You are here
Macroscopic Crash Analysis and Its Implications for Transportation Safety Planning
- Date Issued:
- 2012
- Abstract/Description:
- Incorporating safety into the transportation planning stage, which is often termed as transportation safety planning (TSP), relies on the vital interplay between zone characteristics and zonal traffic crashes. Although a few safety studies had made some effort towards integrating safety and planning, several unresolved problems and a complete framework of TSP are still absent in the literature. This research aims at examining the suitability of the current traffic-related zoning planning process in a new suggested planning method which incorporates safety measures. In order to accomplish this broader research goal, the study defined its research objectives in the following directions towards establishing a framework of TSP- i) exploring the existing key determinants in traditional transportation planning (e.g., trip generation/distribution data, land use types, demographics, etc.) in order to develop an effective and efficient TSP framework, ii) investigation of the Modifiable Aerial Unit Problem (MAUP) in the context of macro-level crash modeling to investigate the effect of the zone's size and boundary, iii) understanding neighborhood influence of the crashes at or near zonal boundaries, and iv) development of crash-specific safety measure in the four-step transportation planning process.This research was conducted using spatial data from the counties of West Central Florida. Analysis of different crash data per spatial unit was performed using nonparametric approaches (e.g., data mining and random forest), classical statistical methods (e.g., negative binomial models), and Bayesian statistical techniques. In addition, a comprehensive Geographic Information System (GIS) based application tools were utilized for spatial data analysis and representation.Exploring the significant variables related to specific types of crashes is vital in the planning stages of a transportation network. This study identified and examined important variables associated with total crashes and severe crashes per traffic analysis zone (TAZ) by applying nonparametric statistical techniques using different trip related variables and road-traffic related factors. Since a macro-level analysis, by definition, will necessarily involve aggregating crashes per spatial unit, a spatial dependence or autocorrelation may arise if a particular variable of a geographic region is affected by the same variable of the neighboring regions. So far, few safety studies were performed to examine crashes at TAZs and none of them explicitly considered spatial effect of crashes occurring in them. In order to understand the clear picture of spatial autocorrelation of crashes, this study investigated the effect of spatial autocorrelation in modeling pedestrian and bicycle crashes in TAZs. Additionally, this study examined pedestrian crashes at Environmental Justice (EJ) TAZs which were identified in compliance with the various ongoing practices undertaken by Metropolitan Planning Organizations (MPOs) and previous research. Minority population and the low-income group are two important criteria based on which EJ areas are being identified. These unique areal characteristics have been of particular interest to the traffic safety analysts in order to investigate the contributing factors of pedestrian crashes in these deprived areas. Pedestrian and bicycle crashes were estimated as a function of variables related to roadway characteristics, and various demographic and socio-economic factors. It was found that significant differences are present between the predictor sets for pedestrian and bicycle crashes. In all cases the models with spatial correlation performed better than the models that did not account for spatial correlation among TAZs. This finding implied that spatial correlation should be considered while modeling pedestrian and bicycle crashes at the aggregate or macro-level. Also, the significance of spatial autocorrelation was later found in the total and severe crash analyses and accounted for in their respective modeling techniques.Since the study found affirmative evidence about the inclusion of spatial autocorrelation in the safety performance functions, this research considered identifying appropriate spatial entity based on which TSP framework would be developed. A wide array of spatial units has been explored in macro-level crash modeling in previous safety research. With the advancement of GIS, safety analysts are able to analyze crashes for various geographical units. However, a clear guideline on which geographic entity should a modeler choose is not present so far. This preference of spatial unit can vary with the dependent variable of the model. Or, for a specific dependent variable, models may be invariant to multiple spatial units by producing a similar goodness-of-fits. This problem is closely related to the Modifiable Areal Unit Problem which is a common issue in spatial data analysis. The study investigated three different crash (total, severe, and pedestrian) models developed for TAZs, block groups (BGs) and census tracts (CTs) using various roadway characteristics and census variables (e.g., land use, socio-economic, etc.); and compared them based on multiple goodness-of-fit measures.Based on MAD and MSPE it was evident that the total, severe and pedestrian crash models for TAZs and BGs had similar fits, and better than the ones developed for CTs. This indicated that the total, severe and pedestrian crash models are being affected by the size of the spatial units rather than their zoning configurations. So far, TAZs have been the base spatial units of analyses for developing travel demand models. Metropolitan planning organizations widely use TAZs in developing their long range transportation plans (LRTPs). Therefore, considering the practical application it was concluded that as a geographical unit, TAZs had a relative ascendancy over block group and census tract.Once TAZs were selected as the base spatial unit of the TSP framework, careful inspections on the TAZ delineations were performed. Traffic analysis zones are often delineated by the existing street network. This may result in considerable number of crashes on or near zonal boundaries. While the traditional macro-level crash modeling approach assigns zonal attributes to all crashes that occur within the zonal boundary, this research acknowledged the inaccuracy resulting from relating crashes on or near the boundary of the zone to merely the attributes of that zone. A novel approach was proposed to account for the spatial influence of the neighboring zones on crashes which specifically occur on or near the zonal boundaries. Predictive model for pedestrian crashes per zone were developed using a hierarchical Bayesian framework and utilized separate predictor sets for boundary and interior (non-boundary) crashes. It was found that these models (that account for boundary and interior crashes separately) had better goodness-of-fit measures compared to the models which had no specific consideration for crashes located at/near the zone boundaries. Additionally, the models were able to capture some unique predictors associated explicitly with interior and boundary-related crashes. For example, the variables- 'total roadway length with 35mph posted speed limit' and 'long term parking cost' were statistically not significantly different from zero in the interior crash model but they were significantly different from zero at the 95% level in the boundary crash model.Although an adjacent traffic analysis zones (a single layer) were defined for pedestrian crashes and boundary pedestrian crashes were modeled based on the characteristic factors of these adjacent zones, this was not considered reasonable for bicycle-related crashes as the average roaming area of bicyclists are usually greater than that of pedestrians. For smaller TAZs sometimes it is possible for a bicyclist to cross the entire TAZ. To account for this greater area of coverage, boundary bicycle crashes were modeled based on two layers of adjacent zones. As observed from the goodness-of-fit measures, performances of model considering single layer variables and model considering two layer variables were superior from the models that did not consider layering at all; but these models were comparable. Motor vehicle crashes (total and severe crashes) were classified as 'on-system' and 'off-system' crashes and two sub-models were fitted in order to calibrate the safety performance function for these crashes. On-system and off-system roads refer to two different roadway hierarchies. On-system or state maintained roads typically possess higher speed limit and carries traffic from distant TAZs. Off-system roads are, however, mostly local roads with relatively low speed limits. Due to these distinct characteristics, on-system crashes were modeled with only population and total employment variables of a zone in addition to the roadway and traffic variables; and all other zonal variables were disregarded. For off-system crashes, on contrary, all zonal variables was considered. It was evident by comparing this on- and off-system sub-model-framework to the other candidate models that it provided superior goodness-of-fit for both total and severe crashes.Based on the safety performance functions developed for pedestrian, bicycle, total and severe crashes, the study proposed a novel and complete framework for assessing safety (of these crash types) simultaneously in parallel with the four-step transportation planning process with no need of any additional data requirements from the practitioners' side.
Title: | Macroscopic Crash Analysis and Its Implications for Transportation Safety Planning. |
145 views
126 downloads |
---|---|---|
Name(s): |
Siddiqui, Chowdhury, Author Abdel-Aty, Mohamed, Committee Chair Abdel-Aty, Mohamed, Committee Member Uddin, Nizam, Committee Member Huang, Helai, Committee Member , Committee Member University of Central Florida, Degree Grantor |
|
Type of Resource: | text | |
Date Issued: | 2012 | |
Publisher: | University of Central Florida | |
Language(s): | English | |
Abstract/Description: | Incorporating safety into the transportation planning stage, which is often termed as transportation safety planning (TSP), relies on the vital interplay between zone characteristics and zonal traffic crashes. Although a few safety studies had made some effort towards integrating safety and planning, several unresolved problems and a complete framework of TSP are still absent in the literature. This research aims at examining the suitability of the current traffic-related zoning planning process in a new suggested planning method which incorporates safety measures. In order to accomplish this broader research goal, the study defined its research objectives in the following directions towards establishing a framework of TSP- i) exploring the existing key determinants in traditional transportation planning (e.g., trip generation/distribution data, land use types, demographics, etc.) in order to develop an effective and efficient TSP framework, ii) investigation of the Modifiable Aerial Unit Problem (MAUP) in the context of macro-level crash modeling to investigate the effect of the zone's size and boundary, iii) understanding neighborhood influence of the crashes at or near zonal boundaries, and iv) development of crash-specific safety measure in the four-step transportation planning process.This research was conducted using spatial data from the counties of West Central Florida. Analysis of different crash data per spatial unit was performed using nonparametric approaches (e.g., data mining and random forest), classical statistical methods (e.g., negative binomial models), and Bayesian statistical techniques. In addition, a comprehensive Geographic Information System (GIS) based application tools were utilized for spatial data analysis and representation.Exploring the significant variables related to specific types of crashes is vital in the planning stages of a transportation network. This study identified and examined important variables associated with total crashes and severe crashes per traffic analysis zone (TAZ) by applying nonparametric statistical techniques using different trip related variables and road-traffic related factors. Since a macro-level analysis, by definition, will necessarily involve aggregating crashes per spatial unit, a spatial dependence or autocorrelation may arise if a particular variable of a geographic region is affected by the same variable of the neighboring regions. So far, few safety studies were performed to examine crashes at TAZs and none of them explicitly considered spatial effect of crashes occurring in them. In order to understand the clear picture of spatial autocorrelation of crashes, this study investigated the effect of spatial autocorrelation in modeling pedestrian and bicycle crashes in TAZs. Additionally, this study examined pedestrian crashes at Environmental Justice (EJ) TAZs which were identified in compliance with the various ongoing practices undertaken by Metropolitan Planning Organizations (MPOs) and previous research. Minority population and the low-income group are two important criteria based on which EJ areas are being identified. These unique areal characteristics have been of particular interest to the traffic safety analysts in order to investigate the contributing factors of pedestrian crashes in these deprived areas. Pedestrian and bicycle crashes were estimated as a function of variables related to roadway characteristics, and various demographic and socio-economic factors. It was found that significant differences are present between the predictor sets for pedestrian and bicycle crashes. In all cases the models with spatial correlation performed better than the models that did not account for spatial correlation among TAZs. This finding implied that spatial correlation should be considered while modeling pedestrian and bicycle crashes at the aggregate or macro-level. Also, the significance of spatial autocorrelation was later found in the total and severe crash analyses and accounted for in their respective modeling techniques.Since the study found affirmative evidence about the inclusion of spatial autocorrelation in the safety performance functions, this research considered identifying appropriate spatial entity based on which TSP framework would be developed. A wide array of spatial units has been explored in macro-level crash modeling in previous safety research. With the advancement of GIS, safety analysts are able to analyze crashes for various geographical units. However, a clear guideline on which geographic entity should a modeler choose is not present so far. This preference of spatial unit can vary with the dependent variable of the model. Or, for a specific dependent variable, models may be invariant to multiple spatial units by producing a similar goodness-of-fits. This problem is closely related to the Modifiable Areal Unit Problem which is a common issue in spatial data analysis. The study investigated three different crash (total, severe, and pedestrian) models developed for TAZs, block groups (BGs) and census tracts (CTs) using various roadway characteristics and census variables (e.g., land use, socio-economic, etc.); and compared them based on multiple goodness-of-fit measures.Based on MAD and MSPE it was evident that the total, severe and pedestrian crash models for TAZs and BGs had similar fits, and better than the ones developed for CTs. This indicated that the total, severe and pedestrian crash models are being affected by the size of the spatial units rather than their zoning configurations. So far, TAZs have been the base spatial units of analyses for developing travel demand models. Metropolitan planning organizations widely use TAZs in developing their long range transportation plans (LRTPs). Therefore, considering the practical application it was concluded that as a geographical unit, TAZs had a relative ascendancy over block group and census tract.Once TAZs were selected as the base spatial unit of the TSP framework, careful inspections on the TAZ delineations were performed. Traffic analysis zones are often delineated by the existing street network. This may result in considerable number of crashes on or near zonal boundaries. While the traditional macro-level crash modeling approach assigns zonal attributes to all crashes that occur within the zonal boundary, this research acknowledged the inaccuracy resulting from relating crashes on or near the boundary of the zone to merely the attributes of that zone. A novel approach was proposed to account for the spatial influence of the neighboring zones on crashes which specifically occur on or near the zonal boundaries. Predictive model for pedestrian crashes per zone were developed using a hierarchical Bayesian framework and utilized separate predictor sets for boundary and interior (non-boundary) crashes. It was found that these models (that account for boundary and interior crashes separately) had better goodness-of-fit measures compared to the models which had no specific consideration for crashes located at/near the zone boundaries. Additionally, the models were able to capture some unique predictors associated explicitly with interior and boundary-related crashes. For example, the variables- 'total roadway length with 35mph posted speed limit' and 'long term parking cost' were statistically not significantly different from zero in the interior crash model but they were significantly different from zero at the 95% level in the boundary crash model.Although an adjacent traffic analysis zones (a single layer) were defined for pedestrian crashes and boundary pedestrian crashes were modeled based on the characteristic factors of these adjacent zones, this was not considered reasonable for bicycle-related crashes as the average roaming area of bicyclists are usually greater than that of pedestrians. For smaller TAZs sometimes it is possible for a bicyclist to cross the entire TAZ. To account for this greater area of coverage, boundary bicycle crashes were modeled based on two layers of adjacent zones. As observed from the goodness-of-fit measures, performances of model considering single layer variables and model considering two layer variables were superior from the models that did not consider layering at all; but these models were comparable. Motor vehicle crashes (total and severe crashes) were classified as 'on-system' and 'off-system' crashes and two sub-models were fitted in order to calibrate the safety performance function for these crashes. On-system and off-system roads refer to two different roadway hierarchies. On-system or state maintained roads typically possess higher speed limit and carries traffic from distant TAZs. Off-system roads are, however, mostly local roads with relatively low speed limits. Due to these distinct characteristics, on-system crashes were modeled with only population and total employment variables of a zone in addition to the roadway and traffic variables; and all other zonal variables were disregarded. For off-system crashes, on contrary, all zonal variables was considered. It was evident by comparing this on- and off-system sub-model-framework to the other candidate models that it provided superior goodness-of-fit for both total and severe crashes.Based on the safety performance functions developed for pedestrian, bicycle, total and severe crashes, the study proposed a novel and complete framework for assessing safety (of these crash types) simultaneously in parallel with the four-step transportation planning process with no need of any additional data requirements from the practitioners' side. | |
Identifier: | CFE0004191 (IID), ucf:49009 (fedora) | |
Note(s): |
2012-05-01 Ph.D. Engineering and Computer Science, Civil, Environmental and Construction Engineering Doctoral This record was generated from author submitted information. |
|
Subject(s): |
transportation safety planning traffic safety engineering crash analysis bayesian analysis |
|
Persistent Link to This Record: | http://purl.flvc.org/ucf/fd/CFE0004191 | |
Restrictions on Access: | campus 2013-05-15 | |
Host Institution: | UCF |