UVM Theses and Dissertations
Format:
Print
Author:
Liu, Qiong
Dept./Program:
Civil and Environmental Engineering
Year:
2013
Degree:
MS
Abstract:
The thesis consists of two journal papers to explore spatial phenomenon from different perspectives. Due to the lack of spatial focus in transportationrelated literature, both papers attempt to bring in spatial concepts and apply quantitative spatial methodologies. Using data from a New England transportation survey, two papers conducted spatial analysis and contributed to the body of transportation literature on spatial exploration.
The first paper examines Spatial Auto-correlation (SA) in rates of obesity and walking. From the survey data, Body Mass Index (BMI) and walking are derived as study variables. By dividing the dataset into strata by individual states and rural/urban areas, spatial scale effects are accounted for. As the widely applied method in examining SA, Moran's I shows inconsistency for the same stratum using different weight matrices, the study introduces semi-variogram analysis to address local spatial variation. Not surprisingly, some scales find SA while others do not, indicating that spatial scale matters. For scenarios with positive SA, the study predicts obesity and walking rates using GIS kriging based on semi-variogram parameters and compares them with actual values to validate the estimation. The predictions turn out to be good. Lastly, Moran's I is recalculated by using semi-variogram parameters. The study reinforces that semivariogram analysis is effective in examining SA, and Moran's I could be improved with its help.
The second paper examines the predicators of walking rates using spatial regression methods. Based on a Vermont subset of the transportation survey, walking variables and 170 independent variables are derived as dependent and independent variables respectively. Among independent variables, people's life style and perception on the built environment are included, which were rarely addressed in the existing literature. A linear regression model is first established as the base model, followed by spatial regression models. Comparison between these models does not find significant differences. The study concludes that it is safe to apply traditional non-spatial statistics on the relationship between walking and its predictors. But with reservations, we suggest researchers examine the existence of spatial effects for studies where observations are defined by location.
The first paper examines Spatial Auto-correlation (SA) in rates of obesity and walking. From the survey data, Body Mass Index (BMI) and walking are derived as study variables. By dividing the dataset into strata by individual states and rural/urban areas, spatial scale effects are accounted for. As the widely applied method in examining SA, Moran's I shows inconsistency for the same stratum using different weight matrices, the study introduces semi-variogram analysis to address local spatial variation. Not surprisingly, some scales find SA while others do not, indicating that spatial scale matters. For scenarios with positive SA, the study predicts obesity and walking rates using GIS kriging based on semi-variogram parameters and compares them with actual values to validate the estimation. The predictions turn out to be good. Lastly, Moran's I is recalculated by using semi-variogram parameters. The study reinforces that semivariogram analysis is effective in examining SA, and Moran's I could be improved with its help.
The second paper examines the predicators of walking rates using spatial regression methods. Based on a Vermont subset of the transportation survey, walking variables and 170 independent variables are derived as dependent and independent variables respectively. Among independent variables, people's life style and perception on the built environment are included, which were rarely addressed in the existing literature. A linear regression model is first established as the base model, followed by spatial regression models. Comparison between these models does not find significant differences. The study concludes that it is safe to apply traditional non-spatial statistics on the relationship between walking and its predictors. But with reservations, we suggest researchers examine the existence of spatial effects for studies where observations are defined by location.