How the Pandemic Shifted NYC Movements
Introduction
The outbreak of the COVID-19 pandemic in 2020 not only caused a public health crisis, the lockdowns and restrictions had a major impact on the economy and led to changes in people’s mobility and lifestyle. he focus of this project is to closely examine those changes in NYC residents’ mobility behaviors and explore how they are impacted by COVID across the five boroughs that make up the city. It is important to make the comparisons through the perspective of boroughs instead of the entire city itself, because populations, cultures and functions are vastly different across boroughs in NYC.
Data
Data are collected from multiple resources, including:
- Mobility data from Google COVID-19 Community Mobility Reports
- COVID-19 data published by NYC Health Department
- Census data from the US Census Bureau
Mobility Data
The mobility data used are from a subset of the Google mobility data on the five counties (New York, Kings, Bronx, Richmond and Queens), equivalent to the five boroughs of New York. The dataset records the daily mobility changes compared to a baseline for six categories:
- Retail and Recreation
- Grocery and Pharmacy
- Parks
- Transit Stations
- Workplaces
- Residential
The mobility change is defined as the percentage change in the number of visits (first five categories) or length of stay (residential). And the baseline is the median for the corresponding day of the week over a five-week period from Jan 3 to Feb 6, 2020.
COVID-19 Data
For the COVID-19 data, I collected the daily borough-specific counts of confirmed cases, hospitalizations, and death counts. I also added the vaccination data over time by borough.
Census Data
The latest population of each NYC borough is collected from the 2020 census. This was joined to the COVID data to normalize the counts.
Data Pre-processing
The pre-processing involves converting data types and handling missing values. Three major additional procedures were applied to data:
- Joined borough population data to covid data
- Generated percentages of the population that are newly confirmed, hospitalized, dead, or fully vaccinated
- Classified weekdays and weekends for Workplaces and Residential
With regard to that classification for day, preliminary analysis indicated that both categories have clear differences in the level of mobility changes between weekdays and weekends. Thus, they were split into weekdays and weekends in the following analysis.
Analysis
The following plot is a brief overview of the city-wide COVID trend.
There were major spikes at the outbreak of the pandemic in early 2021 (Alpha variant) and end of 2021 (rise of Omicron). In the following analysis, each category generally follows the same 3-stage pattern corresponding to the COVID spikes observed here.
Part 1 - NYC by category
Retail & Recreation
- Overall, all boroughs show a clear upward recovery trend but are still lower than the pre-COVID levesl.
- Among all boroughs, Manhattan was most heavily impacted, with the mobility level dropping by nearly 90% at the start of the pandemic, while all other boroughs declined by around 60-70%. Even as time went by, Manhattan’s level of decline was always double of the other boroughs.
Grocery & Pharmacy
- In terms of the trends and scales, this is the least impacted category. This is not surprising, as people visit these places to purchase necessities. Manhattan was the most impacted borough, while Queens was the least affected; its mobility level has recovered to pre-COVID level.
Parks
- Visits to parks are typically correlated with weather and temperature. Even with COVID disrupting people’s habits, the seasonality factor remains, as we observe peaks in summer and early fall time. Brooklyn and Queens especially have very high variations between levels in summer and winter.
- In addition, compared with the baseline (January 2020), the same periods in 2021 and 2022 are still lower, as it takes time to rebuild confidence following the pandemic.
Transit Stations
- Transit stations were heavily impacted during the pandemic, with Manhattan suffering the largest decline from the baseline. The difference among boroughs is not as large as in some categories, and it remains to be seen whether the decline is there to stay as people have new preferences to commute post-COVID.
Workplaces
For workplace mobility change on weekdays, it is not surprising that it was impacted the most. Manhattan’s mobility decreased the most, while the other boroughs dropped off at comparable levels. Starting in mid-2021, the change stabilized at a new level that is below the baseline. It is logical to deduce that negative mobility changes are expected to last for a while as remote and hybrid work models have become the new norm.
As for weekends, it shows a slightly different picture. There was a dramatic fall at first, but since 2H 2020, mobility has already shown a trend of stabilization. A possible explanation for this could be that the majority of those who work on weekends did not have the flexibility to choose a remote work model, such as those who work in the service industry and work as temp workers.
Residential
The only category in which mobility increased is residential. Intuitively, the shorter time you spend outside means longer time at home.
For weekdays, all boroughs experienced major increases at first before plateauing, and the inter-borough variations were the smallest compared to other categories.
Weekends tell a similar story except during summertime. There were some negative changes that occurred during the summer of 2021 & 2022.
To summarize, residents in different boroughs acted differently during the pandemic, with Manhattan being the one with the most changes across categories. Thus, in the next step, the analysis focused on Manhattan and identified the correlations between its changes in mobility behaviors and the development of COVID.
Part 2 – Manhattan in focus
To explore the synchrony of mobility change and development of covid, visualize mobility changes, newly confirmed cases, and daily death count over time.
There were large changes in mobility behaviors upon major COVID spikes for all categories. Meanwhile, consistent with the previous analysis, residential mobility change is positively correlated with COVID stats, while all others are negatively correlated.
As for the death rate, the largest peak occurred as part of the original wave, and the much smaller peak in January 2022 was a result of the Omicron variant. The most recent outbreak caused the case count to spike but only slightly increased the death rate, and this coincided with a mild decline in mobility.
To quantify the synchrony among all the mobility behaviors and COVID categories, I set a cutoff date at end of Nov 2021 and split the data into pre-Omicron and post-Omicron periods. The Omicron variant, which was vastly different from all others in its transmissibility and severity, was first reported in early December in NYC and across the US.
Looking at the pre-Omicron correlation matrix, a strong relationship was found in the top three COVID statistics (death, hospitalization and case %):
Death rate has the largest (most negative) correlation with most of the mobility categories. Hospitalization rate and infection rate (case percentage) follow. All three rates have positive correlations with the Residential category.
Vaccination percentage is the exact opposite, with the highest correlation with Retail & Recreation (+0.92) and Transit (+0.84). And close to -0.9 with Residential.
Since the Omicron outbreak, the correlation of death/hospitalization/infection rates with Parks, Retail & recreation, and Grocery all decreased significantly. The two exceptions were Workplace (esp. weekdays) and Residential (esp. weekdays). Both showed stronger correlations with cases, deaths & hospitalization, and weaker correlations with Vaccination percentage.
R Shiny App
An interactive R Shiny app called “Empire State of Mind” is created to visualize all the analyses and comparisons conducted above, both by category and by borough. It also allows users to easily play around with different boroughs and cut-off dates for correlation analysis with different COVID stats.
Next Steps
- Build up prediction models incorporating time-series effects, COVID conditions and other possible factors.
- Add more interactive functionalities to the shiny app (e.g. allow users to check mobility and covid stats based on one selected date) and even expand to other major cities in US.
About Author
