Significance This paper compares the probabilistic accuracy of short-term forecasts of reported deaths due to COVID-19 during the first year and a half of the pandemic in the United States. Results show high variation in accuracy between and within stand-alone models and more consistent accuracy from an ensemble model that combined forecasts from all eligible models. This demonstrates that an ensemble model provided a reliable and comparatively accurate means of forecasting deaths during the COVID-19 pandemic that exceeded the performance of all of the models that contributed to it. This work strengthens the evidence base for synthesizing multiple models to support public-health action.

We propose a Bayesian model for projecting first-wave COVID-19 deaths in all 50 U.S. states. Our model's projections are based on data derived from mobile-phone GPS traces, which allows us to estimate how socialdistancing behavior is "flattening the curve" in each state. In a two-week look-ahead test of out-of-sample forecasting accuracy, our model significantly outperforms the widely used model from the Institute for Health Metrics and Evaluation (IHME), achieving 42% lower prediction error: 13.2 deaths per day average error across all U.S. states, versus 22.8 deaths per day average error for the IHME model. Our model also provides an accurate, if slightly conservative, assessment of forecasting accuracy: in the same look-ahead test, 98% of data points fell within the model's 95% credible intervals. Our model's projections are updated daily at https://covid-19. tacc.utexas.edu/projections/. : medRxiv preprint Model overviewAt a high level, our model shares some key properties of the IHME model.Similarity 1: a statistical curve-fitting approach. Ours is not an epidemiological model, in the sense that we do not try to model disease transmission, 2 All rights reserved. No reuse allowed without permission.

Forecasting the burden of COVID-19 has been impeded by limitations in data, with case reporting biased by testing practices, death counts lagging far behind infections, and hospital census reflecting time-varying patient access, admission criteria, and demographics. Here, we show that hospital admissions coupled with mobility data can reliably predict severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) transmission rates and healthcare demand. Using a forecasting model that has guided mitigation policies in Austin, TX, we estimate that the local reproduction number had an initial 7-d average of 5.8 (95% credible interval [CrI]: 3.6 to 7.9) and reached a low of 0.65 (95% CrI: 0.52 to 0.77) after the summer 2020 surge. Estimated case detection rates ranged from 17.2% (95% CrI: 11.8 to 22.1%) at the outset to a high of 70% (95% CrI: 64 to 80%) in January 2021, and infection prevalence remained above 0.1% between April 2020 and March 1, 2021, peaking at 0.8% (0.7-0.9%) in early January 2021. As precautionary behaviors increased safety in public spaces, the relationship between mobility and transmission weakened. We estimate that mobility-associated transmission was 62% (95% CrI: 52 to 68%) lower in February 2021 compared to March 2020. In a retrospective comparison, the 95% CrIs of our 1, 2, and 3 wk ahead forecasts contained 93.6%, 89.9%, and 87.7% of reported data, respectively. Developed by a task force including scientists, public health officials, policy makers, and hospital executives, this model can reliably project COVID-19 healthcare needs in US cities.

This paper considers the problem of spatial and temporal mispricing of ride-sourcing trips from a driver perspective. Using empirical data from more than 1.1 million rides in Austin, Texas, we explore the spatial structure of ride-sourcing search frictions and driver performance variables as a function of the trip destination. The spatial information is subject to noise and sparsity, and researchers tend to aggregate the data in large areas, which results in the loss of high-resolution insights. We implemented the graph-fused lasso (GFL), a spatial smoothing or denoising methodology that allows for high-definition spatial evaluation. GFL removes noise in discrete areas by emphasizing edges, which is practical for evaluating zones with heterogeneous types of trips, such as airports, without blurring the information to surrounding areas. Principal findings suggest that there are differences in driver productivity depending on trip type and pickup and drop-off location. Therefore, providing spatio-temporal pricing strategies could be one way to balance driver equity across the network.

SARS-CoV-2 transmission continues to evolve in the United States following the large second wave in the Summer. Understanding how location-specific variations in non-pharmaceutical epidemic control policies and behaviors contributed to disease transmission will be key for designing effective strategies to avoid future resurgences. We offer a statistical analysis of the relative effectiveness of the timing of both official stay-at-home orders and population mobility reductions, offering a distinct (but complementary) dimension of evidence gleaned from more traditional mechanistic models of epidemic dynamics. Specifically, we use a Bayesian hierarchical model fit to county-level mortality data from the first wave of the pandemic from Jan 21 2020 through May 10 2020 to establish how timing of stay-at-home orders and population mobility changes impacted county-specific epidemic growth. We find that population mobility reductions generally preceded stay-at-home orders, and among 356 counties with a pronounced early local epidemic between January 21 and May 10 (representing 195 million people and 32,000 observed deaths), a 10 day delay in population mobility reduction would have added 16,149 (95% credible interval [CI] 9,517 24,381) deaths by Apr 20, whereas shifting mobility reductions 10 days earlier would have saved 13,571 (95% CI 8,449 16,930) lives. Analogous estimates attributable to the timing of explicit stay-at-home policies were less pronounced, suggesting that mobility changes were the clearer drivers of epidemic dynamics. Our results also suggest that the timing of mobility reductions and policies most impacted epidemic dynamics in larger, urban counties compared with smaller, rural ones. Overall, our results suggest that community behavioral changes had greater impact on curve flattening during the Spring wave compared with stay at home orders. Thus, community engagement and buy-in with precautionary policies may be more important for predicting transmission risk than explicit policies.