Accessible Requires Authentication Pre-published online by De Gruyter September 14, 2021

Broadband Internet and Income Inequality

Georges V. Houngbonon and Julienne Liang

Abstract

Digital technologies like the Internet can affect income inequality through increased demand for employment in manual and abstract jobs and reduced demand for employment in routine jobs. In this paper, we combine city-level income distribution and jobs data with broadband data from France to investigate the impact of broadband Internet access on income inequality. Using an instrumental variable estimation strategy, we find that broadband Internet reduces income inequality through increased employment in manual jobs. These effects increase with the availability of skilled workers and are significant in cities with a large service sector or high-speed Internet access. Further, the diffusion of broadband Internet comes with relatively greater benefits in low-income cities compared to high-income cities. Several robustness checks support these findings.

JEL Classification: D31; L96; O15; J20

Corresponding author: Georges V. Houngbonon, Laboratoire de Genie Industrielle, Gif-sur-Yvette, France, E-mail:

Acknowledgments

We thank participants of the Royal Economic Society Conference (2018) and seminars participants at the Paris School of Economics and Telecom Paristech for comments and suggestions. We are greateful to Thomas Piketty, Denis Cogneau, Ekaterina Zhuravskaya, Marc Bourreau, Maya Bacache-Beauvallet, Eva Moreno-Galbis, Christine Zulehner and nine anonymous reviewers for helpful feedbacks. We would also like to thank Marc Lebourges and colleagues at Orange for comments and suggestions on earlier versions of this paper. Georges V. Houngbonon acknowledges financial support from the Chair CapitalDon at CentraleSupelec. All errors remain our own.

Appendix

A Conceptual Framework – Automation and Jobs

The objective of this section is to present a simple model that links broadband adoption to the Gini index of income inequality. Our intention is not to introduce a model for structural estimation, but rather to illustrate this link in a simple way by a graphical and mathematical expressions.

Acemoglu and Restrepo (2018) provides a task-based framework to describe a production process in a single-sector economy. The author assumes that production is composed of a number of tasks. A task is indexed by z, which is between N − 1 and N. Each task can be produced by capita or labor. When, z < I the task is automated, i.e. produced by capital. On the other hand, when z > I, the task cannot be automated and is only performed by labor.

An increase in I represents a greater automation of task z. An increase of N corresponds to the introduction of new labor with new skills. With Internet adoption, the increase in I and N coexist. The increase in I allows for the replacement of some routine tasks. The increase in N, represented by an increase in new jobs related to the Internet or broadly ICT (Information and Communications Technology).

Figure A1: Link between broadband internet adoption and Gini index of income inequality.

Figure A1:

Link between broadband internet adoption and Gini index of income inequality.

As shown in Figure A1, the Gini index of income inequality is represented by the surface between the diagonal line (y = x) (in blue) and the convex curve f(x) (in orange). The Gini is calculated by the integral of (xf(x)) between 0 and 1, i.e. Gini=01(xf(x))1/2dx.

The Internet adoption has potentially two effects in f(x). The first effect is the new tasks effect which makes the curve f(x) more convex by passing through f(x) = xb to f(x) = x(b+w) with b > 1 (b represents the initial level of the Gini), and w > 0 (w represents the extent of skilled web workers), since ICT tends to benefit employees who are highly skilled and whose skills are complementary to ICT. The second effect is the automation effect for x close to 0. Low deciles between 0 and a, mainly routine workers, are reduced by automation (a reflects the share of routine works replaced by automation). The first effect increases the Gini index and the second effect reduces the Gini index.

New jobs created with the arrival of the Internet are represented as “Web works” in Figure A1. These high income jobs increase the surface between y = x and y = f(x). As a result, the Gini index may rise with these new jobs.

Internet adoption has also reduced some routine tasks. In general, these tasks are basic service jobs and are in the low income decile.[20]

This reduction decreases the surface between y = x and y = f(x). Internet adoption is giving rise to a new business which is the delivery of e-Commerce goods. The deliverers, who probably used to be routine workers before, can earn more by working more. The latter can contribute to raising the incomes of low deciles.

These two effects co-exist. While the first effect dominates, inequality increases with Internet adoption. While the second effect dominates, inequality decreases with Internet adoption. We can describe these two effects with simple illustrative mathematical expressions.

First effect: as the new jobs make the function f(x) more convex from f(x) = xb to f(x) = x(b+w), the surface between y = x and y = f(x) increases. It is a matter of simple maths to show that the Gini is increased by 2w(b+1)2 compared to the initial Gini, Gini0=(12b+1), before the arrival of new Web jobs (i.e. w = 0).

Second effect: the loss of some routine jobs represented by the area of the small triangle at the bottom left for x between 0 and a. This loss makes x take value between a and 1. We could make a variable transformation to express f(x) in f(z) with z=(xa)(1a) and z is between 0 and 1 for x between a and 1. So f(z) > f(x), the surface between y = z and y = f(z) decreases, that means the Gini decreases. It is also possible to approximate this effect by a simple calculation without the variable transformation, by performing the integral of (xf(x)) for x between a and 1 with the assumption of a ≪ 1. The Gini is reduced by (a22a(b+1)(b+1)) compared to the initial Gini with (a = 0). Depending on the dominance of one of these two effects, the Gini can increase or decrease according to a (loss of routine works due to automation) and w (new skilled Web works). When both effects coexist:

  1. the Gini increases if 2w(b+1)2>(a22a(b+1)(b+1))

  2. the Gini increases if 2w(b+1)2<(a22a(b+1)(b+1))

These mathematical expressions are simply illustrative and we do not attempt to calibrate the values of b, a or w by the empirical estimations.

B Graphs and tables

Figure A2: Central offices and distance to households – illustration from a French department (Belfort).The squares correspond to central offices and the dots correspond to upgraded central offices to get closer to households.Source: actualisation du schema directeur d’amenagement numerique de l’Aisne – Volet Infrastructures numeriques – February 2016.

Figure A2:

Central offices and distance to households – illustration from a French department (Belfort).

The squares correspond to central offices and the dots correspond to upgraded central offices to get closer to households.

Source: actualisation du schema directeur d’amenagement numerique de l’Aisne – Volet Infrastructures numeriques – February 2016.

Figure A3: Broadband subscriptions by technology (million).Source: authors, using data from the national regulator (ARCEP, 2016).

Figure A3:

Broadband subscriptions by technology (million).

Source: authors, using data from the national regulator (ARCEP, 2016).

Table A1:

Detailed statistics on broadband data – fixed broadband penetration rate (%).

2009201020112012201320142015
Mean income in 2001 (euros)
8500–14,50045.0749.3152.7455.9058.3660.6362.78
1056105510281050105210531044
14,500–16,00050.1954.6658.3561.6064.3366.8869.40
1096109610561096109610951085
16,000–19,00055.4959.8863.3766.8069.6172.3375.02
1102110210781102110211021097
19,000–54,00062.7967.0070.0373.1675.7877.9980.46
1108110810971107110811081107
Mean Gini in 2001
22.2–28.855.2859.4562.9966.1668.7771.1573.54
1118111810891117111811171103
28.8–31.353.1757.3960.8764.1166.7069.2671.81
1112111210841111111211111100
31.3–34.351.6055.9859.5162.7865.6468.2270.73
1083108310611082108310811081
34.3–55.153.8858.4461.7264.8367.4069.6272.09
1049104810251045104510491049
Mean pop. density in 1999 (inhab./km2)
6–12650.2854.8658.8562.5065.4368.4371.25
1176117611501174117611731155
126–26053.0757.5561.4264.8067.6570.4473.04
1176117611381173117611741166
260–66054.6558.8562.4165.5768.2370.6573.03
1176117611471176117611761173
660–44,00056.3460.4562.9265.7067.9669.5071.63
1176117511591171117011741171

  1. Broadband penetration and number of cities at the bottom.

Table A2:

Detailed statistics on broadband data – median download speed (Mbps).

2009201020112012201320142015
Mean income in 2001 (euros)
8500–14,5006.868.6811.7513.4914.3515.2315.77
1124112410671098109410911083
14,500–16,0006.998.0711.1613.0713.9715.1215.89
1129112910721117111211121103
16,000–19,0006.397.229.7811.5012.6013.9014.57
1129112910991127112511271122
19,000–54,0005.836.418.3510.0810.9812.2313.02
1129112911121125112411241124
Mean Gini in 2001
22.2–28.85.976.869.7311.4012.2113.2813.88
1132113210961124112411221108
28.8–31.36.397.289.9211.9412.9514.0614.74
1133113310981127112711261116
31.3–34.36.788.2811.0812.9313.7814.8715.54
1122112210851104109810981101
34.3–55.16.917.9610.2411.8512.9314.2315.03
1124112410711112110611081107
Mean pop. density in 1999 (inhab./km2)
6–1266.838.2011.6313.9114.8716.0116.79
1176117611501174117611731155
126–2606.738.2911.6713.5314.3715.4216.03
1176117611381173117611741166
260–6606.247.239.8011.4512.3913.4414.17
1176117611481176117611761173
660–44,0006.096.578.229.4610.4011.6512.35
1161116111441161116111591156

  1. Median download speed and number of cities at the bottom.

Table A3:

Classification of employment occupations.

Occupation typeExamples
Manual: low-skill Caregivers, nurses, paramedic, security agent,
service workerspolice officer, receptionist, sales agent, waiters,
hairdresser, childminder, cleaner, bricklayers, carpenters,
plumbers, manual welders, gardeners, electrician,
butcher, tailor, baker, taxi drivers, delivery
Routine: low or high skill jobsArchivist, accountant, financial or insurance adviser,
librarian, pharmacy technician, tax controller,
administrative assistant, translators, photographers
Abstract: high-skill Executives, managers, lawyers, surgeon,
non-routine jobsengineers, teachers, professors, medical
doctors, journalists, researchers, marketing officers,
statisticians

  1. This classification follows Moreno-Galbis and Sopraseuth (2014).

Table A4:

Compliers analysis – city characteristics according to level of availability and change in uptake of broadband Internet (2009–2015).

Level of availabilityChange in penetration rate
Below medianAbove median
Below medianIncome in 2001 (euros)17,988.517,698.9
Gini index in 200130.730.4
Population density in 1999959.9494.3
Number of cities375241
Above medianIncome in 2001 (euros)17,545.816,778.5
Gini index in 200132.031.9
Population density in 19991380.9462.1
Number of cities17792141

  1. Compliers are typically cities with above-median availability and above-median increase in penetration rate, or below-median availability and below-median increase in penetration rate.

Table A5:

Identification strategy – two example cities.

Ay-ChampagneGrand-Fort-Philippe
% hh. within 3 kmPenbb (%)Gini% hh. within 3 kmPenbb (%)Gini
200999.8%47.4131.373.58%43.0929.99
201099.8%50.2331.293.58%44.2830.10
201199.8%52.9631.253.58%48.0229.87
201299.8%54.7831.623.58%50.2429.86
201399.8%56.6430.363.58%52.1729.47
201499.8%59.5829.963.58%53.4930.50
201599.8%61.7730.203.58%55.2030.90
Table A6:

First stages of the IV models – estimation results.

Specifications(3)–(6)(11)(14)(15)(20)(21)–(27)(28)–(37)
Dep. var.PenbbPenbbPenbb × high-speedPenbbBusinessBb yearPenbbPenbbPenbb
EstimatorNLLSOLSOLSOLSOLSOLSOLSNLLSOLSNLLSOLSNLLSOLS
Bb avail.0.154***0.077***−0.250***0.076***0.089***1.076***1.043***
(0.015)(0.008)(0.019)(0.006)(0.003)(0.092)(0.110)
Bb avail. × high-speed0.0000.846***
(0.005)(0.005)
Penbb0.438***0.470***0.910***0.953***
(0.107)(0.058)(0.164)(0.215)
Signal attenuation−0.001***
(0.000)
# Fixed telephone lines−0.066***
(0.002)
Diffusion speed0.072***0.218***0.174***0.166***
(0.007)(0.009)(0.046)(0.054)
Inflexion year11.242***4.059***5.258***5.040***
(2.609)(0.057)(0.481)(0.604)
Shbac4.106***−0.680***1.515***1.480***1.518***−1.190***2.522***−0.770***2.149***0.1942.900***0.389
(0.390)(0.211)(0.113)(0.123)(0.111)(0.217)(0.040)(0.116)(0.347)(0.906)(0.447)(0.519)
Shold−0.310***−0.0370.196***0.0140.191***−0.541***−0.281***0.005−0.547***0.390−0.785***1.111*
(0.035)(0.036)(0.057)(0.063)(0.056)(0.119)(0.012)(0.037)(0.105)(0.715)(0.136)(0.610)
Pop_density0.012***0.061*0.048***0.0140.048***−0.0330.005***0.068**0.006***0.311***0.008***0.098
(0.001)(0.032)(0.018)(0.016)(0.018)(0.041)(0.000)(0.030)(0.002)(0.117)(0.002)(0.072)
gdp0.000***−0.001**
(0.000)(0.000)
Table A6:

(continued)

Specifications(3)–(6)(11)(14)(15)(20)(21)–(27)(28)–(37)
Dep. var.PenbbPenbbPenbb × high-speedPenbbBusinessBb yearPenbbPenbbPenbb
EstimatorNLLSOLSOLSOLSOLSOLSOLSNLLSOLSNLLSOLSNLLSOLS
Baseline chars.YesYesYesYes
Constant0.617***0.411***0.193***−0.197***0.196***−0.218***2002.160***0.401***0.391***−0.304***0.005−0.323***−0.273
(0.058)(0.042)(0.031)(0.034)(0.030)(0.044)(0.028)(0.007)(0.043)(0.047)(0.292)(0.057)(0.245)
Observations31,74131,7344216421643194327493232,60531,734602602655655
R-squared0.9800.9680.4020.8900.4050.2770.2920.9800.9680.9920.9820.9900.982

  1. Significant at 10% level (*), 5% level (**) or 1% level (***). Robust standard errors in parentheses. The specifications refer to the ones in Tables 24 and A-8. The first stages of specifications (3)–(6) have been applied to specifications (7)–(10), (12)–(13) and (16)–(19), but restricted to the relevant samples. The logistic diffusion model is estimated by non-linear least squares (NLLS). Baseline characteristics include the Gini index in 2001, high-school completion rates in 1999, share of population above 65 in 1999, and population density in 1999. Variable definitions are as follow: broadband availability (BBavail.) is measured by the percentage of households living within 3 km from the nearest Internet node; dummy for high-speed (High-speed equals 1 if median download speed is above 10 Mbps); (Penbb_) denotes the predicted values from the logistic diffusion model; year when a town got connected to broadband Internet (BbYear); fixed broadband subscribers in percentage of population (Penbb, 0–1); median download speed (Speed, ×10 Mbps); number of business subscribers (Business).

Figure A4: Impact of broadband internet uptake in low-income cities.Source: low-income cities have average income below 16,000 euros in 2001.

Figure A4:

Impact of broadband internet uptake in low-income cities.

Source: low-income cities have average income below 16,000 euros in 2001.

Figure A5: Impact of broadband internet uptake in high-income cities.Source: high-income cities have average income above 16,000 euros in 2001.

Figure A5:

Impact of broadband internet uptake in high-income cities.

Source: high-income cities have average income above 16,000 euros in 2001.

Table A7:

Correlation matrix.

dGinidPenbbdShbacdSholddPopdens
dGini1
dPenbb−0.152*1
dShbac−0.093*0.071*1
dShold−0.082*0.029−0.139*1
dPopdens0.083*0.072*−0.078*−0.1611

  1. Pairwise correlations, significant at 1% level (*). d is the sixth difference operator in 2015. For instance dGini is the change in the Gini index between 2009 and 2015.

Table A8:

Broadband internet and income inequality – department-level results.

(28)(29)(30)(31)(32)(33)(34)(35)(36)(37)
Ginilnd1lnd2lnd3lnd4lnd5lnd6lnd7lnd8lnd9
Penbb_−0.074*0.859**0.424*0.2760.207−0.0700.1460.1400.1330.129
(0.040)(0.374)(0.227)(0.185)(0.158)(0.093)(0.143)(0.147)(0.150)(0.153)
Shbac−0.285**−0.0250.6020.6620.4040.822**0.1720.058−0.036−0.195
(0.143)(1.536)(0.826)(0.635)(0.501)(0.357)(0.411)(0.412)(0.410)(0.418)
Shold0.260***−1.576*−0.978**−0.726*−0.620*−0.301−0.418−0.421−0.336−0.211
(0.090)(0.845)(0.481)(0.388)(0.336)(0.283)(0.322)(0.322)(0.326)(0.344)
Pop_density0.023**−0.152**−0.093**−0.078*−0.052−0.031−0.034−0.028−0.0240.002
(0.009)(0.069)(0.046)(0.042)(0.035)(0.031)(0.033)(0.034)(0.037)(0.038)
Dep. FEYesYesYesYesYesYesYesYesYesYes
Year FEYesYesYesYesYesYesYesYesYesYes
Region × year FEYesYesYesYesYesYesYesYesYesYes
Observations655655655655655655655655655655

  1. Significant at 10% level (*), 5% level (**) or 1% level (***). Robust standard errors in parentheses. Estimations from a panel of 94 departments. All specifications include department and year fixed effects, as well as fixed effects of the interaction between regions and years. Variable definitions are as follow: fixed broadband subscriber in percentage of population (Penbb, 0–1); gross domestic product at the regional level (GDP, billion euros); high-school completion rates (Shbac, 0–1); share of population above 65 years old (Shold, 0–1); and population density (Popdens, thousands).

Figure A6: Impact of broadband internet uptake on department-level income deciles.

Figure A6:

Impact of broadband internet uptake on department-level income deciles.

References

Acemoglu, D., and P. Restrepo. 2018. “The Race between Man and Machine: Implications of Technology for Growth, Factor Shares, and Employment.” The American Economic Review 108 (6): 1488–542. https://doi.org/10.1257/aer.20160696. Search in Google Scholar

Acemoglu, D., and P. Restrepo. 2019. “Automation and New Tasks: How Technology Displaces and Reinstates Labor.” The Journal of Economic Perspectives 33: 3–30. https://doi.org/10.1257/jep.33.2.3. Search in Google Scholar

Akerman, A., I. Gaarder, and M. Mogstad. 2015. “The Skill Complementarity of Broadband Internet.” Quarterly Journal of Economics 130 (4): 1781. https://doi.org/10.1093/qje/qjv028. Search in Google Scholar

Angrist, J. D., and J.-S. Pischke. 2008. Mostly Harmless Econometrics: An Empiricist’s Companion. Princeton University Press. Search in Google Scholar

Autor, D. H., L. F. Katz, and A. B. Krueger. 1998. “Computing Inequality: Have Computers Changed the Labor Market?” Quarterly Journal of Economics 113: 1169–213. https://doi.org/10.1162/003355398555874. Search in Google Scholar

Autor, H. D., and D. Dorn. 2013. “The Growth of Low-Skill Service Jobs and the Polarization of the US Labor Market.” The American Economic Review 103: 1553–97. https://doi.org/10.1257/aer.103.5.1553. Search in Google Scholar

Autor, H. D., F. Levy, and R. J. Murnane. 2003. “The Skill Content of Recent Technological Change: An Empirical Exploration.” Quarterly Journal of Economics 118: 1279–333. https://doi.org/10.1162/003355303322552801. Search in Google Scholar

Bertschek, I., D. Cerquera, and G. J. Klein. 2013. “More Bits–More Bucks? Measuring the Impact of Broadband Internet on Firm Performance.” Information Economics and Policy 25 (3): 190–203. https://doi.org/10.1016/j.infoecopol.2012.11.002. Search in Google Scholar

Bertschek, I., W. Briglauer, K. Huschelrath, B. Kauf, and T. Niebel. 2016. “The Economic Impacts of Telecommunications Networks and Broadband Internet: A Survey.” Review of Network Economics 14: 201–227. Search in Google Scholar

Brice, L., P. Croutte, P. Jauneau-Cottet, and S. Lautie. 2015. “Barometre du Numerique.” In Report. CREDOC. Search in Google Scholar

Czernich, N. 2014. “Does Broadband Internet Reduce the Unemployment Rate? Evidence for Germany.” Information Economics and Policy 29: 32–45. https://doi.org/10.1016/j.infoecopol.2014.10.001. Search in Google Scholar

Czernich, N., O. Falck, T. Kretschmer, and L. Woessmann. 2011. “Broadband Infrastructure and Economic Growth.” The Economic Journal 121: 505–32. https://doi.org/10.1111/j.1468-0297.2011.02420.x. Search in Google Scholar

Falck, O., R. Gold, and S. Heblich. 2014. “E-lections: Voting Behavior and the Internet.” The American Economic Review 104: 2238–65. https://doi.org/10.1257/aer.104.7.2238. Search in Google Scholar

Forman, C., A. Goldfarb, and S. Greenstein. 2012. “The Internet and Local Wages: A Puzzle.” The American Economic Review 102: 556–75. https://doi.org/10.1257/aer.102.1.556. Search in Google Scholar

Goldfarb, A., and C. Tucker. 2019. “Digital Economics.” Journal of Economic Literature 57: 3–43. https://doi.org/10.1257/jel.20171452. Search in Google Scholar

Goos, M., and A. Manning. 2007. “Lousy and Lovely Jobs: The Rising Polarization of Work in Britain.” The Review of Economics and Statistics 89: 118–33. https://doi.org/10.1162/rest.89.1.118. Search in Google Scholar

Haller, S. A., and S. Lyons. 2015. “Broadband Adoption and Firm Productivity: Evidence from Irish Manufacturing Firms.” Telecommunications Policy 39 (1): 1–13. https://doi.org/10.1016/j.telpol.2014.10.003. Search in Google Scholar

Haller, S. A., and S. Lyons. 2019. “Effects of Broadband Availability on Total Factor Productivity in Service Sector Firms: Evidence from Ireland.” Telecommunications Policy 43 (1): 11–22. https://doi.org/10.1016/j.telpol.2018.09.005. Search in Google Scholar

Harrigan, J., A. Reshef, and F. Toubal. 2016. “The March of the Techies: Technology, Trade and Job Polarization in France, 1994-2007.” In Working Paper 22110. NBER. Search in Google Scholar

Hjort, J., and J. Poulsen. 2019. “The Arrival of Fast Internet and Employment in Africa.” The American Economic Review 109 (3): 1032–79. https://doi.org/10.1257/aer.20161385. Search in Google Scholar

Katz, R., and F. Callorda. 2018. “The Economic Contribution of Broadband, Digitization and ICT Regulation.” In Expert Report. International Telecommunications Union. Search in Google Scholar

Krueger, A. 1993. “How Computers Have Changed the Wage Structure: Evidence from Microdata, 1984-1989.” Quarterly Journal of Economics 108: 33–60. https://doi.org/10.2307/2118494. Search in Google Scholar

Malgouyres, C., T. Mayer, and C. Mazet-Sonilhac. 2019. “Technology-induced Trade Shocks? Evidence from Broadband Expansion in France.” In CEPR Working Paper. CEPR. Search in Google Scholar

Michaels, G., A. Natraj, and J. Van Reenen. 2014. “Has ICT Polarized Skill Demand? Evidence from Eleven Countries Over 25 Years.” The Review of Economics and Statistics 96: 60–77. https://doi.org/10.1162/rest_a_00366. Search in Google Scholar

Moreno-Galbis, E., and T. Sopraseuth. 2014. “Job Polarization in Aging Economies.” Labour Economics 27: 44–55. https://doi.org/10.1016/j.labeco.2013.12.001. Search in Google Scholar

Nardotto, M., T. Valletti, and F. Verboven. 2015. “Unbundling the Incumbent: Evidence from UK Broadband.” Journal of the European Economic Association 13: 330–62. https://doi.org/10.1111/jeea.12127. Search in Google Scholar

Roller, L.-H., and L. Waverman. 2001. “Telecommunications Infrastructure and Economic Development: A Simultaneous Approach.” The American Economic Review 91: 909–23. https://doi.org/10.1257/aer.91.4.909. Search in Google Scholar

World Bank Group. 2016. “Digital Dividends.” In World Development Report. World Bank Group. Search in Google Scholar

Received: 2020-08-29
Accepted: 2021-09-06
Published Online: 2021-09-14

© 2021 Walter de Gruyter GmbH, Berlin/Boston