Interrogating supply-side data Helani Galpaya Katmandu, March, 2015 This work was carried out with the aid of a grant from the International Development Research Centre, Canada and UKaid from the Department for International Development, UK.

Much data; many producers/publishers OPERATORS/SUPPLIERS -Financial data -Operational data ( equipment, quality) -Complaints -Transaction Generated Data (Big Data) REGULATOR/ POLICY MAKER -Network level data for decision making and Monitoring -Complaints ITU/UN/WEF/OTHER INT’L/NTL ORGs -Raw data -Composite indices (ranking countries) USERS -Quantitative surveys -Qualitative studies THIRD PARTY RESEARCH -Specialized Studies based on data obtained from all stakeholders

CONNECTIVITY

Is connectivity increasing? Looks impressive

But PK is in middle of pack when compared to S Asian neighbors Benchmarking is a basic, but useful way to understand if performance is ‘good’.

But telecom data changes, fast: Most recent data is a must-have 6 SIMs per 100 population

Who is actually ahead? Why? 7 Aided by multiple millions of SIMs deregistered in PK in 2008 & and change of definition of active SIMs in India in 2011

Which indicator benchmark? Different indicators can tell different stories Mobile SIMs per 100 population, India and Bangladesh Mobiles SIMs: India and Bangladesh

Are the data comparable? E.g., How do you reconcile different financial years? Many countries Jan – Dec (calendar year) – E.g., Sri Lanka But many others differ – India: Apr – Mar – Pakistan : Jul – June So “number of Internet users in per 100 inhabitants in 2013” reported by IN not comparable with PK Having quarterly data eliminates problem Especially important if benchmarks are used for mainstream regulatory work such as interconnection or retail tariff regulation

Prerequisites for comparison 10 Internationally accepted definitions and procedures – ITU (2010) Definitions of World Telecommunication/ICT Indicators, Geneva: ITU Make sure that the definitions are adhered to – ITU has mobile broadband definition; use is inconsistent “Mobile broadband subscribers refer to subscribers to mobile cellular networks with access to data communications (e.g. the Internet) at broadband speeds (here defined as greater than or equal to 256 kbit/s in one or both directions) such as WCDMA, HSDPA, CDMA2000 1xEV-DO, CDMA xEV-DV etc, irrespective of the device used to access the Internet (handheld computer, laptop or mobile cellular telephone etc). These services are typically referred to as 3G or 3.5G and include: Wideband CDMA (W-CDMA), an IMT G mobile network technology, based on CDMA”

And whose data do you use? # of internet subscribers (millions), IndiaDifference between… YearNASSCOM dataTRAI Data Ministry of Statistics & PI NASSCOM & TRAI numbers TRAI & Ministry numbers %1% %4% %5% %4% %11% %5% % Note: Based on Financial Year – e.g. “2000” refers to April 1999 – Mar 2000 Source: NASSCOM Strategic Review 2005; TRAI; Ministry of Statistics and Program Implementation, Govt. of India

Useful Indicators to measure connectivity FIXED Number of fixed lines Number of fixed wireline phones Number of fixed wireless phones Total fixed access paths per 100 inhabitants MOBILE Number of mobile SIM cards Number of mobile SIM cards – prepaid Number of mobile SIM cards – postpaid Total mobile SIMs per 100 inhabitants BROADBAND Number of broadband connections per 100 inhabitants ICT Number of mobile users Number of Internet users IN-COUNTRY ACCESS GROWTH Backbone map for a country Mobile coverage map per operator Base station map per operator Can the method for estimating be improved?

‘Proportion of individuals using the Internet’ Base indicator in composite indices such as: – NRI (Network Readiness Index) – KEI (Knowledge Economy Index) – IDI (ICT Development Index) Best measurement method recommended by ITU: – demand-side survey on proportion of individuals using the Internet (from any location) in the last 12 months (HH7) But only 58% percent of the countries have conducted demand-side surveys by ITU

Various methods can be used to estimate the number of Internet users Internet Users = multiplier x Internet Subs (supply side) Where – The multiplier = a number used to reflect that each subscription is used by more than one individual (e.g. at kiosks) – Internet subscriptions = Internet subscription of all types (speeds, technologies etc. ) Wired, wireless etc.

Building on foundations of sand… Multipliers chosen at discretion of Country administrations – Perverse incentive to use higher multiplier to show high Internet penetration in country Difficulties in counting Internet subscriptions include… – Over-counting (counting all “Internet-capable” SIMs, irrespective of use) – Under-counting (being able to only count SIMs that have subscribed to a data package; SIMs with only voice packages may use Internet, but operators cannot count; impossible for pre-paid) – General difficulty with multiple ownership (one user with fixed and many SIM connections) leading to questionable multipliers

………… ………… ………… ………… Difficult to find rationale for multipliers Country Fixed Internet Subscriptions (000s), 2009 Internet Users (000s), 2009, ITU method ITU multiplier Russia88,06859, Mauritius Liberia Liechtenstein Hong Kong, China3,0424, Côte d'Ivoire Sudan444, Iraq Uganda303, Afghanistan21, Huge variance in Multipliers: 0.68 (Russia) to 500 (Afghanistan) in 2009 “Similar” countries with very different multipliers Afghanistan -2,000 fixed subscriptions; Multiplier=500 Burundi -5,000 fixed subscriptions; Multiplier=13

Improvement proposed by LIRNEasia % of Internet users increase with Education and Income components of Human Development Index (HDI) of a country – Education component - mean of years of schooling for adults and expected years of schooling for children – Income component- Logarithm of GNI per capita (PPP$). – Health component of HDI is not used, due to lack of evidence that internet penetration is correlated with life expectancy Studied the correlation between Internet penetration rate of countries which conducted demand side surveys and the education and income components of HDI 2011 – Data on countries which have conducted demand-side surveys was obtained from ITU and RIA – Sub index Education_GNI Index, consisting of education and income components of the HDI index was calculated using ‘DIY HDI: Build Your Own Index’ on UNDP website. Both Education and Income were given equal weight

Strong correlation between Education_GNI Index and Internet penetration Education_GNI Index 2011

Step 1: If survey is available, use it since survey results are first best If representative survey from regional organization is available, use their data ( e.g. RIA ) If survey from current year is not available, use previous year’s data with adjustment – Adjust by average growth for country grouping (e.g., middle income countries etc.)

Derive model using income and education components of Human Development Index (HDI) vs. Internet penetration rate for countries which have conducted a survey (annually after HDI report has been released) Use this model to impute % of Internet Users for countries which have never conducted a survey If Internet penetration rate provided by country administrator is within +/- 7 percentage point band around calculated estimate -> use country reported figure Else use imputed figure Step 2: In the absence of survey data use Education_GNI Index to estimate proportion of Internet users

Less than 30% countries show different Internet penetration rates ‘X’ Survey data from RIA but not same as ITU Internet penetration rate

PRICE & AFFORDABILITY

In selecting an operator, consumers are likely to think about ALL costs including Connection charge, monthly rental etc. ITU ICT price basket methodology takes these issues into account and has created Broadband Baskets consisting of Monthly cost of 1 GB use per month with at least 256kbps connection for a period of 24 months (includes Initial Connection Fee/24) Affordability measured by dividing the cost of the Broadband basket by National average monthly GNI per capita RIA (Research ICT Africa) has further developed this methodology and also measure the cost of the following baskets in addition to the ITU basket Monthly cost of uncapped use per month with at least 256kbps connection for a period of 24 months. Broadband Baskets: a realistic method of price comparison

Affordability of Fixed BB is increasing in developing countries, but still higher than developed countries Source: ITU, Measuring the Information Society 2013, D/Statistics/Documents/publications/mis2013/MIS2013_without_Annex_4.pdfhttp:// D/Statistics/Documents/publications/mis2013/MIS2013_without_Annex_4.pdf

What about other prices? E.g. BB, wholesale & retail? With 83 footnotes in the most recent publications we did

QUALITY

Measuring BB quality: trade offs across differing approaches (who can measure?) 27 Diagnostics conducted by ProsCons Service Providers (ISPs/operators) Easy to implement Results based on equipment placed in the most optimized points of the network Not representative of the ‘actual’ speeds Users Represents actual user experiences Assumes user’s PC is virus-free, with no parallel process running etc. Dependant on user’s willingness to participate Large files impact user’s data limits National Regulatory Authorities (NRAs) In a position to request operator involvement when necessary Known test locations will prompt operators to optimize networks in selected areas

Download speeds: Indonesia performs among the lowest (2014)

Delivers what was promised (advertised): Indonesia is not bad (2014)

Value for Money: Indonesia highest (2014)

Useful Indicators to measure Quality Voice Quality Call drop rates % of connections with good voice clarity Call success rate Broadband Quality Broadband upload/ download speed (kbps/Mbps) RTT or Round Trip Time (mili-second Jitter (mili-second) Packet-Loss (as a %) Broadband availability (as a %)

Quality is more than download speed +++ Highly relevant; ++ Very relevant; + Relevant; - Irrelevant ServiceDownload (kbps) Upload (kbps) Latency (Round Trip Time, RTT) (ms) Jitter (ms) Packet Loss (%) Browsing (Text) Browsing (Media) Downloading Transactions Streaming media VOIP+++++ Games RTT has implications on client-server interactive systems - Jitter adds to the ‘noise’ of the transmission - Packet Loss affects streaming media Source: Gonsalves, T.A & Bharadwaj, A. (2009) 32 With increased use of cloud resources, this changes. E.g. upload as important as download

INDUSTRY-STRUCTURE INDICATORS

Do you know a competitive market when you see one? “There was only ONE broadband service provider (Company A) in the country until Since then, there has been 4 service providers: A, B, C, D”. – Is this a competitive industry? What if we saw this? YearA’s shareB’s shareC’s shareD’share % %2% 1% %1%2% %.5%2.5%2% %0.25%2.75%2%

HHI (Hirschman Herfindahl Index) is basic measure of market concentration Define Market – Fixed? Mobile? Voice telephony (fixed and mobile)? Internet Services? – Identify market share of each operator M1, M2, M3…. – Subscriber share, revenue share, minute share? HHI = (M 1 ) 2 + (M 2 ) 2, (M 3 ) 2 +…+(M n ) 2 US Dept of Justice says… – Greater than1800  concentrated market – Between moderately concentrated – Less than 1000, concentrated – M&A activity increasing HHI by100+ and HHI >1800  automatic review (etc.)

But what is the ‘market’? Fixed? Mobile? Voice? BB? Substitutability/elasticity India, Mar Mobile HHI = BB HHI = Fixed HHI = 7034 India, fixed market shares India, BB market shares India, mobile market shares

Market share based on SIMs? Revenue? Minutes? What if SMP definition was 60% for price regulation?

Market segmented by wholesale vs. retail – E.g., Ofcom (UK regulator) reports wholesale (BT dominated, HIGHLY concentrated) vs. retail (less concentrated, many ISPs including BT) Other ways

IN UNDERSTANDING THE HEALTH OF THE ICT SECTOR…

Traditionally prominance was for ‘supply side data’ Supply side data = data produced by the suppliers of services: MNOs, equipment manf. Reluctance to share, even when demanded by regulation Incentives to over- or under-report Not comparable due to lack of agreement on definitions – Internet user (good enough for most most people) – ‘daily active internet user (useful to operators)

Demand-side data: great for some measures, not for others. The right mix? Understand perceptions, opinions, feelings, satisfaction levels UK broadband satisfaction survey Combined with crowd-sourced testing LIRNEasia electricity sector customer satisfaction survey in LK, IN, BD IN worst actual performance But highest satisfaction levels

Big Data: bridging the gap between expressed vs. revealed preference How many minutes does it take, on average, for a vehicle to get from point A to B? Can survey ppl  reasonable estimate Can do traffic monitoring  estimate also But most people have cell phones CDR or VLR Analyze the digital trace

In short, when looking at supply-side data, QUESTION EVERYTHING Question EVERYTHING Who reported it? What incentives did reporter of data have? What was the data collection method? Is it a reasonable representation? Does the data in isolation say a story that’s different from the data that’s compared to others (over time, to other countries, etc.) Where possible, triangulate (supply+demand)

COMPOSITE INDEXES

Networked Readiness Index

UN e-Gov survey Expert assessment of online Presense of 193 states + Data from int’l orgs

A4AI/Web Foundation: The affordability index Mix of subjective, perception indicators (survey) + objective data (mostly supply-side) No indicators of price or affordability!. One respondent per country in 2013  now 2

Composite indices … Are great for getting attention of nations – Countries want to lead; Donors love them May not stand scientific interrogation? – Why is one sub-index weighed 1/3 while another is weighted 2/3? Etc. Often suffer from the same issues identified in supply-side and demand-side data – How many respondents in expert survey? What incentives? – How old/new and how comparable is objective data?