Conditional Autoregressive Model Research Articles

Dengue fever mapping in Bangladesh: A spatial modeling approach

AbstractBackgroundEpidemics of the dengue virus can trigger widespread morbidity and mortality along with no specific treatment. Examining the spatial autocorrelation and variability of dengue prevalence throughout Bangladesh's 64 districts was the focus of this study.MethodsThe spatial autocorrelation is evaluated with the help of Moran and Geary . Local Moran was used to detect hotspots and cold spots, whereas local Getis Ord G was used to identify only spatial hotspots. The spatial heterogeneity has been detected using various conventional and spatial models, including the Poisson‐Gamma model, the Poisson‐Lognormal Model, the Conditional Autoregressive (CAR) model, the Convolution model, and the BYM2 model, respectively. These models are implemented using Gibbs sampling and other Bayesian hierarchical approaches to analyze the posterior distribution effectively, enabling inference within a Bayesian context.ResultsThe study's findings show that Moran and Geary analysis provides a substantial clustering pattern of positive spatial autocorrelation of dengue fever (DF) rates between surrounding districts at a 90% confidence interval. The Local Indicators of Spatial Autocorrelation cluster mapped spatial clusters and outliers based on prevalence rates, while the local Getis‐Ord G displayed a thorough breakdown of high or low rates, omitting outliers. Although Chattogram had the most dengue cases (15,752), Khulna district had a higher prevalence rate (133.636) than Chattogram (104.796). The BYM2 model, determined to be well‐fitted based on the lowest Deviance Information Criterion value (527.340), explains a significant association between spatial heterogeneity and prevalence rates.ConclusionThis research pinpoints the district with the highest prevalence rate for dengue and the neighboring districts that also have high risk, allowing government agencies and communities to take the necessary precautions to mollify the risk effect of DF.

Conditional seasonal markov-switching autoregressive model to simulate extreme events: Application to river flow

Extreme events have the potential to significantly impact transportation infrastructure performance. For example, in the case of bridges, climate change impacts the river discharge, hence scouring patterns, which in turn, affects the bridge foundation stability. Therefore, extreme events (river flow) forecasting is mandatory in bridge reliability analysis. This paper approaches this river flow forecasting problem by developing a Markov-Switching Autoregressive model coupled with a conditional hidden seasonal Markov component. In addition, the proposed model is also combined with the deep machine learning neural networks method to forecast river flow from a dataset or from simulations. The proposed method is illustrated by using realistic data: historic river flow values of the Thames River. The results indicate that the proposed model well represented the extreme events within the dataset. In terms of river flow forecasting, the results indicate that the forecasts improve when the training period changes from 20 years to 40 years.

Colombian Ocular Diseases Epidemiology Study (CODES): Prevalence Incidence, and Sociodemographic Characterization of Vision Impairment (2015–2019), A Nationwide Study

ABSTRACT Purpose To describe the prevalence, incidence, and sociodemographic characterization of moderate to severe vision impairment (MSVI) and blindness in Colombia based on the National Health Registry Integrated Social Protection Information System (SISPRO) Database. Methods We performed a nationwide population-based study using SISPRO and the International Classification of Diseases. Total and new cases were identified to calculate the prevalence and incidence per 100,000 inhabitants of MSVI and blindness between 2015 and 2019. Blindness was defined as a best-corrected visual acuity (BCVA) of less than 20/400 in the better-seeing eye. Meanwhile, MSVI is a BCVA from 20/70 to equal or better than 20/400 in the better-seeing eye. An ANOVA test was performed to identify age differences. A conditional autoregressive model was also employed to depict standardized morbidity rate maps. Results From the 50 million inhabitants, the average prevalence and incidence of MSVI were 13.94 and 13.34 between 2015 and 2019, respectively, while for blindness, they were 4.03 and 3.53. Females accounted for most reported cases, and there was a notable shift towards individuals over 50 years (p < 0.001). Valle del Cauca was the region with the most cases reported and the greatest disease burden. Conclusion This is the first nationwide population-based study describing the prevalence, incidence, and sociodemographic characterization of blindness and MSVI in Colombia. In recent years, there has been an increased number of cases, prevalence, and incidence, with females over 50 particularly affected. This research provides insight into the country’s vision impairment epidemiology landscape and contributes to formulating public health policies to improve eye health care.

Powering bikeshare in New York City: does the usage of e-bikes differ from regular bikes?

ABSTRACT In this study, we investigate the difference between shared electric bicycles (e-bikes) and conventional shared bikes operated by Citi Bike in New York City. We examine differences in usage by examining summary statistics and we develop conditional autoregressive models to examine differences in factors associated with trip generation. These factors include bike infrastructure, subway proximity, area demographics, land use, and elevation. We also control for border effects by identifying peripheral stations. The percentage of e-bike trips fluctuates across hours of the day, with a peak around 3 PM but diminishes during peak commuting hours. We find mostly no significant differences in factors associated with the generation of trips between the two modes. Exceptions include a larger effect on e-bike trip generation when the daytime worker population is larger; other differences are due to the racial composition around the station area, some land use categories, and elevation differentials.

Spatial analysis of outdoor indecent assault risk: a study using ambient population data

Spatiotemporal data on ambient populations have recently become widely available. Although previous studies have indicated a link between the spatial patterns of crime occurrence and ambient population distribution, more detailed information, such as the population most likely to be victims by gender and age group, could better predict the risk of crime occurrence. Therefore, this study aimed to analyze the risk of indecent assault, a typical crime with a high number of young female victims, in southern Kyoto Prefecture. We utilized population distribution by gender and age group at different times of the day. After extracting daily patterns (factors) of the population using non-negative matrix factorization, we statistically modeled the risk of indecent assault using a spatial conditional autoregressive model. The results showed that the model, which considered a spatiotemporal ambient population, demonstrated superior performance during nighttime hours. Furthermore, by interpreting the factors significantly associated with the risk of crime occurrence, the findings provided valuable insights into local crime prevention measures that consider daily temporal changes in the gender and age-group composition of individuals present in a specific area.

Restricted spatial models for the analysis of geographic and racial disparities in the incidence of low birthweight in Pennsylvania

The incidence of low birthweight is a common measure of public health due to the increased risk of complications associated with infants having low and very low birthweights. Moreover, many factors that increase the risk of an infant having a low birthweight can be linked to the mother’s socioeconomic status, leading to large racial/ethnic disparities in its incidence. Our objective is thus to analyze the incidence of low and very low birthweight in Pennsylvania counties by race/ethnicity. Due to the small number of births in many Pennsylvania counties when stratified by race/ethnicity, our methods must walk a fine line: While we wish to leverage spatial structure to improve the precision of our estimates, we also wish to avoid oversmoothing the data, which can yield spurious conclusions. As such, we develop a framework by which we can measure (and control) the informativeness of our spatial model. To analyze the data, we first model the Pennsylvania birth data using the conditional autoregressive model to demonstrate the extent to which it can lead to oversmoothing. We then reanalyze the data using our proposed framework and highlight its ability to detect (or not detect) evidence of racial/ethnic disparities in the incidence of low birthweight.

A Bayesian Hierarchical Spatial Longitudinal Model Improves Estimation of Local Macular Rates of Change in Glaucomatous Eyes.

Demonstrate that a novel Bayesian hierarchical spatial longitudinal (HSL) model improves estimation of local macular ganglion cell complex (GCC) rates of change compared to simple linear regression (SLR) and a conditional autoregressive (CAR) model. We analyzed GCC thickness measurements within 49 macular superpixels in 111 eyes (111 patients) with four or more macular optical coherence tomography scans and two or more years of follow-up. We compared superpixel-patient-specific estimates and their posterior variances derived from the latest version of a recently developed Bayesian HSL model, CAR, and SLR. We performed a simulation study to compare the accuracy of intercept and slope estimates in individual superpixels. HSL identified a significantly higher proportion of significant negative slopes in 13/49 superpixels and a significantly lower proportion of significant positive slopes in 21/49 superpixels than SLR. In the simulation study, the median (tenth, ninetieth percentile) ratio of mean squared error of SLR [CAR] over HSL for intercepts and slopes were 1.91 (1.23, 2.75) [1.51 (1.05, 2.20)] and 3.25 (1.40, 10.14) [2.36 (1.17, 5.56)], respectively. A novel Bayesian HSL model improves estimation accuracy of patient-specific local GCC rates of change. The proposed model is more than twice as efficient as SLR for estimating superpixel-patient slopes and identifies a higher proportion of deteriorating superpixels than SLR while minimizing false-positive detection rates. The proposed HSL model can be used to model macular structural measurements to detect individual glaucoma progression earlier and more efficiently in clinical and research settings.

Spatial distribution of arboviruses and its association with a social development index and the waste disposal in São Luís, state of Maranhão, Brazil, 2015 to 2019.

To detect spatial and spatiotemporal clusters of urban arboviruses and to investigate whether the social development index (SDI) and irregular waste disposal are related to the coefficient of urban arboviruses detection in São Luís, state of Maranhão, Brazil. The confirmed cases of Dengue, Zika and Chikungunya in São Luís, from 2015 to 2019, were georeferenced to the census tract of residence. The Bayesian Conditional Autoregressive regression model was used to identify the association between SDI and irregular waste disposal sites and the coefficient of urban arboviruses detection. The spatial pattern of arboviruses pointed to the predominance of a low-incidence cluster, except 2016. For the years 2015, 2016, 2017, and 2019, an increase of one unit of waste disposal site increased the coefficient of arboviruses detection in 1.25, 1.09, 1.23, and 1.13 cases of arboviruses per 100 thousand inhabitants, respectively. The SDI was not associated with the coefficient of arboviruses detection. In São Luís, spatiotemporal risk clusters for the occurrence of arboviruses and a positive association between the coefficient of arbovirus detection and sites of irregular waste disposal were identified.

Accounting conservatism, timeliness and interactions in the Scandinavian stock markets

PurposeThe purpose of this study is to examine within a unified framework the timeliness and conservatism of accounting disclosure accommodating the transmission of news among the Scandinavian stock markets.Design/methodology/approachTo this end the authors have used an augmented ordinary least squares (OLS) approach and univariate generalized autoregressive conditional heteroskedastic and vector autoregressive (VAR) modeling. The sample covers the period from 1987 to 2020, totaling 1452 observations. The sample was collected from the datastream database.FindingsThe empirical results of this study are consistent with previous findings and provide evidence that accounting reporting is timely and conservative while news is transmitted amongst the Scandinavian stock markets.Practical implicationsThe findings could be important for investors, firms and regulators since failure of considering information that is derived from more advanced approaches could result in lower quality of annual reports of companies.Originality/valueThe authors examined the relationship between earnings yield and conditional risk using an augmented OLS model and the transmission of news among Scandinavian stock markets using a VAR model.

Understanding the spread of infectious diseases in edge areas of hotspots: dengue epidemics in tropical metropolitan regions

Identifying clusters or hotspots from disease maps is critical in research and practice. Hotspots have been shown to have a higher potential for transmission risk and may be the source of infections, making them a priority for controlling epidemics. However, the role of edge areas of hotspots in disease transmission remains unclear. This study aims to investigate the role of edge areas in disease transmission by examining whether disease incidence rate growth is higher in the edges of disease hotspots during outbreaks. Our data is based on the three most severe dengue epidemic years in Kaohsiung city, Taiwan, from 1998 to 2020. We employed conditional autoregressive (CAR) models and Bayesian areal Wombling methods to identify significant edge areas of hotspots based on the extent of risk difference between adjacent areas. The difference-in-difference (DID) estimator in spatial panel models measures the growth rate of risk by comparing the incidence rate between two groups (hotspots and edge areas) over two time periods. Our results show that in years characterized by exceptionally large-scale outbreaks, the edge areas of hotspots have a more significant increase in disease risk than hotspots, leading to a higher risk of disease transmission and potential disease foci. This finding explains the geographic diffusion mechanism of epidemics, a pattern mixed with expansion and relocation, indicating that the edge areas play an essential role. The study highlights the importance of considering edge areas of hotspots in disease transmission. Furthermore, it provides valuable insights for policymakers and health authorities in designing effective interventions to control large-scale disease outbreaks.

Ecological drivers for poultry farms predisposed to highly pathogenic avian influenza virus infection during the initial phase of the six outbreaks between 2010-2021: a nationwide study in South Korea.

Highly pathogenic avian influenza (HPAI) has caused substantial economic losses worldwide. An understanding of the environmental drivers that contribute to spillover transmission from wild birds to poultry farms is important for predicting areas at risk of introduction and developing risk-based surveillance strategies. We conducted an epidemiological study using data from six HPAI outbreak events in South Korea. An aggregate-level study design was implemented using third-level administrative units in South Korea. Only regions with high natural reservoir suitability were included. The incidence of HPAI at chicken and duck farms during the initial phase (30 and 45 days after the first case) of each outbreak event was used as the outcome variable, assuming that cross-species transmission from wild birds was the dominant exposure leading to infection. Candidate environmental drivers were meteorological factors, including temperature, precipitation, humidity, and altitude, as well as the proportion of protected area, farm density, deforestation level, and predator species richness. Logistic regression models were implemented; conditional autoregression models were used in cases of spatial autocorrelation of residuals. Lower temperature, higher farm density, and lower predator species richness were significantly associated with a higher risk of HPAI infection on chicken farms. Lower temperature, higher proportion of protected area, and lower predator species richness were significantly associated with a higher risk of HPAI infection on duck farms. The predicted dominant transmission routes on chicken and duck farms were horizontal and spillover, respectively. These results reveal a potential protective effect of predator species richness against HPAI outbreaks. Further studies are required to confirm a causal relationship.

Factors associated with different types of freight crashes: A macro-level analysis

Introduction: Despite evidence showing higher fatality rates in freight-related crashes, there has been limited exploration of their spatial distribution and factors associated with such distribution. This gap in the literature primarily stems from the focus of existing studies on micro-level factors predicting the frequency or severity of injuries in freight crashes. The present study delves into the factors contributing to freight crashes at the neighborhood level, particularly focusing on different types of freight crashes: collisions involving a freight vehicle and a passenger vehicle, crashes between freight vehicles, and freight vehicle–non-motorized crashes. Method: This study analyzes traffic crash data from the urbanized region of Seoul, collected between 2016 and 2019. To effectively deal with spatial autocorrelation and model different types of crashes in a unified framework, a Bayesian multivariate conditional autoregressive model was employed. Results: Findings show substantial differences in the factors associated with various types of freight crashes. The predictors for crashes between freight vehicles diverge significantly from those for freight vehicle–non-motorized crashes. Crashes between freight vehicles are relatively more influenced by road network structure, while freight crashes involving non-motorized users are relatively more affected by the built environment and freight facilities than the other crash types examined. Freight vehicle–passenger vehicle crashes fall into an intermediate category, sharing most predictors with either of the other two types of freight crashes. Conclusions and Practical Applications: The findings of this study offer valuable lessons for transportation practitioners and policymakers. They can guide the formulation of effective land use policies and infrastructure planning, specifically designed to address the unique characteristics of different types of freight crashes.

A Bayesian spatio-temporal study of the association between meteorological factors and the spread of COVID-19

BackgroundThe spread of COVID-19 has brought challenges to health, social and economic systems around the world. With little to no prior immunity in the global population, transmission has been driven primarily by human interaction. However, as with common respiratory illnesses such as influenza some authors have suggested COVID-19 may become seasonal as immunity grows. Despite this, the effects of meteorological conditions on the spread of COVID-19 are poorly understood. Previous studies have produced contrasting results, due in part to limited and inconsistent study designs.MethodsThis study investigates the effects of meteorological conditions on COVID-19 infections in England using a Bayesian conditional auto-regressive spatio-temporal model. Our data consists of daily case counts from local authorities in England during the first lockdown from March–May 2020. During this period, legal restrictions limiting human interaction remained consistent, minimising the impact of changes in human interaction. We introduce a lag from weather conditions to daily cases to accommodate an incubation period and delays in obtaining test results. By modelling spatio-temporal random effects we account for the nature of a human transmissible virus, allowing the model to isolate meteorological effects.ResultsOur analysis considers cases across England’s 312 local authorities for a 55-day period. We find relative humidity is negatively associated with COVID-19 cases, with a 1% increase in relative humidity corresponding to a reduction in relative risk of 0.2% [95% highest posterior density (HPD): 0.1–0.3%]. However, we find no evidence for temperature, wind speed, precipitation or solar radiation being associated with COVID-19 spread. The inclusion of weekdays highlights systematic under reporting of cases on weekends with between 27.2–43.7% fewer cases reported on Saturdays and 26.3–44.8% fewer cases on Sundays respectively (based on 95% HPDs).ConclusionBy applying a Bayesian conditional auto-regressive model to COVID-19 case data we capture the underlying spatio-temporal trends present in the data. This enables us to isolate the main meteorological effects and make robust claims about the association of weather variables to COVID-19 incidence. Overall, we find no strong association between meteorological factors and COVID-19 transmission.

Spatio-temporal dynamics of visceral leishmaniasis in Brazil: A nonlinear regression analysis.

This study describes the spatio-temporal dynamics of new visceral leishmaniasis (VL) cases notified in Brazil between 2001 and 2020. Data on the occurrence of the disease were obtained by means of the Notifiable Diseases Information System of the Brazilian Ministry of Health. Joinpoint, temporal generalized additive models and conditional autoregressive (CAR) models were used to analyse the temporal evolution of the rates in Brazil, states and regions. Spatio-temporal generalized additive and CAR models were used to identify the distribution of annual risks of VL occurrence in the Brazilian territory in relation to variation in the spatial average. There were 63,966 VL cases in the target period (3.198 cases/year), corresponding to a mean incidence rate of 1.68 cases/100,000 inhabitants. Of these, 4451 resulted in deaths, which gives a mean mortality rate of 0.12 deaths/100,000 inhabitants and a case fatality of 6.96%. The highest incidence rate was found in the North region, followed closely by the Northeast region, which presented the second and first highest mortality rates, respectively. For all of Brazil, and in the Northeast region, there were stability in the incidence rates, while the other regions showed an increasing trend in different time segments in the period: Central-West up to 2011, North up to 2008, Southeast up to 2004, and South up to 2010. On the other hand, all regions experienced a reduction in incidence rate during the last years of the series. The Northeast region had the highest number of municipalities with statistically significant elevated relative risks. The spatio-temporal analysis showed the highest risk area predominantly in the Northeast region in the beginning of the time series. From 2002 to 2018, this area expanded to the interior of the country. The present study has shown that VL has expanded in Brazil. However, the North and Northeast regions continue to have the highest incidence, and the risk of infection has decreased in recent years.

Association between greenspace morphology and prevalence of non-communicable diseases mediated by air pollution and physical activity

Neighborhood greenspace is a recognized roborant, yet limited is known regarding the influence of greenspace spatial morphology on the prevalence of non-communicable diseases. Evidence is also scarce as to whether such associations are generalizable. Using census tracts, we examined these relationships in Los Angeles and explored the generalizability of the discovered associations in New York City, San Antonio, Seattle, and Miami, all major metropolitan areas in the US. We used high-resolution satellite imagery to generate greenspace maps, and calculated metrics to measure the average size, fragmentation, connectedness, aggregation, and shape of greenspace. The prevalence of poor mental health, heart disease, stroke, diabetes, COPD, and lack of leisure time physical activity was examined. Conditional spatial autoregression models were used, controlling for geographic, demographic, and socioeconomic factors. Neighborhoods with more connected, aggregated, coherent, and complex-shaped greenspace morphologies had a lower prevalence of non-communicable diseases. Such associations are mediated by air pollution and physical inactivity. A one-interquartile increase in greenspace aggregation in Los Angles was associated with a decreasing prevalence of poor mental health, heart disease, stroke, diabetes, physical inactivity, and COPD by 1.19%, 0.53%, 0.44%, 1.63%, 3.53%, and 0.62%, respectively. A one-interquartile increase in greenspace average size, shape complexity, cohesion, aggregation, and decrease in fragmentation associated with a decreasing prevalence of poor mental health in New York City by 0.12%, 4.29%, 0.68%, 0.44%, and 2.87%, respectively. We offer evidence that the location, shape, and connectivity of urban greenspace have a significant and independent effect on health at the neighborhood level above and beyond that meliorated by sheer amount. The spatial morphology of designed urban greenspace should therefore be incorporated into routine environment-based health planning.

A systematic approach to macro-level safety assessment and contributing factors analysis considering traffic crashes and violations

During rapid urbanization and increase in motorization, it becomes particularly important to understand the relationships between traffic safety and risk factors in order to provide targeted improvements and policy recommendations. Violations and police enforcement are key variables, but the endogenous relationship between crashes and violations has made these variables unreliable and has limited their use. To manage this problem, this study developed a systematic approach for the joint modeling of crashes and violations to identify crash and violation hotspots and examine the mechanisms underlying macro-level contributing factors. Socio-economic, road network, public facility, traffic enforcement, and land use intensity data from 115 towns in Suzhou, China, were collected as independent variables. A bivariate negative binomial spatial conditional autoregressive model (BNB-CAR) and the potential for safety improvement (PSI) method were adopted to identify crash-prone and violation-prone areas, and an interpretable machine learning framework was applied to explore the factors’ effects by area. Results showed that the proposed framework was able to accurately identify problem areas and quantify the impact of key factors, which, in Suzhou, were the number of traffic police and their daily patrol time. Considering such enforcement-related information provided important insights into reducing crash and violation frequency; for example, keeping the number of traffic police and daily patrol time under certain thresholds (number of police lower than 11 and patrol time lower than 2.3 h in this sample) was as effective as increasing these numbers for reducing the probability of high-crash and high-violation areas. The proposed approach can help traffic administrators identify the key contributing factors, especially enforcement factors, in crash-prone and violation-prone areas and provide guidelines for improvement.

Impact and Equity of New York City's Tobacco Retail Reduction Initiative

High levels of tobacco retailer density in communities is associated with a range of tobacco use behaviors and is a key structural driver of tobacco-related disparities. This study evaluates the impacts of New York City's (NYC) novel policy intervention to cap tobacco retail licenses on tobacco retailer density levels and neighborhood inequities in tobacco access. Using geocoded tobacco retail licensing data from 2010 to 2022, Bayesian conditional autoregressive Poisson panel models estimated the association between policy implementation in 2018 and retailer density per 1,000 population, controlling for neighborhood-level sociodemographic factors. Data were analyzed in 2023. The number of tobacco retail licenses decreased from 9,304 in 2010 to 5,107 in 2022, with the rate of decline significantly accelerating post-policy (-14·2% versus -34·2%). Policy effects were stronger in districts with lower income and greater proportions of non-Hispanic Black residents. NYC's policy substantially reduced tobacco retailer density and appeared to close longstanding patterns of inequity in tobacco access, serving as a rare example of a tobacco control policy that may effectively reduce tobacco-related disparities. This emergent approach to restructure tobacco retail in communities may reach populations that have not benefitted from traditional tobacco control policies and should be considered by other localities.

Understanding the importance of spatial correlation in identifying spatio-temporal variation of disease risk, in the case of malaria risk mapping in southern Ethiopia

Malaria remains a major health problem in developing countries despite a significant reduction in incidence in the last few years. Disease mapping thus helps to understand the spatial pattern and identify areas characterized by unusual risks. Several spatial models have been used to analyze the incidence of malaria. We aim to compare the predictive performance of these models and investigate the effect of ignoring spatial correlation. The reported malaria case counts of genus P.falciparum in 149 districts of southern Ethiopia from January 2016 to May 2019 were analyzed using the spatial time series model (STS) that ignores spatial correlation, Spatio-temporal conditional autoregressive model (STCAR), Spatio-temporal geostatistical model (STG) and Spatio-temporal spatial discrete approximation to log Gaussian cox process (STSDALGCP). We assess the predictive performance of the models using root mean square error, mean absolute error, and coverage probability. We found that monthly average rainfall, temperature, humidity, and EVI are significantly associated with malaria risk. The spatial variation of malaria incidence changes with time, in particular, the high incidence was observed from November to December, months after heavy rainfall, and more pronounced in the southwest of the country. STSDALGCP gives a small prediction error in test set and captures the uncertainties better than other models, while the STS model gives a high prediction error. Accounting for spatial correlation is crucial for disease risk mapping and leads to better prediction of disease risk. Since malaria transmission operates in a spatially continuous manner, a spatially continuous model should be considered when it is computationally feasible.

Controlling for spatial confounding and spatial interference in causal inference: modelling insights from a computational experiment

ABSTRACT Causal inference is a rapidly growing field of statistics that applies logical reasoning to statistical inference to estimate causal relationships. Spatial data poses several problems in causal inference – namely, spatial confounding and interference – that require different strategies when designing causal models. In order to obtain valid inferences, existing nonspatial causal models must adjust for such spatial problems. Given the blossoming literature on spatial causal inference, this research analyzes the usage of spatial causal models under a priori knowledge and a priori ignorance of the spatial structure of data. We synthesize existing research directions in noncausal spatial modelling and causal nonspatial modelling by assessing the performance of 28 spatial causal models across 16 spatial data scenarios. We used ordinary least squares (OLS) models, conditional autoregressive (CAR) models, and jointly CAR models for outcome and treatment variables as the basis for the tested models, equipping them with a variety of spatial causal adjustments. We compare our results to principles of noncausal spatial modelling and investigate their implications for spatial causal modelling. Specifically, we show that noncausal spatial modelling guidance holds in causal spatial modelling workflows and demonstrate how researchers can leverage noncausal theory to great effect. In parallel, we introduce the spycause Python package of spatial causal models and data simulators to facilitate the widespread use of these models and to enable reproduction and extension of our work.

Are ride-hailing services safer than taxis? A multivariate spatial approach with accommodation of exposure uncertainty

Despite many research efforts on ride-hailing services and taxis, limited studies have compared the safety performance of the two modes. A major challenge is the need for reliable mode-specific exposure data to model their safety outcomes. Moreover, crash frequencies of the two modes by injury severities tend to be spatially and inherently correlated. To fully address these issues, this study proposes a novel multivariate conditional autoregressive model considering measurement errors in mode-specific exposures (MVCARME). More specially, a classical measurement error structure accommodates the uncertainty of estimated mode-specific exposures, and a multivariate spatial specification is adopted to capture potential spatial and inherent correlations. The model estimation is accelerated by an integrated nest Laplace approximation method. The census tracts in the city of Chicago are set as the spatial analysis unit. The mode-specific exposures (vehicle-mile-traveled) in each census tract are estimated by trip assignments using ride-hailing and taxi trip data in 2019. The modeling results indicate that both ride-hailing crashes and taxi crashes are positively associated with transportation factors (e.g., vehicle-mile-traveled, mode-specific vehicle-mile-traveled, and traffic signal numbers), land use factors (i.e., number of educational and alcohol-related sites), and demographic factors (e.g., median household income, transit ratio, and walk ratio). By comparison, the proposed model outperforms the others (i.e., negative binomial models and multivariate conditional autoregressive model) by yielding the lowest deviance information criterion (DIC), Watanabe-Akaike information criterion (WAIC), mean absolute error (MAE), and root-mean-square error (RMSE). According to the results of t-tests, ride-hailing services are found to be prone to a higher risk of minor injury crashes compared with taxis, despite no significant difference between the risks of severe injury crashes. Methodologically, this study adds a robust safety evaluation approach for comparing crash risks of different modes to the literature. At the same time, practically, it provides researchers, practitioners, and policy-makers insights into the safety management of various mobility alternatives.