Cellphone visit counts for 7 million places lifted a US community health model by 7.5%

The 7.5% is an average across health measures, and some measures gained far more than others. The gains cluster in five outcomes. "In particular, the model showed the biggest predictive gains for place visits related to binge drinking, depression, routine medical checkups, obesity and asthma," said Temitope Akinboyewa, the doctoral student in geography who is first author on the paper ^[10][15]. In urban tracts the binge drinking improvement was 38.8% ^[8]. In rural tracts the depression improvement was 48.9% ^[9].

Those are relative improvements over a baseline built from US Census and state health department variables: race, age, socioeconomic status, access to health care, education levels ^[4]. The added layer was aggregated 2019 cellphone visit counts for roughly 7 million public points of interest, among them parks, restaurants, gyms, casinos, convenience stores, primary care facilities and religious centers ^[5].

What the design tests is prediction. The reported associations run at the tract level: in both urban and rural tracts, frequent visits to places where alcohol is sold were associated with a higher prevalence of binge drinking and of depression ^[16]. That is not evidence that the people making those visits are the people reporting the condition, and it fixes no direction of causation. The paper does not give the accuracy metric behind the 7.5%, or the rule for whether a visit counted toward the visitor's home tract or the destination tract ^[21].

The two strata are lopsided. About 12,800 rural and 56,600 urban tracts make 69,400, so rural tracts are 18.4% of the analysis ^[6][7]. The largest single gain reported, depression in rural areas, comes out of that smaller fifth.

Some category signs hold across both strata. Limited-service restaurants were associated with lower binge drinking prevalence in urban and in rural tracts, and standalone casinos with higher prevalence in both ^[17][18]. For depression in urban tracts, casinos, jewelry stores and general merchandise stores were associated with higher prevalence, and caterers, art dealers and child day care services with lower ^[19]. Socioeconomic status was already in the baseline ^[4], so the visit variables are contributing something past income.

Zhenlong Li, the corresponding author and director of Penn State's Geoinformation and Big Data Research Lab, framed the result as an addition. "Our results show that these place visitation patterns can provide useful additional information for estimating community health measures, complementing traditional demographic and social variables," he said ^[13][14]. The visits were counted in 2019 ^[5]. A health department applying those coefficients now is assuming the pattern of where people go has held for the years since.