Mapping Digital Poverty in the Philippines using AI/Big Data and Machine Learning
The COVID-19 pandemic has further exacerbated existing inequalities accentuated by the widening of the digital divide. As the pandemic continues to reinforce the need for social distancing and continued lockdowns, the need for quality digital access and connectivity that is efficient, inclusive, and sustainable - where users including the poor and marginalized have access to sufficient internet connection even in remote areas – has increased in importance. More generally, broadband internet is seen as a general-purpose technology essential for the competitiveness of nations. Achieving universal access to high-quality internet is an important public policy goal but despite an accelerated shift to the digital space enabling service delivery, over 700 million people around the world remain without digital access. The ADB and other development partners call on the need to expand investments in digital infrastructure and ensure equitable access to technology as economies recover from the coronavirus disease (COVID-19) pandemic.
In the Philippines, connectivity remains higher in urban centers and weak digital infrastructures persist in more rural areas. The Department of Information and Communications Technology (DICT) cites that a higher incidence of urban households have internet compared to rural households, with Metro Manila households having the highest access at 32.3%. The impact of the pandemic, coupled with the regulatory environment and weak competition, have further hindered ICT policy interventions. The digital landscape has made productivity a privilege and those without sufficient access are left behind, losing out on opportunities from basic amenities and quality education, to decent work and reskilling. In a post-COVID-19 world, the challenges to the Philippines and other countries in delivering availability, accessibility and affordability of reliable internet have never been greater.
As part of a joint series by ADB and Thinking Machines on Artificial Intelligence, Big Data, and Machine Learning for Development, we map digital poverty in the Philippines. Using advanced machine learning techniques, we estimate poverty combined with big data (e.g. from Project BASS and Speedtest by Ookla) to analyze spatial patterns of digital inequality to better inform and target strategic investments for digital development.
For our work, we use open crowdsourced information from Project Bandwidth and Signal Strength (BASS) which shows the approximate location of cell site towers, triangulated from user devices. We complement this with Speedtest by Ookla for Q3 2020 which records the download speed, upload speed1, and latency on fixed broadband and mobile internet connections. This data is anonymized, averaged and aggregated per quarter available at a high level of spatial granularity for all countries2. For Speedtest by Ookla in the Philippines, we observe for both urban and rural areas, over 80% of devices running tests for Q3 were on fixed broadband, while just under 20% were on mobile. This is consistent with the recent 2019 National ICT Survey which shows that the majority of households use fixed broadband internet versus mobile internet, and that this is the case for most regions across the country (Government of the Philippines, DICT, June 2020).
Our key findings are enumerated below:
Access to sufficient internet speeds shrink in rural areas.
In 2019, the Philippine DICT released high-level findings from the first National ICT Survey, aimed to establish internationally comparable ICT indicators for the country. Based on the survey, only half of the country’s 42,064 barangays (the smallest administrative level of government) have telecom operators in the area and only 30% have fiber optic cables installed.
But even for those with internet access, is there sufficient bandwidth for remote work or learning? For popular video calling platforms like Zoom and Google, the minimum requirement to conduct quality group calls ranges from 3-3.2 megabytes per second (mbps) for upload and download speeds. Using this as a benchmark for “sufficient internet speed”, we analyze the population for each region in the Philippines cross-referenced to the Speedtests from Q3 2020 to see how many people have access to the minimum required bandwidth for group video calls and how many do not.
From our research, we see that only 83% of the Philippine population live in areas with sufficient fixed broadband speeds, while only 70% for mobile. If we filter this down to the population in barangays categorized by administrative boundary data as rural, this percentage shrinks to 70% of the rural population on fixed broadband and 52% on mobile.
Figures 1 and 2 below show the percent of the population per region living in areas categorized as urban (blue) and rural (green)3, highlighting those with access to at least 3.2mbps for both download and upload speeds. Unsurprisingly, highly urbanized and dense Metro Manila (NCR) has the largest percent of the population with sufficient access to both fixed broadband and mobile. Whereas the Bangsamoro Autonomous Region of Muslim Mindanao (BARMM) has the largest percent of the rural population without sufficient access for both fixed broadband and mobile internet.
The 2019 ICT Survey results support this, showing that BARMM has significant digital inequality, with the lowest prevalence of 4G connectivity and one of the highest percent of barangays without telecom service providers. The decline in access for rural populations in these charts not only shows how many are left without the means to fully participate in this shift to the digital space but indicates the disparity of quality infrastructure between more central and urban areas versus more remote and rural areas.
Poorer areas have less access, slower internet speeds, and fewer cell towers.
In drilling down further, we wanted to check whether we see similar patterns when looking at this through the lens of poverty measures. To do this, we use a machine learning model we developed to estimate poverty incidence by getting the wealth index 4 for the entire Philippines. This model was trained using data from the 2017 Demographic and Health Survey (DHS), satellite data, and social media data with estimations providing a high degree of confidence.
Figure 3 below is a map of the Philippines at the city level showing two variables: wealth index and download speeds. Referring to the legend on the upper right-hand corner of the image, the colors from left to right indicate decreasing wealth/increasing poverty incidence from 1 (wealthiest) to 0 (poorest) according to the machine learning model, while colors from bottom to top indicate increasing download speeds according to Speedtest by Ookla. We can see that areas with high wealth and faster speeds (pink) are mainly located in Metro Manila, Central Luzon, and Cebu, while there are much larger portions colored dark blue in the peripheries (Northern Luzon, Palawan, Eastern Visayas, Northern Mindanao). These pertain to poor areas with slow download speeds.
Comparing the average download speeds for five of the wealthiest versus five of the poorest cities in the Philippines, the disparity is stark. The average download speeds in the wealthiest cities are up to 21 mbps faster than the average in the poorest cities (Figure 4).
Zooming into some of these cities, the map below shows how much of the population in these wealthiest and poorest areas have access to sufficient internet speeds of at least 3.2 mbps. Up to 100% of the population in wealthier cities have access to sufficient internet speed of at least 3.2 mbps, while as small as 6% of the poorest cities do. For instance, 100% of the population in San Juan, Metro Manila have access to sufficient internet speeds, while only 6.22% and 4.14% of the population in Burgos, Surigao del Norte have sufficient access on fixed and mobile respectively (Figure 5).
The location of infrastructure (e.g. cell towers) is also key in understanding how access is distributed across the country (Figure 6). The heat map below shows the approximate location of cell towers triangulated from Project BASS’s crowdsourced data. We see clusters of cell towers in urban areas like Cebu City, Puerto Princesa, and Davao City, but none of these compare to the breadth and density of towers in Metro Manila with the largest concentration of cell sites. Meanwhile large gaps can be seen in areas with more geographical constraints, such as the mountainous region in northern Philippines.
At the last mile, only 15% of Filipinos have access to sufficient internet speeds and only 9.5% live within the serviceable scope of cell towers.
For the population considered to be part of the ‘last mile’, access reduces even further. To measure this, we looked at the population living more than 2 kilometers from a major road network and found that, of the 9.4 million Filipinos at the last mile, only 15% actually have access to sufficient internet speeds on fixed broadband. For mobile, this is just 6.5% of the 9.4 million population group.
If we break this down per region, we find that Eastern Visayas (Region VIII) has the largest last mile population without sufficient internet access for fixed broadband, while Central Visayas (Region VII) has the largest last mile population without access to sufficient speeds on mobile (Figures 7 and 8). Note that this only includes the last mile population with internet access to run a Speedtest.
This indicates that up to 600,000 people on mobile and up to 1.4 million people on fixed broadband either have speeds slower than 3.2mbps or did not run a Speedtest and are thus not included in the Ookla data.
Eastern and Central Visayas are composed of fragmented island groups where infrastructure development is more challenging. From the 2015 national census, Region VIII in particular had the lowest level of urbanization in the country, with only 11.9% of the population residing in urban areas, versus the national average of 51.2% (source). This means that up to 88% of the population live in rural areas. We also see from the map in Figure 6 that cell towers are particularly clustered in urban areas as service providers tend to focus their operations in denser and wealthier areas, leaving out those at the last mile with little to no access. For Eastern Visayas, cell towers are particularly clustered in Tacloban City, while towers in Central Visayas are mostly in Cebu City and Tagbilaran City.
In checking the proximity of the total population at the last mile to cell towers from Project BASS data, only 9.5% of the 9.4 million at the last mile live within 500m from a cell tower. This leaves almost 850,000 Filipinos beyond the scope of cell site serviceability.
By identifying these communities from their distance to infrastructure and access to sufficient internet, the implications amidst the shift to digital could be long-term. The pandemic is catalyzing innovations in work, learning, and access to resources and services, but these are centralized in wealthier, urban areas. We risk further entrenching socioeconomic stratification as communities on the margins continue to lose out on the means to recover and adapt in a post-pandemic world.
Big data and machine learning can help inform better digital infrastructure investments moving forward, else we fail to include populations already being left behind.
By augmenting ground truth data with machine learning and open source big data, our findings show that there is still a significant percent of households left behind despite the overwhelming shift to the digital space. We see that even for those with internet connection to run a Speedtest on Ookla, the population with sufficient internet access reduces as we focus on the margins
Why is internet access in the Philippines so disparate? The insufficiency of digital infrastructures in the country is a result of multiple factors, including stringent restrictions in the telecommunications market leading to a lack of competition and high barriers to entry. With predominantly two major service providers that own and control the entire broadband infrastructure, it is extremely difficult for new firms to compete. In addition, the Philippines is one of the top business process outsourcing (BPO) destinations in the world, further concentrating demand for telecommunication services in Metro Manila. A whole-of-government approach will be needed in identifying and implementing both legal and regulatory reforms and policy measures to bridge the digital divide. Key lessons from better practices across the world point to the need for governments to open market entry, make more spectrum available, ensure the independence of regulators and encourage greater investment into the market. To aid in this, ADB is supporting the strengthening of the institutional capacity of the Philippines Competition Commission (PCC) to foster competition through the Capacity Building to Foster Competition Project.
The Philippine government has ongoing efforts to provide better access to the underserved population, such as the National Broadband Plan and The Free Internet Access in Public Places Act of 2017 (RA 10929). Under the latter, the aim is to provide internet for government offices, public basic education institutions, public hospitals and health centers, public parks, airports and seaports, and transport terminals. However, results have been slow. As of 2019, only 13% of surveyed barangays by the DICT have free public Wi-Fi present. Challenges prevent fast nationwide rollout, including lack of government expertise and synergy with the private service providers. As of 2020, the World Bank cites that only 3% of the total target of Wi-Fi sites have been built since 2015 (source).
Integral to these efforts is human capital development – upskilling the existing labor force to harness digital technologies and improving digital literacy by embedding cognitive, socioemotional, and technical skills as part of foundational learning in educational programs. ADB is also currently supporting Department of Education through EdTech solutions to build capacity and support distance education by developing content, training teachers, and piloting education technology solutions (e.g. low-cost tablets to be connected to a local area network (LAN) supported by a solar-powered battery antenna) to ensure that secondary school students have access to quality education throughout school year (SY) 2020/21 and beyond.
Outside of government, there are a number of private and civic sector efforts to bring better connectivity to Filipinos. Messaging applications like Facebook and Viber are now free for users, depending on the service provider. Innovations to decentralize broadband infrastructure are also in development, with SmartMesh planning to roll out in the Philippines (source). Groups around the world have also focused their efforts to bring digital connectivity to the margins. For instance, Google’s internet balloons are a new technology bringing connectivity to Kenya’s geographically hard-to-reach areas through AI in the stratosphere, while Giga Connect of the UN ITU and UNICEF aims to bring sufficient internet access to every school in the world, with operations in Africa, the Carribean, Central America, and Central Asia.
Without collaborative intervention, the digital shift will only grow stronger and leave at least hundreds of thousands behind in the path to recovery. The UN ITU estimates it will cost up to $251bn in investments for Asian countries to achieve universal internet access. Universal access to high-quality internet is critical to achieving digital transformation, boosting growth, expanding opportunities, and improving service delivery in the Philippines and much of the world. By augmenting ground truth information on digital access with machine learning poverty estimates and open data on broadband speeds and cell tower locations, decision-makers can be better informed on targeted investments and strategic plans in addressing digital poverty. The time to act is now.
This is the final blog in a series by the Asian Development Bank and Thinking Machines Data Science, Inc. where we explore ways we can use big data, artificial intelligence, and machine learning to craft development solutions during the pandemic.
Stephanie Sy, Anica Araneta
Hanif Rahemtulla, Bruno Carrasco, Stella Balgos
Asian Development Bank
2 A caveat to note is that it is possible that users of fixed broadband are more likely to run a Speedtest than those on mobile because of the nature of faster internet speeds with the former, thereby making the Speedtest dataset more biased to fixed broadband. For both urban and rural areas in the Philippines, over 80% of devices running tests for Q3 were on fixed broadband, while just under 20% were on mobile.
3 As of 2003, the [Philippine Statistics Authority PSA defines a barangay as urban if it has a population size of at least 5,000, has at least one establishment with a minimum of 100 employees, has five or more establishments with 10-99 employees, and five or more facilities within the 2-kilometer radius from the barangay hall, among others. Barangays that do not meet any of the criteria are classified as rural.
4 The machine learning estimations of the wealth index are derived from asset-based wealth data, in particular the Demographic and Health Survey (DHS), which cites that “The wealth index is a composite measure of a household’s cumulative living standard. The wealth index is calculated using easy-to-collect data on a household’s ownership of selected assets, such as televisions and bicycles; materials used for housing construction; and types of water access and sanitation facilities.” (The DHS Program)
5 For a review of such strategies, see World Development Report 2016 Digital Dividend, Chapter 4 on Sectoral Policies.
Thank you to Maahid for the image in the cover photo.