Data integration and analysis of qualitative and quantitative data in active travel

Urban cycling behaviour is shaped by interactions between infrastructure, public perceptions, and local socio-economic context, yet these elements are typically recorded in datasets that differ in format and spatial structure. This thesis develops a coherent and reproducible methodology for integrating quantitative and qualitative cycling-related data to investigate spatial variation in cycling activity and perception across Glasgow. The approach adapts an active travel analytical framework and demonstrates how diverse sources can be aligned, aggregated, and modelled within a unified spatial structure. Four datasets are integrated: a routable network from OpenStreetMap; usage data from the Nextbike cycle hire scheme; perception data from Glasgow’s Commonplace public consultation; and contextual indicators from the Scottish Index of Multiple Deprivation. A multi-stage pipeline is developed, including route construction via the R5 routing engine, intersection of routes with Data Zones, spatial assignment of consultation responses, aggregation of perception indicators, and compilation into a Data Zone–level analytic dataset. To derive numeric indicators from consultation text, natural language processing methods are evaluated including latent semantic analysis, latent Dirichlet allocation, and RoBERTa sentiment analysis. LDA topic allocations using Gibbs sampling and RoBERTa sentiment measures are selected as the most suitable for generating interpretable perception metrics. Generalised additive models are then used to relate cycling activity to perception indicators and deprivation context. Incorporating qualitative perception variables alongside traditional usage and contextual measures improves explanatory power and reduces residual spatial autocorrelation, indicating that consultation data capture information not available from quantitative sources alone. Cycling perception is additionally modelled at the level of individual responses using spatial model variants, with a Gaussian process specification providing the best fit and removing short-range residual spatial dependence. These findings demonstrate how integrated qualitative and quantitative data can support richer spatial analysis of cycling behaviour, and provide a replicable framework for combining heterogeneous active travel datasets in other urban contexts.