key: cord-1056133-gxjsv896
authors: Hu, T.; Yue, H.; Wang, C.; She, B.; Ye, X.; Liu, R.; Zhu, X.; Bao, S.
title: Racial segregation, testing sites access, and COVID-19 incidence rate in Massachusetts, USA
date: 2020-07-07
journal: nan
DOI: 10.1101/2020.07.05.20146787
sha: b142d829be731292424a08e32936ba572912d18d
doc_id: 1056133
cord_uid: gxjsv896

The U.S. has merely 4% of the world population but 25% of the world's COVID-19 cases. Massachusetts has been in the leading position of total cases since the outbreak in the U.S. Racial residential segregation is a fundamental cause of racial disparities in health. Moreover, disparities of access to health care have a large impact on COVID-19 cases. Thus, this study estimates racial segregation and disparities in testing sites access and employs economic, demographic, and transportation variables at the city/town level in Massachusetts. Spatial regression models are applied to evaluate the relationships between COVID-19 incidence rate and related variables. This is the first study to apply spatial analysis methods across neighborhoods in the U.S. to examine the COVID-19 incidence rate. The findings are: 1) residential segregations of Hispanic and Non-Hispanic Black/African Americans have a significantly positive association with COVID-19 incidence rate, indicating the higher susceptibility of COIVD-19 infections among minority; 2) The Black has the shortest drive time to testing sites, followed by Hispanic, Asian, and Whites. The drive time to testing sites is significantly negatively associated with the COVID-19 incidence rate, implying the importance of testing location being accessed by all populations; 3) Poverty rate and road density are significant explanatory variables. Importantly, overcrowding represented by more than one person per room is a significant variable found to be positively associated with COVID-19 incidence rate, suggesting the effectiveness of social distancing for reducing infection; 4) Different from previous studies, elderly population rate is not statistically significant with incidence rate because the elderly population in Massachusetts is less distributed in the hot spot regions of COVID-19 infections. The findings in this study provide useful insights for policymakers to propose new strategies to contain the COVID-19 transmissions in Massachusetts.

The U.S. has merely 4% of the world population but 25% of the world's COVID-19 cases. 28 Massachusetts has been in the leading position of total cases since the outbreak in the U.S. 29

Racial residential segregation is a fundamental cause of racial disparities in health. Moreover, 30

disparities of access to health care have a large impact on COVID-19 cases. Thus, this study 31 estimates racial segregation and disparities in testing sites access and employs economic, 32 demographic, and transportation variables at the city/town level in Massachusetts. Spatial 33 regression models are applied to evaluate the relationships between COVID-19 incidence rate 34 and related variables. This is the first study to apply spatial analysis methods across 35 neighborhoods in the U.S. to examine the COVID-19 incidence rate. The findings are: 1) 36 residential segregations of Hispanic and Non-Hispanic Black/African Americans have a 37

significantly positive association with COVID-19 incidence rate, indicating the higher 38 susceptibility of COIVD-19 infections among minority; 2) The Black has the shortest drive time 39

to testing sites, followed by Hispanic, Asian, and Whites. The drive time to testing sites is 40 significantly negatively associated with the COVID-19 incidence rate, implying the importance of 41

The COVID-19 pandemic has severely impacted the socioeconomic activities worldwide 57 since its outbreak in January. As of July 3, there have been 10,719,946 confirmed cases 58 globally, including 517,337 deaths. The United States is the leading country with 2,671,220 59 confirmed cases and 127,858 deaths (WHO, 2020). Since the beginning of April, the US has 60 become the COVID-19 pandemic center and the number of cases is still increasing. Social 61 distancing is one of the most effective ways to reduce COVID-19 infection, but due to residential 62 segregation, the separation of people based on income and/or race, some people from ethnic 63 minority groups cannot practice social distancing. Hence, they are often found in overcrowded 64 urban housing areas and make physical distancing and self-isolation difficult, thus leading to the 65 increased risk for the spread of COVID-19 (Bhala, 2020) . In addition, socioeconomic inequities 66 frequently impact health and healthcare access, resulting in a higher burden of disease and 67 mortality in vulnerable social groups (Daniel, 2020). Therefore, it is necessary to integrate 68 social-economic information and disease statistics to help analyze and understand the spread of 69 Many research findings have highlighted the racial disparities in COVID-19 infections. 71

Across the country, deaths caused by COVID-19 are disproportionately high among African 72

Americans (Dorn, 2020), while Chicago and New York City reported greater COVID-19 mortality 73

among Latinos (Hooper, 2020) . To utilize more detailed information on the racial and 74 socioeconomic disparities, Matthew and Julia (2020) 128  demonstrated that areas with a high incidence of COVID-19 usually have high-income inequality  129 and median household income (Mollalo, 2020) . Liu (2020) collected the number of laboratory-130  confirmed COVID-19 cases in 312 cities in China, and a series of sociodemographic variables  131 such as distance to the epicenter, the total length of built urban metro lines, urban area, 132 population density, the annual quantity of wastewater discharged, and residential garbage 133 connected and transported, per capita public recreational green space, the daily highest 134 temperature, and the capital city. Based on these data, a study of the impacts of COVID-19 135 transmission was conducted from the urban perspective. The statistically significant results 136 revealed that residential garbage connected and transported, and the annual quantity of 137 wastewater discharged could increase the confirmed infection number of COVID-19 (Liu, 2020) . 138

To explore regional patterns and inform local policymakers about the efficient allocation 139 of resources and personnel to e ectively mitigate the spread of Covid-19, making inferences at 140

finer scales, such as the sub-county, census tract, or block group, may produce more accurate 141 results than coarse levels, such as the county or above. In Chicago, more than 50% of COVID-142 19 cases and nearly 70% of COVID-19 deaths involve black individuals. These deaths are 143 concentrated in just 5 neighborhoods in the city's South Side. Thus, it is critical to quantify the 144 interaction between the various factors and disease statistics at a finer scale. Analyzing regional 145 patterns and increasing their awareness of the COVID-19 dangers and preventative measures, 146

will benefit the most a ected communities (Daniel, 2020). Quantifying disparities in risk is 147 important for allocating resources to prevent, identify, and treat COVID-19-related severe illness 148 and limit diverging outcomes for vulnerable subgroups. 149

The research aims to identify the impact of racial segregations and testing sites 150 accessibility on the COVID-19 incidence rate in cities/towns of Massachusetts. To the authors' 151 knowledge, it is the first study to employ spatial analysis methods on neighborhoods' COVID-19 152 data in the US. The objectives are: (1) evaluating the minority racial segregation, such as 153

Hispanics, Non-Hispanic Black Americans, and Non-Hispanic Asians and its socioeconomic 154 characteristics in Massachusetts;

(2) accessing the spatial accessibilities to testing sites across 155 different sociodemographic groups in the study area; (3) investigating whether neighborhoods 156 with higher COVID-19 incidence rate are positively associated with highly segregated areas for 157 minority ethnics; (4) exploring whether testing sites is well distributed for the COVID-19 testing; 158 and (5) examining the association between socioeconomic and COVID-19 incidence rate. 159 160 CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted July 7, 2020. . cities/towns in Massachusetts also publishes testing site locations in the state and we geocode the location to latitude and 189 longitude. Figure 1 demonstrates the spatial distribution of the COVID-19 incidence rate and 190

testing sites as of May 20, 2020. There is an extremely uneven distribution pattern: cities/towns 191 with high incidence rates were aggregated in the Greater Boston area. In the central and west of 192

Massachusetts, a few places are the hotspot of high incidence rates. 193 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted July 7, 2020. . 

This study aims to investigate the relations between COVID-19 incidence rate and racial 199 residential segregation, access to testing sites, and other socio-demographic and economic 200

variables. These data are gathered from different sources. The city/town COVID-19 incidence 201 rate data are collected from the Department of Public Health in Massachusetts. The racial 202 residential segregation and disparities in testing site accessibilities are compiled from racial 203 population data in cities/towns and sub-unit, census block groups. There have been many 204 studies involving various socioeconomic and demographic factors that may impact the COVID-205 19 incidence rate, including the elderly population rate, poverty rate, overcrowding rate, 206

household income, and so on. We collect such variables both from the subcounty and census 207 block group (CBG) 2018 American Community Survey (ACS) provided by the US Census 208

Bureau (https://www2.census.gov/geo/tiger/TIGER_DP/2018ACS/). 209

This study estimates minorities' residential segregations, including Hispanic, Non-210

Hispanic Black or African Americans, and Non-Hispanic Asian. To simplify the name of each 211 race, we use Black and Asian to present Non-Hispanic Black or African American and Non-212

Hispanic Asian in the rest of the paper. Residential segregation can be discussed in five distinct 213 dimensions: centralization, concentration, clustering, unevenness, and isolation (Massey and 214 Denton, 1988) . This study applies the isolation index to associate segregation and health 215

outcomes. The isolation index can effectively reveal the racial differential sizes (Chang, 2006). 216 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted July 7, 2020. . Also, it is different from the racial percentage, which is hard to tell the racial distribution 217

disparities. The higher of the racial isolation index indicates the same racial population has a 218 higher chance to live as neighbors. It is a vital variable to study COVID-19 transmission patterns. 219

Thus, this research applies the isolation index to estimate the segregation of each race. The 220

isolation index works as follows. Assuming city/town j consists of n census block groups, the 221 isolation index for a race within j can be presented as:

where i is the i-th census block group in the city/town j;

ܾ is the race population in i;

is the 225 total race population in j, and ܶ is the total population in i. The isolation index ranges from 0 to 1. 226 0 indicates no segregation and 1 indicates the greatest segregation. The index presents the 227 chance of having the same race live as neighbors. 228

Access to testing sites is measured by proximity, a popular method in geographic 229

information systems (GIS), and the most influential component in health-related studies. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted July 7, 2020. . and public transit rate (percentage of workers 16 years and over who go to work by public 259 transportation (excluding taxicab)). . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted July 7, 2020. 

In this study, we first use Moran's I index to test the spatial autocorrelation pattern of the 266 city/town-level COVID-19 incidence rate in Massachusetts. Then, two classical spatial 267 autoregressive models, including the spatial lag model (SLM) and the spatial error model (SEM), 268 are adopted to determine the factors that influence the COVID-19 incidence rate. 269 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted July 7, 2020. (Tobler, 1970) . This phenomenon is referred to as spatial 273 autocorrelation or spatial dependency, which indicates whether the distribution of data depends 274 on geographical locations (Cliff & Ord, 1973) . If data at adjacent locations are similar to each 275 other, the spatial autocorrelation effect is regarded as positive. If data at adjacent locations differ 276 greatly from each other, the spatial autocorrelation effect is negative. If the distribution of data is 277 irrelevant to geographical locations, there is no obvious spatial autocorrelation effect. A 278 commonly used indicator to identify the spatial autocorrelation effect is Moran's I index (Moran, 279 1948), which is formulated as: 280

283 where ‫ݔ‬ and ‫ݔ‬ mean the attribute of i-th and j-th spatial unit (the COVID-19 incidence rate of a 284 city/town in this study),

indicates the average of all the attributes, n means the number of 285 spatial units, and ‫ݓ‬ is a member of the spatial weight matrix W which represents the spatial 286 relationship between spatial unit i and j. In this research, we adopted the frequently-used rook 287 method to construct the weight matrix, i.e., if spatial unit i and j share a common border, 288 then‫ݓ‬ ൌ 1 , otherwise, ‫ݓ‬ ൌ 0 (Anselin, 1988) . 289

The value of Moran's I index ranges between -1 and 1: value less than 0 means a 290 negative spatial autocorrelation effect, and the smaller the value is, the stronger the negative 291 spatial autocorrelation effect is; a value greater than 0 indicates a positive spatial 292 autocorrelation effect, and the larger the value is, the stronger the positive spatial 293 autocorrelation effect is; and a value equals to 0 means no spatial autocorrelation effect. In 294 addition to this, the Z-value can be applied to assess the statistical significance of the Moran's I 295 index, i.e., if the Z-value is larger than 1.96 or smaller than -1.96, then there is a spatial 296 autocorrelation effect with a confidence level of 95%. 297

298 Spatial data usually presents a certain degree of positive spatial autocorrelation, i.e., 299 attributes of adjacent geographical events tend to be more similar than attributes of 300 geographical events that are further away from each other (Anselin & Bera, 1998) . The 301 existence of spatial autocorrelation effect violates the basic independent identical distribution 302

(IID) assumption of classical regression models, i.e., observations are independent of one 303

another. The significance of estimates will be overestimated if this effect is not taken into 304 account . In this study, we applied two common spatial regression 305 models to tackle the spatial autocorrelation effect, including the spatial lag model (SLM) and the 306 spatial error model (SEM). The SLM and SEM account for the spatial autocorrelation effect in 307 different ways: in the SLM, the dependent variable at a location is influenced by dependent 308

(2) . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted July 7, 2020. disparities in magnitude, we standardized these variables to have a mean of 0 and a variance of 328

1. This could make the parameter estimates independent of units and easy to compare. 329 The COVID-19 incidence rate varies across different socio-demographic groups at the 333 city/town level. As shown in Table 2 , the Black population has the lowest mean value of 2.75% 334 ranging from 0 to 43.95%, followed by Asian and Hispanic populations with mean values of 3.6% 335 and 4.83%. The white population has the highest mean value of 91.06% ranging from 39.27% to 336 100%, reflecting the dominance of white people in Massachusetts. These four groups 337 experience very similar COVID-19 incidence rates across the first three quantiles and then the 338 rate differences enlarge in the fourth quantile. Using the first quantile as a reference, Figure 2  339 shows the variation of the COVID-19 incidence rate ratio of each quantile with that of the first 340

quantile. All four groups exhibit a smooth increase with minor differences and then the ratio 341 dramatically moves up to a range of 9 to 16. Due to the large amount of confirmed COVID-19 342 cases in Boston city, we recalculate the COVID-19 incidence rate in the fourth quantile outside 343 of Boston city in Table 2 and Figure 3 . The rate ratios of Asian and White are consistent with 344

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted July 7, 2020. . those in the entire study area while Black and Hispanic groups exhibit a small increase in 345 incidence rate, which are aligned with the concentrated segregation indices of Black and 346

Hispanic in Boston city in Figure 2 . 347

Racial residential segregation is estimated by the isolation index described in Section2. 348 Figure 2 illustrates the segregations of the minorities in Massachusetts, including Hispanic, 349

Black, and Asian. Higher values indicate higher possibilities that people of the same race are 350 living as neighbors. As shown in Figure is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted July 7, 2020. .

(c) Asian residential segregation in Massachusetts cities/towns 360 Figure 2 . Racial segregation index of Hispanic, Non-Hispanic Black, and Non-Hispanic Asian 361 362

To quantify the health disparities across different racial groups, the weighted travel times 363

to testing sites and hospitals are illustrated in the last two columns of . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted July 7, 2020. positive sign implies the existence of spatial adjacency effect. The COVID-19 incidence rate of a 386 city/town affects that of other nearby cities/towns, partially because adjacent cities/towns share 387 similar attributes. The p-value is considerably lower than 1%, indicating that the existence of 388 spatial autocorrelation in the COVID-19 incidence rate is statistically significant. 389

After correlation analysis (correlation coefficients<0.6), among the 12 candidate 390 explanatory variables, 9 variables were selected to be included in the final models. These 391

variables are the percentage of elderly people (65 years and over), percent of more than 1 392 occupants per room, poverty rate, income inequality, road density, and test site accessibility. 393 Table 3 shows the results of the two spatial regression models. For comparison, the classic 394 OLS model was also calibrated, and results were listed. 395 ult ng nd he f a re of te se 1 ty. ic . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted July 7, 2020. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted July 7, 2020. . Three metrics were used to compare the performances of the OLS, SLM, and SEM: log-402 likelihood at convergence (log-likelihood), Akaike information criterion (AIC), and R-squared (R 2 ). 403

For log-likelihood and R 2 , a higher value means better performance; for AIC, a lower value 404

represents better performance. The results of the model performances suggested the following. 405

First, two spatial models could better fit the observations than OLS could: the AIC of the OLS 406 (5263.18) was much larger than in the SLM (5213.96) and SEM (5210.48), while the log-407 likelihood and R 2 of the OLS (-2621.59, 0.579) were smaller than those of the SLM (-408 2596.979,0.654) and SEM (-2594.24,0.657). Second, the performance of the SEM was slightly 409

better than that of the SLM: the log-likelihood and R 2 of the SEM were larger than those of the 410 SLM, while the AIC of the SEM was smaller than that of the SLM. The highest R 2 is achieved by 411 SEM (0.656657) that explains 65.67% of the total variations of COVID-19 incidence rates. 412

A tenet of regression is that residuals should be independent of each other, that is, they 413

should be randomly distributed in space. Furthermore, the degree to which residuals are 414 autocorrelated in space is another important indicator to judge the performance of regression 415 models. The results showed that the residuals of the OLS revealed a strong positive 416 autocorrelation, while the residual autocorrelations of the SLM and SEM were eliminated, and 417 the residual autocorrelation of the SEM was eliminated more thoroughly than that of SLM (the z-418 score of SEM (0.140) is smaller than that of SLM (1.017). 419

According to the above-mentioned results, by incorporating spatial dependence, the two 420 spatial regression models improve the performance of OLS in modeling the COVID-19 421 incidence rate in Massachusetts. Additionally, the SEM outperformed the SLM, which means 422 that the spatial spillover effect in the data is mainly reflected in the residuals. Therefore, the 423 SEM specification is a more appropriate choice in this research to account for the spatial 424 autocorrelation effect. 425

The results of the three models demonstrate that segregations of Hispanic and Black are 426 significantly positively associated with the COVID-19 incidence rate, while segregation of Asian 427 had a negative and nonsignificant influence on the COVID-19 incidence rate. Rate of the population older than 65 was negatively associated with the COVID-19  436 incidence rate, as demonstrated by the result of OLS, but the results of two spatial regression 437 models showed that this relationship was positive. As to the rate of households with more than 1 438 occupants per room, the results of the three models were all significantly positive. Specifically, 439 the result of SEM indicated that a one-point increase in the rate of households with more than 1 440 occupants per room is associated with a 157.385-point increase in COVID-19 incidence rate. 441

With a parameter estimate of -66.217 (SEM), the rate of the population below the poverty level 442 had a significantly negative influence on the COVID-19 incidence rate. The association between 443 this variable and the COVID-19 incidence rate was also negative in OLS and SLM, although the 444 relationship had no statistical significance in the result of SLM. The results of the three models 445 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted July 7, 2020. . demonstrated that income inequality had a nonsignificant and negative impact on the COVID-19 446 incidence rate. Road density had a strong positive influence on the COVID-19 incidence rate, 447 and the influence was statistically significant, as demonstrated by the results of three models. A 448 one-point increase in road density is associated with a 226.89-point increase in COVID-19  449 incidence rate, according to the result of SEM. 450

Racial/ethnic segregation is regarded as a fundamental cause of disparities in diseases 452 (Williams and Collins, 2001) . This study investigates the association between racial segregation 453 and the COVID-19 incidence rate in Massachusetts, particularly the minority groups, such as 454

Hispanic, Black/African American, and Asian. We find higher Hispanic and Black/African 455

American segregations are more likely to be associated with a higher incidence rate. The areas 456

where many black people reside are in poor areas characterized by high housing densities 457 (Yancy, 2020) . As revealed in the regression model, a higher percentage of more than 1 458 occupants per room and a higher poverty rate are significantly associated with the incidence 459 rate as well. The higher observed incidence and severity in minority groups may be also 460 associated with socioeconomic, cultural, or lifestyle factors, genetic predisposition, or 461 pathophysiological differences in susceptibility or response to infection (Khunti, 2020 Moreover, other studies found obesity and smoking were associated with increased risks 474 (Huang, 2020; ). In Italy, higher risks have also been reported in men than in women, 475

which could be partly due to their higher smoking rates and subsequent comorbidities 476 (Livingston, 2020 CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted July 7, 2020. . dominant neighborhoods (Williams and Copper, 2020 Previous studies demonstrated that people over 65 are at the highest risk (Centers for  497 Disease Control and Prevention, 2020), however, our study suggested that the relationship 498 between COVID-19 incidence rate and the proportion of elderly people had no statistical 499

significance. This may be attributable to the geographical distribution of the senior citizens in 500

Massachusetts. As shown in Figure 4 , cities/towns with high percentages of the elderly 501 population are mainly located in the west and eastern coastal areas, which are far from the 502 Greater Boston area. 503 504 Figure 4 . Geographical distribution of percentage of population 65 years and over. 505 506

Results of SEM suggested a strongly significant and positive relationship between 507 COVID-19 incidence rate and percent of occupied housing units with more than 1 occupants per 508 room. Quarantines are often one of the first responses against new infectious diseases. 509

However, it is estimated that 44% of secondary cases were infected during the index cases' 510 presymptomatic stage, in settings with substantial household clustering, and quarantine outside 511 the home (He, 2020). Several people sharing the same room in households with lower income 512 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted July 7, 2020. . may increase the chance for virus transmission. This study validated this hypothesis by showing 513 that a higher percentage of more than 1 occupants per room and higher poverty rates are 514 significantly associated with the incidence rate. The results of SEM also suggested a strongly 515 significant and negative impact of the poverty rate on the COVID-19 incidence rate in 516

Massachusetts. At first glance, this result is counterintuitive, since those in poverty are generally 517

at risk of losing their health insurance coverage, which makes them vulnerable in the face of 518 emergent epidemics like COVID-19. However, most cities/towns with high poverty rates are 519 located in peripheral areas, which are remote from high COVID-19 incidence regions, as can be 520 seen from Figure 5 . Road density is significantly positively associated with COVID-19 incidence 521 rate, this may be because cities/towns with higher road densities usually have more frequent 522 human activities, which could provide conditions for disease transmission. was conducted at or above the county level, which prevents us from observing how the COVID-530 19 incidence rate interacts with the various factors at a finer spatial scale. This study analyzed 531 how the COVID-19 incidence rate was associated with racial and health accessibility disparities, 532

after controlling for the possible influences of a set of demographics, economic, and 533 transportation factors. The study was conducted at the city/town level in the State of 534

Massachusetts, which is one of the states hit hardest by the pandemic. 535 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted July 7, 2020. . https://doi.org/10.1101/2020.07.05.20146787 doi: medRxiv preprint

The classic OLS regression model is unable to deal with the spatial autocorrelation 536 effect. Therefore, we conducted a further spatial regression analysis based on two models, i.e., 537 SLM and SEM, to obtain more robust estimation of the significances and directions of the 538 influences of racial and health accessibility disparities, demographic, economic and 539 transportation characteristics on COVID-19 incidence rate. Results suggest that residential 540 segregations of Hispanic and Non-Hispanic Black/African Americans are associated with an 541 increased risk of COVID-19 infection. Similarly, road density and the percent of more than 1 542 occupants per room have statistically significant and positive impacts on the COVID-19 543 incidence rate. However, test site accessibility and poverty rate are related to a decreased risk 544 of COVID-19 infection. 545

The empirical findings shown in this paper could provide helpful insight and guidance for 546 policymakers to develop strategies to contain the COVID-19 transmissions in Massachusetts. 547 Firstly, political action is needed to resolve long-standing societal inequalities, addressing the 548 injustices of public health, and tackling the COVID-19 pandemic and its sequence (Bhala, 549 2020). Public health is complicated and social reengineering is complex, but a change of this 550 magnitude does not happen without a new resolve (Yancy, 2020) . Additionally, as road density 551 and the overcrowding are found to have so much influence on the pandemic transmission, then 552 the social distancing or stay-at-home policies should be rigorously followed. As the condition of 553 COVID-19 is still going on and evolving quickly in the United States, it is important to explore the 554 spatial patterns in small units, such as neighborhoods. The research framework and 555

implications of this study can also provide a reference for other areas to help understand the 556 local transmission of COVID-19 and minimize the damage of the terrible disease. 557

Residential segregation and health. 559 Neighborhoods and health

Spatial econometrics: Methods and models

Spatial dependence in linear regression models with an 563 introduction to spatial econo-metrics

Reveal Harsh Truths About Structural Inequality in America

Sharpening the 569 global focus on ethnicity and race in the time of COVID-19

National Vital Statistics System. Provisional 571 Death Counts for Coronavirus Disease (COVID-19). Retrieved from

Spatial autocorrelation. Pion

Testing the association between two spatial processes. Statistics 576 and Decisions

Testing the association between two spatial processes

Racial/ethnic and socioeconomic disparities of Covid-19 attacks 580 rates in Suffolk County communities

COVID-19 and RacialEthnic Disparities

Building an Open 584 Resources Repository for COVID-19 Research

Clinical findings of 586 patients with coronavirus disease

Blacks account for high share of Boston COVID-19 cases

Is ethnicity linked to incidence or 592 outcomes of covid-19

The COVID-19 pandemic: a call to action to identify 594 and address racial and ethnic disparities

Emerging study on the transmission of the Novel Coronavirus (COVID-19) from 597 urban perspective: Evidence from China

Spatiotemporal 600 Patterns of COVID-19 Impact on Human Activities and Environment in Mainland China 601 Using Nighttime Light and Air Quality Data

Coronavirus Disease 2019 (COVID-19) in Italy

MA COVID-19 Testing Sites

MA COVID-19 Testing Sites

The dimensions of residential segregation

GIS-based spatial modeling of COVID-19 incidence 611 rate in the continental United States

The interpretation of statistical maps

Disparities in the population at risk of severe illness 616 from covid-19 by race/ethnicity and income

The coronavirus is infecting and killing black 619 Americans at an alarmingly high rate

A computer movie simulating urban growth in the Detroit region

Clinical characteristics 627 of 138 hospitalized patients with 2019 Novel Coronavirus-Infected pneumonia in Wuhan

Quantitative methods and socio-economic applications in GIS

Estimating O-D travel time matrix by Google Maps API: implementation, 632 advantages, and implications

World Health Organization (WHO)., 2020a. WHO Coronavirus Disease (COVID-19) Dashboard. 634 Retrieved from

World Health Organization (WHO)., 2020b. WHO Director-General's opening remarks at the 636 media briefing on COVID-19

Racial residential segregation: a fundamental cause of racial 640 disparities in health

COVID-19 and Health Equity-A New Kind of "Herd 642 Immunity

Temporal 644 dynamics in viral shedding and transmissibility of COVID-19

COVID-19 and African Americans

Taking the pulse 649 of COVID-19: A spatiotemporal perspective

Prevalence of 651 comorbidities in the novel Wuhan coronavirus (COVID-19) infection: a systematic review 652 and meta-analysis. International journal of infectious diseases

Massachusetts Health Reform 655 and Disparities in Coverage, Access and Health Status