Analytics (2012).

⁴ Prior to implementation of the BTOP the government broadly determined funding priorities by specifying the total allocation of funds to be granted in each category.

⁵ See NARA (2009). For an excellent summary of the funding process, see Rosston and Wallsten (2013). BroadbandUSA also offers documentation explaining the rules and processes of the BTOP: http://www2.ntia.doc.gov/. Finally, the following NTIA website offers answers to frequently asked questions about BTOP: http://www2.ntia.doc.gov/files/nofa2_faqs_5_28_10.pdf

⁶ From BroadbandUSA Connecting America’s Communities, accessed 17 August 2014, at http://www2.ntia.doc.gov/documents/BTOPPolicyReviewPPT.pdf. Unserved areas have less than 10% broadband penetration, and underserved areas have less than 40% penetration. The other purposes of BTOP were to support the missions of schools, libraries, healthcare providers and public safety agencies with broadband access, training and use.

⁷ Eligible Costs for Sustainable Broadband Adoption, 74 Fed. Reg.33113 (9 July 2009).

⁸ Applications in Round 1 were accepted from 14 July 2009 to 14 August 2009. Round 2 applications were accepted from 16 February 2010 to 16 March 2010, with the exception of an extension until 26 March 2010 for awards in the category of comprehensive community infrastructure.

⁹ Which broadband speeds are considered as broadband are inconsistent in these data from ITU, making cross-country comparison problematic.

¹⁰ See Goolsbee and Guryan (2006) for a rare exception.

¹¹ See also Firth and Mellor (2005), Papacharissi and Zaks (2006), and Preston et al. (2007).

¹² Consequently, we make no claims that our empirical results here are necessarily externally valid for other programmes in other countries.

¹³ Rosston and Wallsten attempted to obtain data directly from the NTIA but were given an estimated initial fee of $144 715.09 to process their data request; they concluded that ‘the NTIA does not make such information readily available’ (Rosston and Wallsten, 2013, footnote 10).

¹⁴ Examples of judgement include matching statements such as ‘rural New Mexico’ to noncore RUPRI counties and ‘south central rural North Dakota’ to non-CBSA counties in the south central part of the state.

¹⁵ Logging the variable reduces greatly the skewness in the spending levels, but the inherent bimodality remains. While bimodality does not, in principle, render the spending variable unsuitable for use as a regressor, it does lead us to consider carefully in the econometrics whether the variable’s impact on outcomes of interest changes as spending increases.

¹⁶ In the Broadband Technology Opportunities Program Frequently Asked Questions (28 May 2010, page 10), it is stated that ‘Proposals that would benefit tribal entities receive extra consideration by the programme during review, and the comments of tribes are a selection factor in choosing BTOP awards’.

¹⁷ In July 2011, 71% of non-Hispanic whites and 72% of Asians used the Internet at home, but only 52% of blacks and 49% of Hispanics did. Another way to examine broadband availability for minorities is to compare the number of residential broadband providers available to a representative member of each racial or ethnic group. Sixty-four per cent of white non-Hispanics have the greatest chance of having four or more fixed broadband providers where they live. Blacks have the lowest similar probability, at 46%, followed by Asians and Pacific Islanders (49%) and Hispanics (53%) (all statistics from Prieger (2015)).

¹⁸ The RFC/HH data using the BTOP definition are unavailable after 31 December 2012 due to a change in data collection at the FCC.

¹⁹ By December 2013, 192 projects (86% of all) had been completed (according to NTIA’s 20th Quarterly Status Report, issued 26 June 2014). The remaining 32 projects in progress had received an unspecified extended award period.

²⁰ Given our short period under study, how slowly demographics typically evolve, and our inclusion of county fixed effects in the estimations, we will not worry about year-specific demographic covariates.

²¹ Restricted cubic splines (Harrell, 2001) result in an estimated continuous and differentiable f that is linear before the first knot and after the last knot, and is a piecewise cubic polynomial between adjacent knots.

²² The figure reveals that we have made a slight simplification in identifying y* with RFC/HH. Latent y* takes values on the real line, while actual RFC/HH cannot be negative (although it could, in principle, exceed 100). A more pedantic exposition would define the latent variable y** to take values on the real line, then define y* = RFC/HH to be a left censored version of y* a la the Tobit model, and then finally to define y as the categorical observed variable. For our purposes, the only thing that matters, however, is that y* as we have defined it has units on the appropriate scale of RFC/HH.

²³ We estimate the model using constrained ordered logit estimation, as described in the Appendix. In ordered logistic regression, the scale parameter σ is not identified separately from the coefficients and the cutpoints, but here σ and the constant in β are identified because the cutpoints are known (see Fig. 4).

²⁴ In July 2011, 69.9% of those between the ages of 18 and 64 used broadband Internet at home, while only 39.7% of senior citizens used broadband (computations by the authors using the Computer and Internet Use Supplement to the Current Population Survey).

²⁵ The figure is apparently derived from the new subscribers each grantee claimed in its progress reports to NTIA.

²⁶ In the regression results shown in the last columns of Table 9, the confidence interval for the number of households connected with an extra dollar of sustainable adoption spending span zero due to the insignificance of the γ coefficients.

²⁷ Specifically, Alaska is almost entirely rural and is composed primarily of small, remote villages. Physical remoteness can hinder deployment of infrastructure and adoption (Hudson, 2011). Similarly, Hawaii’s formation as a string of islands presents infrastructure challenges atypical of the mainland states. Inhabitants of Hawaii also differ. While 77.7% of US residents are of white race, in Hawaii less than a quarter of the population is white. Whites have adopted broadband more readily than others in general (Prieger, 2015).

²⁸ Governors vetted applications for funding and influential committee members in Congress may have influenced where awards were granted (Hauge, 2015). On the other hand, the Washington D.C.-based funding agency (NTIA) may have favoured projects put forward on behalf of D.C. residents.

²⁹ Ros (1999) finds that publicly owned telecommunications firms had lower labour productivity than privately owned firms. For other industries, Megginson and Netter’s (2001, p.380) survey article concludes that the body of research ‘supports the proposition that privately owned firms are more efficient’ than state-owned firms.

³⁰ There are likely to be deficient incentives for managers of nonprofits since managers cannot legally share in the surplus their decisions may create (see Weisbrod 1988 and citations therein).

³¹ These are the observations created by prorating each project to its affected counties, before spending is summed over awards for each county.

³² The BTOP took the narrowest approach to accountability, seeking to ensure only that money was spent as intended. Grant recipients’ quarterly progress reports were made public in July 2014 (available at http://www2.ntia.doc.gov/awards) and show little detail with respect to project implementation other than funds spent. Recipients were not held accountable for successful project outcomes. For overall project evaluation, NTIA paid ASR Analytics $5 million to study project success. As cited by Wallsten (2015), ‘ASR Analytics itself wrote that “The selection of grants [to be evaluated] was purposeful and not meant to yield a statistical sample.” Yet, they used those very grants as the basis for the counties they studied statistically.’

³³ The nlcom command in Stata 13.1 returns the estimates $(\hat{\alpha },\hat{\beta },\hat{\sigma })$ and the associated asymptotic variance matrix.