Jump to Main Content
Geographically weighted logistic regression approach to explore the spatial variability in travel behaviour and built environment interactions: Accounting simultaneously for demographic and socioeconomic characteristics
- Nkeki, Felix Ndidi, Asikhia, Monday Ohi
- Applied geography 2019 v.108 pp. 47-63
- data collection, human behavior, regression analysis, socioeconomic factors, spatial variation, statistical models, surveys, travel, Benin
- The relationship between built environment and travel behaviour has been a topical subject of academic debate over the last two decades. This has given rise to a plethora of empirical literature in this area of study. Ultimately, these studies were conducted using statistical models that fail to explain spatial non-stationarity of the processes in their dataset. To improve understanding concerning built environment and travel behaviour interactions, local model against global model is suggested. The aim of this study is to analyze the spatial variation in travel behaviour and built environment interactions using Geographically Weighted Logistic Regression (GWLR) and at the same time accounting for the individual attributes of commuters (e.g., demographic and socioeconomic characteristics). Based on valid responses from 1028 survey points carried out in Benin metropolitan region, a GWLR of travel mode choice was estimated. The result shows that unlike global statistics, local model revealed a significant spatial variation in the association between travel mode choice and the factor scores of demographic and socioeconomic variables across neighbourhoods. GWLR model also revealed the occurrence of spatial mismatch between demographic and socioeconomic characteristics, and this created a dichotomy by demarcating the neighbourhoods into two levels of influence. The result further showed that built environment variables are weak predictors of mode choice in the region. Local model proved to be most suitable for exploring this relationship since it accounted for local variation which is often lost when using global models.