The social-economic impact of fiber broadband penetration: a hype or a reality?

Transcription

1 The social-economic impact of fiber broadband penetration: a hype or a reality? Jie Li, Marco Forzati Networking and Transmission Laboratory, RISE Acreo, SE Kista, Sweden Tel: , Fax: , jie.li@ri.se ABSTRACT The social-economic impact of broadband penetration has been studied extensively. The general consensus is that basic broadband penetration (from none or upgraded from 256 kbps to the order of 2-6 Mbps) indeed has positive impact on economic growth and performance in terms of e.g. GDP and employment. Nevertheless, studies on the social-economic impact of high-speed broadband access (20-30 Mbps and beyond) are far more controversial showing a mixed picture. In this paper, we investigated fiber broadband access impact on the employment rate, population evolution, reduced driving distance per capita and new company registrations at the municipality level in Sweden using multiple regression analysis. Based on the availability of historical fiber network deployment data, analyses can be carried out by regressing current values of social-economic indicators under study on the fiber penetration levels 4 years earlier, hence reverse causality in the regression analyses can be tackled effectively. Our results show that, given all the other significant factors remain the same and with at least 90% confidence interval, 10% increase of fiber broadband penetration would result in 0.17% population increase, 0.32% employment rate enhancement, 28.7 km reduced driving distance per capita per year, and one new company among inhabitants. Keywords: fiber broadband penetration, social-economic impact, multiple regression analysis, social-economic indicators, municipality, causality 1. INTRODUCTION Based on the broadband deployment around the world, Internet has now become the information and communication base for every aspect of people s daily work and life. In parallel, the social-economic impact of broadband penetration had been studied extensively accompanying the expansion of the broadband infrastructure. Among these studies the general consensus is that basic broadband penetration (from none or upgraded from 256 kbps to the order of 2-6 Mbps) indeed has positive impact on economic growth and performance in terms of e.g. GDP and employment [1]-[9]. Nevertheless, with the ever-increasing mainly video-oriented content distributions and services, as well as with the ever-increasing new devices that are connected to the network, the underlying broadband infrastructure has been rapidly evolving to meet the online traffic requirements towards a Gigabit Society where, e.g., all the European households should have access to networks offering a download speed of at least 100 Mbps and can be upgraded to 1 Gbps by 2025 [10]. In the mean time, studies on the socialeconomic impact of high-speed broadband access (20-30 Mbps and beyond) are far more controversial showing a mixed picture [11]-[15]. In [11]-[12] the authors showed that faster broadband speed stimulated higher GDP per capita for OECD countries, and the effects of broadband speed were also greater in countries with lower income. In [13] the author examined if faster broadband was more effective in boosting the employment rate on the countylevel in US, and the results showed that compared to normal-speed broadband, faster broadband did not generate greater positive effects on employment. In another effort [14] the author also investigated relationship between the economic impacts in terms of county-level employment and number of firms in US and fast broadband represented by fiber-based internet access, and the results showed a positive association. Even so, another study investigating the social economic impact of fiber-based broadband access on the municipality level in Sweden showed that increased fiber coverage among households was estimated to have a negative effect on municipal employment whereas increased coverage among workplaces was only weakly positively related to employment [15]. In this study, we continue the effort in investigating the social economic impact of fiber-based broadband access using multiple regression analysis on the municipality level in Sweden. In our previous study we already observed an early stage statistically significant impact [16]. In this work, we further extend our previous study by investigating the fiber broadband access impact on the employment rate, population evolution, reduced driving distance per capita and new company registrations at the municipality level in Sweden. Reverse causality in the regression analyses was tackled effectively by regressing current values of social economic indicators under study on the fiber-penetration levels above a minimum threshold 4 years earlier. Our results show that fiber broadband access penetration indeed has statistically significant and positive impact on the studied social-economic indicators with at least 90% confidence interval. 1

2 2. METHODOLOGY 2.1. Social-economic indicator at the municipality level In this work, we study the social-economic impact of fiber broadband penetration in Sweden using multiple regression analysis. Social-economic indicator and fiber broadband penetration statistics at the municipality level were used in the analysis in order to achieve a good balance between a sufficiently large data sample size (with a maximum of 290 observations corresponding to the total number of municipalities in Sweden) and a reasonably good data aggregation level. Since GDP statistics are only available at the national level, we used the following 4 parameters as the social-economic impact indicators of fiber broadband penetration at the municipality level: Population evolution Employment rate Average driving distance per capita per year New company registration per 1000 capita per year 2.2. Regression modelling and methods For the regression analysis, we adopted the same approach in [1] by regressing the current values of the studied social-economic indicators on fiber penetration levels a few years earlier. More specifically, the multiple regression model is represented as: Y(t) = 0 + FB FB(t T 0 ) + Y Y(t T 0 ) + X i (t T 0 ) + + e (1) where Y(t) denotes the studied social-economic indicator (e.g. average driving distance per capita) at the current level, and FB(t T 0 ) denotes the fiber broadband penetration level at T 0 years earlier. In the meantime, the studied indicator value Y(t T 0 ) at the same year as fiber broadband penetration was used as the major control variable in order to differentiate the impact of fiber broadband penetration (over T 0 years). Furthermore, a set of other control variables X i (t T 0 ) were employed if they were found to be statistically significant to further isolate the impact of fiber broadband penetration. Eq. (1) was used to model the regression analysis for the average driving distance and new company registration per year. For population evolution and employment rate, it has been considered more appropriate to study the growth rate for such indicators that can vary dynamically between municipalities [1]. Accordingly, the multiple regression model may be modified as: Ln(Y(t)/Y(t T 0 )) = 0 + FB FB(t T 0 ) + X i (t T 0 ) + + e (2) where the natural logarithmic of the ratio Y(t)/Y(t T 0 ) for the studied social-economic indicator was used to model the corresponding grow rate (over T 0.years) In doing the regression analyses, ordinary least-squares estimates (OLS) was used in the first place to investigate the significance and extract the regression coefficients of independent variables. In addition, Huber- White heteroscedastic consistent estimates were also carried out to further verify and consolidate the significance of the independent variables with robust error estimation Sample data timing selection In this study the current year t was set at year 2015 at which the latest social-economic indicator statistics were available at the time of study. The time period T 0, on the other hand, was set to 4 years, i.e., the fiber broadband penetration level in year 2011 was used to evaluate its social-economic impact. This is based on two considerations: A 4-year period of time would allow the time required for real end users to actually subscribe to the broadband service available after the fiber network deployment, and the social-economic impact would only manifest after end users had actually used the services that fiber broadband network enabled [1][8][9]; Large scale fiber broadband access network deployment in Sweden was initiated around year 2007, i.e., 4-year earlier than Hence, we expect that the corresponding social-economic indicators and other control variables in year 2011 would be hardly correlated to fiber broadband network deployment earlier. As an illustration, Figure 1 shows the household fiber broadband penetration map at the municipality level in Sweden in year 2011, and Figure 2 shows the corresponding histogram distribution. We see that while the household fiber broadband penetration already reached a highest level up to 88% in one municipalities, the penetration level distribution was broad, with ~ 25% of municipalities at penetration levels even below 8%. 2

3 Figure 1 Municipality household fiber broadband penetration map, 2011 Sweden. Figure 2 Histogram distribution of municipality household fiber broadband penetration, 2011 Sweden Sample municipality selection As mentioned earlier, it is widely recognized that regression analysis at the municipality level retains a good balance between a sufficiently large data sample size and a reasonably good data aggregation level. However, from Figure 1 & Figure 2 one can see that among the 290 municipalities in Sweden, the fiber broadband penetration distribution was relatively broad with ~ 25% of them at levels even below 8% in year Meanwhile, earlier study [15] shows that at relatively low fiber penetration levels, the relationship between fiber broadband coverage and social-economic indicators (e.g. employment) was weak, which suggests that fiber broadband penetration s impact on the social-economic development of the society would only take effect after a certain minimum/threshold level is reached (so that sufficiently large number of end users really start to use the broadband access provided by the deployed fiber broadband network). Based on these considerations, in this work, we set a threshold of fiber penetration level at 8.5%, i.e., all the municipalities that had fiber penetration levels below 8.5% in 2011 were excluded in the analysis. This resulted in a total of 214 municipalities that were above the threshold and hence were selected for the multiple regression analysis in this work. Note here that the threshold at 8.5% was chosen based on the least municipality fiber penetration level in year 2016 (the latest statistics available at the time of study) in Sweden. 3

4 2.5. Variables used in the multiple regression analysis Table 1 lists the dependent variables for the multiple regression analysis in this work. As described in section 2.2, for average driving distance and new company registration per year, we used the respective statistics in year 2015 directly. Meanwhile, for the population evolution and employment rate, we used the natural logarithm of the ratio of corresponding values in year 2015 over year 2011 to model their growth rates. Table 2 lists, apart from the fiber penetration level in year 2011, the independent variables used in the analysis that were found to contribute significantly to at least one of the 4 studied social-economic indicators. Either statistics in year 2011 or the growth rate dating back by 4 years were used in order to tackle the backward causality. Note that during the study other social-economic statistics at the municipality level were also investigated, e.g. the local tax rate etc., but were found insignificant in the regressions to the studied social-economic indicators. Table 1 Dependent variables lnpop1115 lnemp1115 ln (population_2015/population_2011) ln (employment_2015/employment_2011) driven_km_2015 Average driving distance per inhabitant year 2015 new_nompany_2015 New company registration per 1000 inhabitants year 2015 Table 2 Independent variables FP_2011 Fiber penetration household year 2011 delta_pop10_ year population increase rate year 2011 Urb_2011 Municipality urbanization level 2011 gincom0711 Average employee income growth rate from year 2007 to 2011 gemp0711 Employment rate growth rate from year 2007 to 2011 driven_km_2011 Average driving distance per inhabitant year 2011 New_Company_2011 New company registration per 1000 inhabitants year Sample data sources The data used in this study was based on publicly available statistical data from Statistics Sweden (SCB), The Swedish Post and Telecom Authority (PTS), Swedish Association of Local Authorities and Regions (SALAR Kolada), and Swedish Regional Development and Cooperation in the Environmental Objective System (RUS). 3. RESULTS AND DISCUSSIONS Table 3 - Table 6 show the OLS regression results for population evolution, employment rate, average driving distance per capita per year and new company registration per 1000 inhabitants per year, respectively. In addition to ordinary OLS regressions, Huber-White heteroscedastic consistent estimates were also carried out and verified the significance of fiber broadband penetration with robust error estimation. As an illustration, Figure 3 - Figure 6 show the isolated effect of fiber penetration by taking away all the other predicted independent variable s influence on the studied social-economic indicators. We see that for all the four social-economic indicators, fiber broadband access penetration indeed shows statistically significant positive impact with at least 90% confidence interval. More specifically, given all the other significant factors remain the same, 10% increase of fiber broadband penetration would result in: 0.17% population increase; 0.32% employment rate increase; 28.7 km reduced driving distance per capita per year; new company registrations per year among 1000 inhabitants, or equivalently one new company among inhabitants. The attentive readers may notice that our result on employment rate growth contradicts what was reported in [15] where another effort on trying to find a relationship between fiber deployment and employment rate on the municipality level in Sweden was carried out, and it was found that increased fiber coverage among households was estimated to have a negative effect on municipal employment. However, in [15] the work looked for the 4

5 relationship between employment and fiber coverage during As we already described earlier that large scale fiber broadband access network deployment in Sweden was initiated around year 2007, and even at year 2011, the penetration levels of ~ 25% of municipalities were below 8%. In addition, due to the time required for fiber penetration to take effect, one may not expect that the employment level to be enhanced simultaneously with the fiber penetration, especially at early stage of fiber broadband network deployment. Conversely in this work, we looked at the employment level 4 years later than 2011, and took away those municipalities with fiber penetration levels lower than 8% (76 municipalities) so that the potential social-economic benefits of fiber broadband penetration can manifest. Table 3 Regression results - population evolution Dependent variable lnpop1115 R-squared Observations 214 Independent variables Coefficient t-statistic Confidence interval FP_ % delta_pop10_ % Urb_ % beta_ Table 4 Regression results - Employment rate Dependent variable lnemp1115 R-squared Observations 214 Independent variables Coefficient t-statistic Confidence interval FP_ % gemp % gincom % delta_pop10_ % beta_ Table 5 Regression results - Average driving distance Dependent variable driven_km_2015 R-squared Observations 214 Independent variables Coefficient t-statistic Confidence interval FP_ % driven_km_ % delta_pop10_ % beta_

6 Table 6 Regression results - new company registration Dependent variable new_company_2011 R-squared Observations 214 Independent variables Coefficient t-statistic Confidence interval FP_ % New_Company_ % gincom % delta_pop10_ % beta_ Figure 3 Population growth versus fiber penetration by taking away all the other predicted independent variable s influence. Figure 4 Employment rate growth versus fiber penetration by taking away all the other predicted independent variable s influence. 6

7 Figure 5 Driving distance per capita per year versus fiber penetration by taking away all the other predicted independent variable s influence. Figure 6 New company registration per 1000 inhabitants versus fiber penetration by taking away all the other predicted independent variable s influence. The other important aspect in regression analysis is the causal relationship between independent control variables and the dependent variable under study. From the approach we adopted in this work, the causal relationship between fiber broadband penetration and the social-economic indicators under study is expected. This is based on the availability of historical fiber network deployment data, and analyses can be carried out by regressing current values of social economic indicators under study on the fiber-penetration levels 4 years earlier, hence reverse causality in the regression analyses can be tackled effectively. This differs most of previous studies limited by the availability of sample data sets that lacked sufficient time difference between broadband penetration and social-economic indicators (e.g. employment rate etc.), hence a causal relationship can not be claimed without a control-match group test [4][13][14]. 4. CONCLUSIONS In this work, we investigated fiber broadband access impact on the employment rate, population evolution, reduced driving distance per capita and new company registrations at the municipality level in Sweden using multiple regression analysis. Based on the availability of historical fiber network deployment data, analyses can be carried out by regressing current values of social-economic indicators under study on the fiber penetration levels 4 years earlier, hence reverse causality in the regression analyses can be tackled effectively. Our results show that, given all the other significant factors remain the same and with at least 90% confidence interval, 10% increase of fiber broadband penetration would result in 0.17% population increase, 0.32% employment rate enhancement, 28.7 km reduced driving distance per capita per year, and one new company among inhabitants. 7

8 ACKNOWLEDGEMENT This work was supported by The Swedish Broadband Forum. REFERENCES [1] William H. Lehr et al.:, "Measuring Broadband s Economic Impact", 33rd Research Conference on Communication, Information, and Internet Policy (TPRC), Arlington, VA, September 23-25, [2] David Shideler et al.:, The Economic Impact of Broadband Deployment in Kentucky, Federal Reserve Bank of St. Louis Regional Economic Development, 3(2), pp , [3] Katz, Raul L. et al.: The impact of broadband on jobs and the German economy, Intereconomics, ISSN X, Springer, Heidelberg, Vol. 45, Iss. 1, pp , [4] Krishna Jayakar and Eun-A Park, Broadband Availability and Employment: An Analysis of County-Level Data from the National Broadband Map, Journal of Information Policy, Vol. 3, pp , [5] Hilal Atasoy, The Effects of Broadband Internet Expansion on Labor Market Outcomes, ILR Review, V. 66 issue 2, pp , April 1, [6] Bertschek, Irene et al.:, More bits - more bucks? Measuring the impact of broadband internet on firm performance, DICE Discussion Paper, No. 86, ISBN , 2013 [7] Gruber, H. et al.:, Broadband access in the EU: An assessment of future economic benefits, 24th European Regional Conference of the International Telecommunication Society, Florence, Italy, October [8] Brian Whitacre et al.:, Broadband s contribution to economic growth in rural areas: Moving towards a causal relationship, Telecommunications Policy 38, pp , [9] Brian Whitacre et al.:, Does rural broadband impact jobs and income? Evidence from spatial and firstdifferenced regressions, The Annals of Regional Science, Volume 53, Issue 3, pp , November [10] EU Policy: Connectivity for a European Gigabit Society, [11] Rohman, Ibrahim Kholilul and Bohlin, Erik, Does Broadband Speed Really Matter for Driving Economic Growth? Investigating OECD Countries, available at SSRN: April 4, [12] Kongaut, Chatchai and Bohlin, Erik: Impact of broadband speed on economic outputs: An empirical study of OECD countries, 25th European Regional Conference of the International Telecommunications Society (ITS), Brussels, Belgium, June [13] Bai, Yang, The Faster, the Better? The Impact of Internet Speed on Employment. 44th Research Conference on Communication, Information and Internet Policy (TPRC 44), April [14] Paul Lapointe, Does Speed Matter? The Employment Impact of Increasing Access to Fiber Internet, Graduate thesis of Georgetown University-Graduate School of Arts & Sciences, [15] Erik Grenestam, Business or Pleasure: Broadband and Employment in Swedish Municipalities , Master thesis of Department of Economics, Lund University, [16] Marco Forzati et al.:, Early effects of FTTH/FTTx on employment and population evolution population evolution, Proceedings of the 11th Conference of Telecommunication, Media and Internet Techno- Economics (CTTE), Athens,