Federal Communications Commission FCC 20-188 APPENDIX G INTERNATIONAL BROADBAND DATA REPORT APPENDICES APPENDIX G-1: Country List APPENDIX G-2: Broadband Speed and Performance Comparisons APPENDIX G-3: Broadband Pricing Comparisons APPENDIX G-4: High-Speed Broadband Deployment Comparison with Europe Federal Communications Commission FCC 20-188 APPX. G-1 Country List 1. The Commission must include  information comparing the extent of broadband service capability (including data transmission speeds and price for broadband service capability) in a total of 75 communities in at least 25 countries abroad for each of the data rate benchmarks for broadband service utilized by the Commission to reflect different speed tiers. 1 We must choose international communities comparable to various communities in the United States with respect to population size, population density, topography, and demographic profile.2 The Commission is required to include  a geographically diverse selection of countries and  communities including the capital cities of such countries. 3 2. In the Figure below, we list the United States and the 35 foreign countries selected for purposes of the International Broadband Data Report and identify the countries that are included in an Appendix with an  X. These 35 countries and the United States account for 36 of the 37 Organisation for Economic Co-operation and Development (OECD) Member countries.4 We refer to these countries as the  comparison countries. For the fixed and mobile broadband price comparisons, we rely on a smaller subset of 25 comparison countries.5 For the fixed and mobile deployment comparison, we rely on the 26 European comparison countries (EU26). 1 47 U.S.C. § 1303(b)(1); see also Section 401 of the Repack Airwaves Yielding Better Access for Users of Modern Services Act of 2018 (RAY BAUM S Act), Pub. L. No. 115-141, 132 Stat. 1087 (codified at 47 U.S.C. § 163) (RAY BAUM S Act). 2 47 U.S.C. § 1303(b)(2). Fig. III.A.8 depicts how the average proportion of the population with coverage by fixed terrestrial services by speed tier varies with median household income, population density, and household poverty rate at the census block group level. On average, deployment is highest in census blocks with the highest median household incomes, the highest population densities and the lowest household poverty rates. 3 Id. 4 Colombia is the only OECD country not included as a comparison country because of unavailability of the data before it became an OECD member country in April 2020. 5 The countries excluded from the pricing analysis are: Chile, Hungary, Israel, Japan, Lithuania, Poland, Slovakia, Slovenia, South Korea, and Turkey. Due to the time intensive nature of collecting both fixed broadband and mobile broadband pricing data from multiple providers in each country, we limited the pricing analysis to the same countries analyzed in the 2018 International Broadband Data Report except for Chile, Japan, and South Korea. International Comparison Requirements Pursuant to the Broadband Data Improvement Act; International Broadband Data Report, GN Docket No. 17-199, Sixth Report, 33 FCC Rcd 978 (IB 2018) (2018 International Broadband Data Report). 2 Federal Communications Commission FCC 20-188 Appx. G-2: Speed and Appx. G-3: Appx. G-4: Country Performance (Ookla) Price Deployment Australia (AU) X X Austria (AT) X X X Belgium (BE) X X X Canada (CA) X X Chile (CL) X Czech Republic (CZ) X X X Denmark (DK) X X X Estonia (EE) X X X Finland (FI) X X X France (FR) X X X Germany (DE) X X X Greece (GR) X X X Hungary (HU) X X Iceland (IS) X X X Ireland (IE) X X X Israel (IL) X Italy (IT) X X X Japan (JP) X Latvia (LV) X X X Lithuania (LT) X X Luxembourg (LU) X X X Mexico (MX) X X Netherlands (NL) X X X New Zealand (NZ) X X Norway (NO) X X X Poland (PL) X X Portugal (PT) X X X Slovakia (SK) X X Slovenia (SI) X X South Korea (KR) X Spain (ES) X X X Sweden (SE) X X X Switzerland (CH) X X X Turkey (TR) X United Kingdom (GB) X X X United States (US) X X X 3 Federal Communications Commission FCC 20-188 APPX. G-2 Broadband Speed and Performance Comparisons 1. This section of the International Broadband Data Report Appendix presents a comparison of fixed broadband and mobile broadband performance metrics in terms of  data transmission speeds (download and upload speeds) and latency for the United States and 35 other comparison countries. The main analysis relies solely on Ookla Speed Test datasets for both speed and latency. For fixed broadband we consider all technologies, and for mobile broadband we only consider 4G LTE, because it is the baseline industry standard for the marketing of mobile broadband service.6 Compared to previous International Broadband Data Reports, in this report we present a deeper analysis of download and upload speeds, as well as an analysis of latency, with a five-year time horizon for fixed broadband services and a four-year time horizon for mobile broadband services.7 We also present the data visually with new maps and graphs for more countries. We rank speeds from fastest (1st) to slowest (36th) and latency from shortest (1st) to longest (36th). In section IV, we present additional mobile broadband data on download speeds for 3G/4G and 5G and 5G availability as calculated by OpenSignal. I. FIXED BROADBAND SPEED AND LATENCY RESULTS 2. Figure G-1. U.S. mean download speed rankings improved significantly to a ranking of 5th among the 36 comparison countries for the past two years, up from a ranking of 9th in 2017 and 2016, and 14th (of 35 countries) in 2015.8 In 2019, the mean download speed for the United States was 119.6 Mbps which almost tripled the mean download speed in 2015 of 40.4 Mbps. Iceland had the fastest mean download speed of the countries in 2019 with a mean download speed of 164.1 Mbps. 3. Figure G-2. U.S. mean upload speed rankings were relatively stable for the last five years, with the United States ranking 17th of the 36 comparison countries for the past two years, 16th in 2017, 17th in 2016, and 18th (of 35 countries) in 2015.9 The mean upload speed in 2019 for the United States was 46.3 Mbps, compared to the fastest mean upload speed of 169.4 Mbps in Iceland. 4. Figure G-3. U.S. mean latency rankings were consistent over the comparison period, ranking 24th in 2015 and 2019. The mean latency for the United Sates in 2019 was 23.7 ms, compared to Latvia s mean latency of 14.2 ms in 2019, which was ranked the best among the countries. 5. Figure G-4. The mean download speed in Washington D.C. in 2019 was 119.6 Mbps, ranked 30th among the 86 country and state capital cities. The highest ranked U.S. capital city in 2019 was Dover, Delaware which ranked 3rd with a mean download speed of 155.7 Mbps. Other U.S. capital cities in the top ten in 2019 included Austin, Texas at 4th, Raleigh, North Carolina at 5th, Lincoln, Nebraska at 6th, Boston, Massachusetts at 9th and Salt Lake City, Utah at 10th. 6. Figure G-5. This graph shows the distribution of download speeds for each country. The top of each color bar represents the corresponding 25th, 50th, and 75th percentiles. The 25th, 50th, and 75th percentiles of download speeds in the United States were 33.2 Mbps, 77.9 Mbps and 159.5 Mbps, respectively. 6 Prior International Broadband Data Reports considered all mobile technologies available. This report has been updated to only present Ookla Mobile 4G LTE data. 7 We use a shorter time horizon for mobile broadband than for fixed broadband because the Mobile 4G LTE data is only available beginning in 2016. 8 For fixed broadband download speeds, Luxembourg is excluded in 2015. 9 For fixed broadband upload speeds, Luxembourg is excluded in 2015. 4 Federal Communications Commission FCC 20-188 7. Figure G-6. This graph depicts mean download speeds in G710 countries and South Korea from 2015 to 2019. U.S. mean download speeds increased at a similar trajectory as other G7 countries, with download speeds increasing from 40.4 Mbps in 2015 to 119.6 Mbps in 2019. South Korea had the fastest mean download speed of these countries at 151.6 Mbps in 2019. 8. Figure G-7.11 Test counts in the United States increased by 36% from 125.6 million in 2015 to 171.3 million in 2019. The number of cities with fixed broadband tests remained roughly constant in the United States during the five-year time horizon. 9. Figure G-8. Mean download speeds in 2019 in North America ranged from 31.5 to 119.6 Mbps.12 The top six countries had a range of download speeds from 118.4 to 164.1 Mbps, whereas the bottom six countries had a range from 22.8 Mbps to 50.4 Mbps. Western Europe and Scandinavia generally had higher download speeds than Eastern and Southern Europe. 10. Figure G-9. Mean upload speeds in 2019 in North America ranged from 13.2 to 46.4 Mbps.13 The top six countries had a range of download speeds from 87.9 to 169.4 Mbps, whereas the bottom six countries had a range from 6.0 to 16.5 Mbps. 11. Figure G-10. Mean latency in 2019 was between 20.5 ms and 32.3 ms for North American countries.14 Mean latency in 2019 was the lowest in the Northern and Eastern European countries of Iceland, Latvia, and Lithuania, which had latencies ranging from 14 ms to 15 ms. II. MOBILE BROADBAND  4G LTE RESULTS 12. Figure G-11. For mean download speeds, the United States ranked 25th among the 36 comparison countries in 2019, with a mean download speed of 37.0 Mbps, increasing from 21.4 Mbps with a ranking of 35th in 2016. In 2019, Iceland had the highest mean download speed at 78.6 Mbps, whereas Chile had the lowest at 21.2 Mbps. 13. Figure G-12. U.S. mean upload speeds consistently ranked 35th among the 36 comparison countries for the past four years, with the speeds increasing from 8.8 Mbps in 2016 to 11.1 Mbps in 2019. Iceland, the country with the fastest mean upload speed in each of the past four years, had a 22.6 Mbps upload speed in 2019--an increase from 19.3 Mbps in 2016. 14. Figure G-13. U.S. mean latency ranked 34th among the 36 comparison countries in 2019 at 46.7 ms. Iceland ranked 1st in 2019 with latency of 21.1 ms. 15. Figure G-14. The mean download speed in Washington D.C. in 2019 was 44.9 Mbps, which was 25th of the 86 country and state capital cities. The highest ranked U.S. state capital city in 2019 was Annapolis, Maryland which ranked 8th with a mean download speed of 55.6 Mbps. No other U.S. state capitals were among the top ten ranked capital cities. 16. Figure G-15. This graph shows the distribution of download speeds for each country. The top of each color bar represents the corresponding 25th, 50th, and 75th percentiles. The 25th, 50th and 10 The G7 or Group of Seven is an informal group of industrialized democracies whose leaders meet annually to discuss various issues. Council on Foreign Relations, The G7 and the Future of Multilateralism (Aug. 20, 2019), https://www.cfr.org/backgrounder/g7-and-future-multilateralism. 11 In mid-2016, Ookla adjusted the method by which they perform geolocation, resulting in subnational geographies (e.g., cities) being potentially incomparable between 2015 and 2017. This methodological change explains why the number of cities per country varies significantly for some countries between these years. 12 Each country s mean fixed broadband download speed values are reported in Fig. G-1. 13 Each country s mean fixed broadband upload speed values are reported in Fig. G-2. 14 Each country s mean fixed broadband latency values are reported in Fig. G-3. 5 Federal Communications Commission FCC 20-188 75th percentiles for download speed in the United States were 10.7 Mbps, 26.6 Mbps, and 52.3 Mbps, respectively. 17. Figure G-16. U.S. mean mobile broadband download speed increased at a similar pace as in G7 countries during the past few years, most closely mirroring Japan s trend in download speeds. Canada experienced the fastest growth in mean download speed over the last four years, increasing from 36.2 Mbps in 2016 to 71.3 Mbps in 2019. 18. Figure G-17. Test counts in the United States for 4G LTE increased by 25% from 14.3 million in 2016 to 17.9 million in 2019. The number of cities with 4G LTE tests in the United States increased modestly by about 1,900 cities during the same period. 19. Figure G-18. Mean 4G LTE download speeds in 2019 in North America ranged from 27.4 to 71.3 Mbps.15 The top six countries had a range of download speeds from 61.2 to 78.6 Mbps while the bottom six countries had a range from 21.2 to 33.4 Mbps. 20. Figure G-19. Mean 4G LTE upload speeds in 2019 in North America ranged from 11.1 to 15.9 Mbps.16 The top six countries had a range of upload speeds from 17.0 to 22.6 Mbps, whereas the bottom six countries had a range from 9.8 to 12.5 Mbps. 21. Figure G-20. Mean 4G LTE latency in 2019 was between 34.1 and 50.0 ms for North American countries.17 In Europe, the lowest mean latency was concentrated in Eastern European countries, such as Estonia and Hungary. III. DATA AND ANALYSIS 22. Data. The FCC obtains aggregated fixed broadband and mobile broadband speed and latency datasets from Ookla for the United States and the 35 comparison countries. The annual fixed datasets are aggregated to the city-platform level; whereas the annual mobile datasets are aggregated to the city-platform-technology level.18 Prior to aggregating the data, Ookla applies a set of cleaning and filtering rules to ensure the quality of the data and to further control for certain variables and remove invalid test results.19 The Ookla Speed Test data are user-generated, meaning the user manually chooses to run each speed test. Therefore, the results from these tests may represent nontypical situations (e.g. when the user is experiencing congestion issues). Because the tests are not taken randomly, they may not represent consumers typical broadband experience. 23. Analysis. In our analysis, we consistently aggregate the data to higher levels using sample counts as a weight.20 First, we aggregate over platforms for fixed broadband and mobile  4G 15 Each country s mean 4G LTE download speed values are reported in Fig. G-11. 16 Each country s mean 4G LTE upload speed values are reported in Fig. G-12. 17 Each country s mean 4G LTE latency values are reported in Fig. G-13. 18 For 2015, the annual fixed broadband dataset is aggregated to the city-level. 19 We do not report fixed broadband speeds for Luxembourg for 2015, as these values are potentially incomparable with later years. This is due to adjustments in the method by which Ookla performs geolocation, as well as certain methodological changes in their cleaning and filtering rules. Further, for the 2018 and 2019 mobile 4G LTE data, Ookla adopted additional minor changes to their cleaning and filtering methodology. For more information regarding Ookla s methodology, see Brian Connolly, How Ookla Ensures Accurate Reliable Data: A Guide to Our Metrics and Methodology (Updated for 2020), Ookla (Apr. 28, 2020), https://www.speedtest.net/insights/blog/how- ookla-ensures-accurate-reliable-data-2020/. 20 In the 2018 Communications Marketplace Report, we weighted summary statistics by the number of tests because the sample count was unavailable in earlier datasets. Communications Marketplace Report et al., GN Docket No. 18-231, Report, 33 FCC Rcd 12558, 12560-61, paras. 2-4 (2018) (2018 Communications Marketplace Report). Results from prior International Broadband Data Reports will not match exactly due to this change in methodology; (continued& .) 6 Federal Communications Commission FCC 20-188 LTE broadband. Then, we aggregate data over cities to the state or country level. Ideally, we would have an observation for each broadband subscriber or at least a representative sample of all broadband users, but as subscribers choose to opt-in to Ookla s service, this is unlikely to be the case. For example, if the ratio of Ookla users relative to broadband subscribers is greater in urban areas compared to rural areas, it may produce an urban bias in the dataset at the country level. 24. The 2015 fixed broadband speed dataset is aggregated to a higher level (over platforms) by Ookla.21 Given Ookla s aggregation methodology, the 2015 city-level data are not perfectly comparable to the 2016-2019 city-level data. However, we do not suspect these discrepancies to affect the results significantly. Similarly, our city-level and country-level results are not directly comparable to any city-level and country-level results published by Ookla because Ookla applies their aggregation methodology to the given level of aggregation before calculating statistics, whereas we must weight the lower level of disaggregation by sample count to aggregate the data to higher levels. however, in the 2018 Communications Marketplace Report Data Update, we used the new methodology, and those results will be consistent with this report. FCC, Communications Marketplace Report  Updates, https://www.fcc.gov/communications-marketplace-report-updates (last visited Oct. 27, 2020). 21 For fixed broadband, speed tests include technologies such as DSL/Copper, Cable Modem, Fiber, Satellite, Fixed Wireless, and Other (e.g., Electric Power Line). 7 Federal Communications Commission FCC 20-188 Fig. G-1: Fixed Broadband Mean Download Speed by Country (2015-2019) 2015 2016 2017 2018 2019 Country Rank Mbps Rank Mbps Rank Mbps Rank Mbps Rank Mbps Australia 30 17.7 32 18.7 33 23.4 33 30.0 33 38.7 Austria 26 29.5 29 28.4 31 32.2 32 37.4 32 43.8 Belgium 12 41.1 18 43.7 18 51.8 21 59.1 23 72.2 Canada 23 31.3 17 43.8 13 60.6 10 81.9 7 114.3 Chile 31 15.9 31 25.3 30 32.8 26 48.9 19 77.2 Czech Republic 24 31.1 27 32.2 28 34.8 29 41.2 31 50.4 Denmark 8 45.8 10 51.6 10 66.4 13 81.0 12 103.3 Estonia 20 32.4 22 40.2 25 42.4 28 41.8 29 55.4 Finland 15 40.0 20 43.2 23 46.4 25 51.5 25 66.0 France 10 42.7 14 44.6 15 56.5 15 79.1 9 114.0 Germany 19 33.4 24 35.0 21 46.9 22 56.0 24 71.1 Greece 35 10.6 36 11.8 36 13.9 35 18.6 35 23.8 Hungary 13 40.6 12 49.8 6 77.0 3 102.3 3 124.3 Iceland 9 45.3 2 82.4 2 124.1 1 153.5 1 164.1 Ireland 21 32.2 15 44.1 20 50.7 20 59.6 20 76.4 Israel 25 30.5 26 33.0 26 40.6 18 62.2 21 76.3 Italy 34 11.0 34 16.9 32 25.5 31 38.3 30 52.2 Japan 1 85.2 5 60.4 7 72.3 11 81.5 13 97.7 Latvia 6 50.8 8 55.1 17 54.4 19 60.1 16 90.6 Lithuania 7 50.3 6 59.9 3 99.6 9 82.2 17 89.5 Luxembourg 16 43.8 16 56.1 14 79.6 10 109.1 Mexico 32 14.0 33 17.8 34 20.6 34 24.2 34 31.5 Netherlands 5 55.0 7 58.6 8 71.0 12 81.4 14 96.4 New Zealand 27 29.1 21 41.1 14 58.5 16 73.3 15 91.1 Norway 16 38.7 11 49.8 11 65.8 8 85.3 11 105.8 Poland 28 25.7 25 34.7 24 45.7 23 54.5 22 76.0 Portugal 18 35.7 19 43.3 19 51.8 17 69.4 18 88.4 Slovakia 22 31.6 28 31.3 27 38.1 27 45.1 27 58.6 Slovenia 29 24.7 30 28.1 29 33.1 30 39.8 28 57.4 South Korea 2 69.4 1 83.6 1 128.5 2 119.8 2 151.6 Spain 11 41.9 13 48.9 12 61.9 7 87.7 8 114.1 Sweden 3 61.7 3 67.4 4 81.6 4 96.9 6 118.4 Switzerland 4 58.7 4 63.4 5 77.4 6 92.1 4 120.6 Turkey 33 12.6 35 14.8 35 16.0 36 18.4 36 22.8 United Kingdom 17 36.3 23 37.2 22 46.6 24 52.6 26 61.0 United States 14 40.4 9 52.7 9 70.1 5 92.5 5 119.6 8 Federal Communications Commission FCC 20-188 Fig. G-2: Fixed Broadband Mean Upload Speed by Country (2015-2019) 2015 2016 2017 2018 2019 Country Rank Mbps Rank Mbps Rank Mbps Rank Mbps Rank Mbps Australia 28 5.9 33 5.2 32 8.2 31 11.6 30 16.9 Austria 24 7.0 27 8.5 28 9.8 32 11.4 33 14.9 Belgium 30 5.4 28 8.3 26 10.4 29 13.0 32 15.8 Canada 21 8.0 23 12.3 21 18.6 18 30.9 16 46.4 Chile 32 3.9 32 5.3 34 7.1 34 10.1 25 20.5 Czech Republic 12 15.7 20 16.5 20 19.3 21 21.2 23 25.9 Denmark 7 30.0 7 35.8 8 49.0 8 61.1 8 80.1 Estonia 11 18.0 13 23.9 15 27.1 19 28.3 19 40.3 Finland 16 13.9 18 17.3 19 19.9 20 22.0 20 29.0 France 15 14.7 16 18.3 18 24.1 15 37.4 12 66.6 Germany 29 5.5 30 7.2 27 9.8 28 13.5 28 18.6 Greece 35 1.5 36 2.2 36 2.9 36 4.2 36 6.0 Hungary 14 14.9 14 20.1 13 29.6 14 39.7 13 61.4 Iceland 5 36.7 2 78.2 1 129.7 1 160.1 1 169.4 Ireland 20 9.6 21 15.4 22 18.3 22 20.8 21 26.9 Israel 31 4.0 31 5.4 33 7.5 27 13.5 29 16.9 Italy 34 2.2 34 5.1 31 8.4 26 13.8 26 20.1 Japan 1 75.6 3 59.5 4 73.9 3 91.5 2 108.9 Latvia 4 45.8 5 54.8 5 54.3 9 60.5 5 92.2 Lithuania 3 46.2 4 55.3 3 85.7 4 74.4 7 82.7 Luxembourg 12 24.8 12 33.2 11 47.6 11 67.9 Mexico 27 6.4 26 8.6 30 8.9 33 10.3 34 13.2 Netherlands 9 23.3 10 27.4 11 33.4 12 41.3 15 48.6 New Zealand 19 12.3 15 18.6 14 29.2 13 40.1 14 55.2 Norway 8 25.2 8 34.5 7 49.4 7 62.2 9 79.0 Poland 22 7.9 24 11.1 24 14.2 23 17.9 22 26.2 Portugal 26 6.6 19 17.3 17 25.7 16 36.6 18 45.0 Slovakia 13 15.0 22 13.6 23 14.7 24 16.2 24 21.3 Slovenia 23 7.9 25 10.2 25 11.9 25 14.1 27 18.8 South Korea 2 60.8 1 80.9 2 127.9 2 98.4 3 105.1 Spain 17 12.9 11 25.9 10 43.4 5 71.3 4 98.9 Sweden 6 34.1 6 40.6 6 53.3 6 68.3 6 87.9 Switzerland 10 21.3 9 31.0 9 43.8 10 58.0 10 77.1 Turkey 33 3.1 35 3.6 35 3.9 35 5.7 35 7.0 United Kingdom 25 6.7 29 8.1 29 9.7 30 11.9 31 16.5 United States 18 12.7 17 17.9 16 26.9 17 34.6 17 46.3 9 Federal Communications Commission FCC 20-188 Fig. G-3: Fixed Broadband Mean Latency by Country (2015-2019) 2015 2016 2017 2018 2019 Country Rank ms Rank ms Rank ms Rank ms Rank ms Australia 33 48.5 35 49.6 31 40.0 32 32.3 30 24.7 Austria 20 31.6 23 31.3 26 29.6 27 28.9 28 24.2 Belgium 19 31.2 15 27.0 17 24.8 14 21.4 14 18.3 Canada 22 36.4 21 30.8 21 28.7 19 25.0 18 20.5 Chile 34 53.2 31 43.3 33 40.5 31 31.5 20 22.2 Czech Republic 9 26.4 12 25.5 15 24.0 15 22.4 16 19.4 Denmark 3 23.3 8 21.7 7 19.7 6 18.3 7 15.2 Estonia 12 28.3 6 20.7 8 20.3 17 24.2 12 16.7 Finland 11 27.4 14 26.7 19 27.3 24 27.2 27 24.1 France 28 44.0 32 44.3 32 40.4 35 38.7 34 31.6 Germany 25 37.6 27 34.7 27 29.8 21 26.3 23 23.6 Greece 30 45.5 33 48.2 34 43.8 36 40.4 36 36.8 Hungary 13 28.9 13 26.3 13 22.0 12 20.8 13 17.0 Iceland 2 23.2 1 15.6 1 13.6 1 12.9 2 14.4 Ireland 31 45.6 17 27.4 16 24.7 18 24.7 22 23.3 Israel 17 30.3 19 28.7 14 23.0 10 19.6 15 19.0 Italy 36 65.4 36 57.0 35 43.8 33 35.8 33 29.2 Japan 32 47.3 29 37.8 29 33.6 29 30.8 31 28.1 Latvia 4 23.8 2 17.3 4 18.8 7 18.3 1 14.2 Lithuania 7 24.7 3 18.9 3 17.2 3 17.5 3 14.5 Luxembourg 1 18.6 11 24.8 10 20.6 4 17.5 5 14.5 Mexico 35 55.6 34 48.8 36 44.0 34 38.0 35 32.3 Netherlands 6 24.1 4 19.7 5 19.0 5 18.2 6 15.2 New Zealand 23 36.5 25 32.1 22 28.9 20 25.4 19 21.9 Norway 16 29.9 10 23.3 9 20.4 11 20.0 11 16.6 Poland 27 39.8 24 31.3 20 28.2 23 26.8 26 23.8 Portugal 18 30.6 16 27.2 11 21.2 9 19.4 8 16.0 Slovakia 15 29.4 20 30.7 23 28.9 25 27.5 29 24.3 Slovenia 14 29.0 18 27.5 18 25.8 16 23.8 17 19.5 South Korea 8 26.2 5 20.2 2 15.7 2 15.6 4 14.5 Spain 29 45.1 30 41.7 30 36.3 28 29.4 25 23.7 Sweden 5 24.0 7 21.4 6 19.4 8 19.2 10 16.5 Switzerland 10 27.2 9 23.3 12 22.0 13 21.1 9 16.2 Turkey 21 36.3 28 36.9 28 32.6 30 30.9 32 29.0 United Kingdom 26 37.7 26 33.4 24 29.5 22 26.7 21 22.4 United States 24 37.5 22 30.9 25 29.6 26 28.4 24 23.7 10 Federal Communications Commission FCC 20-188 Fig. G-4: Fixed Broadband Mean Download Speed by Country Capital and U.S. State Capital Cities (2015-2019) City, 2015 2016 2017 2018 2019 Country/State Rank Mbps Rank Mbps Rank Mbps Rank Mbps Rank Mbps Canberra, Australia 78 20.0 79 23.9 82 28.7 83 36.8 81 53.2 Vienna, Austria 30 41.9 66 36.9 76 39.1 80 41.6 82 51.7 Brussels, Belgium 56 32.8 69 35.8 72 41.7 76 49.3 79 61.4 Ottawa, Canada 50 35.2 37 48.7 36 65.0 20 101.5 8 147.2 Santiago, Chile 79 18.1 81 23.4 80 30.5 78 42.0 72 71.5 Prague, Czech Republic 34 40.6 57 42.3 69 43.4 75 50.1 78 62.6 Copenhagen, Denmark 23 44.1 25 56.4 32 67.6 39 83.1 35 113.1 Tallinn, Estonia 29 42.1 46 46.7 64 48.0 68 57.3 74 70.8 Helsinki, Finland 25 43.1 53 44.5 70 43.3 72 54.8 76 65.8 Paris, France 2 76.4 2 93.9 5 111.9 8 114.7 2 163.6 Berlin, Germany 66 28.9 70 35.7 71 42.8 65 61.2 65 84.2 Athens, Greece 85 10.8 86 11.6 86 14.0 86 18.4 86 23.5 Budapest, Hungary 14 48.7 15 62.2 14 87.4 10 113.8 18 132.3 Reykjavik, Iceland 15 48.5 4 86.1 3 127.2 1 159.1 1 169.5 Dublin, Ireland 41 38.5 35 50.9 46 57.8 63 64.6 63 87.1 Jerusalem, Israel 70 26.0 78 25.6 78 34.8 81 41.0 83 48.6 Rome, Italy 83 14.5 82 19.7 81 28.8 82 37.2 80 56.5 Tokyo, Japan 4 72.3 10 65.2 23 74.5 62 65.0 48 102.5 Riga, Latvia 7 60.7 14 62.8 45 58.2 52 71.5 45 105.1 Vilnius, Lithuania 6 66.4 6 77.2 1 146.5 19 102.3 47 102.7 Luxembourg City, Luxembourg 41 47.8 49 57.0 42 80.6 36 112.4 Mexico City, Mexico 76 21.6 80 23.6 84 26.3 84 32.1 84 40.7 Amsterdam, Netherlands 24 43.6 30 53.9 35 66.7 47 76.2 56 92.0 Wellington, New Zealand 69 28.5 31 53.4 15 83.1 24 97.7 31 118.1 Oslo, Norway 13 50.7 26 55.3 28 71.9 35 87.1 41 107.7 Warsaw, Poland 67 28.6 52 44.5 42 60.1 66 61.1 55 93.9 Lisbon, Portugal 26 43.1 44 47.1 55 52.3 60 65.4 59 90.4 Bratislava, Slovakia 10 52.2 28 55.0 39 63.2 48 73.4 62 88.8 Ljubljana, Slovenia 62 30.4 68 36.3 74 40.8 73 52.5 75 68.4 Seoul, South Korea 5 68.8 3 87.0 2 136.7 4 127.5 7 150.2 Madrid, Spain 17 46.4 9 65.4 20 77.0 9 114.5 11 140.8 11 Federal Communications Commission FCC 20-188 City, 2015 2016 2017 2018 2019 Country/State Rank Mbps Rank Mbps Rank Mbps Rank Mbps Rank Mbps Stockholm, Sweden 3 72.8 5 79.3 9 96.1 14 111.2 21 130.9 Bern, Switzerland 8 56.3 24 56.7 29 71.6 33 89.0 38 110.8 Ankara, Turkey 82 15.3 84 17.0 85 17.9 85 20.0 85 25.3 London, United Kingdom 63 29.7 72 35.1 68 45.1 74 51.8 77 64.3 Albany, New York 72 25.3 75 28.1 77 38.0 50 71.9 54 96.1 Annapolis, Maryland 18 46.1 27 55.3 21 76.5 13 111.8 20 131.0 Atlanta, Georgia 38 39.1 11 65.2 11 89.4 43 79.3 14 138.6 Augusta, Maine 81 16.8 83 19.4 79 30.7 71 56.6 69 73.5 Austin, Texas 1 80.2 1 96.1 4 115.9 2 136.4 4 154.5 Baton Rouge, Louisiana 52 34.0 60 41.8 38 64.1 44 78.0 39 108.8 Bismarck, North Dakota 16 48.3 45 47.0 27 72.2 22 99.9 28 122.4 Boise, Idaho 73 22.9 36 50.3 51 56.3 57 67.0 58 91.2 Boston, Massachusetts 37 39.1 22 57.0 13 87.6 7 115.8 9 142.8 Carson City, Nevada 71 25.7 71 35.3 59 50.9 61 65.1 66 83.3 Charleston, West Virginia 44 37.2 48 45.7 52 53.5 32 93.5 42 107.4 Cheyenne, Wyoming 53 34.0 56 42.3 67 45.2 58 66.6 61 90.1 Columbia, South Carolina 74 22.7 77 25.6 73 41.3 69 57.2 60 90.2 Columbus, Ohio 68 28.5 67 36.6 57 51.3 54 69.3 51 98.4 Concord, New Hampshire 19 44.5 23 56.8 22 75.4 15 110.0 24 129.8 Denver, Colorado 36 39.8 34 51.4 30 71.6 34 88.9 37 111.7 Des Moines, Iowa 59 31.5 49 45.6 50 56.5 56 68.1 57 92.0 Dover, Delaware 9 52.3 17 61.4 10 93.0 6 120.5 3 155.7 Frankfort, Kentucky 84 13.2 85 13.4 83 27.8 77 43.5 71 72.8 Harrisburg, Pennsylvania 47 36.2 39 48.3 33 67.2 21 100.6 32 117.5 Hartford, Connecticut 48 35.9 47 46.2 56 51.4 49 72.3 50 98.6 Helena, Montana 64 29.2 64 37.6 75 39.8 59 65.9 68 77.0 Honolulu, Hawaii 31 41.5 16 61.5 25 73.6 25 97.7 26 126.7 Indianapolis, Indiana 60 31.0 51 45.3 37 64.3 38 83.7 27 123.1 Jackson, Mississippi 54 33.9 8 67.6 48 57.1 40 82.7 52 97.9 Jefferson City, Missouri 65 29.2 74 30.8 62 49.3 67 60.9 70 72.9 Juneau, Alaska 77 21.2 76 25.6 66 45.6 70 56.7 67 80.0 12 Federal Communications Commission FCC 20-188 City, 2015 2016 2017 2018 2019 Country/State Rank Mbps Rank Mbps Rank Mbps Rank Mbps Rank Mbps Lansing, Michigan 32 41.0 21 57.3 26 73.1 26 96.9 25 127.5 Lincoln, Nebraska 80 17.3 73 31.2 44 59.2 16 109.4 6 151.1 Little Rock, Arkansas 39 38.8 58 42.1 61 49.9 53 69.7 53 97.7 Madison, Wisconsin 51 34.5 62 38.9 60 50.8 36 86.3 34 113.2 Montgomery, Alabama 55 33.9 59 42.1 54 52.3 45 76.8 46 104.3 Montpelier, Vermont 75 22.0 63 37.9 65 46.7 79 42.0 73 71.1 Nashville, Tennessee 45 36.6 12 64.1 12 88.4 18 108.0 15 138.1 Oklahoma City, Oklahoma 22 44.2 20 57.9 18 79.6 31 93.6 16 135.7 Olympia, Washington 11 51.2 18 61.2 19 78.5 12 112.4 17 133.0 Phoenix, Arizona 40 38.5 29 54.6 31 71.4 29 94.2 29 120.8 Pierre, South Dakota 42 38.2 42 47.8 40 61.4 37 84.2 44 105.2 Providence, Rhode Island 35 40.5 40 48.1 41 60.7 27 95.1 23 129.9 Raleigh, North Carolina 49 35.6 13 63.7 8 99.8 3 127.9 5 153.3 Richmond, Virginia 28 42.4 54 43.8 24 73.8 23 99.0 22 130.1 Sacramento, California 33 40.8 38 48.4 34 67.1 30 94.1 19 131.2 Saint Paul, Minnesota 61 30.8 50 45.3 43 59.5 41 80.7 43 106.7 Salem, Oregon 12 51.1 19 60.3 17 79.8 17 109.2 12 140.4 Salt Lake City, Utah 21 44.2 7 70.8 6 109.6 5 120.7 10 141.5 Santa Fe, New Mexico 43 37.8 43 47.8 47 57.8 55 68.5 64 85.9 Springfield, Illinois 46 36.4 55 43.5 53 52.9 64 62.0 33 115.6 Tallahassee, Florida 57 31.8 61 41.0 63 48.7 46 76.7 40 108.8 Topeka, Kansas 58 31.5 65 37.0 58 51.0 51 71.6 49 101.6 Trenton, New Jersey 20 44.4 33 53.1 7 102.1 11 112.7 13 140.1 Washington, District of Columbia 27 42.8 32 53.4 16 80.8 28 94.9 30 119.6 13 Federal Communications Commission FCC 20-188 Fig. G-5: Fixed Broadband Download Speed Percentiles (2019) 250.0 200.0 150.0 Mbps 100.0 50.0 0.0 Italy Chile Israel Spain Japan Latvia Greece France Poland Mexico Iceland Ireland Turkey Austria Estonia Sweden Canada Finland Norway Belgium Slovenia Slovakia Portugal Hungary Australia Denmark Germany Lithuania Switzerland Netherlands South Korea South Luxembourg New Zealand New United States United Czech Republic Czech United Kingdom United 25th Percentile 50th Percentile 75th Percentile 14 Federal Communications Commission FCC 20-188 Fig. G-6: Fixed Broadband Mean Download Speed (2015-2019) 160.0 140.0 120.0 100.0 80.0 Mbps 60.0 40.0 20.0 0.0 2015 2016 2017 2018 2019 Canada France Germany Italy Japan South Korea United Kingdom United States 15 Federal Communications Commission FCC 20-188 Fig. G-7: Fixed Broadband City Count and Test Count by Country (2015-2019) Test Count (1000s) City Count Country 2015 2016 2017 2018 2019 2015 2016 2017 2018 2019 Australia 15,161 19,271 31,912 28,426 27,127 4,350 8,021 9,648 10,939 13,246 Austria 2,571 3,539 6,234 6,267 4,732 5,535 1,403 1,413 1,417 1,422 Belgium 2,914 3,932 5,940 4,996 4,814 3,701 607 606 608 612 Canada 17,720 21,948 31,334 30,081 29,883 4,501 2,637 2,830 2,895 3,225 Chile 5,852 6,089 9,458 9,558 7,902 664 230 231 260 267 Czech Republic 3,119 3,340 5,140 5,267 4,870 4,231 5,736 5,941 5,984 5,955 Denmark 2,326 3,452 5,080 5,160 5,012 3,278 588 586 587 634 Estonia 899 597 996 1,359 1,163 671 1,581 1,893 3,514 3,629 Finland 2,516 2,220 3,967 4,170 3,989 3,549 81 83 83 330 France 13,024 17,328 25,845 23,568 21,586 31,852 34,258 35,131 35,104 35,309 Germany 13,369 24,012 37,897 37,737 37,640 28,765 11,610 11,632 11,617 11,642 Greece 4,003 3,791 6,924 7,761 7,984 1,753 5,466 6,233 6,878 7,775 Hungary 4,541 4,729 7,398 7,954 7,306 3,314 3,011 3,070 3,095 3,113 Iceland 171 157 274 276 235 105 82 99 95 106 Ireland 2,140 1,234 2,394 2,517 2,657 2,815 160 163 160 159 Israel 2,288 2,521 4,320 5,437 5,056 1,095 992 1,007 1,003 1,045 Italy 27,924 32,991 57,872 54,093 43,095 14,173 36,909 40,379 40,802 40,126 Japan 2,458 8,431 16,314 15,445 14,063 3,156 2,014 1,965 2,010 1,905 Latvia 1,036 708 1,260 1,121 1,093 595 1,025 1,257 1,229 1,305 Lithuania 854 893 1,586 1,418 1,303 689 2,200 2,722 2,854 2,760 Luxembourg 360 327 505 547 447 365 421 427 434 431 Mexico 23,903 25,851 39,054 42,458 44,245 6,740 8,212 9,083 10,138 11,034 Netherlands 6,342 11,448 17,843 15,760 15,106 4,048 2,446 2,458 2,457 2,458 New Zealand 2,247 3,191 4,460 3,994 3,551 1,321 2,150 2,223 2,252 2,268 Norway 2,674 2,130 3,486 3,447 3,212 3,251 726 741 755 1,941 Poland 11,160 8,881 13,248 12,608 12,537 16,089 3,953 3,995 4,015 9,734 Portugal 3,909 4,279 7,116 7,946 7,804 4,431 1,176 1,180 1,180 1,353 Slovakia 1,302 1,906 2,941 3,244 3,464 2,519 2,703 2,780 2,797 2,806 16 Federal Communications Commission FCC 20-188 Test Count (1000s) City Count Country 2015 2016 2017 2018 2019 2015 2016 2017 2018 2019 Slovenia 1,018 1,261 1,682 1,720 1,813 1,565 5,210 5,526 5,489 5,553 South Korea 704 1,229 2,686 2,971 3,062 638 162 161 162 162 Spain 9,689 12,335 15,392 14,399 12,943 9,054 12,915 13,739 14,201 14,169 Sweden 1,164 1,132 1,834 1,725 1,921 3,023 397 414 444 507 Switzerland 2,200 3,546 4,884 5,395 5,228 3,866 2,587 2,584 2,579 2,593 Turkey 5,885 8,696 12,025 14,058 13,806 2,984 4,074 4,500 4,652 4,767 United Kingdom 20,135 40,534 47,236 53,479 51,881 11,492 6,467 6,417 6,511 6,624 United States 125,634 125,425 174,228 179,304 171,306 27,595 26,482 27,000 27,433 27,952 17 Federal Communications Commission FCC 20-188 Fig. G-8: Fixed Broadband Mean Download Speed by Country (2019) 18 Federal Communications Commission FCC 20-188 Fig. G-9: Fixed Broadband Mean Upload Speed by Country (2019) 19 Federal Communications Commission FCC 20-188 Fig. G-10: Fixed Broadband Mean Latency by Country (2019) 20 Federal Communications Commission FCC 20-188 Fig. G-11: Mobile Broadband  4G LTE Mean Download Speed by Country (2016-2019) 2016 2017 2018 2019 Country Rank Mbps Rank Mbps Rank Mbps Rank Mbps Australia 5 42.8 5 48.5 4 56.3 5 62.7 Austria 7 41.0 18 36.5 19 39.0 17 45.6 Belgium 19 32.8 11 40.7 7 50.4 10 50.3 Canada 14 36.2 6 44.8 3 59.2 3 71.3 Chile 32 24.7 36 20.9 36 20.0 36 21.2 Czech Republic 30 27.2 21 32.8 15 42.8 16 46.4 Denmark 10 39.2 9 42.2 11 46.9 11 49.4 Estonia 27 29.3 24 31.6 22 35.8 19 44.2 Finland 16 34.3 19 36.3 17 41.8 14 47.5 France 18 33.0 22 32.0 20 38.5 15 46.8 Germany 22 31.7 28 30.0 26 33.3 27 35.7 Greece 8 40.5 14 39.8 14 43.1 18 44.2 Hungary 1 46.1 3 50.5 8 50.2 20 43.2 Iceland 9 40.1 4 49.7 2 69.3 1 78.6 Ireland 28 28.7 25 31.0 29 30.5 32 31.7 Israel 29 27.7 31 26.3 33 26.9 34 27.8 Italy 20 32.3 17 37.6 23 35.3 26 36.6 Japan 36 18.8 35 22.1 34 26.7 33 31.7 Latvia 21 32.0 26 30.9 28 31.8 29 34.3 Lithuania 13 36.3 15 38.6 16 42.7 13 48.3 Luxembourg 11 38.2 10 41.5 9 47.6 12 48.4 Mexico 34 23.5 33 25.0 35 25.2 35 27.4 Netherlands 2 44.7 2 51.7 5 55.7 6 61.2 New Zealand 6 42.0 7 44.6 6 51.6 9 52.0 Norway 4 43.6 1 63.1 1 71.8 2 74.5 Poland 33 24.4 32 25.4 32 28.9 28 35.4 Portugal 24 31.0 30 29.4 25 33.5 23 37.7 Slovakia 26 30.2 23 31.6 27 33.3 30 34.2 Slovenia 31 26.3 27 30.1 24 34.5 22 38.5 South Korea 3 43.7 8 43.4 13 44.5 4 63.2 Spain 15 34.8 16 37.7 21 38.2 24 37.4 Sweden 17 33.8 12 40.5 10 46.9 7 54.3 Switzerland 23 31.4 20 35.7 12 46.3 8 52.1 Turkey 12 38.1 13 40.0 18 39.7 21 38.6 United Kingdom 25 30.3 29 29.6 31 29.5 31 33.4 United States 35 21.4 34 24.4 30 30.4 25 37.0 21 Federal Communications Commission FCC 20-188 Fig. G-12: Mobile Broadband  4G LTE Mean Upload Speed by Country (2016-2019) 2016 2017 2018 2019 Country Rank Mbps Rank Mbps Rank Mbps Rank Mbps Australia 9 15.4 10 15.8 10 16.0 7 16.9 Austria 11 15.4 21 14.1 18 14.4 17 15.3 Belgium 16 14.5 11 15.8 6 16.8 8 16.6 Canada 28 12.2 24 13.1 21 14.0 15 15.9 Chile 27 12.4 31 11.4 32 11.5 25 13.8 Czech Republic 23 13.3 14 14.9 8 16.4 6 17.0 Denmark 2 18.1 3 18.6 2 20.0 3 19.9 Estonia 32 11.0 30 11.6 26 12.6 24 13.8 Finland 13 15.1 12 15.7 11 15.8 10 16.3 France 33 10.9 34 10.3 33 10.6 33 11.4 Germany 31 11.7 29 11.7 27 12.5 27 13.3 Greece 17 14.4 22 14.0 20 14.1 19 15.1 Hungary 3 17.9 4 18.1 4 17.9 14 16.1 Iceland 1 19.3 1 21.5 1 23.0 1 22.6 Ireland 14 15.1 18 14.4 24 13.1 28 13.3 Israel 6 16.3 7 16.3 5 16.9 9 16.3 Italy 24 13.2 20 14.2 22 13.9 22 14.4 Japan 36 8.0 36 8.5 36 9.1 36 9.8 Latvia 19 14.2 23 13.3 25 13.1 29 12.9 Lithuania 18 14.3 17 14.4 17 14.9 21 14.9 Luxembourg 8 15.5 13 15.2 14 15.3 12 16.1 Mexico 15 14.7 8 16.0 19 14.2 23 14.0 Netherlands 7 16.3 9 15.9 13 15.6 13 16.1 New Zealand 4 17.5 6 16.3 12 15.7 11 16.3 Norway 12 15.3 2 19.6 3 19.7 2 20.3 Poland 30 11.7 33 10.6 34 10.4 34 11.3 Portugal 22 13.4 27 12.6 28 12.4 26 13.5 Slovakia 29 11.9 28 12.0 30 12.1 31 12.5 Slovenia 34 9.2 32 11.0 29 12.2 30 12.8 South Korea 10 15.4 19 14.4 15 15.2 16 15.7 Spain 21 13.6 16 14.7 16 15.1 20 15.0 Sweden 26 12.5 26 12.6 23 13.2 18 15.1 Switzerland 20 13.7 15 14.9 7 16.8 4 19.5 Turkey 5 17.3 5 16.8 9 16.3 5 17.1 United Kingdom 25 13.0 25 13.0 31 12.0 32 12.2 United States 35 8.8 35 9.0 35 9.7 35 11.1 22 Federal Communications Commission FCC 20-188 Fig. G-13: Mobile Broadband  4G LTE Mean Latency by Country (2016-2019) 2016 2017 2018 2019 Country Rank ms Rank ms Rank ms Rank ms Australia 16 32.6 11 29.3 13 28.2 17 29.6 Austria 8 28.4 10 28.8 11 27.3 12 27.4 Belgium 10 29.9 8 27.6 9 27.0 18 29.7 Canada 30 41.9 28 38.8 27 35.9 23 34.1 Chile 26 38.4 23 34.6 24 34.2 22 33.9 Czech Republic 17 32.8 13 29.5 8 26.7 8 26.4 Denmark 4 25.6 5 24.8 6 25.6 10 27.1 Estonia 6 27.1 6 25.3 5 24.2 4 24.9 Finland 5 27.0 7 26.7 7 25.7 5 25.3 France 28 40.5 31 40.9 30 41.3 30 41.5 Germany 32 44.7 32 41.7 28 38.1 28 38.2 Greece 13 31.6 21 32.0 12 27.3 11 27.4 Hungary 3 25.5 3 24.0 4 24.0 6 25.3 Iceland 7 27.5 4 24.4 1 21.0 1 21.1 Ireland 20 34.8 20 32.0 22 33.5 24 34.3 Israel 25 38.2 14 30.4 18 29.5 15 29.1 Italy 31 43.3 27 38.4 35 49.7 33 45.3 Japan 36 59.6 35 56.2 36 53.0 36 54.0 Latvia 1 21.7 1 21.3 2 22.5 2 23.4 Lithuania 9 29.3 9 28.3 10 27.2 7 26.3 Luxembourg 11 31.0 18 31.2 15 28.5 9 26.5 Mexico 35 58.2 36 60.1 34 49.2 35 50.0 Netherlands 15 32.6 12 29.4 17 29.1 20 31.0 New Zealand 21 35.5 26 38.1 29 39.3 29 39.4 Norway 23 36.6 24 34.7 26 35.4 27 37.6 Poland 24 36.6 25 35.6 23 33.9 25 34.5 Portugal 18 33.8 15 30.6 16 28.7 16 29.5 Slovakia 14 32.4 17 30.9 20 31.9 21 31.1 Slovenia 2 24.1 2 23.6 3 23.0 3 24.5 South Korea 27 40.3 29 39.3 25 34.5 26 35.4 Spain 33 50.5 33 47.5 32 45.3 32 43.6 Sweden 22 36.4 22 33.4 21 32.8 19 30.9 Switzerland 12 31.4 19 31.5 19 29.5 13 28.9 Turkey 19 34.6 16 30.6 14 28.4 14 29.0 United Kingdom 29 40.5 30 39.8 31 41.4 31 42.0 United States 34 52.5 34 50.4 33 46.4 34 46.7 23 Federal Communications Commission FCC 20-188 Fig. G-14: Mobile Broadband  4G LTE Mean Download Speed by Country Capital and U.S. State Capital Cities (2016-2019) 2016 2017 2018 2019 City, Country/State Rank Mbps Rank Mbps Rank Mbps Rank Mbps Canberra, Australia 15 36.4 7 43.6 11 49.2 4 65.7 Vienna, Austria 3 44.8 21 37.1 28 38.2 29 44.0 Brussels, Belgium 24 31.2 14 39.7 8 49.7 14 49.7 Ottawa, Canada 28 30.6 16 39.4 3 56.0 3 65.9 Santiago, Chile 37 24.3 68 20.0 84 18.9 86 20.2 Prague, Czech Republic 23 32.3 6 43.9 4 55.0 9 55.6 Copenhagen, Denmark 17 35.5 13 41.0 12 47.2 11 51.3 Tallinn, Estonia 25 31.2 25 34.8 26 39.6 20 48.5 Helsinki, Finland 14 36.9 18 38.0 16 44.4 18 49.0 Paris, France 20 34.2 27 33.0 21 41.1 15 49.3 Berlin, Germany 22 32.6 32 30.6 29 37.3 26 44.9 Athens, Greece 8 38.1 15 39.7 19 41.4 30 42.6 Budapest, Hungary 2 45.6 2 53.9 5 54.3 23 46.5 Reykjavik, Iceland 9 38.1 4 48.6 2 71.1 1 82.2 Dublin, Ireland 29 29.7 31 31.0 49 30.2 65 31.8 Jerusalem, Israel 52 20.5 48 25.3 48 30.2 81 24.2 Rome, Italy 21 32.8 19 37.3 36 34.8 52 36.7 Tokyo, Japan 51 20.8 55 23.2 60 27.0 73 29.1 Riga, Latvia 19 34.5 28 33.0 41 33.0 60 35.0 Vilnius, Lithuania 6 39.9 8 43.3 15 44.8 17 49.2 Luxembourg City, Luxembourg 11 37.9 11 42.6 10 49.3 22 47.1 Mexico City, Mexico 39 23.7 58 23.0 72 24.5 78 27.3 Amsterdam, Netherlands 1 46.7 3 50.7 7 53.5 6 58.0 Wellington, New Zealand 16 36.2 5 44.9 6 53.9 13 50.2 Oslo, Norway 4 43.0 1 64.6 1 72.2 2 74.2 Warsaw, Poland 34 25.9 39 27.9 47 30.4 54 36.5 Lisbon, Portugal 10 38.0 23 35.6 25 39.7 31 42.1 Bratislava, Slovakia 18 35.0 20 37.1 22 40.3 34 42.1 Ljubljana, Slovenia 27 31.1 22 36.3 23 40.2 43 38.2 Seoul, South Korea 5 42.9 12 42.3 17 43.7 5 63.3 Madrid, Spain 7 39.3 9 43.1 18 42.7 32 42.1 Stockholm, Sweden 13 37.2 10 42.9 9 49.5 7 57.7 Bern, Switzerland 26 31.1 24 35.6 13 45.4 10 52.8 Ankara, Turkey 12 37.7 17 39.4 24 39.7 50 37.1 London, United Kingdom 30 28.3 40 27.8 55 28.5 51 37.1 Albany, New York 63 19.5 64 21.0 64 26.6 61 34.6 Annapolis, Maryland 48 22.6 29 32.0 14 44.9 8 55.6 Atlanta, Georgia 43 23.0 38 28.3 32 35.9 21 48.4 Augusta, Maine 82 14.8 78 17.8 78 22.1 79 26.1 Austin, Texas 56 20.3 47 25.3 46 31.0 56 36.0 Baton Rouge, Louisiana 67 18.1 59 22.5 58 28.2 57 35.8 24 Federal Communications Commission FCC 20-188 2016 2017 2018 2019 City, Country/State Rank Mbps Rank Mbps Rank Mbps Rank Mbps Bismarck, North Dakota 33 26.0 30 31.1 68 25.4 55 36.1 Boise, Idaho 77 16.7 69 20.0 44 31.4 41 38.7 Boston, Massachusetts 62 19.7 51 24.5 45 31.2 35 41.7 Carson City, Nevada 70 17.8 83 16.6 86 17.8 84 21.5 Charleston, West Virginia 83 12.9 81 16.8 66 25.9 62 34.6 Cheyenne, Wyoming 85 12.2 85 15.0 83 19.3 71 29.5 Columbia, South Carolina 76 16.7 63 21.1 57 28.4 63 33.6 Columbus, Ohio 47 22.7 42 25.8 35 34.9 27 44.7 Concord, New Hampshire 84 12.5 82 16.8 82 19.8 83 23.3 Denver, Colorado 80 14.8 65 20.9 51 29.3 48 37.2 Des Moines, Iowa 49 21.8 53 23.7 71 24.7 74 29.1 Dover, Delaware 44 22.8 36 28.7 27 38.8 16 49.2 Frankfort, Kentucky 45 22.8 66 20.4 65 26.5 38 40.2 Harrisburg, Pennsylvania 53 20.4 54 23.6 33 35.7 28 44.6 Hartford, Connecticut 61 20.0 57 23.1 43 31.7 53 36.6 Helena, Montana 66 18.5 72 19.4 73 24.4 46 37.7 Honolulu, Hawaii 69 18.0 71 19.8 67 25.5 67 31.4 Indianapolis, Indiana 40 23.5 35 29.1 37 34.6 36 40.9 Jackson, Mississippi 73 17.2 80 17.0 76 23.1 80 24.6 Jefferson City, Missouri 74 17.2 75 18.4 77 22.3 68 30.9 Juneau, Alaska 57 20.3 77 18.3 85 17.9 85 21.2 Lansing, Michigan 32 26.6 34 30.2 31 36.7 39 39.0 Lincoln, Nebraska 60 20.1 56 23.1 70 24.8 72 29.2 Little Rock, Arkansas 46 22.8 41 26.9 34 35.4 33 42.1 Madison, Wisconsin 71 17.5 76 18.3 80 20.2 82 24.0 Montgomery, Alabama 35 25.8 37 28.3 52 29.1 64 31.9 Montpelier, Vermont 75 16.9 79 17.4 69 25.2 76 28.9 Nashville, Tennessee 65 18.9 61 22.1 54 29.0 45 38.1 Oklahoma City, Oklahoma 79 16.4 73 19.3 75 23.3 77 27.4 Olympia, Washington 64 19.1 74 19.2 74 24.2 69 30.4 Phoenix, Arizona 72 17.4 62 21.2 63 26.8 42 38.7 Pierre, South Dakota 41 23.1 43 25.7 62 26.8 58 35.8 Providence, Rhode Island 55 20.4 46 25.5 40 33.3 12 51.1 Raleigh, North Carolina 59 20.1 50 24.6 50 29.5 47 37.4 Richmond, Virginia 58 20.1 52 24.4 42 32.2 40 38.9 Sacramento, California 54 20.4 60 22.4 59 28.0 59 35.5 Saint Paul, Minnesota 36 25.6 26 34.5 20 41.1 19 48.7 Salem, Oregon 31 27.5 33 30.4 30 37.0 37 40.9 Salt Lake City, Utah 68 18.1 70 19.9 61 26.9 44 38.2 Santa Fe, New Mexico 86 12.1 86 14.6 81 20.0 66 31.6 Springfield, Illinois 38 23.8 44 25.6 53 29.1 49 37.1 Tallahassee, Florida 42 23.1 45 25.6 39 33.5 24 45.5 Topeka, Kansas 78 16.4 67 20.4 56 28.5 70 30.1 25 Federal Communications Commission FCC 20-188 2016 2017 2018 2019 City, Country/State Rank Mbps Rank Mbps Rank Mbps Rank Mbps Trenton, New Jersey 81 14.8 84 16.4 79 21.8 75 29.0 Washington, District of Columbia 50 21.6 49 24.8 38 34.0 25 44.9 26 Federal Communications Commission FCC 20-188 Fig. G-15: Mobile Broadband  4G LTE Download Speed Percentiles (2019) 120.0 100.0 80.0 60.0 Mbps 40.0 20.0 0.0 Italy Chile Israel Spain Japan Latvia France Greece Poland Mexico Iceland Ireland Turkey Austria Estonia Sweden Canada Finland Norway Belgium Slovenia Slovakia Portugal Hungary Australia Denmark Germany Lithuania Switzerland Netherlands South Korea South Luxembourg New Zealand New United States United Czech Republic Czech United Kingdom United 25th Percentile 50th Percentile 75th Percentile 27 Federal Communications Commission FCC 20-188 Fig. G-16: Mobile Broadband  4G LTE Mean Download Speeds (2016-2019) 80.0 70.0 60.0 50.0 40.0 30.0 20.0 Download Speeds (Mbps) 10.0 0.0 2016 2017 2018 2019 Year Canada France Germany Italy Japan South Korea United Kingdom United States 28 Federal Communications Commission FCC 20-188 Fig. G-17: Mobile Broadband  4G LTE City Count and Test Count by Country (2016-2019) Test Count (1000s) City Count Country 2016 2017 2018 2019 2016 2017 2018 2019 Australia 1,551 2,567 3,310 3,711 9,247 10,240 11,139 12,240 Austria 551 872 912 872 1,380 1,396 1,402 1,398 Belgium 101 165 182 214 600 602 607 610 Canada 773 1,180 1,130 1,255 1,985 2,359 2,395 2,628 Chile 424 768 1,430 1,245 215 227 241 245 Czech Republic 119 187 211 313 4,431 4,838 4,974 5,333 Denmark 364 502 558 559 586 586 586 615 Estonia 118 184 239 200 1,563 1,965 3,388 3,510 Finland 944 1,733 1,823 1,838 84 85 83 396 France 1,436 3,649 4,209 3,187 19,151 27,016 28,838 29,598 Germany 1,206 1,971 2,634 2,907 10,127 10,470 10,679 10,865 Greece 203 408 477 510 2,940 4,649 5,283 5,960 Hungary 211 427 577 618 2,455 2,843 2,922 2,923 Iceland 11 22 30 20 63 80 82 100 Ireland 109 205 291 339 127 140 148 143 Israel 291 477 606 651 743 925 969 1,023 Italy 2,834 5,268 11,786 9,563 23,279 28,550 33,594 34,517 Japan 1,984 2,585 2,186 1,802 1,991 1,930 1,996 1,826 Latvia 126 216 219 247 881 1,084 1,171 1,242 Lithuania 98 156 171 202 1,721 2,207 2,340 2,390 Luxembourg 25 36 35 28 310 349 365 361 Mexico 810 1,498 2,230 2,244 2,864 3,855 4,958 6,018 Netherlands 419 802 850 880 2,324 2,404 2,429 2,428 New Zealand 87 140 138 159 1,058 1,326 1,465 1,574 Norway 226 245 235 209 624 682 685 1,619 Poland 1,324 2,235 2,213 2,013 3,547 3,791 3,856 7,913 Portugal 125 249 316 305 1,072 1,128 1,142 1,264 Slovakia 84 168 198 231 1,756 2,190 2,305 2,399 Slovenia 51 118 130 171 3,201 4,161 4,247 4,261 South Korea 119 159 272 387 161 162 162 162 Spain 498 663 698 727 5,643 7,833 8,677 9,639 Sweden 64 89 105 120 400 405 414 434 Switzerland 350 657 873 970 2,445 2,525 2,542 2,569 Turkey 2,158 1,097 1,513 1,702 2,029 2,208 2,784 3,428 United Kingdom 2,488 3,464 3,772 4,199 6,019 6,331 6,407 6,494 United States 14,332 20,657 18,576 17,941 24,471 25,922 25,975 26,346 29 Federal Communications Commission FCC 20-188 Fig. G-18: Mobile Broadband  4G LTE Mean Download Speed by Country (2019) 30 Federal Communications Commission FCC 20-188 Fig. G-19: Mobile Broadband  4G LTE Mean Upload Speed by Country (2019) 31 Federal Communications Commission FCC 20-188 Fig. G-20: Mobile Broadband  4G LTE Mean Latency by Country (2019) IV. OPENSIGNAL ANALYSIS 25. This section presents mobile download speed data for 3G/4G and 5G as well as 5G availability data, as measured and calculated by OpenSignal.22 Average combined 3G/4G download speeds for the first half of 2019 and the first half of 2020 are presented in Figure G-21 below.23 Figure G- 22 OpenSignal gathers crowdsourced mobile speed data through the use of its mobile app as well as through partner apps. The partners they work with are strategically selected to cover a wide range of users, demographics, and devices. OpenSignal, Methodology Overview: How OpenSignal Measures Mobile Network Experience, https://www.opensignal.com/sites/opensignal-com/files/opensignal_methodology_overview_june_2020.pdf (last visited Oct. 27, 2020). 23 Fig. G-21 presents Download Speed Experience by country and shows the average download speed (Mbps) experienced by OpenSignal users across an operator s 3G and 4G networks. This metric factors in 3G and 4G download speeds along with the availability of each technology. 4G availability measures the proportion of time OpenSignal users with a 4G device have a 4G connection, while 3G availability measures the proportion of time OpenSignal users with a 3G device have a 3G connection. Data for the first half of 2019 were collected from January 1 March 31, 2019, and data for the first half of 2020 were collected from January 1 March 30, 2020. Peter Boyland, The State of Mobile Network Experience: Benchmarking Mobile on the Eve of the 5G Revolution, OpenSignal (May 2019), https://www.opensignal.com/sites/opensignal-com/files/data/reports/global/data-2019- 05/the_state_of_mobile_experience_may_2019_0.pdf; Sam Fenwick and Hardik Khatri, The State of Mobile Network Experience 2020: One Year into the 5G Era, OpenSignal (May 2020), (continued& .) 32 Federal Communications Commission FCC 20-188 22 presents average 5G download speeds as well as 5G availability, which is defined as the proportion of time that OpenSignal users with a 5G device and subscription have a 5G connection, for the first and second half of 2020.24 Fig. G-21: OpenSignal  Mobile Broadband Download Speed by Country (2019-2020) 1H2019 1H2020 Country Mbps Mbps Afghanistan 2.9 Albania 21.4 25.8 Algeria 3.1 4.0 Argentina 12.8 17.4 Australia 37.4 43.0 Austria 27.5 34.6 Azerbaijan 13.4 17.8 Bahrain 13.9 16.4 Bangladesh 5.7 6.8 Belarus 7.7 10.8 Belgium 34.2 37.6 Bolivia 12.5 13.6 Brazil 13.0 15.3 Brunei 16.4 Bulgaria 22.5 Cambodia 5.6 8.0 Cameroon 7.5 Canada 42.5 59.6 Chile 12.0 13.7 Colombia 10.0 13.4 Costa Rica 10.1 14.0 Cote d'Ivoire 7.4 Croatia 26.7 36.6 Czech Republic 31.5 32.7 Denmark 34.6 33.5 Dominican Republic 8.5 11.5 https://www.opensignal.com/sites/opensignal-com/files/data/reports/pdf-only/data-2020- 05/state_of_mobile_experience_may_2020_opensignal_3_0.pdf. 24 Fig. G-22 presents 5G download speed by country, which is the average download speed for each operator on an active 5G connection as experienced by OpenSignal users. This Figure also presents 5G availability, which is the proportion of time OpenSignal users with a 5G device and subscription have a 5G connection. Data for the first half of 2020 were collected from January 22 April 21, 2020, and data for the second half of 2020 were collected from May 16 August 14, 2020. Ian Fogg, 5G Download Speed is Now Faster than Wifi in Seven Leading 5G Countries, OpenSignal (May 6, 2020), https://www.opensignal.com/2020/05/06/5g-download-speed-is-now-faster-than-wifi-in- seven-leading-5g-countries; Ian Fogg, Benchmarking the Global 5G User Experience, OpenSignal (Aug. 26, 2020), https://www.opensignal.com/2020/10/13/benchmarking-the-global-5g-user-experience-october-update. 33 Federal Communications Commission FCC 20-188 1H2019 1H2020 Country Mbps Mbps Ecuador 10.5 13.3 Egypt 8.6 10.7 El Salvador 5.4 5.8 Finland 27.0 29.8 France 25.2 28.6 Germany 22.6 28.7 Ghana 5.1 6.8 Greece 23.8 23.7 Guatemala 10.8 15.0 Honduras 13.4 Hong Kong 16.7 21.8 Hungary 32.7 31.7 India 6.8 8.1 Indonesia 6.9 9.9 Iraq 1.6 1.6 Ireland 16.2 19.2 Israel 13.6 15.2 Italy 19.9 24.3 Ivory Coast 6.7 Japan 33.0 49.3 Jordan 10.4 12.5 Kazakhstan 11.4 11.9 Kenya 10.1 10.9 Kyrgyzstan 10.5 Kuwait 16.2 16.6 Laos 17.1 Lebanon 16.9 23.8 Lithuania 33.3 Malaysia 11.5 11.0 Maldives 19.4 Mexico 14.9 19.6 Morocco 11.2 17.4 Myanmar 16.0 16.4 Nepal 4.4 7.5 Netherlands 42.4 54.8 New Zealand 27.3 35.2 Nigeria 5.4 7.3 North Macedonia 30.0 Norway 48.2 47.5 Oman 20.3 25.2 34 Federal Communications Commission FCC 20-188 1H2019 1H2020 Country Mbps Mbps Pakistan 6.2 8.4 Panama 7.2 8.4 Paraguay 10.6 10.8 Peru 11.7 12.1 Philippines 7.0 8.5 Poland 17.3 20.7 Portugal 21.6 26.3 Puerto Rico 18.0 Qatar 24.6 31.3 Romania 20.6 21.4 Russian Federation 12.0 14.5 Saudi Arabia 13.6 21.4 Senegal 5.1 9.1 Serbia 21.5 25.2 Singapore 39.3 47.5 Slovakia 23.3 25.3 Slovenia 26.0 Somalia 6.4 South Africa 15.0 19.1 South Korea 52.4 59.0 Spain 24.8 26.2 Sri Lanka 10.7 10.2 Sweden 30.8 29.7 Switzerland 35.2 42.8 Tanzania 5.4 Taiwan 26.6 28.9 Thailand 5.7 9.2 Tunisia 13.4 15.5 Turkey 17.1 20.0 Ukraine 11.2 14.0 United Arab Emirates 19.9 32.2 United Kingdom 21.7 22.9 United States 21.3 26.7 Uruguay 20.3 Uzbekistan 5.0 6.2 Vietnam 14.1 20.6 35 Federal Communications Commission FCC 20-188 Fig. G-22: OpenSignal  5G Download Speed and Availability by Country (1H2020, 2H2020) 1H2020 2H2020 Country Mbps Availability Mbps Availability Australia 163.9 6.1% 215.7 8.6% Canada 178.1 8.8% Germany 102.0 10.3% Hong Kong 142.8 26.1% Kuwait 185.1 34.9% 171.5 29.1% Netherlands 79.2 13.2% Saudi Arabia 291.2 30.8% 414.2 34.4% South Korea 224.0 14.2% 312.7 20.7% Spain 146.8 6.9% Switzerland 201.9 8.7% 150.7 7.5% Taiwan 210.2 18.6% United Kingdom 138.1 5.2% 133.5 4.5% United States 52.3 12.7% 50.9 19.3% 36 Federal Communications Commission FCC 20-188 APPX. G-3 Broadband Pricing Comparisons 1. Congress directs the Commission to compare broadband pricing in  communities of a population size, population density, topography, and demographic profile that are comparable to the population size, population density, topography, and demographic profile of various communities within the United States. 25 To meet this directive, we first collected a comprehensive sample of advertised prices and terms for over 1,000 fixed and mobile broadband plans from the largest broadband providers in the United States and 25 other countries.26 We then rank the countries by fixed and mobile broadband prices from the least expensive (1st) to most expensive (26th) according to two different methodologies. The first method calculates weighted average prices for a set of fixed broadband products based on download speeds and for a set of mobile broadband products based on data usage allowances.27 These two weighted average prices are then used to calculate an overall average price, and countries are ranked by this measure.28 To more closely match the characteristics of the comparison communities and their broadband offerings, the second method constructs hedonic fixed and mobile broadband price indexes from a regression of broadband prices on broadband product characteristics and country-level variables to control for differences in broadband market conditions.29 The hedonic method seeks to better assess how U.S. broadband prices compare to prices in other countries after accounting for country-level cost and demographic differences that likely affect broadband pricing, including population density, topography, income, and education levels. The hedonic price index also adjusts for observable differences in broadband plan characteristics across countries (e.g., speed and usage limits) and generates prices for a set of standardized broadband plans to facilitate price comparisons across countries. The results of our fixed and mobile broadband pricing analyses demonstrate that accounting for these country-level differences in cost, demand, and quality factors gives a substantially different assessment of the competitiveness of the U.S. broadband market. I. OVERVIEW AND DATA HIGHLIGHTS 2. Comparing broadband prices across countries presents several challenges. One difficulty is that broadband product offerings are complex and vary widely across countries. Among other aspects, the plans may differ with respect to: (1) download and upload speeds; (2) types of technology used to deliver broadband services; (3) limitations on use, including limits on upload and download volumes; (4) contractual conditions; (5) additional services included; and (6) consequences of exceeding usage limits, with some plans reducing speeds, imposing surcharges, or shutting off service. In addition, broadband 25 47 U.S.C. § 1303(b)(2); see also RAY BAUM S Act. 26 The 2018 International Broadband Data Report included three additional comparison countries: Chile, Japan, and South Korea. These countries were excluded from this Report due to resource limitations and the difficulty of collecting information from Japan and South Korea s providers websites. 2018 International Broadband Data Report, 33 FCC Rcd at 981, para. 6. 27 The data was collected between February and September 2020. The data we use for these comparisons contain the terms and advertised prices for select fixed and mobile broadband plan offerings available on the websites of the largest broadband providers in each country. See infra paras. 39-60. 28 Our broadband price index measures the dollar amount that U.S. broadband subscribers would need to have added or subtracted from their incomes to purchase the same basket of broadband services under the pricing structures in other countries. Quantity weights for the price index are the share of broadband subscribers in the United States that, for fixed broadband, take each of the three broadband speed tiers and, for mobile broadband, take each of the three data usage tiers in the analyses. See infra paras. 61-62. 29 A hedonic regression provides an empirical summary of how prices vary with the characteristics of a good (e.g., download speed). In this Report, the hedonic regression builds on the price index method by allowing adjustment of prices for quality, cost, and demographic differences across countries and then predicting broadband prices for each country at the average U.S. values of these variables. See infra paras. 28-32. 37 Federal Communications Commission FCC 20-188 service is also frequently purchased as part of a discounted bundle of services, making it difficult to identify the price of the broadband service. Lastly, differences across countries in the quality of networks deployed, cost factors (e.g., population density and topography), and demand factors (e.g., demographics and content quality), would be expected to affect pricing, all else equal. Building on the work in the 6th International Broadband Data Report, which was released by the International Bureau,30 our hedonic price index analysis accounts for these differences, with the intention of producing comparisons that are more meaningful for the purposes of assessing which countries have broadband policies that foster competition and provide the greatest consumer benefits.31 A. Fixed Broadband Pricing Results 3. Broadband Price Index Results. This analysis compares broadband prices across countries by calculating weighted average prices within each fixed broadband download speed tier and then aggregating these prices into an overall average fixed broadband price measure. " For broadband service purchased on a standalone basis, we find that the United States ranks 21st out of the 26 countries in our broadband price index, not adjusting for cost and demand factor differences across countries.32 " For broadband service purchased in a bundle with video service, we find that the United States ranks 19th out of the 26 countries. " Overall, we find that the United States ranks 21st out of the 26 countries that does not account for cost and demand differences across countries. 4. Hedonic Price Index Results. The hedonic price index adjusts broadband prices for differences in demographic and cost profiles across countries using a hedonic regression framework. The hedonic regression also adjusts for observable differences in broadband plan characteristics across countries (e.g., the speed and usage limits of each plan) and generates prices for a set of standardized broadband plans in every country to facilitate price comparisons. Based on the predicted prices for these standardized plans, we then calculate a hedonic price index to serve as our price comparison measure across countries. This index estimates what the average U.S. consumer would expect to pay for service in each country if that country had the same demographics, cost structure, and broadband plan characteristics as the United States.33 " After adjusting for differences in cost and demographic factors across countries, as well as differences in broadband plan characteristics, our hedonic price index estimates that the United States ranks 12th out of the 26 countries.34 30 2018 International Broadband Data Report, 33 FCC Rcd 978. 31 Using standard discrete choice consumer demand models, it is simple to construct examples where consumers in a country with higher broadband prices receive greater consumer surplus (i.e., are better off) from their broadband services, compared to consumers in a country with lower prices. Similarly, higher prices may not indicate that one market is less competitive than another in terms of the economic profits earned by broadband firms. As such, simple broadband price comparisons may not be appropriate for comparing the effectiveness of competition and regulatory policies across countries. 32 See infra Fig. G-24. 33 The country rankings would not change if, instead of using the United States as our baseline country, we predicted prices at the values of the country-level variables for any other country or at the average of these variables across all countries. The only difference in our results would be in the levels of the predicted prices. Due to the provider-level random coefficients in the hedonic model, changing the values of the plan characteristics used to predict prices would change the country rankings. 34 See infra Fig. G-26. 38 Federal Communications Commission FCC 20-188 " The U.S. ranking remains unchanged at 12th after adjusting for our measure of fixed broadband network quality. " After further adjusting prices for measures of broadband content quality, the United States ranks 2nd among the 26 countries. B. Mobile Broadband Pricing Results 5. Our mobile broadband price comparison methodology is the same as our fixed broadband price methodology with two exceptions. First, because nearly all mobile broadband plans are sold by data usage allowance rather than speed, we classify mobile broadband products by data usage allowances rather than by download speeds. Second, we account for bundling in this sector by analyzing multi-line data plans (i.e., family plans) rather than the video and broadband bundling that is more common in the fixed broadband market. 6. Broadband Price Index. This analysis compares countries by calculating weighted average prices for mobile plans that fall within specified data usage tiers and then aggregates these prices into an overall average mobile broadband price. " The United States ranks 22nd in single-line plan pricing and 21st in multi-line pricing out of the 26 countries.35 " Overall, we find that the United States ranks 21st out of the 26 countries in our mobile broadband price index, not adjusting for cost and demand factor differences across countries. 7. Hedonic Price Index Results. As in our fixed broadband analysis, we calculate a hedonic index that estimates what the average U.S. consumer would expect to pay for her level of mobile broadband service in each country if that country had the same demographics, cost structure, and broadband plan characteristics as the United States. " After adjusting for differences across countries in the cost and demographic factors, as well as differences in broadband plan characteristics, our hedonic price index estimates that the United States ranks 22nd out of the 26 countries.36 " Adjusting for mobile network quality measures, the United States ranks 17th out of 26 countries. " After we further adjust the mobile hedonic price index for our measures of content quality, the United States is ranked 7th. 8. Combining Fixed and Mobile Hedonic Price Index Rankings. Typical consumers in the United States subscribe to both fixed and mobile broadband services, so we also measure overall broadband affordability by calculating the average monthly cost that U.S. consumers would pay to subscribe to both services in each country. After accounting for differences in content quality, costs, demographics and broadband plan characteristics, we find that the United States ranks 2nd overall by this measure, at $121.49 per month for a mobile and fixed broadband connection.37 35 See infra Fig. G-28. 36 See infra Fig. G-30. 37 See infra Fig. G-32. 39 Federal Communications Commission FCC 20-188 II. FIXED BROADBAND PRICING ANALYSIS 9. Many studies compare advertised prices for  similar telecommunications services.38 While such price comparisons are appropriate for descriptive assessments of price levels, they are less useful for identifying which countries have industry structures and policies that produce the greatest broadband consumer benefits.39 Rankings that account for these factors are necessary to inform government competition and regulatory policy because the determinants of price that are outside the scope of competition policy may differ across countries and distort comparisons. The challenge in comparing prices across markets is that the supply and demand factors which generate different broadband prices and offerings vary widely from one market to the next. An analysis that seeks to make normative comparisons of broadband prices across countries would, at a minimum, need to account for: (1) the different costs of deploying and operating broadband networks; (2) demographic differences that affect demand for broadband service; (3) multi-product bundling in broadband pricing; (4) different product offerings in each country; and (5) the availability and quality of complementary content and applications. The 2018 International Broadband Data Report described in detail how each of these factors would be expected to affect international price comparisons and why these should be accounted for when comparing prices across countries.40 10. As in the 2018 International Broadband Data Report, we attempt to adjust for these cost and demand factor differences by estimating a hedonic regression.41 Our approach extends a standard hedonic framework by controlling for cost and demand factors instead of only adjusting prices for differences in product characteristics.42 The first step of constructing the index is to use our model to predict broadband prices for a set of standardized plans for each provider in our study, setting the country and demographic characteristic variables at the U.S. values but using the estimated provider-specific product characteristic random coefficients and random intercepts.43 From these predicted prices, we then construct a hedonic price index that facilitates comparisons by adjusting for observable differences in 38 For example, see Carol Corrado and Olga Ukhaneva, Hedonic Prices for Fixed Broadband Services: Estimation Across OECD Countries (Oct. 20, 2016), https://www.oecd-ilibrary.org/docserver/5jlpl4sgc9hj- en.pdf?expires=1603997556&id=id&accname=guest&checksum=1D0A776B692D8F368F8A696A24A0E702. 39 In the language of economics, price indexes are positive analyses that describe what the price differences are across countries or what the typical consumer would be expected to pay for broadband in each country. However, cross-country price differences are frequently used to normatively rank countries and interpreted as meaningful differences in industry performance or regulatory policies. In order to provide a more normative assessment, our analysis also accounts for potentially exogenous supply and demand differences across countries that would result in price differences regardless of broadband policy differences. However, given the limited number of country-level variables that we can include in the analysis, even our results should still be interpreted with caution when comparing country rankings. 40 2018 International Broadband Data Report, 33 FCC Rcd at 980-81, paras. 5-6, Appx. C, paras. 7-13. 41 A hedonic regression provides an empirical summary of how prices vary with the characteristics of a good and is a standard technique used to adjust prices for differences in quality in price indexes. U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index, Quality Adjustment in the CPI (Nov. 20, 2017), https://www.bls.gov/cpi/quality-adjustment/home.htm. 42 In a standard hedonic broadband pricing analysis, a country fixed effect would be included to account for country- level differences in cost and demand factors. However, since the country fixed effect is used to predict prices from the model, these cost and demand differences remain in the predicted price levels. Our approach differs by decomposing the fixed effect into observable cost components and an unobserved random effect to remove the effect of exogenous country-level observable cost and demand differences from predicted prices. See infra paras. 28-32. 43 All plan characteristics of the standardized plans we generate to predict prices have the exact same characteristics (other than download speed) in order to make prices comparable across countries. These features of the standardized plans are as follows: no contract, no phone service, and an unlimited data usage allowance. 40 Federal Communications Commission FCC 20-188 broadband plan characteristics across countries (e.g., speed and data usage limits), as well as differences in market cost and demand conditions (e.g., population density and income). A. Fixed Broadband Price Index 11. To compare broadband pricing across countries, we need an estimate of  the price of broadband in each country. Our approach is to follow well-established practices in the price index literature. Price indexes calculate measures of price changes for goods and services by comparing the prices in a base period to those in a comparison period. One such index is the U.S. CPI, calculated by the Bureau of Labor Statistics of the U.S. Department of Labor.44 While the CPI involves measuring price changes across time periods, our application to price changes across countries is analogous, with the two periods now corresponding to two different countries. 45 12. Our goal is to calculate the following Laspeyres broadband price index, where pj,t represents the price of product j in comparison country t, pj,0 is the price of product j in the base country and qj,0 is the market share of product j in the base country. The index is therefore the ratio of the weighted average price of all of the j broadband products sold in the comparison country to the weighted average price of these same products in the base country, where the weights are the percentage of broadband consumers who choose each product in the base country.46 13. Ideally, the price index would be calculated over every broadband plan offered in every country. However, there are at least two difficulties in doing so. First, we would need to know the number of households that subscribe to each base country plan, and we do not have these data. Second, the broadband products available in each country are not the same. Even if we had such quantity weights for the base country, they would not be applicable in the comparison countries. To deal with these issues, we classify all available broadband plans into j = 6 products based on download speed categories for which we have information on the U.S. broadband product shares.47 We define three standalone products 44 U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index Frequently Asked Questions (FAQs) (Jan. 15, 2019), https://www.bls.gov/cpi/questions-and-answers.htm. 45 The Laspeyres price index yields an upper bound for the average compensating variation from a price change. Compensating variation measures the dollar amount by which a given consumer would need to have their income adjusted to obtain the same level of utility, or well-being, under the comparison prices and product choice set. See Ariel Pakes, A Reconsideration of Hedonic Price Indexes with an Application to PCs, 93 American Economic Review 1578-96 (2003). 46 The United States is used as the base country for several reasons. First, the focus of this Report is to evaluate how the prices of broadband products purchased in the United States compare to those of other countries. Second, we have better estimates of the subscriber quantity weights for the United States than for any other country. Finally, this index ensures that U.S. broadband consumers would be at least as well-off as in higher ranked countries by measuring the dollar amount that U.S. broadband subscribers would need to have added or subtracted from their incomes to purchase the same basket of broadband services under the pricing structures in the other countries. 47 Aggregating products in this manner is common in the differentiated products demand model literature. See Steven Berry, James Levinsohn, and Ariel Pakes, Automobile Prices in Market Equilibrium, 63 Econometrica 841 (1995), http://people.stern.nyu.edu/wgreene/Econometrics/BLP.pdf; Aviv Nevo, Measuring Market Power in the Ready-to-Eat Cereal Industry, 69 Econometrica 307 (2001), https://economia.uniandes.edu.co/files/profesores/jorge_tovar/docs/Seminario%20de%20Tesis%20PEG/apuntes%20 de%20clase/Nevo_2001_Measuring_Mkt_Pwr_Econometrica.pdf; Austan Goolsbee and Amil Petrin, The Consumer (continued& .) 41 Federal Communications Commission FCC 20-188 classified by the following download speed tiers: less than 25 Mbps; at least 25 Mbps but less than 100 Mbps; and at least 100 Mbps but no more than 1000 Mbps.48 We also define three additional products when these speed tiers are purchased in a bundle with video service. 14. Fixed Product Shares. To calculate the U.S. quantity weights for each of the six products in our price indexes, we use the FCC Form 477 data49 to estimate the share of U.S. households that subscribe to each of the three broadband speed tiers and an estimate from S&P Global that about 65% of all U.S. broadband households purchase their service in a bundle.50 The resulting broadband products and their estimated U.S. market shares are shown in Figure G-23 below. Fig. G-23: Fixed Broadband Product Shares Bundle Speed Tier Product Product Download Speed Tier Share Share Share Plans 1 Standalone: 0 < Mbps < 25 34.69% 28.95% 10.04% 67 2 Standalone: 25 d" Mbps < 100 34.69% 31.58% 10.96% 105 3 Standalone: 100 d" Mbps d" 1000 34.69% 39.47% 13.69% 253 4 Bundle: 0 < Mbps < 25 65.31% 28.95% 18.90% 77 5 Bundle: 25 d" Mbps < 100 65.31% 31.58% 20.62% 133 6 Bundle: 100 d" Mbps d" 1000 65.31% 39.47% 25.78% 319 Sources: S&P Global; Preliminary December 2019 FCC Form 477 data. 15. Calculating comparable prices for each of our six broadband products for each country is more difficult. We again follow the price-index literature in implementing two common approaches: a standard price index and hedonic analysis. The standard price index approach, discussed in section IV.B, calculates a price for each of the six products in a country by calculating the weighted average price of all plans that fall within that product category, and then constructs a Laspeyres price index using the U.S. product shares as weights.51 To calculate the broadband price index, we first calculate simple unweighted average prices for each provider s offerings that fall into each of the six product categories. We then use the market share of each provider to calculate a country-level weighted average for each of the six broadband products from these provider-level prices.52 Finally, we calculate an average broadband price Gains from Direct Broadcast Satellites and the Competition with Cable TV, 72 Econometrica 351 (2004), https://onlinelibrary.wiley.com/doi/abs/10.1111/j.1468-0262.2004.00494.x. 48 The speed tier cutoffs were chosen to correspond to quantity data available in the FCC Form 477 broadband subscription data collection. 49 FCC, Form 477 Resources, https://www.fcc.gov/economics-analytics/industry-analysis-division/form-477- resources (last visited Oct. 27, 2020). All FCC Form 477 data used in this Report have been certified as accurate by the filers. We note that the Report s analysis may understate or overstate consumers options for services to the extent that broadband providers fail to report data or misreport data. See FCC, Explanation of Broadband Deployment Data, https://www.fcc.gov/general/explanation-broadband-deployment-data (last visited Oct. 27, 2020) (describing quality and consistency checks performed on providers submitted data and explaining any adjustments made to the FCC Form 477 data as filed). 50 S&P Global, Estimated broadband-only homes as a percentage of wireline broadband households, Q1'18 vs. Q1'19 vs. Q1'20 (last accessed July 21, 2020). We used preliminary December 2019 FCC Form 477 subscription data for these calculations. We again note that the year-end FCC Form 477 data are preliminary only and are subject to corrections as appropriate by the service provider, and the final data will be published in due course by the agency. 51 See infra paras. 33-38, 45-47. 52 If a provider does not offer any plans in the product category, that provider s market share is distributed proportionally to the providers that do offer plans in the product category (i.e., the logit assumption). If no providers (continued& .) 42 Federal Communications Commission FCC 20-188 for each country by weighting these six product level prices by the estimated percentage of consumers in the United States that subscribe to each product category. The hedonic analysis, discussed in section IV.A, extends this analysis by constructing a price index that accounts for missing product prices, quality differences within product categories, and differences in the broadband cost and demand structures in each country. 16. Fixed Broadband Price Index Results. In Figure G-24 below, we present country rankings based on the fixed broadband price index, as well as this index divided by the average monthly data usage per subscriber to calculate a unit price measured in dollar per gigabyte of data consumption ($/GB).53 The United States ranks 21st out of 26 countries in standalone pricing but the ranking improves to 19th for broadband bundled with video service, due to more extensive bundle discounting.54 Combining standalone and bundled pricing, the overall ranking of the United States is 21st out of 26. On a price per GB of data consumed basis, the United States ranks 3rd out of the 18 countries for which we have usage data. However, it may not be appropriate to divide the monthly price by average monthly data consumption. The problem with doing so is that data consumption affects broadband pricing, and broadband pricing also likely affects data consumption in other words, data consumption is endogenous to price. For broadband services without usage allowances, the monthly subscription price should arguably not affect usage because the cost of additional data is zero once the access price is paid. The flaw in this reasoning is that consumers likely choose whether or not to adopt broadband based on their expected monthly data usage and how much they value that usage.55 If prices were higher in a country, then we would expect that consumers with lower expected data usage would be less likely to subscribe to broadband. Conversely, in countries with lower prices, we would expect more low-usage consumers to subscribe. As a result, dividing price by usage may unfairly advantage countries with higher prices and disadvantage those with lower prices. To account for higher data usage that may result from better applications and content, in our hedonic analysis we control for content quality using a proxy measure that is less susceptible to this reverse causality issue.56 This approach isolates the effect of content quality on prices and allows us to predict prices from the hedonic regression holding content quality fixed. in the country offer the highest product, we assign the next highest available product price to the highest missing product price(s). If no providers in a country offer any plans in a product category, we assign the next closest available product price to the missing category prices. See infra para. 36. 53 All reported prices for the broadband index are adjusted using a measure of PPP to make the results comparable to the income-adjusted hedonic index results. The figure presents the weighted average prices in each country for the indicated products. The Laspeyres index for each country would be calculated by dividing the given country s weighted price by the U.S. weighted price. 54 To calculate the price of broadband for each bundle offering, we first calculate the bundle discount as the difference between the total price of the standalone offerings for each service and the bundle. We then assume that this bundle discount is allocated to each component of the bundle in proportion to the standalone costs of each component. In this manner, we remove the video component price from the broadband bundle price. We also note that the bundle and standalone pricing measures are not strictly comparable in Fig. G-24 because the plans that are included in each calculation may be different. For this reason, the bundle price in a country may be higher than the standalone price. See infra Fig. G-33. 55 This is known as  selection bias in the econometrics literature. See James J. Heckman, Sample Selection Bias as a Specification Error, 47 Econometrica 153 (1979). 56 Access to a broad range of valuable applications and content over both fixed and mobile connections increases the value that each user derives from broadband service (i.e., content is a complement). To construct our measure of content quality, we perform a principal components factor analysis on the following four measures of content quality and availability: number of web pages in the country s primary domain(s), number of web sites in the top-level domain(s) (TLDs), the percentage of all web sites in the country s primary language, and English proficiency of the country. We then predict the first factor component based on the estimated factor loadings and use this as our measure of content quality. See infra paras. 64-65. 43 Federal Communications Commission FCC 20-188 Fig. G-24: Fixed Broadband Price Indexes (PPP Adjusted) Standalone Bundled Overall $/GB Country Mean Rank Mean Rank Mean Rank Mean Rank Australia 61.73 16 61.19 16 61.37 16 0.34 9 Austria 59.91 13 49.29 8 52.97 12 0.42 13 Belgium 50.90 9 50.18 11 50.43 10 0.34 11 Canada 69.93 23 67.39 22 68.27 22 0.35 12 Czech Republic 48.74 6 45.84 5 46.85 5 0.32 7 Denmark 48.53 5 48.53 6 48.53 6 0.22 4 Estonia 68.01 20 64.23 20 65.54 20 Finland 38.68 2 37.53 2 37.93 2 France 38.76 3 38.76 3 38.76 3 Germany 49.21 7 48.82 7 48.95 7 0.42 14 Greece 67.31 19 62.01 17 63.85 18 0.71 17 Iceland 72.82 24 72.82 24 72.82 24 0.27 5 Ireland 51.11 10 50.78 12 50.89 11 0.98 18 Italy 44.02 4 44.02 4 44.02 4 0.34 10 Latvia 35.34 1 33.10 1 33.88 1 0.15 1 Luxembourg 72.92 25 72.92 25 72.92 25 Mexico 69.87 22 69.87 23 69.87 23 Netherlands 63.57 17 63.57 18 63.57 17 New Zealand 59.95 14 59.95 14 59.95 14 0.34 8 Norway 84.50 26 74.51 26 77.98 26 Portugal 56.03 12 53.80 13 54.57 13 0.43 15 Spain 64.66 18 64.66 21 64.66 19 0.46 16 Sweden 51.28 11 49.90 10 50.38 9 Switzerland 60.05 15 60.05 15 60.05 15 0.32 6 United Kingdom 49.74 8 49.74 9 49.74 8 0.16 2 United States 68.74 21 64.23 19 65.80 21 0.19 3 Sources: International Telecommunications Union (ITU), World Telecommunications/ICT Indicators Database 2020 (24th Edition/July 2020) (last accessed Aug. 19, 2020); TeleGeography, GlobalComms Database (last visited Oct. 27, 2020); OpenVault, Broadband Industry Report 4Q 2019, Quarterly Advisories (Feb. 11, 2020), https://openvault.com/ovbi-median-broadband-usage-on-pace-to-surpass-250-gb-per-month-in-2020/. Note: To make the results comparable to the income-adjusted hedonic analysis, prices are reported in purchasing power parity (PPP) adjusted U.S. dollars. B. Fixed Broadband Hedonic Price Index 17. We estimate four hedonic regression models and then construct hedonic price indexes from each model. Our hedonic regression is a multilevel random coefficients model that allows the coefficients on some of the broadband plan characteristics (e.g., download speeds) to vary by broadband provider.57 From the regression model, the hedonic index is constructed by predicting provider-specific 57 See infra paras. 59-64. 44 Federal Communications Commission FCC 20-188 prices for each of our six standardized broadband products based on each provider s estimated coefficients. While the details of the hedonic modeling are contained in section IV.A, we summarize the basic approach here. The first model regresses the logarithm of broadband plan price on the plan characteristics to account for how plan characteristics explain differences in plan prices across countries. The second model builds upon the first by adding income per capita, a measure of terrain ruggedness, population density, and educational attainment into the model to capture how country-level differences in these broadband demand and cost factors influence observed pricing.58 The third model adds the percentage of households in the country that have access to speeds of at least 100 Mbps as a measure of network quality and investment.59 The final model adds our proxy measure for content availability and quality. 18. To calculate the hedonic price index, we predict provider-specific prices from the estimated hedonic regression for six standardized broadband plans. For these price predictions, we set the income per capita, terrain ruggedness, population density, education, and content quality variables at the U.S. values, and use the estimated provider-specific coefficients on product characteristics to predict prices. This procedure effectively estimates what each provider s price would be for each of the six standardized broadband products in each country if income per capita, terrain, population density, education, and content quality were at U.S. levels.60 We then aggregate these provider-specific price predictions for each of the six products using U.S. product share weights and the previously described Laspeyres price index formula, to arrive at the price that U.S. consumers would have to pay in each country for their broadband services if those countries had U.S. broadband cost and demand conditions. 19. Fixed Hedonic Price Index Results. The estimated coefficients for the four fixed broadband hedonic models are shown in Figure G-25 below.61 Before reviewing the estimates, we first note that the estimated coefficients in our models are reduced form estimates of how prices are correlated with product characteristics and country-level factors, so they should not be given a causal interpretation for how we would expect price to change if, for example, the income level of a country increased. Despite this issue, the coefficients generally align with expectations and are often statistically significant. The model estimates that higher speed plans cost more and the rate of increase in price (i.e., slope) is higher for plans at a higher speed tier.62 Bundling broadband with other services is estimated to lower the price of the broadband service by approximately 4.7% on average across all countries.63 A 1% higher 58 Our measure of terrain in each country is the population weighted terrain ruggedness index calculated in Nathan Nunn and Diego Puga, Ruggedness: The Blessing of Bad Geography in Africa, 94 Review of Economics and Statistics 20-36 (2012). See infra Section IV. 59 We do not control for observed broadband performance characteristics in each country (e.g., actual download and upload speeds, latency, etc.) because the general practice of pricing fixed broadband access by speed tier would influence these observed network performance measures. Lower prices for higher speed tiers would tend to increase measured download speed and vice-versa. This would create an endogeneity problem in the regression and bias the estimated coefficients. Network deployment measures are less susceptible to this issue because such measures are not directly affected by broadband pricing. 60 We predict prices from the hedonic regression for broadband plans at the following download speeds for both standalone and bundled plans: 25 Mbps, 100 Mbps, and 1000 Mbps. All other plan characteristics are the same in order to make prices comparable across countries. The other features of the plans used to predict prices are as follows: no contract, no phone service, and an unlimited data usage allowance. 61 The estimated random coefficient variances are provided in Fig. G-36. 62 The effect of download speeds on broadband prices is estimated as a piecewise linear spline with three download speed cutoffs. A linear spline allows the estimated coefficients to be different between for the range of download speeds between each cutoff. For example, our estimated coefficients imply that price of fixed broadband increases more steeply for plans with download speeds above 100 Mbps compared to those below 25 Mbps. 63 When a dependent variable is measured in log form, the percentage change in the dependent variable for a change (continued& .) 45 Federal Communications Commission FCC 20-188 data usage allowance is estimated to increase price by about 0.1% in all models. For the country-level control variables, we find that the per capita income in a country has a large and statistically significant effect on prices. Both the population density and educational attainment variables are statistically insignificant. However, our other broadband cost proxy variable, terrain ruggedness, has a large and statistically significant effect on fixed broadband prices. In Model 4, we estimate that a 1% increase in terrain ruggedness increases broadband prices by nearly 0.2%, and this is statistically significant at the 1% level. Finally, as observed in Model 4, the proxy variable for content availability and quality also has a strong positive effect on broadband prices, and this is also significant at the 1% level. in a dummy variable from 0 to 1, or a logged continuous independent variable, is calculated as 100[exp(²)  1]. A dummy, or indicator, variable refers to a binary variable that can take only the values 0 and 1. See, e.g., James H. Stock & Mark W. Watson, Introduction to Econometrics 145 (4th ed. 2019). 46 Federal Communications Commission FCC 20-188 Fig. G-25: Fixed Broadband Hedonic Regressions Model 1 Model 2 Model 3 Model 4 Log Average Monthly Price (USD) Coef. SE p Coef. SE p Coef. SE p Coef. SE p Spline: 0 < Mbps < 50 0.068 0.018 0.000 0.072 0.018 0.000 0.071 0.018 0.000 0.071 0.018 0.000 Spline: 50 d" Mbps < 100 0.122 0.038 0.002 0.118 0.038 0.002 0.118 0.038 0.002 0.123 0.038 0.001 Spline: 100 d" Mbps d" 1000 0.196 0.022 0.000 0.192 0.022 0.000 0.192 0.022 0.000 0.193 0.022 0.000 Bundle Dummy -0.047 0.013 0.000 -0.047 0.013 0.000 -0.047 0.013 0.000 -0.047 0.013 0.000 Fixed Voice Dummy -0.012 0.040 0.762 -0.006 0.040 0.880 -0.003 0.040 0.933 0.000 0.040 0.999 Log Contract Length -0.033 0.017 0.055 -0.034 0.017 0.041 -0.032 0.017 0.058 -0.033 0.017 0.051 Unlimited Data Dummy -0.096 0.070 0.172 -0.087 0.070 0.215 -0.087 0.070 0.212 -0.081 0.070 0.248 Log Data Cap Allowance 0.110 0.023 0.000 0.104 0.023 0.000 0.105 0.023 0.000 0.103 0.023 0.000 Log GNI Per Capita 0.426 0.109 0.000 0.410 0.113 0.000 0.318 0.101 0.002 Log Non-Rural Population Density -0.033 0.049 0.501 -0.029 0.049 0.560 -0.001 0.043 0.974 Educational Attainment 1.173 0.884 0.184 1.066 0.896 0.234 0.568 0.777 0.465 Log Terrain Ruggedness Weighted by Population 0.113 0.062 0.067 0.121 0.063 0.054 0.174 0.056 0.002 Coverage (% Households with > 100 Mbps) 0.135 0.226 0.550 0.218 0.192 0.257 Content Quality (1st Principal Component) (Standardized) 0.134 0.044 0.002 Constant 2.678 0.145 0.000 -1.902 1.030 0.065 -1.822 1.043 0.081 -0.859 0.944 0.363 Number of Observations 954 954 954 954 Log Likelihood 82.4 94.1 94.3 98.1 Likelihood Ratio Test vs. Linear Model P-Value 0.000 0.000 0.000 0.000 Note: The estimated random coefficient variances and measures of goodness of fit are provided in Fig. G-36 of this appendix. 47 Federal Communications Commission FCC 20-188 20. The resulting country rankings under each model are shown in Figure G-26 below. This figure reports the overall rankings that aggregate over the three standalone and three bundled products in each country. In Model 1, after adjusting for only broadband plan characteristics, we find that the United States ranks 19th out of the 26 countries in our sample, with an average broadband price of $65.54. Countries with lower average incomes like Latvia, the Czech Republic, and Estonia rank near the top before we correct the price levels for per capita income. In Model 2, after we correct price levels for differences in income, terrain, education, and population density, we find that the United States ranks 12th. The change in ranking from the first model is due to the United States having relatively high income and educational levels and more rugged terrain compared to the other countries in our sample.64 Model 3 includes the percentage of households with access to broadband connection speeds of at least 100 Mbps, and the U.S. ranking remains at 12th. Model 4 adds our content quality proxy variable into the hedonic regression and results in the United States ranking 2nd least expensive out of the 26 countries. Fig. G-26: Fixed Broadband Hedonic Price Indexes Model 1 Model 2 Model 3 Model 4 Country Price Rank Price Rank Price Rank Price Rank Australia 87.41 25 101.53 25 105.94 25 125.38 23 Austria 59.28 16 73.73 18 74.23 18 90.85 18 Belgium 57.69 13 68.48 14 68.13 14 96.74 20 Canada 65.76 20 78.88 22 79.03 20 86.59 13 Czech Republic 31.87 2 55.78 5 56.50 5 71.81 6 Denmark 50.11 11 58.58 7 58.49 6 81.38 8 Estonia 48.02 8 78.44 21 80.57 21 119.62 22 Finland 47.25 7 56.12 6 58.94 7 88.99 14 France 35.29 4 48.75 2 50.22 3 69.51 4 Germany 49.65 10 62.78 11 63.48 11 84.39 10 Greece 58.51 15 90.20 23 98.14 24 129.93 24 Iceland 68.67 22 61.78 9 63.35 10 89.74 16 Ireland 64.37 18 69.70 15 72.63 16 81.93 9 Italy 33.06 3 49.80 3 52.34 4 69.76 5 Latvia 17.88 1 36.74 1 36.24 1 51.58 1 Luxembourg 76.36 24 67.97 13 67.47 13 96.11 19 Mexico 46.12 6 122.31 26 120.94 26 142.07 26 Netherlands 61.48 17 91.12 24 91.41 23 132.63 25 New Zealand 67.46 21 74.98 19 76.51 19 86.24 12 Norway 89.96 26 72.66 17 73.32 17 101.99 21 Portugal 38.13 5 62.61 10 61.71 9 75.41 7 Spain 49.30 9 70.76 16 68.96 15 89.36 15 Sweden 53.47 12 59.71 8 60.40 8 85.32 11 Switzerland 69.44 23 50.33 4 49.79 2 68.02 3 United Kingdom 58.48 14 77.67 20 81.63 22 90.16 17 United States 65.54 19 65.48 12 65.63 12 65.61 2 64 See infra Fig. G-43. 48 Federal Communications Commission FCC 20-188 III. MOBILE BROADBAND PRICING ANALYSIS 21. The issues confronted when comparing mobile broadband pricing across countries are similar to those encountered in our fixed broadband pricing analysis with two exceptions. First, mobile plans are generally sold by data usage allowances instead of download speed, so we classify mobile products by data allowance rather than download speed. Second, the most prevalent form of bundling in mobile broadband involves the number of lines on a given plan rather than bundling mobile broadband with other telecommunications services. Cisco estimates that 79% of U.S. subscribers obtain their mobile service through multi-line data plans (i.e.,  family plans ).65 These bundled plans are offered at greatly discounted rates and need to be properly accounted for to reflect the prices that U.S. consumers actually pay for their mobile services. As in our fixed analysis, for mobile broadband we also define three single- line products, which are classified by the following data usage limits: less than or equal to 5GB per line; greater than 5GB but less than or equal to 20GB per line; greater than 20GB per line. We also define three additional multi-line products when these products are bundled with additional lines. A. Mobile Broadband Price Index 22. In this section, we compare mobile broadband prices by calculating a mobile broadband price index using the same Laspeyres formula and price index construction methodology we used for fixed broadband.66 23. Mobile Product Shares. To construct our mobile price indexes, we need to estimate the percentage of U.S. consumers who subscribe to each of our six mobile products defined by data usage allowance and number of lines. To estimate these product shares, we assume that consumers choose the optimal amount of data given their expected usage. We use Cisco data coupled with an assumption on the shape of the usage distribution to estimate the percentage of U.S. consumers who would find each usage allowance optimal.67 Based on the estimated log-normal distribution,68 in Figure G-27 below, we calculate the product shares for each of our six standardized mobile products. The column  Data Usage (Per Line) Share provides the estimated percentage of all subscribers from the estimated log-normal distribution that consume an amount of data within the corresponding ranges of data usage and number of lines on the plan. For example, 38% of all single-line plans in the United States are estimated to consume between 0 and 5 GB of data per line (product 1), while 50% of multi-line plans would be expected to consume this amount of data per line (product 4).69 We then multiply these estimated single-line and 65 See Cisco, Annual Internet Report (2018-2023) White Paper, Fig. 17 (2020), https://www.cisco.com/c/en/us/solutions/collateral/executive-perspectives/annual-internet-report/white-paper-c11- 741490.html. We are treating the share of  shared data plans as equivalent to the share of  multi-line plans in the United States. 66 We again calculate a Laspeyres price index that estimates how much consumers in the United States would pay for their mobile broadband plans in each of the comparison countries. The formula is identical to that used for fixed broadband. See supra paras. 12-15. 67 See infra Section IV and Cisco, Annual Internet Report (2018-2023) White Paper, Fig. 17 (2020), https://www.cisco.com/c/en/us/solutions/collateral/executive-perspectives/annual-internet-report/white-paper-c11- 741490.html. 68 See infra Fig. G-31 for the estimated log-normal parameters and distribution. 69 We use the terms  shared plan,  multi-line plan, and  family plan interchangeably in this report. However, some multi-line plans may have shared data among the lines, but some other multi-line plans have separate data allowances for each line. We do not distinguish between shared data and separate data allowances for multi-line plans. 49 Federal Communications Commission FCC 20-188 multi-line data usage shares by the percentage of all U.S. plans that are single versus multi-line to arrive at our final mobile product shares.70 Fig. G-27: Mobile Broadband Product Shares Data Data Allowance Usage (Per (Per Line) Bundling Line) Product Product Lines Tier Shares Share Share Plans 1 1 0 < GB d" 5 21.0% 38.0% 8.0% 101 2 1 5 < GB d" 20 21.0% 44.0% 9.2% 122 3 1 GB > 20 21.0% 18.0% 3.8% 182 4 2 0 < GB d" 5 79.0% 50.0% 39.5% 113 5 2 5 < GB d" 20 79.0% 39.0% 30.8% 124 6 3 GB > 20 79.0% 11.0% 8.7% 169 Sources: Cisco, Annual Internet Report (2018-2023) White Paper, Fig. 17 (2020), https://www.cisco.com/c/en/us/solutions/collateral/executive-perspectives/annual-internet-report/white-paper-c11- 741490.html. 24. Mobile Broadband Price Index Results. In Figure G-28 below, we present the country rankings based on the Laspeyres broadband price index formula. We present an index for single-line plans, another for multi-line plans, and an overall index that is a weighted average of the single- and multi-line plan indexes.71 The United States ranks 22nd out of the 26 countries in single-line plan pricing at $70.22, and is in the 21st place for multi-line pricing at $47.70 per line. Iceland ranks 1st in single-line plan pricing and multi-line pricing, at $26.52 per line per month and $23.63 per line per month, respectively. Combining single-line and multi-line data plan pricing, the overall ranking of the United States is 21st. Finally, due to the relatively high data usage of U.S. subscribers, on a dollar per GB basis, the U.S. ranking improves substantially to 15th place.72 70 For multi-line plans, we assume that the number of lines increases with the data usage allowance. We assume plans with over 20 GB of data have three lines on average while those below 20 GB have two lines. 71 The product prices by country that were used in the mobile broadband price index calculations are presented in Fig. G-40 of section IV.F and adjusted using a measure of PPP. 72 The same caveat given in the fixed analysis regarding the potential problems with dividing price by data usage also applies to mobile, although now the plans are sold by usage allowances so the endogeneity problem may be even more severe. 50 Federal Communications Commission FCC 20-188 Fig. G-28: Mobile Broadband Price Indexes (PPP Adjusted) Single-Line Multi-Line Overall $/GB Country Mean Rank Mean Rank Mean Rank Mean Rank Australia 33.11 8 31.67 9 31.97 10 9.43 14 Austria 30.43 5 27.73 6 28.30 6 1.73 2 Belgium 37.04 12 34.07 11 34.70 11 17.26 21 Canada 87.96 26 81.64 25 82.97 26 33.73 25 Czech Republic 69.70 21 64.25 23 65.39 23 19.29 22 Denmark 28.31 2 24.26 2 25.11 2 3.29 6 Estonia 33.51 9 25.37 3 27.08 3 2.76 3 Finland 31.97 7 31.97 10 31.97 9 1.65 1 France 29.76 3 27.58 5 28.04 5 4.97 8 Germany 49.25 20 37.73 16 40.15 17 15.74 20 Greece 76.12 24 66.08 24 68.19 24 44.57 26 Iceland 26.52 1 23.63 1 24.24 1 3.11 5 Ireland 36.57 11 36.57 14 36.57 14 5.40 9 Italy 44.52 18 44.52 20 44.52 20 10.43 16 Latvia 37.24 13 35.37 13 35.76 12 2.80 4 Luxembourg 30.77 6 26.31 4 27.25 4 6.83 10 Mexico 71.99 23 61.46 22 63.67 22 30.18 23 Netherlands 39.12 14 37.15 15 37.57 15 14.56 17 New Zealand 40.10 15 35.33 12 36.33 13 15.01 19 Norway 42.97 16 38.82 17 39.69 16 8.20 12 Portugal 83.92 25 82.69 26 82.95 25 31.42 24 Spain 47.23 19 39.87 18 41.42 18 14.64 18 Sweden 35.46 10 28.60 8 30.04 8 4.10 7 Switzerland 44.21 17 41.70 19 42.23 19 6.93 11 United Kingdom 29.79 4 27.97 7 28.36 7 8.44 13 United States 70.22 22 47.70 21 52.43 21 9.73 15 Note: To make the results comparable to the income-adjusted hedonic analysis, prices are reported in PPP adjusted U.S. dollars. B. Mobile Hedonic Price Index 25. The mobile broadband price index in Figure G-28 does not account for several factors that likely affect the observed price levels in each country, so we again extend the analysis by estimating four hedonic regression models to adjust prices for country-level differences in cost and demographic factors, differences in mobile broadband product characteristics, and content quality. We then predict prices out of these hedonic models for a standardized set of mobile broadband products at the U.S. averages of the country-level control variables. This approach again seeks to estimate the mobile broadband prices that would be observed in each country if that country had the mobile broadband cost and demand characteristics of the United States.73 To calculate our mobile hedonic price index, these 73 We predict prices from the hedonic regression for mobile broadband plans at the following data allowances for both single-line and multi-line plans: 5 GB, 20 GB, and 50 GB per line. For the multi-line products, the 5 GB and 20 GB plans have two lines each and the 50 GB plan has three lines. Both the single-line and three-line 50 GB plan (continued& .) 51 Federal Communications Commission FCC 20-188 predicted prices are then weighted in the same manner that we used to calculate the fixed hedonic price index. 26. The estimated coefficients for the four mobile broadband hedonic models are shown in Figure G-29 below.74 The four models presented in this section mirror the models in our fixed pricing analysis with the exception that the network quality variables now include measures of both network coverage and average download speeds.75 As expected, the regression coefficients imply that higher data usage allowances increase the expected price per line of a mobile broadband plan, while adding more lines to the plan is expected to lower the average price per line. Increasing the number of minutes on a plan by 1% is expected to raise the expected price per line by approximately 0.17%, while increasing the contract duration by a month would be expected to lower the price per line by about 0.14% across all four models. For mobile broadband, the estimated effects of the country-level variables on broadband prices differ from the patterns we observed in our fixed hedonic analysis. Surprisingly, the estimated effect of income on mobile broadband prices is negative, but this result is not statistically significant in any specification. However, educational attainment, a measure closely related to income, is found to increase expected mobile broadband prices, and this result is significant at the 5% level in Models 3 and 4. The estimated impact of our two cost proxy variables (terrain variability and population density) are similar to our findings for fixed broadband. Population density is again found to have weak and statistically insignificant effects on mobile broadband prices, while greater terrain variation in a country has a statistically significant positive effect on mobile broadband prices. As we would expect, higher network quality is associated with higher prices; however, only the 4G availability measure is statistically significant. Finally, in Model 4 we again find that our measure of content quality has a positive and statistically significant effect on mobile broadband prices, implying that consumers are willing to pay higher mobile broadband prices when they have access to higher quality and more diverse broadband content. are set to unlimited data without throttling. The other plan features for the price predictions are as follows: no contract, unlimited minutes, and unlimited texts. 74 The estimated random coefficient variances and measures of goodness of fit are provided in Fig. G-41 of section IV.F. 75 Mobile plans are not generally sold by speed, so the endogeneity issues regarding the inclusion of observed network performance measures are less of a concern in mobile than fixed broadband pricing analysis. 52 Federal Communications Commission FCC 20-188 Fig. G-29: Mobile Broadband Hedonic Regressions Log Average Monthly Price Per Line Model 1 Model 2 Model 3 Model 4 (USD) Coef. SE P Coef. SE p Coef. SE p Coef. SE p Spline: 0 < GB d" 5 0.185 0.025 0.000 0.185 0.025 0.000 0.184 0.025 0.000 0.182 0.025 0.000 Spline: 5 < GB d" 20 0.251 0.022 0.000 0.251 0.022 0.000 0.252 0.022 0.000 0.252 0.022 0.000 Spline: 20 < GB 0.218 0.036 0.000 0.218 0.036 0.000 0.219 0.036 0.000 0.218 0.036 0.000 Number of Lines -0.032 0.006 0.000 -0.032 0.006 0.000 -0.032 0.006 0.000 -0.032 0.006 0.000 Unlimited Data Dummy 0.122 0.016 0.000 0.122 0.016 0.000 0.122 0.016 0.000 0.122 0.016 0.000 Log Contract Length -0.140 0.034 0.000 -0.145 0.035 0.000 -0.141 0.035 0.000 -0.138 0.034 0.000 Unlimited Minutes Dummy -0.453 0.091 0.000 -0.453 0.091 0.000 -0.450 0.091 0.000 -0.449 0.091 0.000 Log Minutes 0.170 0.022 0.000 0.170 0.022 0.000 0.169 0.022 0.000 0.169 0.022 0.000 Unlimited Text Messages Dummy 0.022 0.087 0.798 0.028 0.087 0.748 0.021 0.087 0.810 0.018 0.087 0.835 Log Text Messages -0.070 0.016 0.000 -0.069 0.016 0.000 -0.069 0.016 0.000 -0.069 0.016 0.000 Throttle Dummy -0.235 0.028 0.000 -0.235 0.028 0.000 -0.236 0.028 0.000 -0.235 0.028 0.000 Log GNI Per Capita -0.090 0.226 0.689 -0.219 0.236 0.354 -0.345 0.230 0.133 Log Country Population Density 0.012 0.067 0.858 -0.012 0.064 0.852 0.031 0.063 0.623 Educational Attainment 3.318 1.792 0.064 4.452 1.765 0.012 3.721 1.672 0.026 Log Terrain Ruggedness Weighted by Population 0.230 0.126 0.067 0.277 0.119 0.020 0.319 0.113 0.005 4G Availability 2.557 1.212 0.035 2.231 1.133 0.049 Download Speed 0.001 0.003 0.748 0.002 0.003 0.430 Content Quality (1st Principal Component) (Standardized) 0.194 0.101 0.054 Constant 2.355 0.163 0.000 2.405 2.156 0.265 1.244 2.150 0.563 2.802 2.161 0.195 Number of Observations 1639 1639 1639 1639 Log Likelihood 708.4 711.0 713.0 714.7 Likelihood Ratio Test vs. Linear Model P-Value 0.000 0.000 0.000 0.000 Note: The estimated random coefficient variances and measures of goodness of fit are provided in Fig. G-41 of this appendix. 53 Federal Communications Commission FCC 20-188 27. Mobile Hedonic Price Index Results. Our hedonic price indexes based on the four estimated hedonic regressions are provided in Figure G-30. For mobile broadband service, adjusting for cost and demographic factors does not have as large of an impact on the U.S. ranking as we observed for fixed broadband service. In Model 1, before adjusting for income, terrain, educational attainment, and population density factors, the United States ranks 24th among the 26 countries in mobile broadband pricing. Correcting for these factors in Model 2 changes the U.S. ranking to 22nd. Adding the network performance measures in Model 3 improves the U.S. ranking to 17th. And finally, the United States ranks 7th in mobile broadband pricing after adding the content quality proxy measure in Model 4. Fig. G-30: Mobile Broadband Hedonic Price Indexes Model 1 Model 2 Model 3 Model 4 Country Price Rank Price Rank Price Rank Price Rank Australia 29.96 10 38.17 15 43.02 11 62.78 12 Austria 26.50 8 36.15 13 51.42 15 75.21 15 Belgium 34.79 13 31.09 9 35.30 5 59.61 8 Canada 69.23 25 78.95 25 81.06 24 100.36 22 Czech Republic 38.59 15 45.97 18 55.78 18 81.21 17 Denmark 24.05 6 30.30 8 39.13 7 61.42 10 Estonia 18.07 1 19.82 1 26.19 1 45.54 2 Finland 22.70 5 25.75 4 29.61 2 54.83 6 France 26.58 9 36.53 14 52.83 16 77.66 16 Germany 46.18 18 54.38 21 78.64 23 115.21 24 Greece 138.24 26 114.81 26 138.31 26 204.90 26 Iceland 31.23 12 24.61 3 51.41 14 83.35 19 Ireland 22.29 4 21.21 2 39.68 9 49.59 3 Italy 21.52 2 29.09 6 41.28 10 62.31 11 Latvia 22.18 3 29.94 7 39.52 8 63.95 14 Luxembourg 49.64 20 42.61 16 62.80 20 96.90 21 Mexico 46.36 19 59.11 23 78.13 22 110.74 23 Netherlands 44.96 17 69.46 24 82.01 25 126.39 25 New Zealand 36.34 14 34.22 12 47.43 12 53.57 4 Norway 52.22 22 46.78 19 48.06 13 82.77 18 Portugal 50.70 21 52.52 20 65.61 21 90.79 20 Spain 24.44 7 26.31 5 30.52 3 44.77 1 Sweden 30.55 11 33.45 11 36.00 6 59.75 9 Switzerland 54.29 23 32.87 10 33.85 4 53.61 5 United Kingdom 40.99 16 43.72 17 56.24 19 63.04 13 United States 55.65 24 55.65 22 55.70 17 55.88 7 IV. DATA AND METHODOLOGY A. Hedonic Model 28. While the classic hedonic framework involves adjusting for changing product quality over time, accounting for product quality differences across firms and countries is analogous. In the 54 Federal Communications Commission FCC 20-188 equation below, we present a standard linear hedonic regression of prices on product characteristics.76 The dependent variable, ln(CÔVÔXÔ), is the logarithm of the price of plan i in country k, Xi is a vector of plan characteristics, and ×VÔXÔ is a scalar idiosyncratic error term. Under this approach, the country specific intercepts, üÖXÔ, estimate the differences in the average quality-adjusted price levels across countries. This framework has been widely used in making temporal and spatial price comparisons; however, it is not ideal for cross-country broadband pricing comparisons because it assumes that coefficients on product characteristics (the slope parameters ýÖ) are the same for each country.77 While it is plausible that the supply and demand conditions that generate the ýÖ coefficients could be similar in adjacent time periods, or even cities, within the same country, it is highly unlikely that these conditions are similar across countries. If broadband cost structures, determinants of demand (e.g. demographics), product offerings, ownership structures, regulatory conditions, subsidies, or other conditions that impact prices vary across countries, then we would expect the slope parameters to reflect these differences. ln(CÔVÔXÔ) = üÖXÔ + KÔVÔýÖ + ×VÔXÔ 29. We estimate a more flexible model that allows the slope coefficients for certain characteristics to differ across providers. However, due to sample size limitations in our pricing data, we do not estimate all of the j possible slope parameters for each product characteristic at the provider level but rather use multilevel modeling techniques similar to those recently proposed in broadband price hedonic work at the OECD.78 The multilevel model recognizes that plans are nested within providers which are nested within countries and that prices are likely correlated within these nests. Rather than estimating separate parameters for each provider and product characteristic, the model assumes normally distributed zero-mean random coefficients on some product characteristics at the provider level and then estimates the variance of each random coefficient. The model is therefore more parsimonious because it estimates a single unknown variance parameter for each product characteristic rather than a separate slope parameter for each provider by product characteristic combination. Our base multilevel hedonic pricing equation (Model 1 in Figures G-25 and G-29 above) is as follows. 30. To explain why prices may differ across countries, we also include some exogenous supply and demand shifters into the model that we expect to explain why broadband quality-adjusted price levels may differ by country. In the standard model, these factors are absorbed in the country fixed effect, so instead of including this fixed effect we parametrize the more traditional country effect as a random effect plus country-level supply and demand factors that we expect to be correlated with average price levels. This allows us to remove the effect of these country-level supply and demand conditions when predicting prices rather than having them remain in the price predictions as they would in a fixed effect specification. ln(CÔVÔWÔXÔ) = KÔVÔýÖ + MÔXÔþÖ + KÔVÔýÖWÔ + ×WÔ + ×XÔ + ×VÔWÔXÔ, where " CÔVÔWÔXÔ is the price for plan i, offered by provider j, in country k; 76 See Zvi Griliches, Hedonic Price Indexes for Automobiles: An Econometric Analysis of Quality Change, National Bureau of Economic Research (NBER) (1961), https://www.nber.org/system/files/chapters/c6492/c6492.pdf. 77 See W. Erwin Diewert et al., Hedonic Imputation versus Time Dummy Hedonic Indexes in Price Index Concepts and Measurement, NBER (Dec. 2009), https://www.nber.org/system/files/chapters/c5073/c5073.pdf. 78 See Carol Corrado and Olga Ukhaneva, Hedonic Prices for Fixed Broadband Services: Estimation Across OECD Countries (Oct. 20, 2016), https://www.oecd-ilibrary.org/science-and-technology/hedonic-prices-for-fixed- broadband-services_5jlpl4sgc9hj-en;jsessionid=yPSoFOaGChbj-Yk8Cf8ZedL3.ip-10-240-5-72. These models are also called  random effects models,  hierarchical linear models, and  mixed models. 55 Federal Communications Commission FCC 20-188 79 " KÔVÔ is a vector of plan characteristic variables; " ýÖ is a vector of unknown fixed coefficients; " MÔXÔ is a vector of country characteristics (e.g., measures of income and population density) for the country in which the given plan is offered; " þÖ is a vector of unknown, fixed coefficients for the country characteristics; " KÔVÔ is a subset of the variables in KÔVÔ for which the coefficients will be treated as random realizations for each provider in each country; " ýÖWÔ is a vector of random coefficients for the variables included in KÔVÔ. These random coefficients apply to all plans of provider j. We assume that 8Ô[ýÖ] = 0, 6Ô\ÔcÔ[ýÖ, ×] = 0, NÔ[ÔQÔ IÔNÔ_Ô[ýÖ] = :Ô;80 " ×WÔ is a random coefficient applying to all plans offered by provider j; " ×XÔ is a random coefficient applying to all plans offered in country k; and " ×VÔWÔXÔ is an idiosyncratic error term. 31. The multilevel model is estimated by maximum likelihood estimation (MLE) as follows. In matrix form, the model can be written as:81 ln(]Ô) = KÔýÖ + KÔýÖ + MÔþÖ + × 32. The n × 1 vector of errors × is assumed to be distributed mean zero multivariate normal 2 with variance-covariance matrix à In. We also assume that ýÖ is mean zero, orthogonal to ×, and has variance-covariance matrix G. This implies the following: ýÖ :Ô 0 Var [ ] = [ 2 ] × 0 Ã× <Ô[Ô 33. Letting bÔ = KÔýÖ + × be the combined error term, we see that ln(p) is distributed multivariate normal with mean KÔýÖ + MÔþÖ and the following variance-covariance matrix. 2 2 IÔ = MÔ:ÔMÔ + Ã× <Ô[Ô 79 The plan characteristics included in Xi for fixed broadband are three splines of download speed, a dummy variable for whether the plan is bundled with video service, a dummy for whether fixed voice is included, the log of contract length (in months), a dummy variable for whether more than 2000 GB of data is included (i.e., unlimited data), and the log of the data usage allowance. For mobile broadband, they are three splines of data usage allowances, the number of lines, an unlimited data dummy, the log of contract length, an unlimited minutes dummy, the log of the number of minutes, an unlimited text messages dummy, the log of the number of text messages, and a dummy for whether the plan throttles speed. Since the inclusion of too many variables can result in the statistical problem of  overfitting the data, we did not include all observed product characteristics in the model and limited the random coefficients to only those we determined were key product characteristics that likely had the greatest impact on consumer choices. 80  The model does not estimate the random coefficients ýÖ, ×WÔ, or ×XÔ, but instead estimates the diagonal variance elements of the variance-covariance matrix G, known as the variance components. The off-diagonal covariances are assumed to be zero. When predicting prices for each provider, we use the best linear unbiased predictors (BLUPs) of the random coefficients based on the estimated variance components. 81 In the matrix representation, the provider and country random effects are now included in the vector of random coefficients ýÖ. 56 Federal Communications Commission FCC 20-188 34. Letting × be a vector of the unknown variance components of G, we have the following 2 likelihood function that is used to find the unique vectors ýÖ, × and Ã× that maximize this likelihood of observing our data sample.82 1 ?Ô(ýÖ, ×, Ã2) = {" n ln(2 ×) + ln|IÔ| + (ln(]Ô) " KÔýÖ " MÔþÖ)2 IÔ"1(ln(]Ô) " KÔýÖ " MÔþÖ)} × 2 35. Following estimation of the model, we predict broadband prices for each provider for a set of standardized plans. Since the random effects ýÖ are not directly estimated, we calculate them post- estimation by using the following best linear unbiased estimator of the random effects, where variables with ^ denote estimated objects from the MLE. OÔ = :Ô2 KÔ2 IÔ "1(ln(]Ô) " KÔýÖ " MÔþÖ) 36. The predicted price for any one of the six standardized plans used to compare prices across countries is then given by the following formula. ln(CÔVÔWÔXÔ) = KÔVÔýÖ + MÔXÔþÖ + KÔVÔOÔWÔ + ×WÔ + ×XÔ 37. The random coefficients on product characteristics measure how each provider s pricing of the characteristic differs from the pricing of the average provider in the sample as measured by the coefficient ýÖ.83 In our fixed broadband hedonic models, the product characteristics with random coefficients are three download speed splines, the bundling dummy variable, and the logarithm of the plan s contract length.84 In our mobile broadband hedonic models, there are random coefficients on three data usage allowance splines, the number of lines, and the logarithm of contract length.85 38. In an imperfectly competitive market such as broadband, there is no meaningful interpretation of the hedonic regression coefficients. Under perfect competition, the coefficient vector ýÖ estimates both the marginal consumer value and marginal production costs for each product characteristic.86 However, in markets like broadband with substantial fixed costs, the coefficient also includes the markup over cost for that characteristic, and these markups are complex functions of the characteristics of competing products, firm costs, consumer preferences, and market structure.87 As such, in imperfectly competitive markets, hedonic coefficients should only be considered a reduced-form 82 We use the Stata mixed command to estimate the model. For further details on the maximum likelihood estimation routine, see StataCorp LP, STATA Multilevel Mixed-Effects Reference Manual Release 13, https://www.stata.com/manuals13/me.pdf (last visited Oct. 27, 2020). 83 See infra Fig. G-36 and Fig. G-41 for the fixed and mobile broadband, respectively, estimated variances of the random coefficients. 84 We control for download speed using a linear spline in the logarithm of download speed with knot points at the top-end of our speed categories used to define the six broadband products (i.e., knots at 50 and 100 Mbps). 85 We control for data allowance using a linear spline in the logarithm of the data allowance with knot points at the top-end of our data allowance categories used to define mobile broadband products with the three highest data allowances (i.e., knots at 5 and 10 GB). 86 See Sherwin Rosen, Hedonic Prices and Implicit Markets: Product Differentiation in Pure Competition, 82 Journal of Political Economy 34-55 (1974), https://www.jstor.org/stable/1830899?seq=1#metadata_info_tab_contents. 87 See Ariel Pakes, A Reconsideration of Hedonic Price Indexes with an Application to PCs, 93 American Economic Review 1578-96 (2003); Robert C. Feenstra and Gordon H. Hanson, Foreign Investment, Outsourcing and Relative Wages, 5121-NBER (May 1995), https://www.nber.org/system/files/working_papers/w5121/w5121.pdf; Diane Bruce Anstine, How Much Will Consumers Pay? A Hedonic Analysis of the Cable Television Industry, 19 Review of Industrial Organization 129-147 (2001), https://www.jstor.org/stable/41799034?seq=1. Even if the broadband market is competitive in a country, pricing will still need to be above marginal cost for firms to recover their fixed deployment costs. 57 Federal Communications Commission FCC 20-188 description of how prices (costs plus markups) vary with changes in product characteristics. The focus should not be on the particular value, sign, or precision of any one coefficient but rather on how predictive the hedonic pricing function is of provider prices in each country.88 We therefore follow a standard hedonic approach, except we correct price levels for exogenous country-level factors that we expect to be correlated with costs and markups. 39. The last issue that we need to account for in the hedonic regression is product bundling. As noted above, most U.S. consumers purchase broadband and video service in a bundle at steeply discounted rates.89 Further, it is very difficult to compare multichannel video products across countries. The product offerings in terms of channels included are completely different across countries and the same content may be highly watched in some countries (e.g., American football in the United States) but uninteresting to most viewers in another country (e.g., American football in Europe). Therefore, unlike broadband, where a download speed of 25 Mbps is a product characteristic where more of the characteristic is always better (i.e. vertical characteristics), there is no standardized video product that would be comparable across countries that would hold consumer utility fixed. While many studies attempt to control for video quality differences based on observable product characteristics and because we do not believe the observable measures adequately capture quality differences across countries, we calculate a bundle discount and allocate this across the standalone component pricing as described below to isolate the price of broadband when purchased in a bundle. B. Fixed and Mobile Broadband Price Index Calculations 40. We use the same general methodology to calculate the fixed broadband and mobile broadband price indexes in Figure G-24 and Figure G-28, respectively. The supplementary figures of broadband prices by product referenced here are available in section IV.F Supplementary Figures. 41. Step 1. We calculate the unweighted average price of all plans for each provider within each product type.90 Therefore, each provider has up to six product prices. 42. Step 2. Next, we calculate a weighted average price of each product category across providers, using provider market shares as the weight. If a provider does not offer any plans in a particular product category, the weight is proportional to only those providers that do offer a product in the given product category.91 Figure G-35 and Figure G-40 display the country-level product prices for fixed broadband and mobile broadband, respectively. 43. Step 3. There are cases in which no provider in a country offers plans in a product category, so we make assumptions about missing country-level product prices. First, if a bundled product price is missing, we replace it with the corresponding standalone product price (i.e., setting the bundle discount to zero). Next, if the highest tiered product(s) are not offered, we set the missing product prices to the next available product price. For example, if no providers in the country offer products 2 and 3, then we set product 2 s and product 3 s prices to product 1 s price. Finally, for any remaining missing product prices, we set these to the next highest available product price.92 For example, if a country s providers only offer products 1 and 3, then product 2 s price is set to product 3 s price. 88 See Ariel Pakes, A Reconsideration of Hedonic Price Indexes with an Application to PCs, 93 American Economic Review 1578-96 (2003). 89 See supra para. 14. 90 This calculation includes  synthetic plans. See infra paras. 46, 59 for a discussion of synthetic plans. 91 If only one provider in a country offers plans in a product category, that provider s unweighted average price would represent 100% of the country level product price. 92 This assures that U.S. consumers are at least as well off with the product provided as they would have been with the product available in the United States. 58 Federal Communications Commission FCC 20-188 44. Step 4. Finally, we calculate the price indexes using the full set of country-level product prices from Step 3, and the product shares in Figure G-23 for fixed broadband and Figure G-27 for mobile broadband.93 For fixed broadband, we calculate the overall standalone price and overall bundled price using the download speed shares in Figure G-23. For mobile broadband, we calculate the overall single- line price and overall multi-line price using the data usage shares in Figure G-27. To calculate the overall broadband price, we use the bundle shares to weight the overall standalone price and overall bundle price. 45. Step 5. To produce per GB rankings, we divide the overall broadband price calculated in the prior step by the average monthly data usage in each country.94 C. Fixed Broadband Pricing Data Collection 46. Collection of Broadband Prices and Timeframe. We collected fixed residential broadband plan prices and terms from 82 providers in 26 countries, including the United States, between April and July 2020. To determine which providers to sample in each comparison country, we used the TeleGeography GlobalComms Database to select providers with broadband market shares of at least 10% nationally as of December 2019.95 This threshold was chosen to balance data collection costs against the desire to obtain a representative sample of broadband pricing.96 For each provider, we collected plans from 10 randomly selected addresses from the country s capital city.97 These addresses were then entered into providers websites to determine the product offerings at each address. While many providers websites displayed general  promotional splash page plan offerings, entering an address allowed us to capture the variation in product availability within a city, as well as more detailed pricing information.98 Where we could not collect address-level plan data, we collected  promotional splash page plans.99 47. For each provider, we recorded each combination of download speed, upload speed, data usage allowance, and technology (D/U/A/T). For example, a provider offering a fiber-based plan with 100 Mbps download, 100 Mbps upload, and no data cap; a fiber-based plan with 100 Mbps download, 50 93 See supra para. 12 for the price index formula. TeleGeography GlobalComms Database, (last visited Oct. 27, 2020). International Telecommunications Union, World Telecommunications/ICT Indicators Database 2020 (24th Edition/July 2020) (last accessed Aug. 19, 2020). OpenVault, Broadband Industry Report 4Q 2019, Quarterly Advisories (Feb. 11, 2020), https://openvault.com/ovbi-median-broadband-usage-on-pace-to-surpass-250-gb-per- month-in-2020/. 94 For fixed broadband, we only have monthly average usage per subscriber data for 18 of the 26 countries. For mobile broadband, we rely on OECD monthly average usage per subscriber. OECD, Broadband Portal, https://www.oecd.org/sti/broadband/broadband-statistics/ (last visited Oct. 27, 2020). 95 TeleGeography, GlobalComms Database (last visited Oct. 27, 2020). We obtained these data as of February 2020. There is one exception to the 10% rule: Verizon is estimated to have a national broadband market share below 10% in the United States, but it was sampled as it is the largest Fiber to the Premises (FTTP) provider as well as the second largest Incumbent Local Exchange Carrier. 96 On average, our sample covers about 90% of all broadband subscribers over all 26 comparison countries. The lowest total market share is just under 70% while most countries have over 90% total market share covered in our sample. 97 In some cases, a provider did not offer service in the capital city (e.g., AT&T in Washington, D.C.), this required collecting some providers plans from another city. Additionally, when capital cities were not major cities in the given country (e.g., Canberra, Australia), we collected plans from another major city, in addition to the capital city. See 2018 International Broadband Data Report, 33 FCC Rcd at 1027-28, para. 14. 98 If we were able to collect address level plans, we only collected plans that were available for at least one address. Therefore, plans that were advertised on  promotional splash pages may not have been collected if these plans were not available at any of the 10 addresses. 99 Some providers do not provide an option to enter an address to check available plans but instead require customers to call or e-mail to receive more information about availability of plans. 59 Federal Communications Commission FCC 20-188 Mbps upload, and no data cap; and a cable-based plan with 100 Mbps download, 100 Mbps upload, and no data cap has three separate plans recorded.100 Both standalone broadband plans as well as double play packages of broadband bundled with multichannel video services were collected.101 With some exceptions, we did not collect information on  triple play bundles of fixed voice phone, Internet, and video because the extent of the bundle discount received did not tend to increase with the addition of phone service and doing so would have greatly increased the data collection burden.102 In cases where a provider only offered Internet service to customers who also subscribed to fixed voice phone services, we collected Internet bundled with fixed voice phone service plans and any relevant bundled plans of Internet, fixed voice phone service, and television.103 In such cases, we collected triple play bundles from the provider that included that particular phone plan to isolate the bundled broadband price using the methodology described below. Finally, if the provider did not offer video service, bundle discounts, or standalone TV plans, we did not collect bundled plans for the particular D/U/A/T combinations for the provider.104 48. Given the large number of countries, providers, and product offerings, we limited the scope of the collection along several additional dimensions. First, we assumed customers were new to the provider and did not receive any special discounts that were not available to all new customers (e.g., student discounts). Second, we only recorded information for the combination of features that resulted in the lowest price for a given plan.105 For example, we did not include optional add-on features (e.g., HBO, security software, etc.), always chose the lowest priced equipment required for the plan, and assumed consumers were willing to sign up for a two-year contract if this offered the lowest price.106 Also, we did not include any plans with spectrum-based technologies (e.g., fixed wireless, satellite, 4G) and any plans with an advertised download speed of more than 1000 Mbps. 49. We collected three types of data for each plan: (1) general information; (2) pricing data; and (3) non-pricing data. General information captures information such as the name of the plan, date of collection, and currency of prices. For pricing data, we collected all pricing information available on the provider s website including promotions, equipment fees, installation fees, and rebates, in order to calculate the total cost of the broadband service plan over a two-year time horizon. Non-pricing data includes information such as download and upload speeds, data usage allowances, number of channels (if 100 We excluded plans with download speeds above 1000 Mbps as these are generally non-residential offerings. 101 By multichannel video services, we mean linear television packages usually offered using cable, satellite, or Internet with regularly scheduled programs. Over the Top services, which stream programs to specific users, that are bundled with a broadband plan are not considered in our analysis and are thus unobserved product characteristics if they are included in any plans. See supra Section II.D.1. 102 Additionally, we did not collect fixed broadband plans bundled with mobile voice and data services. 103 In cases where fixed voice phone plans are bundled in the plan, we always chose the lowest priced fixed voice phone package and indicated that fixed voice phone service is included in the bundled plan. 104 In the 2018 International Broadband Data Report, we collected bundled plans even when providers did not offer bundle discounts (i.e., add-on pricing), resulting in bundle discounts of 0%, and when providers did not offer standalone TV plans that were needed in our bundle discount calculation, requiring making assumptions about standalone TV price. In this report, we only collected information of bundled plans when the provider offered a clear discount for bundling Internet and TV service (e.g., a plan with a bundle discount due to duplicative installation or activation fees was not eligible for collection). 105 Essentially, if a provider offered multiple plans that would have appeared identical within our data framework, we recorded the lowest priced plan. This approach would exclude any optional add-on products. 106 More generally, if a provider offered the same plan with different contract length options with discounts for longer contracts, we chose the longest contract length available (up to 24 months). 60 Federal Communications Commission FCC 20-188 applicable), and contract length. A unique plan is defined by country, city, provider, broadband plan, TV plan, phone service, technology, download speed, upload speed, and data allowance. 50. Data Review and Cleaning Process. Upon completion of the data collection, we reviewed the data for accuracy and completeness. When the variables essential for the analysis were unavailable, we made the following assumptions to impute the missing data: " If a provider did not explicitly state the length of the contract, we assumed the plan was month-to-month (i.e., one month). " When generally advertised download speeds were not reported, but providers displayed address-specific download speeds, we used the average download speed across addresses for which the plan was available. " If the provider s website did not list a data allowance, we assumed the plan offered an unlimited data allowance. " If a plan advertised a promotional price without specifying duration, we assumed the promotion lasted 12 months. " If the regular monthly price was not found, we assumed that the last available promotional price stayed in effect for the remaining period. " If equipment prices were not available, we assumed the relevant equipment was included. " If activation fees, installation fees, and other recurring and non-recurring fees and rebates were not listed clearly on a provider s website, we assumed that these fees were included or did not apply to the plan. " For Canada and the United States, if taxes were not explicitly stated as included in the list prices and not reported separately, we added a percentage to the total pre-tax prices.107 For all other countries, we assumed taxes were included.108 51. We also made two other assumptions that apply to only two specific providers: " For one of Iceland s providers that did not display download speeds, we assumed the same download speed as all the plans offered by Iceland s other two providers (1000 Mbps). In Iceland, plan prices varied by data usage allowance, not download speed. " For one of New Zealand s providers that did not display a download speed for its two ADSL plans, we assumed the same download speed as another of New Zealand s provider s ADSL plan (20 Mbps). 52. Broadband Price Calculation. After cleaning the data, we calculated the total cost of each plan over the first 24 months. A 24-month price was selected to produce a comparable pricing measure across plans that accounted for all promotional and regular pricing and to amortize one-time fees over a sufficiently long-term horizon. This total 24-month price was calculated using the formula below: 107 International Telecommunications Union, World Telecommunications/ICT Indicators Database 2020 (24th Edition/July 2020) (last accessed Aug. 19, 2020). 108 Outside of the United States and Canada, most providers note that listed prices included taxes (VAT). In the United States and Canada, providers generally stated prices that did not include taxes. In some cases, taxes were not included in prices but were reported separately, in which case we were able to add the reported tax (i.e., we didn t apply a percentage of the pre-tax total price to estimate the tax). 61 Federal Communications Commission FCC 20-188 CÔ_ÔVÔPÔRÔ24@Ô\Ô[ÔaÔ! = (CÔ_Ô\ÔZÔ\ÔCÔ_ÔVÔPÔRÔ1 " CÔ_Ô\ÔZÔ\Ô7ÔbÔ_ÔNÔaÔVÔ\Ô[Ô1) + (CÔ_Ô\ÔZÔ\ÔCÔ_ÔVÔPÔRÔ2 " CÔ_Ô\ÔZÔ\Ô7ÔbÔ_ÔNÔaÔVÔ\Ô[Ô2) + (24 " CÔ_Ô\ÔZÔ\Ô7ÔbÔ_ÔNÔaÔVÔ\Ô[Ô1 " CÔ_Ô\ÔZÔ\Ô7ÔbÔ_ÔNÔaÔVÔ\Ô[Ô2) " AÔ\Ô[ÔCÔ_Ô\ÔZÔ\ÔCÔ_ÔVÔPÔRÔ + 24 " (@Ô\ÔQÔRÔZÔCÔ_ÔVÔPÔRÔ + FÔGÔ5ÔCÔ_ÔVÔPÔRÔ " EÔRÔOÔNÔaÔRÔ@Ô\Ô[ÔaÔ!YÔfÔ + BÔaÔ!RÔ_Ô@Ô\Ô[ÔaÔ!YÔfÔ + GÔNÔeÔ) + <Ô[Ô`ÔaÔNÔYÔYÔNÔaÔVÔ\Ô[Ô9ÔRÔRÔ + 4ÔPÔaÔVÔcÔNÔaÔVÔ\Ô[Ô9ÔRÔRÔ " EÔRÔOÔNÔaÔRÔBÔ[ÔRÔGÔVÔZÔRÔ + BÔaÔ!RÔ_Ô9ÔRÔRÔ`Ô 53. We then divided this price by 24 months to calculate the average monthly price. We converted all currencies to U.S. dollars using Purchasing Power Parity (PPP) for the broadband price index and Currency Exchange Rate conversion factors for the hedonic price index.109 Next, we matched all bundled plans with their corresponding standalone Internet and standalone video component plans to calculate a bundle discount percentage. The formula below calculates the bundle discount percentage 7Ô5Ô based on the standalone Internet price CÔ<Ô, the standalone video price CÔIÔ, and the bundle price CÔ5Ô. For most bundled plans, we were able to collect the exact corresponding standalone Internet and video component plans.110 However, for bundled plans without corresponding standalone Internet plans and for standalone Internet plans without corresponding bundled plans, we created  synthetic plans with the same product characteristics but with a price to set the bundle discount equal to zero. Synthetic plans that correspond with collected bundled plans may represent bundled plans that could be available without a bundle discount (i.e., add-on pricing). (CÔ<Ô + CÔIÔ) " CÔ5Ô CÔ5Ô 7Ô5Ô = = (1 " ) (CÔ<Ô + CÔIÔ) CÔ<Ô + CÔIÔ 54. After calculating the discount percentage from the standalone Internet and standalone video prices for each bundled plan, we applied the percentage equally to the standalone broadband and video component plan prices to arrive at the implied price of broadband when purchased in a bundle.111 To illustrate, suppose the standalone prices for a particular video and Internet broadband plan are $100 and $50, respectively, but the two can be purchased in a bundle for $120. Then the bundle discount percentage is 20% and the implied price of the video plan when purchased in a bundle is $80, while the implied price of broadband when bundled is $40. This implied broadband price when bundled and the associated broadband characteristics would then be included as a plan in the dataset. In this manner, our analysis does not compare video and broadband bundles across countries, but rather isolates an implied price of broadband when bundled to avoid video product comparability issues across countries. 55. In Figure G-33, we calculate country level average bundle discounts over all bundled plans (including synthetic plans). First, we take a simple unweighted average of the bundle discount and bundle discount rates over all plans for each provider s product categories.112 Then, we aggregate over providers, weighting by their market shares. Finally, we aggregate over country level products using the download speed tier shares to arrive at our bundle discount estimate for each country. The results of this analysis confirm that bundling discounts vary widely across countries and therefore accounting for 109 OECD, PPP, https://data.oecd.org/conversion/purchasing-power-parities-ppp.htm (last visited Oct. 27, 2020); OECD, Exchange rates, https://data.oecd.org/conversion/exchange-rates.htm#indicator-chartt (last visited Oct. 27, 2020). 110 In one case, a provider offered standalone broadband without fixed voice but bundled plans with fixed voice. We collected broadband plans with fixed voice to match with these bundled plans, but we excluded the broadband plans with fixed voice from the analysis. 111 Allocating the bundle discount percentage equally to each of the standalone components is equivalent to allocating the bundle discount amount in proportion to the standalone component prices. 112 In some cases, a plan may change data usage tiers as the number of lines increases. For example, if a provider offers an 8 GB single-line plan that allows a customer to add lines to the plan and share the data allowance, the single-line plan with 8 GB is in the 5 to 20 GB data usage (per line) tier and the 2-line plan with 4 GB per line is in the 0 to 5 GB data usage (per line) tier. 62 Federal Communications Commission FCC 20-188 product bundling is important in order to accurately reflect the prices actually paid by consumers for broadband services in each country. D. Mobile Broadband Pricing Data Collection 56. Collection of Broadband Prices and Timeframe. We collected mobile broadband plan prices and terms from 83 providers from 26 countries including the United States between February and September of 2020. To determine which providers to sample in each comparison country, we used the TeleGeography GlobalComms Database to select providers with national broadband market shares of at least 10% as of March 2019.113 Given the wide scope of offerings by mobile providers, we limited the collection to 4G postpaid smartphone plans that allowed unlimited voice calling and texting for up to four lines (when adding lines provided a discount).114 However, where providers did not offer plans with unlimited minutes or unlimited text messages, we collected plans with the highest number of minutes and text messages for a particular data allowance. 57. We collected mobile plan information in three broad categories: (1) general information including country, provider, plan name, and date of collection; (2) pricing information including all types of recurring and non-recurring costs such as promotional prices, activation fees, and rebates; and (3) non- price information such as data usage allowance, number of minutes and text messages (when not unlimited), and consequence of exceeding data allowance.115 We only collected plans available online and to new customers without any special discounts (e.g., student discounts). A unique plan is defined by the country, provider, data allowance, number of lines, contract duration, data allowance consequence, number of minutes, and number of text messages.116 58. We sought to collect pricing information excluding the cost of handsets due to both the complexity that handsets introduce in measuring price and because most providers allow customers to bring their own devices. Generally, providers either sold handsets separately from the service plan and/or allowed customers to bring their own devices (i.e., customers received a SIM card from the provider). Although handsets are a significant portion of the cost of mobile broadband services, we chose not to consider these costs in our pricing analysis due to the additional complexity and in order to keep prices comparable across countries. 59. One of the most important price factors for mobile broadband service is the data usage allowance.117 We recorded the monthly data allowance for each plan.118 In general, providers set a  soft data allowance per month before the provider imposes a consequence for exceeding these usage 113 We obtained these data as of February 2020. TeleGeography GlobalComms, Company Broadband Statistics, (last visited Oct. 27, 2020). 114 By postpaid plans, we refer to plans that are paid after usage (i.e., not prepaid or  pay-as-you-go plans). By smartphone plans, we refer to plans that have a data component. We did not collect plans marketed as 5G-only plans, since most countries providers did not market any plans as 5G, or marketed 4G plans with access to 5G where available. 115 All price variables are recorded as the total for all lines for the plans (i.e., not on a per-line basis). 116 We did not collect all possible mix-and-match combinations of plans. For example, a provider may offer a 5 GB plan that can be combined with a 2 GB plan for a discount, but we only collected multi-line plans of identical data allowances. 117 We only consider data that can be consumed within the customer s country. In some cases, particularly European providers plans, customers can use the main data allowance in several countries and/or have a separate international data allowance. International data allowances are not considered in our analysis as each provider has different policies regarding international data usage. 118 We do not consider promotional (i.e., limited time) data allowances unless the data allowances are included for the entire length of the contract. 63 Federal Communications Commission FCC 20-188 allowances.119 If a customer exceeds the allowance, the provider may decrease mobile broadband speeds for the remainder of the month, charge overage fees (i.e., a consumer pays for additional data use), or stop service entirely (i.e., a  hard data limit). The structure of the data allowance policies varies by provider and can be quite complex, so we record the default consequence for exceeding the first data allowance.120 60. We encountered a few issues unique to a small number of providers that required making assumptions about customer preferences. For providers that offered a plan with a set number of units to allocate between talk and text messages, we split these equally across the services and recorded the exchange rate among the services (e.g., 1 unit = 1 minute = 1 text).121 If a provider offered multiple plans that would appear identical within our data framework, we recorded the cheapest of these plans.122 If a provider did not offer any plans with included text messages, we set the number of text messages equal to one.123 61. Since the 2018 International Broadband Data Report s Mobile Broadband Pricing Data Collection in 2017, the prominence of unlimited plans has expanded greatly, especially for the U.S. providers. Two U.S. providers offer unique unlimited plans in that customers do not have a specified data allowance but can be throttled at any time due to network congestion.124 These providers also offer more expensive plans with  premium data that will not experience throttling until the customer has used beyond the allotted premium data and the network is experiencing congestion.125 Two other U.S. providers offer variations of unlimited plans where the  soft data cap is the same for each plan, but because these more expensive plans have other characteristics outside our data framework (e.g. 1080p video), we only recorded the cheapest of each of the provider s unlimited plans.126 62. Some other countries providers have similar issues. Finland s providers offer only unlimited data plans with prices varying by speeds. In this case, we set each provider s highest speed plan (150 Mbps) as unthrottled and each provider s slowest speed plan (with unlimited data) as throttled.127 One German provider offered an unlimited data plan with a maximum download speed of 2 Mbps so we set these plans as throttled. Each of Portugal s providers unlimited plans have a maximum download speed of 10 Mbps so we treated these plans as throttled. One of the United Kingdom s providers has an unlimited plan with a maximum download speed of 2 Mbps which we also define as throttled. 119 In our regressions,  unlimited is reserved for plans that have at least 50 GB per line per month before there is a consequence imposed. 120 For example, some providers have several data allowance thresholds with different consequences for exceeding each one, while other providers limit the amount of extra data a customer can buy. Some providers allow customers to choose from various data allowance consequences, so there is no clear default data cap consequence. 121 Luxembourg s providers typically have this structure for units of minutes and text messages. 122 For example, a provider may offer an Unlimited Talk/Text plan with 50 GB of data with varying levels of international data or with or without a streaming service included. As we do not have variables for international data or other services, we recorded the cheapest of these plans. 123 Two of Spain s providers only offer plans with Pay-As-You-Go Text Messages. 124 For the regression models, we account for these  Anytime Throttling plans with a dummy that equals one for throttled plans. 125 We have treated these  premium data plans as plans with  soft data caps. 126 For example, these more expensive unlimited plans have more hotspot data or higher hotspot speeds, inclusion of streaming services such as Hulu and Tidal, and/or HD video streaming. 127 Finland s providers offered several higher speed plans marketed as 5G plans so we did not collect these plans. 64 Federal Communications Commission FCC 20-188 63. Data Review and Cleaning Process. After completing the data collection, we reviewed the data for any issues. When certain essential variables were missing, we made the following assumptions to complete the analysis: " If a provider did not explicitly state the length of the contract, we assumed the plan was month-to-month (i.e., one month). " If a plan advertised a promotional price without specifying duration, we assumed the promotion lasted 12 months. " If the regular monthly price was not found, we assumed that the last available promotional price stayed in effect for the remaining period. " If activation fees, access fees, other recurring and non-recurring fees, and rebates were not listed clearly on a provider s website, we assumed that these fees were included or did not apply to the plan. " For Canada and the United States, if taxes were not explicitly stated as included in the list prices and not reported separately, we added a percentage to the total pre-tax prices.128 For all other countries, we assumed taxes were included. 64. Broadband Price Calculation. After cleaning the data, we then calculated the total cost of each plan over the first 24 months. A 24-month price was selected to produce a comparable pricing measure across plans that accounted for all promotional and non-promotional pricing and to amortize one- time fees over a sufficiently long-term horizon. This total 24-month price was calculated using the formula below: CÔ_ÔVÔPÔRÔ24@Ô\Ô[ÔaÔ! = (CÔ_Ô\ÔZÔ\ÔCÔ_ÔVÔPÔRÔ1 " CÔ_Ô\ÔZÔ\Ô7ÔbÔ_ÔNÔaÔVÔ\Ô[Ô1) + (CÔ_Ô\ÔZÔ\ÔCÔ_ÔVÔPÔRÔ2 " CÔ_Ô\ÔZÔ\Ô7ÔbÔ_ÔNÔaÔVÔ\Ô[Ô2) + (24 " CÔ_Ô\ÔZÔ\Ô7ÔbÔ_ÔNÔaÔVÔ\Ô[Ô1 " CÔ_Ô\ÔZÔ\Ô7ÔbÔ_ÔNÔaÔVÔ\Ô[Ô2) " AÔ\Ô[ÔCÔ_Ô\ÔZÔ\ÔCÔ_ÔVÔPÔRÔ + 24 " (4ÔPÔPÔRÔ`Ô`Ô9ÔRÔRÔ " EÔRÔOÔNÔaÔRÔ@Ô\Ô[ÔaÔ!YÔfÔ + BÔaÔ!RÔ_Ô@Ô\Ô[ÔaÔ!YÔfÔ + GÔNÔeÔ) + 4ÔPÔaÔVÔcÔNÔaÔVÔ\Ô[Ô9ÔRÔRÔ " EÔRÔOÔNÔaÔRÔBÔ[ÔRÔGÔVÔZÔRÔ + BÔaÔ!RÔ_Ô9ÔRÔRÔ`Ô 65. Next, we divided the price by the number of lines in the plan to get the total 24-month price per line. Then, we divided the price per line by 24 months to calculate the average monthly price per line. We converted all currencies to U.S. dollars using PPP for the broadband price index calculations and Currency Exchange Rate conversion factors for the hedonic price index.129 66. Similar to our fixed broadband analysis, we also created mobile broadband synthetic plans when a provider did not offer a particular plan at a discounted price for bundling additional lines, up to four lines. The simplest example is when a provider offers only a single-line plan without any discounts for bundling more lines; in this example, we would create a 2-line synthetic plan, a 3-line synthetic plan, and a 4-line synthetic plan with the same product characteristics and price per line (i.e., no bundle discount relative to the single-line plan). As a slightly more complex example, suppose a provider offers a plan as a single-line plan and a 2-line plan but offers no discount for three or four lines. In this example, we create a synthetic 3-line plan with the per line price set to a weighted average of the single- line and 2-line plan prices (i.e., the total price of purchasing a 2-line plan and a single-line plan divided by three) and a synthetic 4-line plan with the per line price set to the per line price of the 2-line plan (i.e., the total price of purchasing two 2-line plans divided by four). We made other similar synthetic plan 128 International Telecommunications Union, World Telecommunications/ICT Indicators Database 2020 (24th Edition/July 2020) (last accessed Aug. 19, 2020). 129 OECD, PPP, https://data.oecd.org/conversion/purchasing-power-parities-ppp.htm (last visited Oct. 27, 2020); OECD, Exchange rates, https://data.oecd.org/conversion/exchange-rates.htm#indicator-chart (last visited Oct. 27, 2020). 65 Federal Communications Commission FCC 20-188 calculations for plans that are not available with bundle discounts with up to four lines, but in all cases synthetic plans are plan combinations that consumers are able to purchase from the provider.130 67. In Figure G-38, we present country-level average mobile broadband bundle discounts (relative to single-line plans). The calculations include all plans (including synthetic plans), except for plans that do not have a single-line option. We calculated the bundle discount relative to the corresponding single-line plan, and then we took a simple unweighted average of the bundle discount and bundle discount rate over all plans for each provider s product categories. We then aggregated over providers, weighting by their market shares. Finally, we aggregated over country level products using the data usage product shares. We again find that bundle discounts vary widely across countries and must be accounted for to properly measure the prices consumers are paying for their mobile services in each country. Many countries, such as the United States, offer large bundle discounts when multiple lines are purchased, but some other countries offer no discounts. E. Data Sources and Variable Construction 68. Fixed Product Shares. To calculate the U.S. quantity weights for each of the six products in our price indexes, we use the FCC Form 477 data to estimate the share of U.S. broadband subscribers that subscribe to each of the three broadband download speed tiers and an estimate from S&P Global that about 65% of all U.S. broadband subscribers purchase their service in a bundle.131 The resulting broadband products and their estimated U.S. market shares are shown in Figure G-23 above. 69. Mobile Product Shares. Based on Cisco data, we know that 18% of all U.S. mobile subscribers use less than two GB of data per month, 23% of mobile subscribers use between two GB and five GB, 41% of mobile subscribers use between five GB and 20 GB, and 18% use more than 20 GB. Cisco also finds that 79% of users subscribe to shared plans with an average usage of approximately 10 GB per line, while 21% of users subscribe to non-shared plans with an average usage of approximately 14 GB of data per month.132 We assume that the percentage of shared data plans is equal to the percentage of multi-line plans (and the percentage of non-shared plans is equal to the percentage of single-line plans).133 However, we do not have an estimate of the percentage of single-line and multi-line plan customers who fall into each of our data usage allowance categories we only know the overall average usage for single and multi-line customers. 70. The log-normal distribution has been shown to approximate consumer usage over nearly every communications network, including broadband.134 This makes estimating the distribution of data usage simple because a log-normal distribution is entirely determined by only two parameters: a location 130 In some cases where a provider does not offer a single-line plan, we cannot calculate some combinations of number of lines. For example, if a plan was only offered as a 2-line plan, then we would calculate a 4-line plan price with the same per line price as the 2-line plan, but we would not have corresponding single-line and 3-line plans. 131 S&P Global, Estimated broadband-only homes as a percentage of wireline broadband households, Q1'18 vs. Q1'19 vs. Q1'20 (last accessed July 21, 2020). We use preliminary December 2019 FCC Form 477 subscriber data collection for these calculations. We again note that the year-end FCC Form 477 data are preliminary only and are subject to corrections as appropriate by the service provider, and the final data will be published in due course by the agency. 132 See Cisco, Annual Internet Report (2018-2023) White Paper, Fig. 18 (2020), https://www.cisco.com/c/en/us/solutions/collateral/executive-perspectives/annual-internet-report/white-paper-c11- 741490.html. 133 Some providers may have multi-line plans with separate data allowances. However, for the limited data plans collected, the two U.S. providers offered a set amount of data shared among lines on the plan. 134 Ioannis Antoniou et al., On the Log-Normal Distribution of Network Traffic, 167 Physica D: Nonlinear Phenomena 72 (2002), https://www.sciencedirect.com/science/article/abs/pii/S0167278902004311. 66 Federal Communications Commission FCC 20-188 parameter that pins down the mean and a scale parameter that determines the shape of the usage distribution.135 Another important property of the distribution is that percentiles are preserved if the mean of the distribution is shifted up or down.136 Combining the Cisco data with a log-normal distribution assumption, we are able to estimate the percentage of subscribers in the United States that have usage between the data usage allowances of our standardized mobile broadband products. The results of this approach are summarized in Figure G-31 below. The column with the heading  Cisco presents Cisco s estimates of the percentage of all U.S. mobile broadband consumers who have usage between the specified ranges of data usage. The next column provides our estimates using a log-normal distribution calibrated to the Cisco percentiles data based on the reported distribution parameters at the bottom of the figure.137 We find that our estimates are a close match and that the log-normal assumption fits these data well, although the Cisco distribution appears to have more mass in the tails. The next two columns provide our estimates for the percentage of single-line and multi-line plan subscribers that fall into each usage category.138 These values multiplied by the percentage of consumers who take single and multi-line products serve as the product shares in our price indexes. Fig. G-31: Mobile Broadband Data Usage Shares Cisco Log-Normal Estimates Usage Tier Overall Overall Single-Line Plan Multi-Line Plan Usage Usage Usage Usage 0 < Usage (GB) d" 2 18.0% 16.0% 14.0% 21.0% 2 < Usage (GB) d" 5 23.0% 26.0% 24.0% 29.0% 5 < Usage (GB) d" 10 23.0% 24.0% 24.0% 22.0% 10 < Usage (GB) d" 20 18.0% 19.0% 20.0% 17.0% 20 < Usage (GB) d" 50 14.0% 13.7% 16.2% 10.2% 50 < Usage (GB) 4.0% 1.3% 1.8% 0.8% Distribution Parameters Plan Type Mean Standard Deviation Overall 1.844 1.15 Individual 1.978 1.15 Shared 1.641 1.15 71. Content Quality Variable. In Figure G-44, we report various proxy measures for content quality as well as each country s primary language. The number of websites in top-level domains (TLDs) shows the count of all domains in each country s main TLD (e.g., Germany uses .de) according to DomainTools.com. For the United States, we aggregate over several major domains: .com, .net, .org, and .us. Similarly, we used the same TLDs to report the number of web pages in the TLDs by searching Google s search engine ( site:.de ) and recording the number of search results. We divide the number of domains and the number of webpages by the country s population to get per capita measures. Also, we 135 See George S. Ford, Approximating the Distribution of Broadband Usage from Publicly-Available Data, 7, n.5 (2012), https://www.phoenix-center.org/perspectives/Perspective12-03Final.pdf. A random variable is log-normally distributed if the logarithm of the variable is normally distributed. 136 Id. 137 The calibration chooses the standard deviation that results in the closest approximation to the data usage percentiles observed in the Cisco White Paper data: Cisco, Annual Internet Report (2018-2023) White Paper, (2020), https://www.cisco.com/c/en/us/solutions/collateral/executive-perspectives/annual-internet-report/white- paper-c11-741490.html. 138 These calculations assume that the standard deviation is the same as the overall usage distribution, but the mean is shifted to match the mean per line usage of multi-line and single-line plan subscribers. 67 Federal Communications Commission FCC 20-188 report each country s English Proficiency Index (EPI) score as a measure of access to English language content. Another proxy measure is the percent of the top 10 million websites in each country s primary language.139 From this data, we find that English-based websites represent over 50% of the top 10 million websites. Although these statistics are not perfect measurements of content quality, they demonstrate that English language content is the dominant form of content available to broadband subscribers.140 72. To construct the content quality measure used in our hedonic regressions, we perform a principal components analysis of the four content quality proxy variables (webpages by TLD per capita, domains by TLD per capita, EPI, and content language percentage), using the 26 country-level observations.141 We keep only the first principal component from this analysis, which explains about 53% of the variation in the 4 content quality measures. We then standardized the first principal component so that the mean value is zero and the standard deviation is one across the 26 country level values. This standardized first principal component is then used as a proxy measure of content quality in both the fixed broadband and mobile broadband hedonic analyses. 73. Purchasing Power Parity. To convert pricing data collected in local currency (LCU) to U.S. dollars, we use the OECD s 2019 PPPs which are defined as  the rates of currency conversion that try to equalise the purchasing power of different currencies, by eliminating the differences in price levels between countries. The basket of goods and services priced is a sample of all those that are part of final expenditures: final consumption of households and government, fixed capital formation, and net exports. 142 74. Exchange Rates. To convert pricing data collected in LCU to U.S. dollars, we also used the OECD s 2019 exchange rates which are defined as  the price of one country's' currency in relation to another country's currency. 143 75. Gross National Income Per Capita. The Gross National Income (GNI) data are used as a demographic control variable in the hedonic regression models and are from the OECD.144 We use the most recently available value for each country and convert all values to 2019 U.S. dollars using the PPP conversion factors. 76. Educational Attainment. These data are used as a demographic control variable in the hedonic regression models and are from the OECD.145 We used the 2018 percentage of 25 to 64-year-olds with Bachelor s (or equivalent education), Master s (or equivalent education), or Doctoral (or equivalent education) degrees. 77. Non-Rural Population Density. For the fixed broadband hedonic analysis, we constructed a measure of non-rural population density using four OECD datasets: (1) National 139 W3Techs, Usage Statistics of Content Languages for Websites, https://w3techs.com/technologies/overview/content_language (last visited Oct. 27, 2020). 140 We have found our results to be robust to using different measures of content quality as well as dropping the United States from the sample and then running the estimation. 141 Principal components analysis is a standard method used in statistics for reducing a large set of variables into a smaller set of variables that retain most of the information contained in the larger variable set. 142 OECD, PPP, https://data.oecd.org/conversion/purchasing-power-parities-ppp.htm (last visited Oct. 27, 2020). 143 OECD, Exchange rates, https://data.oecd.org/conversion/exchange-rates.htm#indicator-chartt (last visited Oct. 27, 2020). 144 OECD, Gross national income, https://data.oecd.org/natincome/gross-national-income.htm (last visited Oct. 27, 2020). 145 OECD, OECD.Stat, https://stats.oecd.org/ (last visited Oct. 27, 2020). 68 Federal Communications Commission FCC 20-188 Population Distribution (NPD),146 (2) National Area Distribution (NAD),147 (3) land area, and (4) population. The NPD is the percent of the population living in three categories: urban, intermediate, and rural areas. The NAD is the percent of the area in three categories: urban, intermediate, and rural. The NPD and NAD are from 2014, therefore we multiplied the percentages by the 2014 population and 2014 land area, respectively, to get the total population and total land area in each category. Then, we divided the total population by category by the total land area in the corresponding category. Non-rural population density is the sum of urban and intermediate population divided by the sum of urban and intermediate land area. 78. Population Density. For the mobile broadband hedonic analysis, we calculated the overall national population density using the OECD s population and land area datasets.148 We divided the most recently available national population (2018) by the most recently available land area (2016) to get 2018 population density.149 79. Coverage (Fixed). For the fixed broadband hedonic analysis, we included a variable measuring the percentage of households with access to download speeds of greater than 100 Mbps in each country. For the 21 European comparison countries, we used data reported in the EC s 2019 Broadband Coverage in Europe Report on the percentage of households living in areas where the download speed of greater than 100 Mbps was deployed as of June 2018.150 For the United States, we relied on FCC Form 477 data for the same measure, as of December 2018.151 For Canada, we used the percentage of households with fixed broadband service of at least 100 Mbps available as of 2018.152 80. For the remaining three countries, we relied on proxy measures of coverage. For Australia, the National Broadband Network Company reports the number of premises ready for service by technology as of June 2018.153 We assumed that Fiber to the Premises (FTTP), Fiber to the Node/Basement/Curb (FTTN/B/C), and Hybrid Fiber Coaxial technologies are capable of achieving at least 100 Mbps, while Fixed Wireless and Satellite are not.154 We divided the number of premises designated as ready for service155 by the total number of premises as our network coverage measure for 146 OECD, National population distribution, https://data.oecd.org/popregion/national-population- distribution.htm#indicator-chart (last visited Oct. 27, 2020). 147 OECD, National area distribution, https://data.oecd.org/popregion/national-area-distribution.htm#indicator-chart (last visited Oct. 27, 2020). 148 OECD, OECD.Stat, https://stats.oecd.org/ (last visited Oct. 27, 2020). 149 Land area rarely changes from year to year in the dataset, and when it does, the changes are minimal, so we believe that 2016 land area is reasonable to use with 2018 population data. 150 European Commission, Broadband Coverage in Europe 2019 (Sept. 4, 2020), https://op.europa.eu/en/publication-detail/-/publication/077cc151-f0b3-11ea-991b-01aa75ed71a1, (2019 Broadband Coverage in Europe Report). 151 FCC Form 477. See infra Fig. G-53. 152 Canadian Radio-television and Telecommunications Commission, Communications Monitoring Report 2019, (2020), https://crtc.gc.ca/pubs/cmr2019-en.pdf. 153 NBN Corporation, Annual Report 2018, (Oct. 31, 2018), https://www.nbnco.com.au/content/dam/nbnco2/2018/documents/media-centre/nbn-co-annual-report-2018.pdf (NBN Annual Report 2018). 154 NBN Annual Report 2018 reports that the wholesale products maximum speeds as 1 Gbps / 400 Mbps for FTTP, 100/40 Mbps for FTTN/B/C and Hybrid Fiber Coaxial, 50/20 Mbps for Fixed Wireless, and 25/5 for Satellite. 155 NBN Annual Report 2018 defines  ready for service as  A Rollout Region is ready for service when the majority of premises are passed by the nbn access network and RSPs are able to begin selling services over the nbn access network in that Rollout Region. 69 Federal Communications Commission FCC 20-188 Australia.156 For Mexico, we used data from Instituto Federal de Telecomunicaciones - Banco de Informacion de Telecomunicaciones which reports the percentage of accesses by technology as of June 2018.157 We assumed that Fiber and Cable Coaxial are the only technologies that could achieve 100 Mbps; and that DSL, Satellite, Fixed Wireless, and Other Technologies are below this threshold. For New Zealand, we relied on data from the country s Ministry of Business, Innovation, and Employment on progress of their Ultra-Fast Broadband (UFB) initiative.158 In particular, we used the percentage of New Zealanders with access to UFB as of March 2019. 81. Mobile Download, Upload, and Latency. For the mobile broadband hedonic analysis, we used 2019 country-level mean download speeds based on our analysis of Ookla Speed Test data.159 82. Mobile 4G Availability. For the mobile broadband hedonic analysis, we used OpenSignal s measure of 4G Availability which is defined as  the proportion of time users with a 4G device and subscription have a 4G LTE connection. 160 For most countries, we used the value from OpenSignal s most recent (May 2020) report, but when some countries were not reported, we used the most recently reported value. 161 Specifically, we relied on the February 2018 report for Estonia, Latvia, and Luxembourg,162 and the November 2016 report for Iceland.163 83. Fixed Data Usage. For the fixed broadband analysis, we calculated the average monthly data usage from several data sources. Our primary source is the International Telecommunications Union (ITU) Database, which provides the total data usage by fixed broadband subscribers in each country.164 We converted the total annual data in exabytes to monthly average data usage in gigabytes. Because the ITU Database does not have 2019 values for all 26 comparison countries, we supplement the data from two other sources. For Austria, New Zealand, and the United Kingdom, we relied on the TeleGeography GlobalComms Database s Fixed Data Traffic Volume dataset which has a 2019 monthly average.165 We 156 NBN Annual Report 2018 reports that  as of 30 June 2018, 7.0 million premises had been declared RTC, an increase of 29 per cent year-on-year. This means that about 60% of Australian premises were able to order a service over the nbn access network at the end of the financial year. This implies that the total number of premises in Australia is about 11.7 million. 157 Instituto Federal de Telecomunicaciones, Banco de Informacion de Telecomunicaciones, Indicadores Internacionales - Comparativo Entre Paises Miembros de Regulatel - Indicadores Por Pais, https://bit.ift.org.mx/BitWebApp/indicadoresInternacioanles.xhtml (last visited Oct. 27, 2020). 158 Crown Infrastructure Partners, Quarterly Connectivity Update, (Mar. 2019), https://www.mbie.govt.nz/assets/quarterly-connectivity-update-q1-31-march-2019.pdf. 159 See supra Appx. G-2, Fig. G-11. 160 Sam Fenwick and Hardik Khatri, The State of Mobile Network Experience 2020: One Year into the 5G Era, Open Signal (May 2020), https://www.opensignal.com/reports/2020/05/global-state-of-the-mobile-network. 161 Id. 162 OpenSignal, State of LTE (February 2018), https://www.opensignal.com/reports/2018/02/state-of-lte (last visited Oct. 27, 2020). 163 OpenSignal, State of LTE (November 2016), https://www.opensignal.com/reports/2016/11/state-of-lte (last visited Oct. 27, 2020). 164 International Telecommunications Union, World Telecommunications/ICT Indicators Database 2020 (24th Edition/July 2020) (last accessed Aug. 19, 2020),  Fixed (wired)- broadband Internet traffic (exabytes) refers to traffic generated by fixed-broadband subscribers measured at the end-user access point. It should be measured adding up download and upload traffic. This should exclude wholesale traffic; walled garden; IPTV and cable TV traffic. 165 TeleGeography GlobalComms, Company Broadband Statistics, (last visited Oct. 27, 2020). Fixed data traffic covers the number of bytes of data traffic originating on fixed broadband networks (xDSL, Cable, FTTx, WiMAX, etc.) within a given country. These volumes include download and upload wherever possible. 70 Federal Communications Commission FCC 20-188 divided both the ITU and TeleGeography monthly averages by the total number of fixed broadband subscribers, according to the OECD, to get the monthly fixed broadband data usage per subscriber.166 84. Mobile Data Usage. For the mobile broadband analysis, we used average monthly data usage reported by the OECD as of December 2018.167 85. Terrain Roughness (Weighted by Population). Our measure of terrain roughness is a population weighted terrain ruggedness index.168 The index is constructed by calculating the terrain ruggedness index for each 30 by 30 arc-second cell using elevation data across the surface of the Earth. Let RÔ_Ô,PÔ denote the elevation at the point located in row _Ô and column PÔ of a grid of elevation points: _Ô+1 PÔ+1 2 GÔEÔ<Ô_Ô,PÔ = " " (RÔVÔ,WÔ " RÔ_Ô,PÔ) VÔ=_Ô"1 WÔ=PÔ"1 86. These values are then weighted by the share of the country population in each cell to calculate the weighted average terrain ruggedness index for the country. The values calculated are reported in 100s of meters.169 87. Domains by Top-Level Domains Per Capita. First, we determined the TLD(s) for each country, and then aggregated the counts of all domains in each TLD over the country s TLD(s).170 Next, we divided the total domains by the country s population to get the domains per capita.171 Figure G-44 reports the TLD(s) assigned to each country. 88. Webpages by Top-Level Domains Per Capita. Using the same TLDs for each country, we determined the number of webpages using Google s search engine for each TLD (for example,  site:.com ).172 Then, we aggregated over TLDs for each country and divided the total webpages for each country by the country s population to get the webpages per capita. 89. English Proficiency Index. We used a measure of a country s English proficiency from Education First, called the EPI.173 In the most recent EPI report, Education First reports an EPI score for each country except Australia, Canada, Iceland, Ireland, New Zealand, the United Kingdom, and the United States. Besides Iceland, we assumed that these countries are all native English-speaking countries 166 OECD, Broadband Portal, https://www.oecd.org/sti/broadband/broadband-statistics/ (last visited Oct. 27, 2020). 167 Id. The OECD has released December 2019 data, but the data do not include a value for the United States. Therefore, we use December 2018 values. 168 Nathan Nunn and Diego Puga, Ruggedness: The Blessing of Bad Geography in Africa, 94 Review of Economics and Statistics 20-36 (2012). 169 Nathan Nunn and Diego Puga, Data and Replication Files for  Ruggedness: The Blessing of Bad Geography in Africa, https://diegopuga.org/data/rugged/ (last visited Oct. 27, 2020). 170 DomainTools, Domain Count Statistics for TLDs, https://research.domaintools.com/statistics/tld-counts/ (last visited Oct. 27, 2020). 171 OECD, OECD.Stat, https://stats.oecd.org/ (last visited Oct. 27, 2020). The most recently available country population data is for 2018. 172 Google, https://www.google.com/ (last visited Oct. 27, 2020). 173 Education First, EF English Proficiency Index, (2019), https://www.ef.com/assetscdn/WIBIwq6RdJvcD9bc8RMd/legacy/__/~/media/centralefcom/epi/downloads/full- reports/v9/ef-epi-2019-english.pdf. 71 Federal Communications Commission FCC 20-188 and set the EPI score to 100% for our analyses. For Iceland, we assumed a  Very High Proficiency and set the EPI score to the average EPI score of other sampled countries in this category.174 90. Content Language. For both the fixed broadband and mobile broadband hedonic analyses, we used the percentage of websites with different content languages.175 A content language is defined as the natural language of the text on a website. The primary language spoken in each country is shown in Figure G-44. F. Supplementary Figures 91. This section provides the supplementary figures referenced in the text. Fig. G-32: Fixed Broadband and Mobile Broadband Combined Hedonic Price Indexes Model 1 Model 2 Model 3 Model 4 Country Price Rank Price Rank Price Rank Price Rank Australia 117.37 20 139.69 22 148.96 22 188.16 21 Austria 85.78 10 109.88 15 125.66 17 166.06 16 Belgium 92.49 14 99.57 12 103.44 9 156.35 14 Canada 134.99 24 157.83 23 160.09 23 186.95 20 Czech Republic 70.46 6 101.74 13 112.28 11 153.02 12 Denmark 74.17 8 88.88 7 97.62 6 142.80 8 Estonia 66.09 4 98.26 11 106.76 10 165.17 15 Finland 69.95 5 81.87 3 88.55 3 143.82 9 France 61.87 3 85.28 5 103.05 8 147.17 11 Germany 95.83 15 117.16 18 142.12 21 199.60 23 Greece 196.75 26 205.01 26 236.45 26 334.83 26 Iceland 99.90 17 86.39 6 114.76 13 173.09 18 Ireland 86.66 11 90.91 8 112.31 12 131.52 4 Italy 54.58 2 78.89 2 93.62 4 132.07 5 Latvia 40.06 1 66.68 1 75.76 1 115.53 1 Luxembourg 126.00 23 110.57 16 130.27 19 193.01 22 Mexico 92.48 13 181.42 25 199.07 25 252.81 24 Netherlands 106.44 19 160.57 24 173.41 24 259.02 25 New Zealand 103.80 18 109.20 14 123.95 16 139.81 7 Norway 142.18 25 119.44 19 121.38 15 184.76 19 Portugal 88.83 12 115.13 17 127.32 18 166.21 17 Spain 73.73 7 97.08 10 99.48 7 134.13 6 Sweden 84.02 9 93.16 9 96.40 5 145.07 10 Switzerland 123.72 22 83.20 4 83.64 2 121.63 3 United Kingdom 99.47 16 121.38 21 137.87 20 153.20 13 United States 121.19 21 121.12 20 121.33 14 121.49 2 174 In Iceland, English is the  first foreign language in the Icelandic National Curriculum for compulsory schools. See Iceland, Ministry of Education, Science and Culture, The Icelandic National Curriculum Guide for Compulsory Schools  with Subjects Areas, 50 (2014), https://www.government.is/library/01-Ministries/Ministry-of- Education/Curriculum/adalnrsk_greinask_ens_2014.pdf. 175 W3Techs, Usage Statistics of Content Languages for Websites, https://w3techs.com/technologies/overview/content_language (last visited Oct. 27, 2020). 72 Federal Communications Commission FCC 20-188 Fig. G-33: Fixed Broadband Average Bundle Discounts and Discount Rates (PPP Adjusted) Discount Country Discount Rate Australia 2.55 3.4% Austria 15.28 18.6% Belgium 4.54 5.3% Canada 18.26 13.3% Czech Republic 19.02 20.1% Denmark 0.00 0.0% Estonia 37.13 30.9% Finland 11.20 15.2% France 0.00 0.0% Germany 4.99 7.0% Greece 10.47 9.5% Iceland 0.00 0.0% Ireland 19.13 15.3% Italy 0.00 0.0% Latvia 11.36 22.4% Luxembourg 0.00 0.0% Mexico 0.00 0.0% Netherlands 0.00 0.0% New Zealand 0.00 0.0% Norway 33.25 19.9% Portugal 44.66 41.9% Spain 0.00 0.0% Sweden 11.74 13.3% Switzerland 0.00 0.0% United Kingdom 0.00 0.0% United States 24.12 14.4% Note: Prices are reported in PPP adjusted U.S. dollars. 73 Federal Communications Commission FCC 20-188 Fig. G-34: Fixed Broadband Unweighted Average Prices by Product (PPP Adjusted) Standalone Bundled Country 0 < Mbps < 25 25 d" Mbps < 100 100 d" Mbps d" 1000 0 < Mbps < 25 25 d" Mbps < 100 100 d" Mbps d" 1000 Mean Count Mean Count Mean Count Mean Count Mean Count Mean Count Australia 34.41 9 50.93 15 70.87 7 34.41 9 50.94 17 71.75 9 Austria 39.36 1 42.17 4 75.56 5 32.62 2 33.11 6 60.65 5 Belgium 35.79 1 56.94 3 35.79 1 55.27 3 Canada 42.67 5 62.15 12 76.97 18 42.67 5 58.70 14 72.77 19 Czech Republic 34.27 3 40.09 4 54.27 6 28.79 5 32.75 8 47.46 10 Denmark 38.48 4 39.90 6 54.16 7 38.48 4 39.90 6 54.16 7 Estonia 33.13 7 44.28 6 81.31 11 25.73 13 33.85 12 57.85 23 Finland 32.42 4 45.65 6 32.42 4 43.38 6 France 28.11 1 34.53 1 40.70 7 28.11 1 34.53 1 40.70 7 Germany 38.42 3 34.82 4 50.06 11 38.42 3 33.84 5 48.03 14 Greece 42.78 4 52.50 5 77.39 6 42.63 5 53.59 8 63.60 14 Iceland 72.73 8 72.73 8 Ireland 31.33 1 37.73 1 65.26 5 31.33 1 37.73 1 67.28 8 Italy 50.37 2 42.00 10 50.37 2 42.00 10 Latvia 39.30 1 21.79 1 36.52 7 39.30 1 17.44 2 30.45 9 Luxembourg 51.91 4 67.28 9 51.91 4 67.28 9 Mexico 53.33 2 48.22 5 92.18 8 53.33 2 48.22 5 92.18 8 Netherlands 56.55 2 66.63 5 56.55 2 66.63 5 New Zealand 57.25 3 54.62 6 61.55 11 57.25 3 54.62 6 61.55 11 Norway 68.23 2 65.69 7 80.67 17 68.23 2 62.83 7 73.11 17 Portugal 50.27 3 46.94 4 60.00 14 50.27 3 46.94 4 56.52 14 Spain 62.40 1 68.55 8 62.40 1 68.55 8 Sweden 36.67 6 46.65 4 61.25 33 36.67 6 40.50 6 55.49 45 Switzerland 46.84 1 53.35 2 57.32 7 46.84 1 53.35 2 57.32 7 United Kingdom 34.12 2 41.53 7 57.91 7 34.12 2 41.53 7 57.91 7 United States 68.11 2 43.40 4 74.27 17 68.11 2 47.07 9 67.38 36 Total 67 105 253 77 133 319 Note: Prices are reported in PPP adjusted U.S. dollars. 74 Federal Communications Commission FCC 20-188 Fig. G-35: Fixed Broadband Weighted Average Prices by Product (PPP Adjusted) Standalone Bundled Country 0 < Mbps < 25 25 d" Mbps < 100 100 d" Mbps d" 1000 0 < Mbps < 25 25 d" Mbps < 100 100 d" Mbps d" 1000 Mean Count Mean Count Mean Count Mean Count Mean Count Mean Count Australia 37.23 9 53.24 15 71.28 7 37.23 9 52.66 17 70.59 9 Austria 39.36 1 44.67 4 70.23 5 32.62 2 33.70 6 58.59 5 Belgium 35.79 1 59.61 3 35.79 1 58.48 3 Canada 42.54 5 65.15 12 79.25 18 42.54 5 62.43 14 76.04 19 Czech Republic 34.84 3 41.75 4 54.78 6 32.43 5 39.09 8 51.67 10 Denmark 38.48 4 40.50 6 53.74 7 38.48 4 40.50 6 53.74 7 Estonia 34.76 7 45.95 6 83.99 11 31.90 13 42.58 12 79.83 23 Finland 33.40 4 41.73 6 33.40 4 39.90 6 France 28.11 1 34.53 1 43.07 7 28.11 1 34.53 1 43.07 7 Germany 42.01 3 40.82 4 53.69 11 42.01 3 41.22 5 52.97 14 Greece 46.76 4 55.07 5 76.78 6 46.59 5 53.57 8 68.91 14 Iceland 72.82 8 72.82 8 Ireland 31.33 1 37.73 1 60.69 5 31.33 1 37.73 1 60.16 8 Italy 50.37 2 42.58 10 50.37 2 42.58 10 Latvia 39.30 1 21.79 1 38.06 7 39.30 1 17.44 2 35.79 9 Luxembourg 48.98 4 78.33 9 48.98 4 78.33 9 Mexico 47.54 2 48.07 5 82.60 8 47.54 2 48.07 5 82.60 8 Netherlands 56.63 2 67.58 5 56.63 2 67.58 5 New Zealand 57.38 3 54.54 6 62.24 11 57.38 3 54.54 6 62.24 11 Norway 68.23 2 69.31 7 93.58 17 68.23 2 65.92 7 78.80 17 Portugal 50.45 3 46.43 4 60.39 14 50.45 3 46.43 4 56.89 14 Spain 62.40 1 65.17 8 62.40 1 65.17 8 Sweden 40.64 6 45.76 4 55.96 33 40.64 6 44.09 6 54.25 45 Switzerland 46.84 1 55.12 2 65.30 7 46.84 1 55.12 2 65.30 7 United Kingdom 34.24 2 42.71 7 56.26 7 34.24 2 42.71 7 56.26 7 United States 66.93 2 48.36 4 75.10 17 66.93 2 46.39 9 68.55 36 Total 67 105 253 77 133 319 Note: Prices are reported in PPP adjusted U.S. dollars. 75 Federal Communications Commission FCC 20-188 Fig. G-36: Fixed Broadband Estimated Variances of Random Coefficients and Likelihood Ratio Tests Model 1 Model 2 Model 3 Model 4 Random Coefficient Parameters Estimate SE Estimate SE Estimate SE Estimate SE Country: Variance(Constant) 0.104 0.037 0.025 0.015 0.025 0.015 0.010 0.012 Provider: Variance(0 < Mbps < 50) 0.001 0.001 0.000 0.001 0.000 0.001 0.001 0.001 Provider: Variance(50 d" Mbps < 100) 0.032 0.014 0.033 0.014 0.033 0.014 0.031 0.013 Provider: Variance(100 d" Mbps d" 1000) 0.021 0.006 0.021 0.006 0.021 0.006 0.021 0.006 Provider: Variance(Bundle Dummy) 0.002 0.002 0.002 0.002 0.002 0.002 0.002 0.002 Provider: Variance(Log Contract Length) 0.003 0.001 0.003 0.001 0.003 0.001 0.003 0.001 Provider: Variance(Constant) 0.044 0.014 0.047 0.013 0.046 0.014 0.048 0.014 Variance(Residual) 0.030 0.002 0.030 0.002 0.030 0.002 0.030 0.002 Likelihood Ratio Tests 1 vs. 2 2 vs. 3 2 vs. 4 P-Value 0.000 0.498 0.019 76 Federal Communications Commission FCC 20-188 Fig. G-37: Fixed Broadband Country Random Coefficients Model Model Model Model Country 1 2 3 4 Australia 0.245 0.128 0.142 0.052 Austria -0.154 -0.091 -0.093 -0.052 Belgium 0.029 -0.015 -0.024 0.009 Canada 0.306 0.190 0.178 0.046 Czech Republic -0.518 -0.163 -0.165 -0.095 Denmark 0.029 -0.026 -0.034 0.002 Estonia -0.533 -0.194 -0.193 -0.066 Finland -0.135 -0.121 -0.102 -0.013 France -0.255 -0.121 -0.112 -0.040 Germany -0.069 -0.051 -0.051 -0.017 Greece -0.107 0.045 0.082 0.056 Iceland 0.192 -0.019 -0.016 0.008 Ireland 0.088 -0.021 -0.009 -0.037 Italy -0.125 0.001 0.019 0.021 Latvia -0.675 -0.188 -0.196 -0.075 Luxembourg 0.259 -0.001 -0.010 0.014 Mexico -0.352 0.132 0.117 0.039 Netherlands 0.222 0.176 0.167 0.101 New Zealand 0.395 0.174 0.173 0.058 Norway 0.392 0.003 0.000 0.025 Portugal -0.078 0.078 0.063 0.020 Spain 0.233 0.190 0.172 0.097 Sweden -0.025 -0.103 -0.103 -0.026 Switzerland 0.340 -0.072 -0.086 -0.030 United Kingdom 0.020 0.037 0.054 -0.016 United States 0.277 0.033 0.028 -0.078 Overall 0.000 0.000 0.000 0.000 77 Federal Communications Commission FCC 20-188 Fig. G-38: Mobile Broadband Average Discount Rates by Number of Lines Relative to Single-Line Plan (PPP Adjusted) 2-Lines 3-Lines 4-Lines Country Discount Discount Rate Discount Discount Rate Discount Discount Rate Australia -0.29 -1.0% -0.59 -1.9% -0.88 -2.9% Austria 0.00 0.0% 0.00 0.0% 0.00 0.0% Belgium 0.00 0.0% 0.00 0.0% 0.00 0.0% Canada -3.20 -3.7% -4.27 -5.0% -6.23 -7.0% Czech Republic 0.00 0.0% 0.00 0.0% 0.00 0.0% Denmark -3.70 -10.3% -5.32 -15.0% -6.16 -17.5% Estonia -7.17 -13.8% -9.51 -18.3% -12.26 -22.0% Finland 0.00 0.0% 0.00 0.0% 0.00 0.0% France 0.00 0.0% 0.00 0.0% 0.00 0.0% Germany -7.35 -15.4% -9.00 -18.5% -11.29 -20.9% Greece -6.52 -7.5% -7.88 -9.1% -6.52 -7.5% Iceland -4.00 -10.6% -6.74 -15.6% -7.71 -17.9% Ireland 0.00 0.0% 0.00 0.0% 0.00 0.0% Italy 0.00 0.0% 0.00 0.0% 0.00 0.0% Latvia 0.00 0.0% -0.21 -0.4% -0.31 -0.7% Luxembourg 0.00 0.0% 0.00 0.0% 0.00 0.0% Mexico -10.09 -8.1% -4.48 -5.6% -5.83 -6.9% Netherlands 0.00 0.0% 0.00 0.0% 0.00 0.0% New Zealand -6.56 -14.9% -8.74 -19.9% -9.83 -22.4% Norway -1.03 -1.7% -1.38 -2.3% -1.55 -2.6% Portugal -0.93 -1.4% -1.24 -1.8% -1.39 -2.0% Spain -3.76 -7.2% -5.02 -9.6% -5.64 -10.8% Sweden -7.74 -17.2% -10.69 -23.5% -12.39 -26.7% Switzerland 0.00 0.0% 0.00 0.0% 0.00 0.0% United Kingdom 0.00 0.0% 0.00 0.0% 0.00 0.0% United States -10.65 -14.9% -16.61 -23.3% -19.72 -27.7% Note: Plans that are not available as Single-Line Plans are not included. Prices are reported in PPP adjusted U.S. dollars. 78 Federal Communications Commission FCC 20-188 Fig. G-39: Mobile Broadband Unweighted Prices by Product (PPP Adjusted) Single-Line Plans Multi-Line Plans Country 0.2 < GB d" 5 5 < GB d" 20 20 < GB 0.2 < GB d" 5 5 < GB d" 20 20 < GB Mean Count Mean Count Mean Count Mean Count Mean Count Mean Count Australia 29.22 3 44.35 11 28.77 3 43.18 11 Austria 19.24 2 29.18 6 50.72 3 19.24 2 29.18 6 50.72 3 Belgium 24.40 1 40.91 5 55.74 3 24.40 1 40.91 5 55.74 3 Canada 81.75 10 124.04 5 78.93 10 120.28 5 Czech Republic 49.59 7 72.51 5 107.05 8 49.59 7 72.51 5 107.05 8 Denmark 25.09 5 37.78 10 23.48 6 27.92 8 Estonia 18.23 5 33.11 6 61.17 5 17.31 6 29.17 6 47.54 4 Finland 32.18 6 32.18 6 France 21.56 3 29.91 2 44.04 12 21.56 3 29.91 2 44.04 12 Germany 32.94 3 50.53 8 64.91 10 29.35 3 41.93 8 61.32 10 Greece 57.42 17 79.15 8 105.23 15 55.01 20 69.90 9 95.71 14 Iceland 19.88 3 24.10 2 48.16 10 20.40 4 24.72 3 38.50 6 Ireland 36.24 3 36.24 3 Italy 44.28 7 44.28 7 Latvia 28.29 7 38.78 2 47.28 3 28.29 7 38.78 2 45.48 3 Luxembourg 15.11 5 29.36 2 61.86 4 15.11 5 29.36 2 61.86 4 Mexico 32.83 9 72.86 14 136.07 2 32.11 10 66.86 14 Netherlands 31.02 10 40.42 12 51.51 9 31.02 10 40.42 12 51.51 9 New Zealand 32.09 3 40.11 3 55.52 5 28.19 5 38.25 4 50.50 4 Norway 31.26 5 46.56 6 60.34 1 31.26 5 46.56 6 47.84 1 Portugal 83.88 12 82.49 12 Spain 29.33 2 44.03 6 71.04 3 29.33 2 40.08 6 55.95 3 Sweden 23.78 4 35.26 6 53.90 9 24.09 6 30.26 6 37.06 7 Switzerland 30.31 3 48.58 3 30.31 3 48.58 3 United Kingdom 22.67 8 29.39 8 42.39 14 22.67 8 29.39 8 42.39 14 United States 63.63 4 70.17 3 78.53 9 48.69 6 44.91 1 54.94 9 Total 101 122 182 113 124 169 Note: The three multi-line products include 2, 2, and 3 lines, respectively; all other plans are excluded. Prices are reported in PPP adjusted U.S. dollars. 79 Federal Communications Commission FCC 20-188 Fig. G-40: Mobile Broadband Weighted Prices by Product (PPP Adjusted) Single-Line Plans Multi-Line Plans Country 0.2 < GB d" 5 5 < GB d" 20 20 < GB 0.2 < GB d" 5 5 < GB d" 20 20 < GB Mean Count Mean Count Mean Count Mean Count Mean Count Mean Count Australia 30.23 3 46.24 11 29.97 3 45.42 11 Austria 20.62 2 29.82 6 52.61 3 20.62 2 29.82 6 52.61 3 Belgium 24.40 1 40.47 5 55.35 3 24.40 1 40.47 5 55.35 3 Canada 80.26 10 123.06 5 77.05 10 118.79 5 Czech Republic 48.33 7 71.93 5 109.33 8 48.33 7 71.93 5 109.33 8 Denmark 25.28 5 42.08 10 23.67 6 28.99 8 Estonia 19.34 5 34.77 6 60.36 5 17.75 6 30.01 6 43.59 4 Finland 31.97 6 31.97 6 France 21.22 3 30.68 2 45.52 12 21.22 3 30.68 2 45.52 12 Germany 33.08 3 52.39 8 75.68 10 28.12 3 41.60 8 67.66 10 Greece 57.99 17 79.61 8 105.89 15 56.27 20 71.20 9 92.50 14 Iceland 19.19 3 24.07 2 47.99 10 20.02 4 24.67 3 36.38 6 Ireland 36.57 3 36.57 3 Italy 44.52 7 44.52 7 Latvia 30.22 7 39.04 2 47.63 3 30.22 7 39.04 2 45.74 3 Luxembourg 14.75 5 29.36 2 68.05 4 14.75 5 29.36 2 68.05 4 Mexico 33.64 9 78.89 14 136.07 2 33.55 10 76.21 14 Netherlands 30.93 10 40.98 12 51.87 9 30.93 10 40.98 12 51.87 9 New Zealand 32.74 3 40.03 3 55.83 5 28.81 5 39.30 4 50.89 4 Norway 31.19 5 46.04 6 60.34 1 31.19 5 46.04 6 47.84 1 Portugal 83.92 12 82.69 12 Spain 34.99 2 47.88 6 71.48 3 34.99 2 41.30 6 57.01 3 Sweden 24.13 4 37.11 6 55.36 9 24.31 6 32.11 6 35.67 7 Switzerland 31.25 3 52.15 3 31.25 3 52.15 3 United Kingdom 22.50 8 30.90 8 42.49 14 22.50 8 30.90 8 42.49 14 United States 61.41 4 76.23 3 74.15 9 49.02 6 44.91 1 51.62 9 Total 101 122 182 113 124 169 Note: Prices are reported in PPP adjusted U.S. dollars. 80 Federal Communications Commission FCC 20-188 Fig. G-41: Mobile Broadband Estimated Variances of Random Coefficients and Likelihood Ratio Tests Model 1 Model 2 Model 3 Model 4 Random Coefficient Parameters Estimate SE Estimate SE Estimate SE Estimate SE Country: Variance(Constant) 0.209 0.079 0.164 0.066 0.130 0.056 0.102 0.050 Provider: Variance(0 < GB d" 5) 0.018 0.005 0.019 0.005 0.019 0.005 0.019 0.005 Provider: Variance(5 < GB d" 20) 0.022 0.005 0.021 0.005 0.021 0.005 0.021 0.005 Provider: Variance(20 < GB) 0.057 0.014 0.057 0.014 0.057 0.014 0.057 0.014 Provider: Variance(Number of Lines) 0.002 0.000 0.002 0.000 0.002 0.000 0.002 0.000 Provider: Variance(Log Contract Length) 0.022 0.008 0.022 0.009 0.022 0.008 0.021 0.008 Provider: Variance(Constant) 0.113 0.034 0.112 0.034 0.111 0.034 0.113 0.034 Variance(Residual) 0.013 0.001 0.013 0.001 0.013 0.001 0.013 0.001 Likelihood Ratio Tests 1 vs. 2 2 vs. 3 2 vs. 4 P-Value 0.278 0.128 0.057 81 Federal Communications Commission FCC 20-188 Fig. G-42: Mobile Broadband Country Random Coefficients Model Model Model Model Country 1 2 3 4 Australia -0.123 0.008 -0.078 -0.073 Austria -0.375 -0.175 -0.086 -0.079 Belgium -0.127 -0.260 -0.325 -0.206 Canada 0.520 0.534 0.350 0.216 Czech Republic 0.060 0.137 0.093 0.083 Denmark -0.210 -0.078 -0.064 -0.010 Estonia -0.646 -0.603 -0.547 -0.389 Finland -0.333 -0.276 -0.327 -0.154 France -0.327 -0.119 -0.017 -0.012 Germany 0.172 0.233 0.314 0.293 Greece 1.226 0.968 0.876 0.821 Iceland 0.077 -0.175 0.207 0.266 Ireland -0.363 -0.432 -0.144 -0.228 Italy -0.405 -0.218 -0.132 -0.101 Latvia -0.378 -0.181 -0.147 -0.066 Luxembourg -0.122 -0.292 -0.173 -0.126 Mexico -0.369 -0.221 -0.186 -0.193 Netherlands 0.281 0.554 0.469 0.474 New Zealand 0.121 0.010 0.071 -0.113 Norway 0.155 0.011 -0.139 -0.025 Portugal 0.277 0.232 0.203 0.152 Spain -0.426 -0.406 -0.458 -0.428 Sweden 0.107 0.117 -0.023 0.074 Switzerland 0.444 -0.035 -0.194 -0.124 United Kingdom 0.118 0.105 0.111 -0.102 United States 0.646 0.561 0.347 0.049 Overall 0.000 0.000 0.000 0.000 82 Federal Communications Commission FCC 20-188 Fig. G-43: Summary Statistics for Independent Variables First Terrain Principal Non- Ruggedness Component Rural Index of Content Exchange Fixed Mobile GNI/ Pop. Pop. Educ. Fixed (Weighted Quality Country PPP Rate Usage Usage Capita Density Density Attnmnt. Coverage 4G Avail. by Pop.) Variables Australia 1.47 1.44 180.51 3.39 54,910 155 8 33.9% 60.9% 94.0% 0.18 1.56 Austria 0.77 0.89 126.34 16.40 51,300 686 277 17.7% 57.5% 91.4% 1.15 -0.36 Belgium 0.77 0.89 147.31 2.01 47,350 1,093 975 40.1% 95.5% 92.6% 0.26 -0.46 Canada 1.20 1.33 197.72 2.46 46,370 187 11 31.8% 84.9% 93.5% 0.37 1.46 Czech Republic 12.56 22.93 147.53 3.39 22,000 368 356 24.1% 58.2% 91.7% 0.58 -0.59 Denmark 6.75 6.67 216.83 7.64 63,240 715 357 33.0% 92.6% 90.5% 0.19 -0.13 Estonia 0.55 0.89 9.82 23,220 89 79 35.1% 68.5% 84.2% 0.19 -0.64 Finland 0.86 0.89 19.39 49,580 232 47 34.0% 51.8% 93.0% 0.27 -0.52 France 0.75 0.89 5.64 42,400 440 317 22.5% 47.5% 86.0% 0.50 -0.78 Germany 0.74 0.89 116.43 2.55 48,520 822 615 28.5% 66.3% 85.8% 0.41 -0.31 Greece 0.56 0.89 90.37 1.53 20,320 464 216 30.0% 0.4% 86.5% 1.29 -0.77 Iceland 140.66 122.61 268.12 7.79 72,850 540 9 41.5% 74.3% 60.7% 0.56 -0.25 Ireland 0.78 0.89 52.16 6.77 62,210 3,695 183 40.2% 55.4% 70.1% 0.28 1.42 Italy 0.67 0.89 129.17 4.27 34,460 664 532 19.3% 23.9% 89.6% 0.75 -0.87 Latvia 0.50 0.89 229.26 12.78 17,730 168 80 30.4% 87.8% 84.1% 0.14 -0.82 Luxembourg 0.85 0.89 3.99 73,910 593 648 39.5% 94.0% 80.0% 0.58 -0.36 Mexico 9.28 19.26 2.11 9,430 607 167 17.5% 57.5% 86.4% 0.82 -1.09 Netherlands 0.79 0.89 2.58 53,200 1,297 1,325 36.2% 93.0% 95.9% 0.04 0.12 New Zealand 1.45 1.52 178.52 2.42 42,670 44 48 35.5% 75.0% 81.7% 0.45 1.61 Norway 9.60 8.80 4.84 82,500 122 38 31.8% 82.0% 95.7% 1.25 -0.32 Portugal 0.57 0.89 128.20 2.64 23,080 775 291 25.0% 70.2% 87.6% 0.97 -0.67 Spain 0.63 0.89 139.90 2.83 30,390 382 242 25.9% 87.2% 90.7% 0.81 -0.83 Sweden 8.92 9.46 7.32 55,840 319 65 33.5% 78.4% 93.5% 0.34 -0.29 Switzerland 1.16 0.99 190.25 6.09 85,500 800 558 43.7% 98.5% 93.1% 1.45 -0.26 United Kingdom 0.69 0.78 306.39 3.36 42,370 893 711 36.1% 48.0% 89.2% 0.21 1.63 83 Federal Communications Commission FCC 20-188 First Terrain Principal Non- Ruggedness Component Rural Index of Content Exchange Fixed Mobile GNI/ Pop. Pop. Educ. Fixed (Weighted Quality Country PPP Rate Usage Usage Capita Density Density Attnmnt. Coverage 4G Avail. by Pop.) Variables United States 1.00 1.00 344.00 5.39 65,760 252 93 36.7% 91.0% 96.1% 0.33 2.51 Analysis Both Both Fixed Mobile Both Fixed Mobile Both Fixed Mobile Both Both Nunn & Source OECD OECD Various OECD OECD OECD OECD OECD Various OpenSignal Puga Various Most Most Year 2019 2019 2019 2018 Recent 2014 2018 2018 2018 Recent 2000/2001 Various GB/ GB/ USD LCU/ LCU/ Month/ Month/ 2019 People/ People/ Unit USD USD Sub Sub (PPP) Mile2 Mile2 % % % 100s Meters Standardized Note: See supra section IV: Data and Methodology for discussion of data sources, variable construction, and details of data issues. 84 Federal Communications Commission FCC 20-188 Fig. G-44: Content Quality Variables Webpages Domains by TLD by TLD Per Per Content Language Country Capita Capita TLDs EPI Language Assumed Australia 53.22 0.12 .au 100.0% 59.8% English Austria 64.04 0.15 .at 64.1% 2.6% German Belgium 60.51 0.14 .be 63.1% 0.6% Dutch Canada 42.37 0.08 .ca 100.0% 59.8% English Czech Republic 89.12 0.12 .cz 59.3% 0.4% Czech Denmark 72.02 0.22 .dk 67.9% 0.2% Danish Estonia 195.92 0.09 .ee 58.3% 0.1% Estonia Finland 70.89 0.09 .fi 65.3% 0.1% Finnish France 38.24 0.05 .fr 57.3% 2.6% French Germany 42.94 0.18 .de 63.8% 2.6% German Greece 31.79 0.04 .gr 59.9% 0.7% Greek Iceland 215.47 0.18 .is 65.6% 0.0% Icelandic Ireland 49.41 0.06 .ie 100.0% 59.8% English Italy 39.56 0.05 .it 55.3% 0.9% Italian Latvia 43.12 0.06 .lv 56.9% 0.1% Latvian Luxembourg 90.30 0.15 .lu 64.0% 2.6% German Mexico 4.80 0.01 .mx 49.0% 4.0% Spanish Netherlands 64.42 0.31 .nl 70.3% 0.6% Dutch New Zealand 56.49 0.14 .nz 100.0% 59.8% English Norway 81.14 0.14 .no 67.9% 0.1% Norwegian Portugal 41.72 0.03 .pt 63.1% 2.0% Portuguese Spain 31.88 0.04 .es 55.5% 4.0% Spanish Sweden 68.11 0.14 .se 68.7% 0.3% Swedish Switzerland 90.10 0.24 .ch 60.2% 2.6% German United Kingdom 42.00 0.15 .uk 100.0% 59.8% English United States 112.96 0.53 .us/.com/.net/.org 100.0% 59.8% English Analysis Both Both Both Both Both Both Domain Education Source Google Tools First W3Techs Year 2020 2020 * 2020 Webpages Domains by TLD by TLD Unit Per Capita Per Capita % % Loading Factor 0.0227 0.3524 0.6728 0.6501 85 Federal Communications Commission FCC 20-188 APPX. G-4 High-Speed Broadband Deployment Comparison with Europe 1. In this Appendix, we compare fixed high-speed and mobile broadband deployment176 in the United States177 and 26 European comparison countries (EU26).178 To conduct the comparison, we rely on the European Commission (EC) deployment data published in the 2019 Broadband Coverage in Europe Report. To match the EC definition of fixed high-speed broadband, we examine U.S. fixed broadband deployment with download speeds of 30 Mbps or higher.179 To match the fixed technologies used in the 2019 Broadband Coverage in Europe Report, we do not include satellite technology.180 We also compare mobile high-speed broadband deployment in the United States and EU26 by focusing exclusively on 4G LTE, which is the baseline industry standard for the marketing of mobile broadband service.181 For our primary fixed and mobile deployment analysis, we rely on data gathered by the FCC and the EC in June 2018, December 2018 (US), and June 2019. We also present a historical overview of fixed deployment in the United States and the EU26 countries from 2015 to 2019. Finally, we provide maps that show fixed high-speed broadband deployment in the United States and Europe. 176 Prior International Broadband Data Reports released by the International Bureau, as part of the annual Broadband Deployment Report and the 2018 Communications Marketplace Report, included comparisons of broadband deployment in the United States and Europe. See, e.g., 2018 Communications Marketplace Report; see also RAY BAUM S Act. 177 We note that our analysis does not include U.S. Territories until December 2018, due to anomalies in the historical data for Puerto Rico and the U.S. Virgin Islands, whose population account for over 92% of the total combined population of the U.S. Territories. The historical data suggest a 21.7 percentage point increase in deployment between 2015 and 2016. 2020 Broadband Deployment Report, GN Docket No. 19-285, Report, 35 FCC Rcd 8986, 8998, para. 25 & n.90 (2020). The year-end 2017 deployment data most likely significantly overstate deployment in Puerto Rico and the U.S. Virgin Islands at that time because the data do not reflect infrastructure damage caused by Hurricanes Maria and Irma. We include data from the U.S. Territories in figures that report data since 2018 only as we believe these FCC Form 477 data collections provide reliable estimates for the U.S. Territories. 178 We refer to the set of countries that we compare here as the EU26, as we selected only 26 of the 31 European countries addressed in the 2019 Broadband Coverage in Europe Report for our analysis. The 2019 Broadband Coverage in Europe Report discusses the 28 member countries of the European Union (EU), as well as Iceland, Norway, and Switzerland. The 26 countries included in our analysis are: Austria (AT), Belgium (BE), Czech Republic (CZ), Denmark (DK), Estonia (EE), Finland (FI), France (FR), Germany (DE), Greece (EL), Hungary (HU), Ireland (IE), Italy (IT), Latvia (LV), Lithuania (LT), Luxembourg (LU), Netherlands (NL), Poland (PL), Portugal (PT), Slovakia (SK), Slovenia (SI), Spain (ES), Sweden (SE), United Kingdom (UK), Iceland (IS), Norway (NO), and Switzerland (CH). 179 2019 Broadband Coverage in Europe Report at 19. We rely on the same data sources, technologies, and methodology as described in the 2018 Communications Marketplace Report International Broadband Data Report Appendices. Communications Marketplace Report, 33 FCC Rcd at 12558, Appx. E-4. As in the 2018 Communication Marketplace Report, we rely on the FCC s Form 477 fixed and mobile 4G LTE deployment data to estimate U.S. broadband deployment as of June 2018, December 2018, and June 2019. FCC, Fixed Broadband Deployment Data from FCC Form 477, https://www.fcc.gov/general/broadband-deployment-data-fcc-form-477 (last visited Oct. 27, 2020); FCC, Mobile Deployment Form 477 Data, https://www.fcc.gov/mobile-deployment-form- 477-data (last visited Oct. 27, 2020). For fixed historical analysis, we also rely on data from the Form 477 data collection. For U.S. fixed technologies capable of at least 30 Mbps download speed, we include: DSL Asymmetric xDSL, ADSL2, symmetric xDSL, VDSL; Cable Modem DOCSIS 1, 1.1, 2, 3.0, and 3.1; Optical Carrier/Fiber to the End User; Copper Wireline; and Fixed Wireless. 180 2019 Broadband Coverage in Europe Report at 7, 17, 24. 181 2018 Communications Marketplace Report, 33 FCC Rcd at 12684, paras. 239-40. In this Appendix, we analyze mobile 4G LTE coverage regardless of minimum advertised speeds or actual speeds to match the 2019 Broadband Coverage in Europe Report. 86 Federal Communications Commission FCC 20-188 I. FIXED HIGH-SPEED BROADBAND COMPARISON A. Total and Rural Household Fixed High-Speed Broadband Deployment Fig. G-45: Fixed High-Speed Broadband Deployment, All Households (EU June 2018, US December 2018, and US/EU June 2019) *EU data from June and US data from December. 87 Federal Communications Commission FCC 20-188 Fig. G-46: Fixed High-Speed Broadband Deployment, All Rural182 Households (EU June 2018, US December 2018, and US/EU June 2019)183 *EU data from June and US data from December. 182 Within the United States, the designation of a census block as urban is based upon the 2010 Census. An urban census block encompasses all population, housing, and territory included within a census block categorized as in an urban area or urban cluster. A rural census block encompasses all population, housing, and territory not included within urban census blocks. The European Commission defines rural households in square kilometers with a population of less than one hundred. U.S. Census, Urban and Rural, https://www.census.gov/programs- surveys/geography/guidance/geo-areas/urban-rural.html (last visited Oct. 27, 2020); 2019 Broadband Coverage in Europe Report at 22. 183 The 2019 Broadband Coverage in Europe Report presents broadband connections capable of at least 30 Mbps at a national level, defined as follows:  This category encompassed VDSL (including VDSL2 Vectoring), FTTP, FWA (4G TD LTE standard) and DOCSIS 3.0 (including DOCSIS 3.1) cable broadband access technologies. However, as not all connections utilizing these technologies can achieve 30 Mbps and higher actual download speeds (for example, VDSL connections with distance from the exchange point higher than 500m see radical decrease in actual speeds), respondents were asked to exclude those connections which did not meet the criteria from their answers. However, this category is not available for rural areas. Therefore, in these areas, we consider next-generation access (NGA) availability.  The NGA combination category is comprised of VDSL (including VDSL 2 Vectoring), FTTP, and cable modem DOCSIS 3.0 (including DOCSIS 3.1) technologies, all typically capable of delivering a service speed of at least 30 Mbps. 2019 Broadband Coverage in Europe Report at 24, 33. 88 Federal Communications Commission FCC 20-188 B. High Speed Rural and Non-Rural Household Broadband Deployment Fig. G-47: United States and EU26 Rural vs. Non-Rural (Households) Fixed High-Speed Broadband Deployment (June 2018) Fig. G-48: United States and EU26 Rural vs. Non-Rural Households, Fixed High-Speed Broadband Deployment (June 2019) 89 Federal Communications Commission FCC 20-188 C. Total High-Speed Broadband Deployment by Country Fig. G-49: Fixed High-Speed Broadband Deployment by Country for All Households (EU June 2018 and US December 2018) *EU data from June and US data from December. Fig. G-50: Fixed High-Speed Broadband Deployment by Country for All Households (EU and US June 2019) 90 Federal Communications Commission FCC 20-188 D. Rural High-Speed Broadband Deployment by Country Fig. G-51: Fixed High-Speed Broadband Deployment by Country for All Rural Households (EU June 2018 and US December 2018) *EU data from June and US data from December. Fig. G-52: Fixed High-Speed Broadband Deployment by Country for All Rural Households (EU and US June 2019) 91 Federal Communications Commission FCC 20-188 E. Comparison of 2 Mbps, 30 Mbps, and 100 Mbps Fixed Broadband Deployment in the United States and the EU26 Fig. G-53: Fixed High-Speed Broadband Deployment for All Households by Speed (EU June 2018 and US December 2018) Fig. G-54: Fixed High-Speed Broadband Deployment for All Households by Speed (EU and US June 2019) 92 Federal Communications Commission FCC 20-188 II. MOBILE HIGH-SPEED BROADBAND COMPARISON Fig. G-55: 4G LTE Mobile Broadband Coverage for All Households (EU and US June 2018 and June 2019) Note: Due to rounding, values of 100% should be interpreted as at least 99.5%. Fig. G-57: 4G LTE Mobile Broadband Coverage for All Rural Households (EU and US June 2018 and June 2019) Note: Due to rounding, values of 100% should be interpreted as at least 99.5%. 93 Federal Communications Commission FCC 20-188 III. HISTORICAL OVERVIEW OF FIXED HIGH-SPEED DEPLOYMENT, 2015-2019 Fig. G-59: Fixed High-Speed Deployment, All Households *EU data from June and US data from December. Fig. G-60: Fixed High-Speed Deployment, All Rural Households *EU data from June and US data from December. 94 Federal Communications Commission FCC 20-188 Fig. G-61: Fixed High-Speed Deployment, Non-Rural Households *EU data from June and US data from December. 95