Potential dissemination of epidemics based on Brazilian mobile geolocation data. Part I: Population dynamics and future spreading of infection in the states of Sao Paulo and Rio de Janeiro during the pandemic of COVID-19.

Pedro S. Peixoto; Diego R. Marcondes; Cláudia M Peixoto; Lucas Queiroz; Rafael Gouveia; Afonso Delgado; Sérgio M Oliva

As we lack information about consecutive uses which occurred in a same location, the proportion of no-movement of a region is underestimated, as we consider it as only movement within the city, disregarding devices which have not moved at all within a day, as we do not have this information. This causes the proportion of movement from A to other regions to be overestimated, as we observe proportions as high as 45% of the movements from a region being to out of it, what is an unrealistic estimation of the proportion of people which move to outside of a region. A more realistic number is no more than 5%, what we believe would be obtained if we had the number of recordings in which the uses occurred in a same location. This overestimation will be corrected in the models simulating the disease spread, but when analysing raw data we disregard any correction, as we are only interested in determining common movements, and are not interested in how common they are . Even though this proportion is not a consistent estimator, in a statistical sense, of the proportion of a population which travels from a region to another within 24 hours, as a same device may be recorded twice in the period of a day, it is a good proxy for the mobility between two regions, as represents in reality a person which traveled from one region to another within 24 hours, or stayed in a same region. As the data is anonymized and each device is followed for only two uses of the app, we do not actually know if the movement is that of a person which is returning to a location or going there the first time, for example. However, this proportion gives a good idea of possible patterns followed by a population in general, as if a pattern is recurrent in the population it may also be in our dataset, although the proportion of movements in the population may be distinct of the one we calculated, i.e., we may be able to identify common patterns of mobility, even though we cannot estimate properly, in a statistical sense, the proportion of the population which leave one region and go to another in the period of a day.