We assemble a daily database of economic indicators and COVID-19 incidences covering 314 regions/states in 17 countries between December 21, 2020–February 28, 2021 (Table A8-9). Our main independent variable of interest is the number of vaccine recipients per capita (of at least one dose). The data are collected directly from national governmental websites.
We examine the effect of vaccines on four regional indicators. First, to test the impact of vaccination on coronavirus spread, we use the number of new COVID-19 cases per capita for a given region, extracted from different sources1,2,3. To construct the measures of economic activity, we use daily night-time lights (NTL) data from the National Aeronautics and Space Administration4 (NASA). The satellite images are taken at 01:30AM local time with a spatial resolution of 500m x 500m. For each region, we divide the sum of all daily night-time light pixels by population. The second measure of economic activity is based on Aerosol Optical Depth (AOD) at 550nm, a proxy for emission levels. The data are based on satellite images provided by the Copernicus satellite program of the European Union5, which comes with spatial and temporal resolution of 125 x 125 km and six-hour intervals, respectively. We aggregate these intra-daily records to a daily aerosol emission indicator and divide by population for a given region. Finally, using the Google mobility indicators6, we construct a general mobility index which is defined as the average of workplace, transport, and the inverse of residential mobility.
Additional data is sourced for the instrumental variable approach: vaccine procurement data7, which records the share of vaccinations ordered on a given date for different vaccination suppliers. For each country, we sum the share of procurement over population over all vaccination suppliers at a specific date. Vaccine procurement is interacted with hospitals per capita from OpenStreetMap Overpass API8.
In the baseline specification, we regress health (new COVID-19 cases) or economic (changes in nighttime lights, emissions, mobility) outcomes per capita on the vaccine recipients of at least one dose per capita by estimating
We extend equation (1) to examine the additional effects stemming from the second vaccine dose:
INSTRUMENTAL VARIABLE (IV) ESTIMATION
There are two main challenges in identifying the causal effects of vaccines on health and economic indicators. The first is reverse causality. For example, regions with a higher number of new COVID-19 cases or weaker economic activity may act faster to roll out vaccines. The second is joint determinacy: vaccine rollouts could be deployed together with NPIs, which in turn affect the evolution of the pandemic and economic activity. Our baseline specification tries to mitigate these concerns through the use of lagged explanatory variables (for reverse causality), the inclusion of the lagged outcome variables as controls, and the extensive set of regional fixed effects (to control for cultural factors that can facilitate (impede) vaccine rollouts) and country-time fixed effects (to control for time-varying factors at the country level affecting the evolution of the pandemic and economic activity—such as the imposition of lockdown measures, mask requirements and other factors affecting social distancing). But there could still be unobserved regional-time varying factors that correlate with vaccine rollout and affect new cases and economic activity.
We thus rely on an Instrumental Variable (IV) approach, which consists of interacting a country-time variable and a constant region-specific variable. The country-time variable we consider is the cumulative number of vaccine deals/procurement from June 1, 2020–August 21, 2020—that is, six months before the administration of vaccines covered in our estimation sample for all countries with four exceptions: for Argentinian, Turkey, and Chile, we consider vaccine procurement three months before; and for India, we consider vaccine procurement 5 month before. Using smaller lags for these four countries provides the strongest instruments in terms of the F-Statistics. We test for the sensitivity of this lag structure through several robustness checks (see below). In general, the identifying assumption is that these deals/procurement decisions are strongly correlated with vaccine rollouts in January–March 2021 and are not correlated with daily shocks affecting the evolution of the pandemic and economic activity—that is, current-year vaccine administration attributable to project approval decisions made last year are unlikely to be correlated with current-year daily shocks to COVID-19 cases and economic activity. The regional term we consider is the (log of the) number of hospitals per capita. The identifying assumption is that regions with a higher number of hospital per capita are able to rollout vaccines faster.
Our empirical framework controls for region and country-time fixed effects. As such, this IV approach follows the same logic of a difference-in-difference estimator12. In particular, our IV strategy reads as follows:
ROBUSTNESS CHECK FOR INSTRUMENTAL VARIABLE ESTIMATION
While most of the countries in our sample had started securing vaccine deals already in the first part of 2020, some (such as Chile and Argentina) signed most procurement deals only towards the end of 2020. To check the sensitivity of our results, we repeated the analysis considering deals signed 150 and 120 days before. While the instruments constructed using these combinations are weaker, the second stage results are qualitatively similar to those obtained with the baseline instrument (EX. Table A4-5, Figure A10-11).
Our identifying assumption is that the interaction between vaccine procurement and number of hospitals per capita are not correlated with daily shocks affecting the evolution of the pandemic and economic activity. A potential concern is that past vaccine procurement may be correlated with current NPIs—such as containment measures, mask requirements (Hale et al., 2021), testing capacity (Roser et al, 2020)—and these interventions affects current new cases and economic indicators through regionals factors correlated with hospitals per capita. To address this issue, we augment the baseline IV specification by including the interaction between these NPIs and hospitals per capita. The results presented in Ex. Table A6 and Figure A12 show that the IV results remain robust.
Another similar concern is that past vaccine procurement may be correlated with past NPIs and health and economic indicators (new cases, mobility, emissions and NTLs)—that is, that countries that signed vaccine contracts were also able to better control the pandemic, We check the robustness of our results by including in the IV specifications the interaction between the number of hospital per capita and NPIs, health and economic indicators 180 days before—that is, at the time when vaccines were procured. The results presented in Ex. Table A7 and Figure A13 show that the IV results remain robust.
Finally, we considered the number of hospitals per capita as the regional factor affecting ability to administer vaccines. We find that other regional characteristics such as (the log of) the inverse of population and (the log of) the number of doctors per capita contribute to facilitating vaccine rollout at the regional level. While the first stage results are considerably weaker, we find that for the combinations of dependent variable/horizon for which the instrument is strong—that is, the Kleibergen‒Paap rk Wald F statistic exceeds the Stock-Yogo critical value—the IV results obtained using these regional characteristics are broadly similar to those obtained with hospitals per capita.
Data and code availability statement
The datasets generated and analyzed for the empirical analysis of the present research article is available from the corresponding author on request or via Github (@MGanslmeier/covid19vaccinationProject): https://github.com/MGanslmeier/covid19vaccinationProject
- Microsoft bing COVID-19 tracker. (n.d.). Microsoft Corporation. Retrieved March 19, 2020, from https://www.bing.com/covid.
- Abbott, S., Sherratt, K., Bevan, J., Gibbs, H., Hellewell, J., Munday, J., et al.,2020. Covidregionaldata: Subnational data for the covid-19 outbreak.
- Wahltinez, O., et al. 2021. Open COVID-19 Dataset. Available at https://github.com/open-Covid-19/data(accessed April 01, 2021).
- Román, M.O., Wang, Z., Shrestha, R., Yao, T., and Kalb, V., 2019. Black marble user guide version 1.0. NASA: Washington, DC, USA.
- ECMWF. CAMS Near-real-time. Retrieved April 01, 2021, from https://apps.ecmwf.int/datasets/data/cams-nrealtime/levtype=sfc/
- Google LLC, Google COVID-19 Community Mobility Reports. Retrieved from: 'https://www.google.com/COVID-19/mobility/' [Online Resource], (2021)
- Duke Global Health Innovation Center, 2020. Launch and Scale Speedometer. Duke University. Retrieved from: https://launchandscalefaster.org/speedometer-data
- OpenStreetMAP. Retrieved from: https://wiki.openstreetmap.org/wiki/Overpass_API
- Dagan, N., Barda, N., Kepten, E., Miron, O., Perchik, S., Katz, M.A., Hernán, M.A., Lipsitch, M., Reis, B., and Balicer, R.D., 2021. “BNT162b2 mRNA Covid-19 Vaccine in a Nationwide Mass Vaccination Setting”. N Engl J Med, NEJMoa2101765.
- Polack, F.P., Thomas, S.J., Kitchin, N., Absalon, J., Gurtman, A., et al., 2020. “C4591001 Clinical Trial Group. Safety and Efficacy of the BNT162b2 mRNA Covid-19 Vaccine”. N Engl J Med, vol. 383(27), pp. 2603-2615.
- Logunov, D.Y., Dolzhikova, I.V., Shcheblyakov, D.V., Tukhvatulin, A.I., Zubkova, O.V., et al., 2021. “Gam-COVID-Vac Vaccine Trial Group. Safety and efficacy of an rAd26 and rAd5 vector-based heterologous prime-boost COVID-19 vaccine: an interim analysis of a randomised controlled phase 3 trial in Russia,” Lancet, vol. 397(10275), pp. 671-681.
- Nunn, N., and Qian, N., 2014. “US food aid and civil conflict,” American Economic Review, vol. 104(6), pp. 1630-66.
- Imbens G.W., Angrist J.D., 1994. “Identification and Estimation of Local Average Treatment Effects”. Econometrica, Vol. 62, No. 2. (Mar., 1994), pp. 467-475.