Introduction

Beginning in March 2020, the COVID-19 pandemic significantly altered global mobility patterns. Public health agencies implemented curfews, traffic restrictions, and stay-at-home orders across various regions to mitigate the disease spread. During the pandemic, a global reduction in traffic volume was noted1. In 2021, restrictions began to be lifted and individuals started to receive COVID-19 vaccines2,3,4,5. By January 31st, 2022, 73.1% of individuals in the United States received their first dose of the Moderna or Pfizer-BioNTech vaccine or a single dose of the Johnson & Johnson vaccine6. After the restrictions were lifted7 and COVID-19 vaccination programs started globally, increases in mobility volume have been observed worldwide8,9.

Are COVID-19 vaccinations associated with traffic recovery? This remains unclear and needs further investigation. One hypothesis is that vaccinated individuals may feel a greater sense of protection, leading to more outgoing behavior and, consequently, an increase in traffic activities10. To study this potential association, we can exploit natural variations in vaccination rates and traffic volume recovery to examine how these variations may be related to each other. Fortunately, prior work often documents substantial variation in traffic recovery after the early phase of COVID-1911,12,13, and vaccination rates are known to vary by region14,15,16,17,18,19,20.

However, differences in vaccination uptake are not random, and are often correlated with socioeconomic, political, and demographic factors21,22,23. It is therefore additionally useful to understand how these factors are associated with vaccination rates, and therefore potentially traffic recovery patterns. It is also useful to control for variation in these factors, along with disease spread across regions, as the COVID-19 burden also varied substantially geographically24. Since vaccination can limit disease spread through herd immunity, it is important to estimate models with and without COVID-19 cases. Such factors may influence disease prevention behavior25; for instance, individuals who are cautious of being infected may reduce their outdoor activities when the COVID-19 prevalence level is higher thus leading to a decrease in traffic.

Additionally, when examining the relationship between traffic recovery and vaccination rates, it is crucial to control for other regional characteristics such as population size, income, racial composition, and employment status. Past studies have identified these variables in assessing how different elements contribute to traffic reductions and recoveries during and after COVID-19 lockdown scenarios13,26,27. For example, counties in the US with a higher household income were more likely to reduce their traffic during the early stages of COVID-1926. These demographic factors are also associated with the recovery in traffic – counties with a lower population density had a slower increase in the traffic volume during the recovery stage of COVID-1913. Therefore, these observations suggest that both regional socioeconomic and demographic characteristics play a crucial role in influencing traffic pattern changes during periods of the COVID-19 pandemic.

It is critical for a study examining traffic patterns to use traffic volume data from local roads. This data is more representative of regional traffic trends than information gathered from highway sensors. For instance, within LAC, the count of vehicles detected by a highway sensor in downtown may not accurately represent the actual car flow within the area, given that a substantial number of commuters traversing from the east side to the west side also travel through downtown. However, collecting local road traffic data can be difficult since it requires deploying a network of sensors that covers major roads and intersections in each region and continuously measuring data for consistent measurement. In this work, we leverage a unique dataset that draws on embedded local road sensors in LAC in the years 2020 and 2022. These local sensor recordings serve as estimates of traffic patterns in the neighborhoods surrounding these sensors.

Our primary objective in this study is to use local road sensor data in LAC and ZIP code level demographic, socioeconomic, and COVID-19 infections and vaccination data to study the association of vaccination rate on the changes in the traffic during the recovery phase of COVID-19 while controlling for demographic and socioeconomic variables. Our secondary goal is to assess the variability in vaccination rates across LAC and identify the factors contributing to this disparity.

Results

The study included 3,767 local road sensors in LAC maintained by the Archived Data Management System (ADMS) at the University of Southern California28. The study period includes May and June in the years 2020 and 2022. Due to the relocation of sensors, we cannot accurately recover pre-2020 traffic volumes through correction, scaling, or conversion. These sensors covered 55 ZIP codes after merging with demographic and socioeconomic data from the American Community Survey (ACS)29 and COVID-19 data from the LA Department of Public Health30,31. Table 1 shows the summary statistics of the traffic, demographic, socioeconomic, and COVID-19 variables. The average traffic increase across 55 ZIP codes from 2020 to 2022 is 103%, with increases ranging from a minimum of 35% to a maximum of 171% over the analysis period. The vaccination rate (at least one dose) by the beginning of May 2022 ranges from 68% to 100% with an average of 81%.

There exists variability in other variables including COVID-19 cumulative case rate (cumulative cases per person), income, proportions of each race, working industry, and commuting patterns. There is little variation in the percentage of the population employed. The average (standard deviation) of the employment percentage is 93% (2%).

Table 1 Summary statistics of variables used in this analysis.
Full size table

Variations in traffic volume recovery and vaccination rate

Figure 1 shows in detail the comparison of the traffic volume in May and June between 2020 and 2022. Due to COVID-19 and work-from-home policy in many companies32, a low traffic volume was observed across most ZIP codes in LAC (less than 150 vehicles passing over per 5 minutes on average). By 2022, the average ZIP code traffic volume had increased substantially, and the variability also increased.

Fig. 1
figure 1

Distribution of traffic volume across LAC in May and June between 2020 and 2022.

Full size image

Differences in traffic volume changes and vaccination rates are both observed throughout LAC. Figure 2a shows the distribution of the traffic increases from May and June in 2020 to May and June in 2022. Among the 55 ZIP codes examined, only El Sereno (90032) and Mar Vista (90066) reported traffic volume increases of less than 50%. Traffic increased in all ZIP codes in both May and June between 2020 and 2022, and the magnitude of the increase varies among them.

On average, ZIP codes included in this study show that 81.02% of the population has received at least one dose of the COVID-19 vaccine by May 2022. Additionally, we noted variations in the vaccination rates for the first dose across LAC in Figure 2b. For example, the vaccination rate in ZIP code 90044, a densely populated region of South LA with a majority of Black or African American residents, is under 70%. On the other hand, the vaccination rate in ZIP code 90045, a region of West LA with a majority of White residents, is 100%. Given the variations in both increases in traffic volume and vaccination rates, we seek to explore the explanations of these variations and, importantly, to determine if there is an association between the two variables.

Fig. 2
figure 2

There is variability in traffic pattern change and vaccination rates across LAC.

Full size image

Correlation between traffic recovery rate and vaccination rate

Fig. 3
figure 3

Univariate relationship between traffic increase relative to vaccination rates.

Full size image

To examine the association between the vaccination rate and the traffic increase, we ran the univariate linear regression model where the dependent variable is the traffic increase and the independent variable is the vaccination rate. Figure 3 shows the regression result for traffic increase relative to vaccination rates. There exists a positive relationship between the two variables, which indicates that a higher vaccination rate is associated with higher traffic increase. Next, we examine whether demographic, socioeconomic, and COVID-19 case data correlate with variations in traffic increases and vaccination rates. We aim to control for these variables in our regression analysis to accurately assess the relationship between increases in traffic volume and vaccination rates.

Correlation between vaccination rate and other variables of interest

We compare the demographic and socioeconomic variables across ZIP codes by vaccination rate to understand whether they statistically differed among ZIP codes that had lower than median vaccination rates and those that had higher. Table 2 shows the results of these t-test comparisons. ZIP codes with lower vaccination rates tended to have larger populations (5.36 vs 4.00, p value=0.018), Hispanic ratios (0.57 vs 0.41, p value = 0.022), proportions of the population who works in construction (0.08 vs 0.05, p value <0.001), manufacturing (0.09 vs 0.08, p value = 0.037), retail trade (0.11 vs 0.10, p value = 0.012), transportation and warehousing, and utilities (0.07 vs 0.05, p value = .001), and ratios of the population who drive when commuting (0.83 vs 0.76, p value = 0.003). Workers in construction, manufacturing, retail trade, transportation, warehousing, and utilities are crucial for the core functioning of societal infrastructure and thus can be defined as essential workers33. This suggests that ZIP codes with a higher proportion of essential workers had lower vaccination rates. Even though these workers were prioritized, we observe a lower vaccination rate and this might be due to vaccine hesitancy among essential workers.

Table 2 T-tests between low vaccination rate group and high vaccination rate group.
Full size table

We found that ZIP codes with lower vaccination rates tended to have lower mean incomes (8.07 vs 10.12, p value = 0.027), non-Hispanic Asian ratios (0.09 vs 0.19, p value = 0.002), and ratios of the population who works in professional, science, and management, and administrative and waste management services(0.12 vs 0.15, p value = 0.001). All other variables of interest were not statistically significant at an alpha level of 0.05.

Correlation between traffic volume recovery and other variables of interest

We compare the demographic and socioeconomic variables across ZIP codes by traffic volume change to understand whether they statistically differed among ZIP codes that had lower than median traffic volume recovery and those that had higher. Table 3 shows the results of these t-test comparisons. Unsurprisingly, we found that ZIP codes experiencing smaller increases in traffic also report lower vaccination rates (0.79 compared to 0.83, p value = 0.009), indicating that areas with higher vaccination coverage tend to see greater traffic growth from May and June of 2020 to 2022.

In addition to vaccination rates, we found that factors such as higher average income, greater proportions of non-Hispanic White and non-Hispanic Other populations, and a higher ratio of finance and technology workers, are also associated with increased recovery in traffic volumes. Specifically, ZIP codes with lower traffic recovery had lower mean incomes (7.98 vs 10.21, p value=0.016), non-Hispanic White ratios (0.17 vs 0.35, p value = 0.002), non-Hispanic other ratios (0.03 vs 0.05,p value<0.001), diverse index (0.45 vs 0.60, p value = 0.001), ratios of the population who works in information (0.03 vs 0.07, p value = 0.001), finance and insurance, and real estate and rental and leasing (0.05 vs 0.07, p value = 0.001), professional, scientific, and management, and administrative and waste management services(0.12 vs 0.15, p value = 0.008). This indicates that areas with more people who work in industries including science, technology, and finance experience a higher increase in traffic volume.

ZIP codes with lower traffic increases had higher populations (5.42 vs 3.94, p value=0.009), Hispanic ratios (0.61 vs 0.37, p value = 0.022), proportions of the population who works in agriculture, forestry, fishing and hunting, and mining (0.006 vs 0.003, p value=0.0013), construction (0.08 vs 0.05, p value <0.001), manufacturing (0.10 vs 0.07, p value = 0.001), transportation and warehousing, and utilities (0.07 vs 0.05, p value = 0.002), public administration (0.04 vs 0.03, p value = 0.005) and ratios of the population who drive when commuting (0.83 vs 0.76, p value=0.006). Workers in the above industries may also be classified as essential workers, given that public administration, agriculture, forestry, fishing, hunting, and mining play a critical role in supporting societal infrastructure as well. This indicates that a ZIP code with a higher proportion of essential workers has a lower traffic increase. As shown in Table 3, we found no other variables were statistically significant at an alpha level of 0.05.

Table 3 T-tests between low traffic recovery group and high traffic recovery group.
Full size table

Regression results

Table 4 shows the regression results with different model specifications. We found that the vaccination rate is significant at 0.05 \(\alpha\)-level across the 8 different models. This indicates a consistent association between vaccination rates and increases in traffic. The coefficient of the vaccination rate shows that a 1% increase in the vaccination rate is associated with a traffic increase ranging from 1.42% to 2.46% across the different regression models.

Additionally, we found differences in traffic volume recovery across races. The proportion of non-Hispanic Asian individuals is statistically significant in all three regression models that include it (p values of 0.013, 0.050, 0.026, respectively), with coefficients of − 1.6126 to − 1.1640. This indicates that a 1% increase in the non-Hispanic Asian population is associated with nearly 1% less traffic increase from May and June 2020 to 2022 compared with non-Hispanic Whites (the reference group). Consider ZIP code 90745 near Long Beach, where the non-Hispanic Asian population makes up 35%; here, traffic increased by 70% from 2022 to 2020. On the other hand, in ZIP code 90018, close to downtown LA and with only 5% non-Hispanic Asian populace, traffic volume increased by 115%.

We observe differences in traffic volume recovery by work industry. A 1% increase in the proportion of individuals who work in finance, insurance, and real estate is associated with a 10.62% increase (relative to the proportion of individuals who work in public administration) in the traffic recovery in model 6 and a 6.49% increase in the traffic recovery in model 8. From model 7, there is insufficient evidence to conclude that there is a true relationship between the commuting patterns and the increase in traffic from selected 55 ZIP codes. Moreover, model 8 shows that a 1% increase in the vaccination rate, the proportion of individuals who work in finance, and the proportion of individuals who work in public administration would correspond to a 1.71%, 6.49%, and − 9.66% change in the traffic increase, respectively.

Table 4 Regression results.
Full size table

Discussion

This study analyzed the association between COVID-19 vaccination rate and traffic recovery between 2020 and 2022 across 55 ZIP codes in LAC. We found that the vaccination rate is positively associated with traffic recovery. This could be because vaccinated individuals may perceive a greater sense of safety in socializing more often, or it could be that those who have a higher frequency of socializing are required to be vaccinated. This finding is robust in various regression models which we control for other variables of interest. These findings suggest that COVID-19 vaccines not only saved lives but also helped increase mobility which might have important effects on social and economic outcomes. Thus, a comprehensive assessment of the benefits and costs of vaccines must consider not only improvements in health but also spillover effects on social and economic outcomes.

Race/ethnicity and employment industry were also significantly associated with traffic recovery. We found that the proportion of the non-Hispanic Asian population had a negative association with the traffic recovery compared with the reference group (non-Hispanic Whites), as evidenced by models 5, 6, and 7 from regression results (see Table 4). This could be because of some unobserved differential mobility preferences, living patterns, and cultural and behavioral factors that need to be further investigated. Also, as evidenced by regression model 6, the proportions of individuals who work in finance had a positive association with the increase in traffic recovery compared with the proportions of individuals who work in public administration. This could be because the commuting patterns for individuals who work in public administration are more consistent than those who work in finance from early pandemic to post-pandemic. This hypothesis is then solidified by the findings in regression model 8, where the increase in traffic recovery is positively correlated to the proportions of individuals who work in finance while negatively correlated to the proportions of those who work in public administration. This result indicates a connection between the types of jobs held by individuals in a neighborhood and the traffic volume recovery in those neighborhoods.

Similar findings were drawn from the t-test results on correlations between traffic recovery and work industries. Areas with more people who work in industries including science, technology, and finance experience a higher increase in traffic. It could be because many workers in the above industries were able to work from home during the early pandemic but were asked to return to the office more often as the severity of the disease was alleviated34,35. Therefore their commuting patterns were shifted from less to more travel in May and June between 2020 and 2022. Conversely, we found that the areas with essential workers experienced less traffic increase. This is not surprising, as many essential workers were not able to work from home during the early pandemic. Thus, regions with more essential workers may not experience a larger increase in traffic in 2022 compared to 2020, given that essential workers’ job requires their physical presence at work, leading to consistent commuting patterns between summer 2020 and summer 2022.

We must acknowledge several limitations of this study. In our analysis, we focus exclusively on post-2020 data because the relocation of local sensors and the change in reading frequency from every 30 seconds to every 5 minutes since 2020 made the pre-2020 traffic data unsuitable for constructing traffic patterns prior to the pandemic. The relocation of sensors makes traffic flow data within the same region incomparable, as traffic volume measurements depend on the locations of these sensors. Therefore, we were not able to test whether a larger traffic recovery after the pandemic was due to a larger reduction in traffic during the early pandemic or not. To answer this question, pre-COVID traffic data for the same sensors selected for this study is required. We are aware that a study has found that traffic in May-June 2020 was about 24% lower than pre-COVID levels in Southern California36. However, these findings are at the county level on highway roads, and there is a lack of zip code-level analysis in current research, highlighting a potential direction for future studies.

We examined the correlation between each pair of variables to assess the presence of multicollinearity in the regression models. The details of these correlation tests are provided in the supplementary material. We found that work industries are strongly correlated with variables like vaccination rates, mean income, and race. These correlations affect the robustness of models 7 and 8 compared to others. However, we did not find strong correlations among variables in the other models, ensuring that our main conclusions and findings remain valid and robust.

Our dataset includes roughly 3,767 sensors across 55 zip codes. Although some roads may not be monitored, the sensors are evenly distributed28, offering a representative view of local traffic patterns in these areas. This suggests that any missed roads are likely small streets, with their traffic flow captured by nearby sensors. Additionally, there should be no bias across ZIP codes due to non-monitored streets, as one ZIP code is unlikely to have more non-monitored streets than another, ensuring valid comparisons between ZIP codes. Moreover, there are over 400 ZIP codes within LAC, and we only used 55 ZIP codes to perform our analysis. Many ZIP codes that we did not include in this analysis are either not covered by arterial road sensors or do not have socioeconomic or demographic information. These 55 ZIP codes might not be representative of the excluded regions. However, given that the 55 ZIP codes cover key regions such as downtown, west LA, south LA, and residential areas like El Monte, the ZIP codes monitored by the chosen road sensors could be considered a representative sample of the broader LAC.

The LAC lockdown and reopening policies homogeneously affected all zip codes in our analysis, as they were part of the same health jurisdiction. However, some businesses may have permanently closed or chosen not to fully reopen when allowed, leading to variation in business operations over time7. These geographic differences in reopening levels could potentially influence traffic patterns. Nevertheless, it is extremely difficult to account for these effects due to limited data. Evidence suggests that COVID-19 cases are predictive of business reopening7. Therefore, we included COVID-19 cases as a covariate in our regression models when examining the relationship between vaccination rates and traffic recovery. Our findings indicate that there remains a strong positive relationship between vaccination rates and traffic recovery even after using COVID-19 cases as a proxy for business reopening.

Moreover, 13 work industry categories included in the survey are based on the North American Industry Classification System (NAICS) codes37. However, the mobility behavior during the pandemic could be substantially different across sub-categories within the same category. For instance, while entertainment and food services fall under the same category, their behaviors during the pandemic were quite distinct. Food service workers often had to continue working outside the home, whereas those in entertainment were more likely to face work stoppages. Additionally, we cannot account for the effects of remote work as we lack the ubiquity of work-from-home across zip codes. Finally, this is an observational study and we can not definitely establish that higher vaccination rates led to traffic recovery. We recognize that vaccination rates might be endogenous, that is, unobserved confounders might be correlated with vaccination rates and traffic which might bias our estimates. To address this issue we show that our results are stable when we account for a variety of observed confounders. This gives us confidence that unobserved confounders which are likely to be correlated with observed confounders are unlikely to introduce significant bias. To fully address the issue we need valid instruments that are correlated with vaccination rates and only affect traffic through their impact on vaccination rates. Unfortunately, we don’t have such instrumental variables. Therefore, we acknowledge potential bias due to unobserved confounders in the limitation section of the paper.

Despite these limitations, we believe that our findings illuminate a significant association between traffic recovery and vaccination rate in LAC. Even after controlling for other COVID-19, demographic, and socioeconomic variables, we found that regions with higher vaccination rates tend to have a greater recovery in traffic. This suggests that an increased vaccination rate could reduce the public’s perception of the risks of disease infection, consequently changing their mobility behaviors. Future research should focus on collecting more comprehensive data that is able to cover more regions in LAC, identifying additional factors linked to traffic patterns in an epidemic context, considering traffic mix changes, and offering a deeper understanding of how policies implemented during an epidemic impact mobility.

Methods

Traffic data processing

This study used local road sensor data maintained by Archived Data Management System (ADMS) at the University of Southern California28. The ADMS collected both arterial and highway road sensor readings on occupancy, volume, and speed for the past five-minute intervals at the rate of one reading per sensor per minute.

In this study, we used the arterial road sensor data to proxy the traffic pattern around each sensor. There are roughly 3,767 arterial traffic sensors (loop-detectors) covering almost all directions at each intersection in many regions28. We focus our analysis on local road traffic, considering it the best proxy for resident mobility patterns within each ZIP code, given the limitations of our data. Although methods exist for estimating traffic flow between origins and destinations, they require more comprehensive coverage of all ZIP codes, whereas our sensors only cover 55 ZIP codes. Additionally, implementing these methods with our study’s nearly 4,000 nodes is challenging due to computational tractability issues.

Sensor data is aggregated temporally and spatially. We first calculate the daily average volume, representing the number of vehicles that passed the sensor during each five-minute interval. This is followed by estimating the average number of vehicles per five-minute interval on a monthly basis for each sensor. Lastly, we aggregate these monthly averages for sensors within the same ZIP codes to obtain the monthly average number of vehicles passing over per five-minute interval for each ZIP code. We exclude sensors with abnormal recordings or consistent zero readings (sensor malfunctioning).

Demographic, socioeconomic, and COVID-19 data

We obtain demographic and socioeconomic data from the American Community Survey (ACS) on the United States Census Bureau website29. We collected data for all ZIP Code tabulation areas within LAC that also have traffic data available.

In the analysis, we also obtained the reported COVID-19 case numbers and vaccination data from LAC COVID-19 Locations & Demographics dashboard30. Because the COVID-19-related data is at the community level, we mapped each community to ZIP codes and then merged data to create a dataset that included demographic, socioeconomic, and traffic data.

Statistical tests

In this study, we utilize data from ADMS arterial road sensors collected during May and June of both 2020 and 2022. The road sensors are aggregated by zip code. For COVID-19 cases and vaccinations, we used the cumulative number reported by the beginning of May 2022. After merging all datasets, we obtained a total of 55 ZIP codes for our analysis.

We first performed t-tests to study the relationship between heterogeneity in vaccination rate and other variables. We divided all ZIP codes into two groups by vaccination rates–above-median and below-median vaccination groups. We tested if there was a significant difference in each of the demographic, socioeconomic, and COVID-19 variables between the two groups.

We then study the association between traffic volume change from 2020 to 2022 and vaccination rate while considering other factors considered in the previous t-test analysis. We define the change in traffic from 2020 to 2022 as the average of the percentage changes in traffic volume between May 2020 and May 2022 and between June 2020 and June 2022 (percentage change in traffic). We divided all ZIP codes into two groups by percentage change in traffic–above-median increase and below-median increase groups. We then performed a t-test on each demographic, socioeconomic, and COVID-19 variable between these two groups and identified the variables that are significantly different across groups.

Regression analysis

To assess the relationship between traffic volume change and vaccination rate using regression models, we estimated eight ordinary least-square regression (OLS) models with the percentage change in traffic as the dependent variable. OLS regression is suitable for our analysis because it has the lowest sampling variance within the class of linear unbiased estimators38. Moreover, since our variable of interest is continuous, OLS is appropriate for modeling continuous outcomes. Additionally, OLS allows us to easily control for covariates, making it a robust choice for our analysis. For the independent variables, we used (1) vaccination rate, (2) vaccination rate + case rate, (3) vaccination rate + case rate + population, (4) vaccination rate + case rate + population + socioeconomic variables, (5) vaccination rate + case rate + population + socioeconomic variables + racial variables, (6) vaccination rate + case rate + population + socioeconomic variables + racial variables + working industry, (7) vaccination rate + case rate + population + socioeconomic variables + racial variables + working industry + commuting pattern, (8) variables that are significantly different across high and low traffic change groups.