Practical geospatial and sociodemographic predictors of human mobility

Abstract Understanding seasonal human mobility at subnational scales has important implications across sciences, from urban planning efforts to disease modelling and control. Assessing how, when, where populations move over the course of year, however, requires spatially temporally resolved datasets spanning large periods time, which can be rare, contain sensitive information, or may proprietary. Here, we aim explore how a set broadly available covariates describe typical in Kenya pre-COVID-19, therefore enabling better low- middle-income country (LMIC) settings non-pandemic settings. To do this, used Google Aggregated Mobility Research Dataset, containing anonymized flows aggregated users who have turned on Location History setting, is off by default. We combined this with socioeconomic geospatial 2018 2019 quantify changes domestic international patterns years. undertook spatiotemporal analysis within Bayesian framework identify relevant explaining movement patterns, while accounting for spatial temporal autocorrelations. Typical pre-pandemic mostly consisted shorter, within-county trips, followed longer travel between counties travel, establishing changed post-pandemic. peaked August December, closely corresponding school holiday seasons, was found an predictor our model. further that variables including urbanicity, poverty, female education strongly explained addition such as accessibility major population centres temperature. These findings derived novel data sources elucidate broad beyond Kenya, easily generalized other LMIC before COVID-19 pandemic. provides crucial baseline interpret both these result pandemic, well whether been permanently altered once pandemic subsides. Our outline key correlates using covariates, alleviating bottlenecks highly proprietary mobile phone datasets, many researchers not access to. results provide insight monitoring proxies context surveillance control through