The impact of internet access on COVID-19 spread in Indonesia

Kunz, Johannes; Propper, Carol; Trinh, Trong-Anh Working Paper The impact of internet access on COVID-19 spread in Indonesia ADB Economics Working Paper Series, No. 723 Provided in Cooperation with: Asian Development Bank (ADB), Manila Suggested Citation: Kunz, Johannes; Propper, Carol; Trinh, Trong-Anh (2024) : The impact of internet access on COVID-19 spread in Indonesia, ADB Economics Working Paper Series, No. 723, Asian Development Bank (ADB), Manila, https://doi.org/10.22617/WPS240232-2 This Version is available at: https://hdl.handle.net/10419/298169 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by/3.0/igo/ ASIAN DEVELOPMENT BANK ASIAN DEVELOPMENT BANK 6 ADB Avenue, Mandaluyong City 1550 Metro Manila, Philippines www.adb.org THE IMPACT OF INTERNET ACCESS ON COVID-19 SPREAD IN INDONESIA Johannes S. Kunz, Carol Propper, and Trong-Anh Trinh ADB ECONOMICS WORKING PAPER SERIES NO. 723 April 2024 The Impact of Internet Access on COVID-19 Spread in Indonesia This study examines the impacts of 3G internet connectivity on COVID-19 case rates across districts in Indonesia. By analyzing geographical variations in mobile internet access and employing lightning strikes as an instrumental variable, the study establishes a causal link between improved internet connectivity and reduced transmission of COVID-19. The findings suggest that investments in digital infrastructure might be a crucial and effective tool in pandemic prevention and response. About the Asian Development Bank ADB is committed to achieving a prosperous, inclusive, resilient, and sustainable Asia and the Pacific, while sustaining its efforts to eradicate extreme poverty. Established in 1966, it is owned by 68 members —49 from the region. Its main instruments for helping its developing member countries are policy dialogue, loans, equity investments, guarantees, grants, and technical assistance. ASIAN DEVELOPMENT BANK The ADB Economics Working Paper Series presents research in progress to elicit comments and encourage debate on development issues in Asia and the Pacific. The views expressed are those of the authors and do not necessarily reflect the views and policies of ADB or its Board of Governors or the governments they represent. ADB Economics Working Paper Series The Impact of Internet Access on COVID-19 Spread in Indonesia Johannes S. Kunz, Carol Propper, and Trong-Anh Trinh No. 723 | April 2024 Johannes S. Kunz ([email protected]) and Trong-Anh Trinh ([email protected]) are research fellows at the Centre for Health Economics, Monash University. Carol Propper (c.propper@ imperial.ac.uk) is a professor of economics at Imperial College London. Creative Commons Attribution 3.0 IGO license (CC BY 3.0 IGO) © 2024 Asian Development Bank 6 ADB Avenue, Mandaluyong City, 1550 Metro Manila, Philippines Tel +63 2 8632 4444; Fax +63 2 8636 2444 www.adb.org Some rights reserved. Published in 2024. ISSN 2313-6537 (print), 2313-6545 (PDF) Publication Stock No. WPS240232-2 DOI: http://dx.doi.org/10.22617/WPS240232-2 The views expressed in this publication are those of the authors and do not necessarily reflect the views and policies ofthe Asian Development Bank (ADB) or its Board of Governors or the governments they represent. ADB does not guarantee the accuracy of the data included in this publication and accepts no responsibility for any consequence of their use. The mention of specific companies or products of manufacturers does not imply that they are endorsed or recommended by ADB in preference to others of a similar nature that are not mentioned. By making any designation of or reference to a particular territory or geographic area inthis document, ADB does not intend to make any judgments as to the legal or other status of any territory or area. This publication is available under the Creative Commons Attribution 3.0 IGO license (CC BY 3.0 IGO) https://creativecommons.org/licenses/by/3.0/igo/. By using the content of this publication, you agree to be bound bytheterms of this license. For attribution, translations, adaptations, and permissions, please read the provisions andterms of use at https://www.adb.org/terms-use#openaccess. This CC license does not apply to non-ADB copyright materials in this publication. If the material is attributed toanother source, please contact the copyright owner or publisher of that source for permission to reproduce it. ADB cannot be held liable for any claims that arise as a result of your use of the material. Please contact [email protected] if you have questions or comments with respect to content, or if you wish toobtain copyright permission for your intended use that does not fall within these terms, or for permission to use theADB logo. Corrigenda to ADB publications may be found at http://www.adb.org/publications/corrigenda. Note: In this publication, ADB recognizes “China” as the People’s Republic of China and “Hong Kong” as Hong Kong, China. ABSTRACT The coronavirus disease (COVID-19) pandemic has highlighted the crucial role of Internet access in pandemic prevention and response. Internet access has facilitated the rapid dissemination of vital information, provided telemedicine services, and enabled remote work and education. This study uses a wide range of data sources to investigate the geographic variation of Internet access proxied by 3G mobile broadband during the COVID-19 pandemic in Indonesia. We employ several approaches to account for potential confounding factors, including using lightning strikes as an instrumental variable, to confirm the significant role that the Internet played in the spread of COVID-19 cases. Our findings suggest that increasing Internet access could positively impact pandemic prevention and response efforts, particularly in regions with limited connectivity. Therefore, improving Internet infrastructure in developing countries may be crucial in preventing future pandemics. Keywords: health emergencies, Internet access, information, COVID-19 spread JEL codes: I12, I15, I31, O18, L96, H41 __________________ We thank Arief Ramayandi and Daniel C. Suryadarma for their outstanding support throughout this project. We thank Kevin Staub, Giovanni van Empel, Paul Raschky and seminar participants at SoDa Lab-Monash, CHE-Monash, and ADB Conference 2023 (Manila) for their useful comments. Financial support from the TA-6989 REG: Asian Development Outlook Update 2023 Theme Chapter: Asia’s Preparedness on Health Emergencies (55278-003) is gratefully acknowledged. Neither of the authors have any conflict of interest to declare. 1 Introduction The emergence of the coronavirus disease (COVID-19) in December 2019 and its rapid spread into a global pandemic has profoundly impacted societies worldwide, causing widespread anxiety and concern among the public. The need to control the spread of the virus has necessitated strict measures such as nationwide quarantines and increased sanitation protocols, leading to social and economic disruption in various sectors (Brodeur, Gray, Islam, and Bhuiyan 2021). In addition, fear and panic have fueled prejudice and discrimination against certain groups (Lu et al. 2021). At the same time, remote working and teaching have become the norm for many, necessitating a major shift in working and educational practices (Bloom, Davis, and Zhestkova 2021). The pandemic has also significantly impacted mental health, with the isolation and anxiety caused by the pandemic becoming a major concern (Brodeur et al. 2021, Butterworth et al. 2022). Moreover, the pandemic has exposed deep-seated poverty and health disparities within many societies, highlighting the urgent need to address these issues (Ahmed et al. 2020). As disease outbreaks are not likely to disappear in the near future, it is crucial to understand preparedness strategies that can reduce the impact of future diseases. The COVID-19 pandemic has brought the importance of Internet access into sharper focus. Prior research suggests that Internet access may play a crucial role in health emergencies (Barrero, Bloom, and Davis 2021; Alipour, Fadinger, and Schymik 2021; Barrero et al. 2021; Whitelaw et al. 2020). With the rapid spread of the virus, there was an urgent need for the quick and effective dissemination of information to the public. The Internet has enabled the rapid sharing of updates on the virus through websites, social media, and messaging platforms. Additionally, the Internet has provided health authorities 2 a platform to communicate directly with the public, sharing guidance on preventive measures such as wearing masks, washing hands, and social distancing. Telemedicine has emerged as a vital tool in the fight against COVID-19, allowing healthcare professionals to provide virtual consultations, monitor patients remotely, and reduce the risk of exposure to the virus in healthcare facilities. However, the impact of Internet access on pandemic prevention may be partially positive, as it can also facilitate the spread of misinformation. For example, false claims about the effectiveness of certain treatments or the safety of vaccines have been widely circulated online, hampering efforts to contain and mitigate the disease (Cinelli et al. 2020, Bursztyn et al. 2020). Therefore, the overall effect of Internet access on pandemic prevention is an empirical question that requires further investigation. In developing countries where the health care system is often underdeveloped, Internet access is crucial in reducing the spread of COVID-19. The World Bank reports that only 19.1% of the population in low-income countries has access to the Internet, compared to 87.7% in high-income countries (Kelly and Rossotto 2011). This limited access to the Internet exacerbates the challenges faced by the health systems in these countries, making it difficult to manage the pandemic effectively. Access to the Internet is essential for healthcare delivery, disease surveillance, and public health interventions. It allows people to access reliable information about the virus, preventive measures, and telemedicine services that reduce the burden on overworked health systems. Additionally, Internet access enables people to work from home, reducing the need for physical contact and minimizing the risk of spreading the virus. Therefore, higher access to the Internet is likely associated with a reduction in COVID-19 transmission in 3 developing countries. Figure 1 plots the relationship between country-level average mobile Internet speed and the number of COVID-19 cases for several economies in Asia and the Pacific countries, suggesting that stronger Internet connections are associated with lower COVID-19 transmission rates and, consequently, deaths. This study provides empirical evidence of the role of Internet access during the COVID-19 pandemic by exploiting subnational data on mobile broadband, and COVID19 spread in Indonesia. Indonesia became the epicenter of the pandemic in Southeast Asia, with the highest number of confirmed cases and deaths in the region by the end of 2022 (Figure 1 and Appendix A.6). Several factors contributed to the severity of the situation, including limited testing and tracing capacity, inadequate healthcare infrastructure, and public skepticism towards the government’s pandemic response measures. Moreover, the variation in Internet access across regions and provinces posed a challenge in Indonesia, with implications for education, healthcare, and economic opportunities. Although some areas had higher levels of Internet access, particularly in urban areas and on the islands of Java and Bali, other areas, particularly rural and remote ones, had limited Internet access due to inadequate infrastructure and resources. This limited access could exacerbate the risk of infection in those regions, especially during the COVID-19 pandemic, where access to government information about the disease became increasingly crucial. Figure 2 shows that areas with higher Internet access are associated with lower disease transmission rates, providing further evidence of the importance of Internet access in controlling the spread of COVID-19 in Indonesia and other developing economies. 4 Our empirical strategy builds on the variation in 3G mobile broadband access across districts in Indonesia in 2019. While the pre-COVID period is less likely to be related to the spread of diseases, concerns remain about the correlation between Internet access and other factors that could influence the spread of the disease. Therefore, we adopt an instrumental variable approach to address this issue, using the incidence of lightning strikes as an instrument (Guriev, Melnikov, and Zhuravskaya 2021; Manacorda and Tesei 2020; Do, Gomez-Parra, and Rijkers 2023). Our strategy assumes that areas with higher incidences of lightning display slower adoption of mobile phone technology. At the same time, it is reasonable to assume that lightning strikes are uncorrelated with the spread of COVID-19. However, this assumption may not hold unconditionally as lightning strikes could be correlated with geographical and climatic variables or the availability of infrastructures or services that might have an independent effect on COVID-19 outcomes. Thus, we control for a wide range of potential determinants of COVID-19, such as demographic factors and health infrastructure, to isolate the effect of 3G access on the spread of the disease. We find that access to 3G Internet plays a vital role in reducing the transmission rate of COVID-19 in Indonesia. Our study shows that the number of COVID-19 cases is approximately 25% lower in areas where 3G Internet access is available. Our findings hold even when we control for other factors that could affect the virus’s spread and consider the potential endogeneity of Internet access. One possible explanation for this result is that Internet access provides individuals with easy access to accurate information about the virus, such as how to prevent infection, where to get tested, and what to do if they become ill. This information can help people make informed decisions about their 11 geographic data. The OSM data for Indonesia are instrumental in quantifying the number of healthcare facilities across various geographic locations, specifically clinics, and hospitals. Clinics are defined as health facilities where outpatient medical care is provided, often by general practitioners or specialists for family or internal medicine. Conversely, hospitals are more extensive facilities that provide a broader range of services, including inpatient care, specialized surgeries, and emergency services. This dataset offers a comprehensive spatial distribution of healthcare infrastructures across the nation, thus enabling the identification of potential disparities in the provision of health services. In conjunction with the OSM data, we also incorporate local healthcare spending data at the regency (or city) level, available from 2017 to 2021. Economic Indicators To estimate local economic activity in Indonesian districts, we utilize satellite nightlight data from the Visible Infrared Imaging Radiometer Suite (VIIRS) administered by the National Oceanic and Atmospheric Administration (NOAA). This methodology has gained popularity in economic research as a reliable proxy for growth outcomes (Hodler and Raschky 2014; Henderson, Storeygard, and Weil 2012) and was found to be superior to other economic variables in some circumstances (Martinez 2022).2 We calculate the nightlight density for all Indonesian districts by aggregating satellite images from daily grids to yearly data.3 2 In the context of Indonesia, Gibson et al. (2021) find a positive relationship between district-level VIIRS data and gross domestic product (GDP). 3 Atmospheric conditions may impact the ability of satellite sensors to capture night lights. We follow the Copernicus program’s recommendation to exclude results from pixels with above 10% cloud fraction by performing cloud masking to address this. For more details: Copernicus. 2020. “Flawed Estimates of the Effects of Lockdown Measures on Air Quality Derived from Satellite Observations.” March 26. https://atmosphere.copernicus.eu/flawed-estimates-effects-lockdown-measures-air-quality-derivedsatellite-observations?q=flawed-estimates-effects-lockdown-measures-air-quality-satellite-observations. 12 We add the recently made available sub-national human development index compiled for Indonesia in 2019 on the regency level, capturing economic status, human well-being, and flourishing more broadly.4 Weather Data We obtain weather data from ERA5, which is the fifth generation of reanalysis dataset produced by the European Centre for Medium-Range Weather Forecasts (ECMWF). Using reanalysis data has several benefits for our study. Firstly, weather stations’ spatial and temporal coverage is often limited in many developing countries. In contrast, reanalysis data covers a larger geographical area and is available over a more extended period. Secondly, reanalysis data may resolve issues related to weather data, such as endogeneity concerns associated with weather stations placement and variations in the quality and quantity of data collection, as noted in previous studies (Auffhammer et al. 2013, Donaldson and Storeygard 2016). ERA5 incorporates information from various sources, including ground stations, satellites, weather balloons, and climate models, to provide weather data from 1979 onwards. The data has a high spatial resolution of 31 kilometers (km) and has been standardized to a regular latitude-longitude grid of 0.25 degrees. For this study, we use the yearly temperature (measured in Celsius) and precipitation (measured in millimeters) data, which we aggregate at the regency level for Indonesia using the inverse-distance weighting approach (Deschenes and Greenstone 2011). As an instrument for Internet coverage, we employ data on lightning strikes sourced from the World Wide Lightning Location Network (WWLLN) dataset (Kaplan and Lau 4 The sub-national indexes are still relatively rare. An alternative approach for other contexts, a recent machine learning approach using satellite data, have been proposed (Sherman et al. 2023). 13 2021). This dataset provides detailed grid-level data of lightning strikes, with a precision of 0.5◦ × 0.5◦ per grid. We then aggregate these data from the grid level up to the regency level for Indonesia. Our measure of lightning strike is the mean stroke power in megawatts (MW), aggregated at the yearly basis. This measure provides a unique perspective on the intensity and energy of lightning activity, which we hypothesize to be a relevant proxy for assessing Internet coverage in the region. 3 Empirical Model We aim to investigate the effects of Internet access, proxied by pre-pandemic (t0) 3G availability (G3accessr,t0), on reducing COVID-19 cases and deaths at the regency level r in Indonesia. We adopt the methodology proposed by Kunz and Propper (2022) to assess the over-time association between COVID-19 outcomes and pre-pandemic access to the Internet, such that 𝐸𝐸�𝑦𝑦𝑟𝑟,𝑡𝑡�𝑋𝑋� =exp(𝛼𝛼𝑡𝑡+ 𝜏𝜏𝑡𝑡𝐺𝐺3𝑓𝑓𝑓𝑓𝑓𝑓𝑎𝑎𝑎𝑎𝑎𝑎𝑟𝑟,𝑡𝑡0+ 𝑥𝑥𝑟𝑟,𝑡𝑡0 ′𝛽𝛽𝑡𝑡+ 𝛿𝛿𝑝𝑝,𝑡𝑡 (1) where 𝑦𝑦𝑟𝑟,𝑡𝑡 is the cumulative cases (or deaths) in regency (or city) r in month t, adjusted by 10,000 population. We control for a set of covariates (𝑥𝑥𝑟𝑟,𝑡𝑡0 ′), all measured on the regency level and before the pandemic’s start. These include demographic factors, health facilities, economic activities, and weather conditions. Additionally, we include province fixed effects (δp) to control for unobservable factors affecting our outcomes at the province level (the main level of local policymaking). We employ robust standard errors to account for potential heteroskedasticity when estimating repeated cross-sections and clustered at the regency level when estimating pooled models. 14 In our primary analysis, we concentrate on the last date in our dataset (i.e., 18 February 2023) and execute a single cross-sectional regression. However, we also estimate Equation (1) for each month separately to illustrate the time trends in the association between Internet access and COVID-19 outcomes. We utilize a Poisson regression model as our primary specification, acknowledging the heavily skewed nature of the cases (Figure A.2). To ensure the robustness of our findings, we evaluate alternative model specifications, including log-transformed OLS and negative binomial regressions. We ultimately choose the Poisson model owing to its beneficial properties, such as being part of the linear, exponential family, demonstrating robustness to misspecification, not necessitating ad-hoc transformation, and avoiding issues related to incidental parameter prediction (Silva and Tenreyro 2006 discuss this in detail). Like any other infrastructure, the expansion of mobile network coverage is likely influenced by endogenous factors. For example, providers of network services tend to prioritize areas with high economic activity and potential demand. However, this trend may be linked to other factors that may, in turn, determine the spread of the pandemic. As a result, simple conditional associations between mobile phone penetration and COVID-19 spread might not correspond to the causal impact. To address this issue, we estimate the following first stage of the Poisson-IV model to mitigate the issue of endogeneity: 𝐺𝐺3𝑓𝑓𝑓𝑓𝑓𝑓𝑎𝑎𝑎𝑎𝑎𝑎𝑟𝑟,𝑡𝑡0=𝛽𝛽+𝛾𝛾1𝑓𝑓�𝐿𝐿𝑓𝑓𝐿𝐿ℎ𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝐿𝐿𝑟𝑟,𝑡𝑡0�+𝛾𝛾2𝑓𝑓𝑎𝑎𝑡𝑡𝑝𝑝𝑎𝑎𝑓𝑓𝑓𝑓𝑓𝑓𝑡𝑡𝑓𝑓𝑎𝑎𝑟𝑟,𝑡𝑡0+𝛾𝛾3𝑝𝑝𝑓𝑓𝑎𝑎𝑓𝑓𝑓𝑓𝑝𝑝𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑟𝑟,𝑡𝑡0+ 𝑥𝑥𝑟𝑟,𝑡𝑡0 ′𝛽𝛽+ 𝛿𝛿𝑝𝑝+ 𝜖𝜖𝑟𝑟,𝑡𝑡0 (2) 15 where the instrumental variable, 𝐿𝐿𝑓𝑓𝐿𝐿ℎ𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝐿𝐿𝑟𝑟, is measured by the average lightning stroke power at the regency level. Our empirical strategy is thus in line with a recent study using the Poisson-IV estimator (Graff Zivin et al. 2023). We now discuss the validity of the instrument employed in our analysis. A good instrumental variable should satisfy several conditions, including exogeneity, relevance, and exclusion. Regarding the exogeneity condition, lightning strikes are a natural phenomenon that occurs independently of COVID-19 spread, making them an exogenous factor in relation to our outcome of interest. Additionally, it is important to note that we measure lightning strikes during the pre-pandemic period, further ensuring the exogeneity of our instrument. In terms of the relevance condition, our instrument is strongly correlated with Internet access, as supported by previous studies (Andersen et al. 2012; Guriev, Melnikov, and Zhuravskaya 2021; Manacorda and Tesei 2020; Do, Gomez-Parra, and Rijkers 2023). Specifically, lightning strikes can cause significant damage to digital infrastructure, leading to higher costs associated with digital technology diffusion. This correlation is particularly relevant in areas with a higher frequency of lightning strikes, where challenges in setting up and maintaining Internet infrastructure may arise. Furthermore, it is worth noting that this IV is ideally suited to our context, as to countries around the equator, as much of the IV’s variation is from this region (Figure B.1). However, there is the possibility of lightning strikes affecting other infrastructure that could indirectly influence the spread of COVID-19 during the pandemic. While the exclusion condition is not directly testable, we have taken measures to control for potential confounding factors, such as weather conditions, humidity, the number of cell 16 towers, and economic activities in our models. These additional controls help to mitigate the potential indirect effects of lightning strikes on COVID-19 spread by capturing other relevant factors that could mediate this relationship. Finally, we modify Equation (1) by including interaction terms between Internet access and a range of factors to examine the heterogeneous effects of Internet access on COVID-19 transmission. 𝐸𝐸�𝑦𝑦𝑟𝑟,𝑡𝑡�𝑋𝑋� =exp(𝛼𝛼𝑡𝑡+ 𝜏𝜏𝑡𝑡𝐺𝐺3𝑓𝑓𝑓𝑓𝑓𝑓𝑎𝑎𝑎𝑎𝑎𝑎𝑟𝑟,𝑡𝑡0+ 𝜏𝜏𝑡𝑡 𝑥𝑥𝐺𝐺3𝑓𝑓𝑓𝑓𝑓𝑓𝑎𝑎𝑎𝑎𝑎𝑎𝑟𝑟,𝑡𝑡0× 1[𝑥𝑥 ≥ 𝑡𝑡𝑎𝑎𝑚𝑚𝑥𝑥]+ 𝑥𝑥𝑟𝑟,𝑡𝑡0 ′𝛽𝛽𝑡𝑡+ 𝛿𝛿𝑝𝑝,𝑡𝑡 (3) In this equation, 1[𝑥𝑥 ≥ 𝑡𝑡𝑎𝑎𝑚𝑚𝑥𝑥] represents an indicator for whether variable 𝑥𝑥 is larger or equal to the median. We concentrate on the primary hypothesis concerning how Internet access might affect the spread of COVID-19 at the regency level. For the heterogeneity analysis, we employ both the Poisson regression as in Equation (1) and the analogous Poisson-IV regressions, using 𝑓𝑓�𝐿𝐿𝑓𝑓𝐿𝐿ℎ𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝐿𝐿𝑟𝑟,𝑡𝑡0�× 1[𝑥𝑥 ≥ 𝑡𝑡𝑎𝑎𝑚𝑚𝑥𝑥] as an additional instrument. We focus on variables deemed important in the literature cited here, such as education level, living areas, health facilities, economic activities, and labor force characteristics. 4 Results Main Findings Table 1 presents results from Equation (1) for increasing sets of covariates, providing a comprehensive analysis of the role of Internet access in reducing COVID-19 transmission. In column 1, the unadjusted association is large, negative, and statistically highly significant. This finding suggests that Internet access, proxied by 3G availability, is 17 indeed associated with a reduction in COVID-19 cases at the regency level in Indonesia. However, the association slightly reduces when including province-level fixed effects, which account for policy-making differences (column 2). Subsequently, we add covariates likely to affect the spread of COVID-19. In doing so, we aim to control for potential confounding factors and isolate the specific effect of Internet access on pandemic outcomes. Columns 3 to 7 results demonstrate that the association between Internet access and COVID-19 transmission remains robust even when accounting for various demographic, healthcare, economic, ethnic, and labor force characteristics. The most substantial drop in the association occurs when including demographic variables, after which the association remains relatively stable. The pseudo R2 increases considerably, particularly when the analysis employs fixed effects and demographic variables. A few of these variables contain missing values, which reduces the sample size. To address this issue, we include these variables either by imputing zeros along with indicators for missing values (column 8) or using the multiple imputation method proposed by Rubin (1996) with 99 random draws (column 9). Both approaches yield nearly identical results, confirming the robustness of our findings. Our preferred model is column 8, and we will focus on this specification in the remainder of the text. By accounting for missing data in our analysis, we ensure a more comprehensive assessment of the relationship between Internet access and COVID-19 transmission, further enhancing the validity and generalizability of our conclusions. In terms of interpretation, we demonstrate that having access to 3G Internet is associated with a substantial reduction in COVID-19 cases by approximately 25% using 18 our preferred estimation in column 7 of Table 1. The magnitude of the effect is noteworthy, particularly in the context of a developing country such as Indonesia, where resources for public health interventions may be limited. Furthermore, the magnitude of the effect observed in our study is particularly relevant compared to the effectiveness of other non-pharmaceutical interventions (NPIs) in reducing COVID-19 transmission. While the impact of specific NPIs varies across studies, the 25% reduction associated with Internet access is comparable to, or even greater than, the effects reported for some widely implemented measures such as social distancing, mask-wearing, and travel restrictions (VoPham et al. 2020, Leech et al. 2022, Kwok et al. 2021). For example, VoPham et al. (2020) show that higher social distancing was associated with a 29% reduction in COVID-19 incidence. These results have important policy implications, especially for developing countries that face challenges in containing the spread of infectious diseases like COVID-19. Our findings suggest that improving Internet access can reduce disease transmission, as it facilitates better dissemination of health-related information, enables remote work and education, and promotes telemedicine services, which collectively reduce the need for physical interaction and help limit the spread of the virus. Figure 4 supports our main findings by consistently demonstrating the effects of Internet access on COVID-19 transmission throughout the pandemic. Furthermore, the stability of these effects over time suggests that factors such as vaccine distribution and potential data inconsistencies at the pandemic’s beginning are unlikely to significantly influence the observed relationship between Internet access and COVID-19 outcomes. 19 Notably, the stable estimates over time indicate that the benefits of Internet access in mitigating the spread of the virus have persisted throughout the various stages of the pandemic. This consistency reinforces the importance of digital connectivity as a crucial factor in public health strategies, both during the initial response to an outbreak and in ongoing efforts to control and manage the spread of infectious diseases. As the pandemic evolves and new challenges arise, such as new variants and changes in public health guidelines, the persistent effects of Internet access on reducing COVID-19 transmission underscore its enduring significance in supporting effective public health interventions. Robustness Table 2 demonstrates that the effects presented in our main analysis are highly robust to various specifications and alternative approaches. These include Log-OLS (column 2), Negative Binomial regression (column 3), alternative measures of 3G access (imputed versus not imputed) (columns 4 and 5), not using population weights (column 6), and imputing the outcomes for areas that did not report any cases (columns 7 and 8).5 Our preferred specification is arguably the most conservative among these approaches. Using the log-transformed OLS approach is similar in terms of the marginal effect. The robustness of our findings across various specifications and alternative methods strengthens our confidence in the highly stable association between Internet access and the spread of COVID-19. Speed or Access Table 3 presents the results for different Internet speeds, including 2G, 3G, and 4G. In the context of the pandemic, these differences in capabilities and data speeds may have 5 Columns 7 and 8 differ as for some, we observe province and population. For the remainder, we impute population in column 8. 20 significant implications for the effectiveness of various non-pharmaceutical interventions, such as remote work, online learning, and the dissemination of public health information. While 2G primarily supports basic services such as SMS and voice calls, 3G enables mobile Internet access, and 4G offers improved data speeds and enhanced mobile Internet experiences. Access to higher-speed Internet, such as 3G and 4G, allows individuals and communities to better adapt to the challenges posed by the pandemic and adhere to public health guidelines while maintaining social and economic activities. Our preferred measure, 3G, exhibits the most substantial relationship with the pandemic experience. When we include all three types of Internet access simultaneously, 2G loses its significance, dropping close to zero, and appears less relevant regarding its impact on COVID-19 transmission. Conversely, 4G demonstrates a similar effect to our main measure, 3G, although its association with the pandemic experience is again weaker. Another aspect worth considering is the potential non-linear effect of Internet access on COVID-19 transmission. This may be important because the relationship between Internet access and COVID-19 outcomes could be characterized by diminishing returns or threshold effects, where the impact of increased Internet access on transmission rates plateaus or even reverses beyond a certain point. For instance, very high levels of Internet access could exacerbate the spread of misinformation or contribute to complacency in following public health guidelines, ultimately leading to adverse effects on COVID-19 transmission. However, our analysis does not find clear evidence of non-linear effects (Table A.1). Including a second-order polynomial term does not reveal a non-linear relationship. This may be due to data limitations, given that we only have 454 observations and considerable covariance among the variables. When we use an 27 our findings underscore the potential role of digital infrastructure in public health responses to pandemics and other health crises. In conclusion, our study underscores the vital role of technology in tackling the COVID19 pandemic and advancing public health in the digital age. The importance of digital connectivity in today’s world cannot be overstated, and our research calls for a renewed focus on this essential facet of modern life in future public health strategies. 28 TABLES AND FIGURES Table 1: Associations between G3 Internet Exposure and COVID-19 Cases: Late Pandemic Stage Dependent variables: Cumulative cases by 10T population, 18 Feb 2023 Adding covariates Imputation Raw +Prov. FE +Demographics +Health facilities +Econ. status +Ethnic comp. +Labor force via: Set 0 Multiple (1) (2) (3) (4) (5) (6) (7) (8) (9) Internet exposure -1.495 -1.125 -0.651 - 0.606 -0.611 - 0.690 - 0.787 - 0.630 - 0.630 (0.222) (0.417) (0.323) (0.315) (0.285) (0.290) (0.313) (0.255) (0.255) Semi-elasticity (1sd) - 0.32 - 0.28 - 0.20 -0.19 -0.19 - 0.21 - 0.23 - 0.20 N 454 454 454 345 323 321 321 454 454 Mean dep. 0.50 0.50 0.50 0.40 0.42 0.42 0.42 0.50 0.50 SD dep. 1.45 1.45 1.45 0.87 0.89 0.89 0.89 1.45 1.45 pR 2 0.05 0.41 0.51 0.37 0.38 0.38 0.38 0.53 Province FE ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ Demographics ✓ ✓ ✓ ✓ ✓ ✓ ✓ Health facilities ✓ ✓ ✓ ✓ ✓ ✓ Economic status ✓ ✓ ✓ ✓ ✓ Ethnic composition ✓ ✓ ✓ ✓ Labor force ✓ ✓ ✓ COVID-19 = coronavirus disease, FE = Fixed effects, GSMA = Mobile Communications Association, SD = standard deviation. Notes: This table presents coefficients estimates from equation (1), of regressions of cumulative COVID-19 cases per 10,000 population on the regency level (18 February 2023) and increasing various sets of covariates all measure before the pandemic begins, for a description of the covariates (Table B.1). At the bottom, we present pR2 pseudo (ml) R2, the mean and standard deviation of the outcome variable, and the semi-elasticity for 1 sd change in internet exposure, that is (exp(τ ) − 1) ∗ sd(Internet exposure). Column (1) depicts the raw association, Column (2) adds province (38) fixed effects, three demographic variables (population density, population, male-female ratio, the share of the population aged 65 and older, the share of people without any education, the share of people with graduate degrees, log household size, rural area), (4) adds health infrastructure measures (number of clinics, number of hospitals, and log of health care spending), (5) adds economic indicators (second order polynomial of nightlight, sub-national human development index - score), (6) ethnicity (ethnic diversity, and polarization), and (7) labor force variables (share of workers able to telework, the share of workers with a long distance (>1hour) commute, and that use public transportation, and the share working in agriculture). The final two columns replace missing values in covariates with 0 and an indicator for missing values (8), and (9) implements the multiple imputation approach by Rubin (1996). All regressions are weighted by population and use robust standard errors. Sources: COVID-19 data, adjusted for population, was sourced from the Indonesian COVID-19 Task Force, using measurements taken as of 18 February 2023. Mobile internet data was obtained from Collins Bartholomew’s GSMA Mobile Coverage Explorer database. 29 Table 2: Testing Additional Robustness Dependent variables: Cumulative cases by 10T population, 18 Feb 2023 Log-cases Neg. Alt. 3G measure Without Missing Impute Base OLS Bin. Plain Imputed weights Outcomes All (1) (2) (3) (4) (5) (6) (7) (8) Internet exposure - 0.630 - 0.365 - 0.630 - 1.005 - 0.627 - 0.627 (3G) (0.255) (0.200) (0.255) (0.226) (0.255) (0.255) Internet exposure - 1.090 (3GMCE ) (0.491) Internet exposure -1.126 (3G-MCE)-imputed (0.496) Semi-elasticity (1sd) - 0.20 -0.13 - 0.20 -0.13 -0.13 - 0.26 - 0.20 -0.19 N 454 454 454 365 454 454 465 510 Mean dep. 0.50 0.50 0.50 0.60 0.50 0.50 0.49 0.45 SD dep. 1.45 1.45 1.45 1.60 1.45 1.45 1.43 1.37 pR2 0.53 0.46 0.54 0.53 0.57 0.53 Province FEs ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ Demographics ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ Health facilities ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ Economic status ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ Ethnic composition ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ Labor force ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ COVID-19 = coronavirus disease, FE = Fixed effects, GSMA = Mobile Communications Association, OLS = ordinary least squares, SD = standard deviation. Notes: Table 1 column (8) presents the main model, see notes therein, represented in column (1). Column (2) presents an OLS regression on the logged outcome measure—the marginal effect is adjusted via exp(τ ) − 1. Column (3) alternatively presents a negative binominal model, (4) uses the alternative exposure measure and (Mobile Coverage Explore - MCE) is sourced directly from the network operators and thus incurs gaps in coverage, which we impute with 0 and add missing indicator in (5). Column (6) drops the population-weighting and (7) and (8) impute areas with missing observation in the outcome (all variables, respectively) with 0. Sources: COVID-19 data, adjusted for population, was sourced from the Indonesian COVID-19 Task Force, using measurements taken as of 18 February 2023. Mobile internet data was obtained from Collins Bartholomew’s GSMA Mobile Coverage Explorer database. 30 Table 3: Association between Different Internet Speeds and Potential Non-Linear Effects Dependent variables: Cumulative cases by 10T population 18 Feb 2023 Base 2G 4G Jointly (1) (2) (3) (4) Internet exposure - 0.630 - 0.530 (3G) (0.255) (0.319) Internet exposure - 0.366 0.104 (2G) (0.294) (0.296) Internet exposure - 0.472 - 0.372 (4G) (0.202) (0.200) Semi-elasticity (1sd) - 0.20 - 0.09 -0.13 [1em] N 454 454 436 436 Mean dep. 0.50 0.50 0.52 0.52 SD dep. 1.45 1.45 1.47 1.47 pR2 0.53 0.53 0.53 0.54 Province FEs ✓ ✓ ✓ ✓ Demographics ✓ ✓ ✓ ✓ Health facilities ✓ ✓ ✓ ✓ Economic status ✓ ✓ ✓ ✓ Ethnic composition ✓ ✓ ✓ ✓ Labor force ✓ ✓ ✓ ✓ COVID-19 = coronavirus disease, FE = Fixed effects, GSMA = Mobile Communications Association, SD = standard deviation. Notes: The Table presents coefficients analogous to Table 1-Column (8)—represented in Column (1). Column (2) replaces our main measure with 2G exposure, and (3) with 4G, Column (4) estimates the model jointly. Sources: COVID-19 data, adjusted for population, was sourced from the Indonesian COVID19 Task Force, using measurements taken as of February 18th, 2023. Mobile internet data was obtained from Collins Bartholomew’s GSMA Mobile Coverage Explorer database. 31 Table 4: IV Results Dependent variables: Cumulative cases by 10T population, 18 Feb 2023 IV Reduced 5 km rad. Main form dec 2019 (1) (2) (3) Panel A. Cases Internet exposure - 0.660 - 1.440 (3G) (0.257) (0.394) Lightning strike frequency (5 km) 7.919 (2.429) Semi-elasticity (1sd) - 0.20 - 0.32 Panel B. First stage Lightning (5 km, Dec 2019) - 4.050 (0.518) N 454 454 454 Difference per million for 1sd increase -1.66 - 2.50 Fstat 61.05 Province FEs Demographics Health facilities Economic status Ethnic composition Labor force Weather controls ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ COVID-19 = coronavirus disease, FE = Fixed effects, GSMA = Mobile Communications Association, km = kilometer, SD = standard deviation. Notes: The Table presents coefficients estimates from equation (1) using IV-Poisson with control function (2). Column (1) adjusts Table 1 Column (8) by adding weather covariates. Column (2) shows the reduced form of lightning strikes in the 5 km radius in December 2019, contemporaneously with the 3G mobile exposure. The bottom of the Table predicts the difference in cases for one std. Deviation in internet exposure and the F-statistic of the first stage. Sources: COVID-19 data, adjusted for population, was sourced from the Indonesian COVID19 Task Force, using measurements taken as of 18 February 2023. Mobile internet data was obtained from Collins Bartholomew’s GSMA Mobile Coverage Explorer database. 32 Figure 1: Mobile Broadband Speed and COVID-19 Cases, Asia and the Pacific COVID-19 = coronavirus disease, Lao PDR = Lao People’s Democratic Republic, PRC = People’s Republic of China. Notes: COVID-19 data were measured as of 1 January 2023, and adjusted for population. Mobile broadband speed (in Mbps) was measured in 2020. Sources: COVID-19 data from the Indonesian COVID-19 Task Force. Mobile broadband speed at countrylevel from https://worldpopulationreview.com/country-rankings/internet-speeds-by-country 33 Figure 2: Mobile Broadband Access and COVID-19 cases, Indonesia COVID-19 = coronavirus disease, PRC = People’s Republic of China. Notes: COVID-19 data were measured as of February 18th, 2023, and adjusted for population. Mobile broadband speed (3G access) was measured in 2019. Sources: COVID-19 data from the Indonesian COVID-19 Task Force. Mobile broadband speed from https://worldpopulationreview.com/. 34 Figure 3: Number of COVID-19 Cases and Mobile Coverage, Indonesia COVID-19 = coronavirus disease, GSMA = Mobile Communications Association. Notes: This figure shows cumulative COVID-19 cases on 18 February 2023 (top) and 3G mobile network exposure (bottom) across regencies in Indonesia. The boundaries, colors, denominations, and any other information shown on this map do not imply, on the part of the Asian Development Bank, any judgment on the legal status of any territory, or any other endorsement or acceptance of such boundaries, colors, denominations, or information. Sources: COVID-19 data from the Indonesian COVID-19 Task Force. Mobile internet data from Collins Bartholomew’s GSMA Mobile Coverage Explorer database. 35 Figure 4: Association between Number of COVID-19 Cases and Deaths, and Mobile Coverage, Indonesia throughout Pandemic COVID-19 = coronavirus disease, GSMA = Mobile Communications Association. Notes: The figure presents regression coefficients of regressions (90-dark and 95-light level confidence intervals) presented in Table 1 (see notes therein) column 8, separately for various dates throughout the pandemic covering various waves of new COVID-19 variants. Round markers depict cases and squares deaths. Sources: COVID-19 data, adjusted for population, was sourced from the Indonesian COVID-19 Task Force. Mobile internet data was obtained from Collins Bartholomew’s GSMA Mobile Coverage Explorer database. 36 Figure 5: Placebo Test—Local Health Care Spending by 10,000 Population GSMA = Mobile Communications Association. Notes: This figure presents analogous regressions to Table 1 Column (8) using as outcome regional health care spending across years before and during the pandemic. Spending is accordingly dropped from the set of controls. Sources: Health care spending data, adjusted for population over the years. Mobile internet data was obtained from Collins Bartholomew’s GSMA Mobile Coverage Explorer database. 43 Figure A.7: Mobile Broadband Access and COVID-19 Deaths, Indonesia COVID-19 = coronavirus disease. Notes: COVID-19 data were measured as of 18 February 2023, and adjusted for population. Mobile broadband speed (3G access) was measured in 2019. Sources: COVID-19 data from the Indonesian COVID-19 Task Force. Mobile internet data from Collins Bartholomew’s GSMA Mobile Coverage Explorer database. 44 Figure A.8: Number of COVID-19 Deaths, Indonesia COVID-19 = coronavirus disease. Notes: COVID-19 data come from the Indonesian COVID-19 Task Force, using measurements taken as of 18 February 2023. The boundaries, colors, denominations, and any other information shown on this map do not imply, on the part of the Asian Development Bank, any judgment on the legal status of any territory, or any other endorsement or acceptance of such boundaries, colors, denominations, or information. Sources: COVID-19 data from the Indonesian COVID-19 Task Force. Mobile internet data from Collins Bartholomew’s GSMA Mobile Coverage Explorer database. 45 Figure B.1: Example of Worldwide Lightning Strikes Note: The boundaries, colors, denominations, and any other information shown on this map do not imply, on the part of the Asian Development Bank, any judgment on the legal status of any territory, or any other endorsement or acceptance of such boundaries, colors, denominations, or information. Source: World Wide Lightning Location Network. https://wwlln.net/. 46 Table A.1: Association between Different Internet Speeds and Potential Non-linear Effects Dependent variables: Cumulative cases by 10T population, 18 Feb 2023 Replicated from main text Base 2G 4G Jointly Polynomial Indicator (1) (2) (3) (4) (5) (6) Internet exposure (3G) -0.630 -0.530 -1.138 (0.255) (0.319) (0.712) Internet exposure (2G) -0.366 0.104 (0.294) (0.296) Internet exposure (4G) -0.472 -0.372 (0.202) (0.200) Internet exposure2 (3G) 0.508 (0.580) 1[Internet exposure > p75] (3G) -0.252 (0.159) N 454 454 436 436 454 454 pR2 0.53 0.53 0.53 0.54 0.53 0.53 Mean dep. 0.50 0.50 0.52 0.52 0.50 0.50 SD dep. 1.45 1.45 1.47 1.47 1.45 1.45 Marginal effect -0.24 -0.14 -0.18 -0.43 Province FEs ✓ ✓ ✓ ✓ ✓ ✓ Demographics ✓ ✓ ✓ ✓ ✓ ✓ Health facilities ✓ ✓ ✓ ✓ ✓ ✓ Economic status ✓ ✓ ✓ ✓ ✓ ✓ Ethnic composition ✓ ✓ ✓ ✓ ✓ ✓ Labor force ✓ ✓ ✓ ✓ ✓ ✓ COVID-19 = coronavirus disease, FE = Fixed effects. Notes: See Table 3’s notes. Sources: COVID-19 data, adjusted for population, was sourced from the Indonesian COVID-19 Task Force, using measurements taken as of 18 February 2023. Mobile internet data was obtained from Collins Bartholomew’s GSMA Mobile Coverage Explorer database. 47 Table A.2: Role of Internet over Different Pandemic Stages Dependent variables: Cumulative cases by 10T population, different pandemic stages 23 Aug 2020 23 Feb 2021 23 Aug 2021 23 Feb 2022 23 Aug 2022 18 Feb 2023 (1) (2) (3) (4) (5) (6) Panel A. Cases Internet exposure -1.160 -1.093 - 0.705 - 0.536 - 0.493 - 0.630 (0.303) (0.274) (0.271) (0.247) (0.245) (0.255) N 442 453 453 401 402 454 pR2 0.07 0.42 0.77 0.91 0.94 0.94 Mean dep. 0.01 0.10 0.32 0.42 0.48 0.50 SD dep. 0.04 0.29 0.80 1.19 1.41 1.45 Panel B. Deaths Internet exposure -2.133 -1.387 -1.216 - 0.851 - 0.843 - 0.987 (0.466) (0.321) (0.316) (0.300) (0.299) (0.296) N 294 397 408 375 376 417 pR2 0.01 0.02 0.04 0.04 0.04 0.05 Mean dep. 0.00 0.00 0.01 0.01 0.01 0.01 SD dep. 0.00 0.01 0.02 0.02 0.02 0.03 Province FEs ✓ ✓ ✓ ✓ ✓ ✓ Demographics ✓ ✓ ✓ ✓ ✓ ✓ Health facilities ✓ ✓ ✓ ✓ ✓ ✓ Economic status ✓ ✓ ✓ ✓ ✓ ✓ Ethnic composition ✓ ✓ ✓ ✓ ✓ ✓ Labor force ✓ ✓ ✓ ✓ ✓ ✓ COVID-19 = coronavirus disease, FE = Fixed effects, GSMA = Mobile Communications Association, km = kilometer, SD = standard deviation. Notes: See Table 3’s notes. Sources: COVID-19 data, adjusted for population, was sourced from the Indonesian COVID-19 Task Force. Mobile internet data was obtained from Collins Bartholomew’s GSMA Mobile Coverage Explorer database. 48 Table A.3: Heterogeneity: Poisson Dependent variables: Cumulative cases by 10T population, 18 Feb 2023 (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) Internet exposure -0.518 -0.710 -0.509 -0.842 -0.750 -0.420 -0.203 -0.743 -0.565 -0.754 -0.675 -0.552 -0.446 -0.761 (0.284) (0.301) (0.251) (0.268) (0.248) (0.308) (0.270) (0.289) (0.256) (0.260) (0.267) (0.286) (0.297) (0.271) × Share of population with no education -0.256 (0.198) × Share of population with grad. degree 0.243 (0.207) × Log average household size -0.264 (0.197) × Share of population living in rural areas 0.352 (0.180) × Number of health sites 0.343 (0.161) × Log health care spending -0.367 (0.222) × Average nightlight -0.499 (0.261) × Sub-national HDI 0.104 (0.190) × Ethnic diversity -0.113 (0.184) × Ethnic polarization 0.086 (0.195) × Share of population commuting over an hour 0.047 (0.161) × Share of population commuting public transport -0.143 (0.199) × Share of population telework potential -0.300 (0.183) × Share of population working agriculture 0.203 (0.195) Province FEs ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ Demographics ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ Health facilities ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ Economic status ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ Ethnic composition ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ Labor force ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ Weather controls ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ FE = Fixed effects, HDI = Human Development Index. Sources: COVID-19 data, adjusted for population, was sourced from the Indonesian COVID-19 Task Force, using measurements taken as of 18 February 2023. Mobile internet data was obtained from Collins Bartholomew’s GSMA Mobile Coverage Explorer database. 49 Table A.4: Heterogeneity: Poisson IV Dependent variables: Cumulative cases by 10T population, 18 Feb 2023 (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) Internet exposure -2.371 -1.190 -0.627 -1.631 -1.472 -1.648 -1.460 -1.406 -0.924 -1.728 -2.133 -1.921 -0.876 -1.723 (0.596) (0.468) (0.715) (0.412) (0.398) (0.439) (0.376) (0.695) (0.517) (0.485) (0.625) (0.585) (0.516) (0.486) × Share of population with no education 1.006 (0.441) × Share of population with grad. degree -0.308 (0.207) × Log average household size -1.052 (0.882) × Share of population living in rural areas 0.326 (0.280) × Number of health sites -0.121 (0.205) × Log health care spending 0.242 (0.189) × Average nightlight 0.238 (0.513) × Sub-national HDI -0.034 (0.834) × Ethnic diversity -0.879 (0.617) × Ethnic polarization 0.464 (0.315) × Share of population commuting over an hour 0.565 (0.352) × Share of population commuting public transport 0.561 (0.351) × Share of population telework potential -0.740 (0.408) × Share of population working agriculture 0.520 (0.282) Province FEs ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ Demographics ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ Health facilities ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ Economic status ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ Ethnic composition ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ Labor force ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ Weather controls ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ FE = Fixed effects, HDI = Human Development Index. Sources: COVID-19 data, adjusted for population, was sourced from the Indonesian COVID-19 Task Force, using measurements taken as of 18 February 2023. Mobile internet data was obtained from Collins Bartholomew’s GSMA Mobile Coverage Explorer database. 50 Table B.1: Variable Overview Variable N Mean SD Description Source Cumulative cases 18 Feb 2023 454 10942.60 31381.30 Total number of cases at the regency level as of 18 Feb 2023 Indonesian COVID-19 Task Force Cumulative deaths 18 Feb 2023 417 305.66 504.09 Total number of deaths at the regency level as of 18 Feb 2023 Indonesian COVID-19 Task Force Mobile 3g OCI 500 0.37 0.45 Share of areas at the regency level having access to 3G as of December 2019 Collins Bartholomew’s Mobile Coverage Explorer (OpenCellID) Mobile 3g MCE 499 62.93 44.91 Share of areas at the regency level having access to 3G as of December 2019 Collins Bartholomew’s Mobile Coverage Explorer Mobile 4g OCI 482 0.76 0.36 (Mobile Coverage Explorer) Share of areas at the regency level having access to 4G as of December 2019 Collins Bartholomew’s Mobile Coverage Explorer (OpenCellID) Mobile 2g OCI 500 0.52 0.34 Share of areas at the regency level having access to 2G as of December 2019 Collins Bartholomew’s Mobile Coverage Explorer (OpenCellID) Population density 465 0.89 2.43 Population density at the regency level Indonesian COVID-19 Task Force Population 465 2.85 2.14 Total population at the regency level Indonesian COVID-19 Task Force Population over 65 510 0.08 0.03 Share of population with a university degree or higher at the regency level 2010 Indonesian population census Population with no education 510 0.19 0.11 Share of population with a university degree or higher at the regency level 2010 Indonesian population census Population with high education 510 0.09 0.05 Share of population with a university degree or higher at the regency level 2010 Indonesian population census Household size 510 4.16 0.46 Average household size at the regency level 2010 Indonesian population census Rural areas 510 0.60 0.31 Share of population living in rural areas 2010 Indonesian population census Number of clinics 363 14.00 42.60 Number of clinics in 2019 at the regency level Open Street Map (OSM) Number of hospitals 363 8.00 14.03 Number of hospitals in 2019 at the regency level Open Street Map (OSM) Health spending 484 27.01 18.76 Amount of (realized) health spending in 2019 at the regency level Open Street Map (OSM) Continued on the next page 51 Variable N Mean SD Description Source Nighttime light 500 2.14 5.07 Average nightlight (Visible Infrared Imaging Radiometer Suite - VIIRS) as National Oceanic and Atmospheric Administration (NOAA) of December 2019 Human development index 465 2.97 1.42 Human development index at the regency level in 2019 (quintiles) Badan Pusat Statistik Ethnic fractionalization 510 0.46 0.31 Probability that two randomly selected people in a regency belong to different ethnic groups 2010 Indonesian population census Ethnic polarization 510 0.45 0.25 Probability that a group of individuals in a regency is divided into different 2010 Indonesian population census ethnic groups Long-distance work 510 0.05 0.03 Share of population traveling ¿=1hour to work at the regency level National Labor Force Survey 2019 Public transport work 510 0.06 0.05 Share of population using public transport to work at the regency level National Labor Force Survey 2019 Teleworkability 510 0.22 0.05 Teleworkability by industry at the regency level National Labor Force Survey 2019 Agriculture share 510 0.38 0.22 Share of population working in agriculture National Labor Force Survey 2019 Temperature 500 25.80 1.92 Average monthly temperature as of December 2019 (Celsius degree) ERA5 reanalysis data Rainfall 500 0.21 0.08 Average monthly precipitation as of December 2019 (millimeters) ERA5 reanalysis data Lightning strike 500 0.01 0.01 Average lightning stroke power as of December 2019 (5-minute resolution) World Wide Lightning Location Network COVID-19 = coronavirus disease, MCE = Mobile Coverage Explorer, OCI = OpenCellID. Source: Authors’ compilation. 52 REFERENCES Ahmed, Faheem, Na’eem Ahmed, Christopher Pissarides, and Joseph Stiglitz. 2020. “Why Inequality Could Spread COVID-19.” The Lancet Public Health, 5 (5): e240. Aker, Jenny C., and Isaac M. Mbiti. 2010. “Mobile Phones and Economic Development in Africa.” Journal of Economic Perspectives, 24 (3): 207–32. Aksoy, Cevat Giray, Jose Maria Barrero, Nicholas Bloom, Steven J Davis, Mathias Dolls, and Pablo Zarate. 2022. “Working from Home Around the World.” NBER Working Paper No. 30446. National Bureau of Economic Research (NBER). Alesina, Alberto, and Eliana La Ferrara. 2005. “Ethnic Diversity and Economic Performance.” Journal of Economic Literature, 43 (3): 762–800. Alipour, Jean-Victor, Harald Fadinger, and Jan Schymik. 2021. “My Home is My Castle– The Benefits of Working from Home During a Pandemic Crisis.” Journal of Public Economics, 196: 104373. Amaral-Garcia, Sofia, Mattia Nardotto, Carol Propper, and Tommaso Valletti. 2022. “Mums Go Online: Is the Internet Changing the Demand for Health Care?” Review of Economics and Statistics, 104 (6): 1157–73. Andersen, Thomas Barnebeck, Jeanet Bentzen, Carl-Johan Dalgaard, and Pablo Selaya. 2012. “Lightning, IT Diffusion, and Economic Growth across US States.” Review of Economics and Statistics, 94 (4): 903–24. Asian Development Bank (ADB). 2021. “COVID-19 and Labor Markets in Southeast Asia: Impacts on Indonesia, Malaysia, the Philippines, Thailand, and Viet Nam.” Manila. Auffhammer, Maximilian, Solomon M. Hsiang, Wolfram Schlenker, and Adam Sobel. 2013. “Using Weather Data and Climate Model Output in Economic Analyses of Climate Change.” Review of Environmental Economics and Policy, 7 (2). Bargain, Olivier, and Ulugbek Aminjonov. 2020. “Trust and Compliance to Public Health Policies in Times of COVID-19.” Journal of Public Economics 192: 104316. Barrero, Jose Maria, Nicholas Bloom, and Steven J. Davis. 2021. “Internet Access and Its Implications for Productivity, Inequality, and Resilience.” NBER working Paper No. 29102. NBER. Barrero, Jose Maria, Nicholas Bloom, Steven J. Davis, and Brent H. Meyer. 2021. “COVID-19 is a Persistent Reallocation Shock.” AEA Papers and Proceedings, 111: 287–91.