Research paper: "Staying at Home Is a Privilege: Evidence from Fine-Grained Mobile Phone Location Data in the United States during the COVID-19 Pandemic"

Hi all, our article titled “Staying at Home Is a Privilege: Evidence from Fine-Grained Mobile Phone Location Data in the United States during the COVID-19 Pandemic” has been published out on the Annals of the American Association of Geographers. Taking advantage of the home-dwelling time records in the U.S, our study reveals the luxury nature of stay-at-home orders with which lower-income groups cannot afford to comply. Such disparity in responses under stay-at-home orders reflects the long-standing social inequity issues in the U.S., potentially causing unequal exposure to COVID-19 that disproportionately affects vulnerable populations. In this article, we argue that we must confront systemic social inequity issues and call for a high-priority assessment of the long-term impact of COVID-19 on geographically and socially disadvantaged groups. It is never too late to act. We will continue to fight for the socially disadvantaged groups and try our best to make their voice heard. Use the link: Staying at Home Is a Privilege: Evidence from Fine-Grained Mobile Phone Location Data in the United States during the COVID-19 Pandemic: Annals of the American Association of Geographers: Vol 112, No 1

@Xiao_Huang_University_of_Arkansas Interesting study. The colorful graphs make reading the study more enticing. I’m looking at the Feature Importance figure and it appears that perhaps MSAs with higher income inequality tend to place greater value on median hhinc? Is that true? Interestingly, 2 of the 12 MSAs don’t have median hhinc as the most important feature - Chicago and Los Angeles. Any suggestion why? While reading the study, I continued to have the question - so what? What could policymakers have done that would have considered socioeconomic status more? I also think it’d be interesting to see some of the numerical output of the random forest model. Anyways, interesting study.

Thanks so much @Xiao_Huang_University_of_Arkansas for sharing this interesting paper!
I really enjoyed reading it and had a few questions. Please feel no urgency to respond!

  1. I understand that you had to exclude some CBGs due to insufficient data available. I’d guess that the CBGs you have to exclude tend to be smaller population and lower-income CBGs (I may be wrong). Is it possible that you are under-estimating some of these correlations because the data is missing from a non-random subset of lower-income or lower-population CBGs?
  2. In Fig 4 you noted that SF, Philly, and LA show bi-modal distributions and that this may be explained by “the differing strictness in mitigation measures due to their administrative-polycentric nature and heterogeneity”. Would it be interesting to compute a geospatial metric to capture the distance between a CBG and the administrative center of the MSA? Even a crude measure of physical distance or spread of an MSA might help explain some of the variance in delta-HDT responses. Did you explore anything like this, or did I misunderstand your interpretation?
  3. Figure 7 is fascinating. I previously did some analysis of median household income (med hhinc) and how it varied with pandemic mobility behavior. One thing I didn’t consider until I analyzed the data is, of course, there are unequal sample sizes across med hhinc segments (i.e., there are more CBGs with <100k hhinc than CBGs with >= 100k). In your study, I suspect that the estimates of delta-HDT for high hhinc CBGs have less certainty (due to smaller samples) than the estimates of lower hhinc segments. Could some of the strange behaviors reported in Figure 7 (non-linear jumps, trend reversals) be explained simply by increased uncertainty in the estimate as you get into higher household income levels? Another way to ask this is: what would Figure 7 look like if you included “confidence intervals” or some measure of uncertainty of the delta-HDT for each income level? Is this a relevant concern for Figure 7, or am I missing some of the nuance? Thanks so much for the interesting work!

I encountered jumps when plotting “median time outside the home” over time in areas where most (>50%) of the population sheltered in place. When most of the population shelters in place, the median time is close to 0. When people started going out again, I suspect the median time can shoot up abruptly when the fraction that shelter in place drop below 50%, which might not happen with means. But mean times were not available in earlier social distancing data.