Box 3.

Methods used to collect and analyze sub-national data on NPI

The information was collected by reviewing official government websites and examining the implementation of each of the public policy variables in Mexico's 32 states and Brazil's 26 states and the Federal District of Brasilia. In addition, we scanned and collected data from official newspapers, local newspapers, and social networks such as Twitter and Facebook. The specific coding of each variable is presented in the Appendix and website http://observcovid.miami.edu/.

For each state, we assign a discrete value for each of the 10 NPIs, ranging from 0 to 1, for each day after the first officially registered case in each country, based on the strength of the policy implemented each day: not implemented (0), moderately implemented (0•5) (e.g., public transport reduced capacity by 60%), or fully implemented (1) (e.g., full school closure). For some NPI there is a more detailed coding scheme, with other possible daily values including 0•33 and 0•66 (A detailed list is described in the Appendix; See http://observcovid.miami.edu/ for additional information).

We summarized the daily data on each of the 10 NPIs separately and in two ways to visualize changes over time and across states in each country. We present a time-weighted approach using the square root of the ratio of the number of days since the policy was first implemented over the number of days since the first case was reported. Values range from 0% to 100%. This gives greater weight to policies that were implemented earlier relative to the first reported case, with relaxation of policies appearing as declines in the value of the NPI. Contrary to the time-weighted approach, the cumulative daily sum would not decrease or drop to 0 when the policy changes. Rather, the slope would decrease or become flat if the policy is abandoned or increase in slope when re-enacted.

We use the time-weighted approach to illustrate the variance over time across states and between the two countries, with the cumulative sum in the Appendix. We use the cumulative sum to evaluate the differences by state political orientation (see the Appendix for the time-weighted data).