Data aggregation blurs inferred temporal trends in bird sampling

Ellis-Soto et al.1 reported that disparity in bird-sampling density between U.S. neighbourhoods rated as risky versus safe for real estate investment (a practice known as “redlining”) increased by 35.6% between 2000 and 2020. We show that this reported trend arises from data aggregation and linear model misspecification. Using the original neighbourhood-level yearly data and mixed-effects models that account for spatial and temporal non-independence, we show that temporal disparities are strongly non-linear and exceed 200% across the study period, while absolute differences remain small (~zero for most of the time and 25 observations per km² per year at maximum). These non-linearities temporally coincide with major shifts in citizen-science participation, including smartphone adoption and COVID-19-related increases in urban greenspace use.