Abstract
In car orientated nations, most commuters living close to work typically do not commute by bicycle. Empirical scholarship seeking to delineate the various barriers to cycling-to-work present a set of somewhat inconsistent findings. This study seeks to demystify this lack of clarity by introducing the concept of “cycling dissonance”—the mismatch between cycling potential and cycling reality—and place an empirical focus on non-cycling commuters who travel a distance to work, deemed “cyclable.” By introducing the concept of cycling dissonance embedded within a spatial modeling approach, the relationship between cycling dissonance and the natural and built environment is captured whilst controlling for the socio-demographic characteristics of commuters. Our findings reveal important spatial variations highlighting commuters working in areas with hillier terrains, sparser populations and lower employment densities, or commuters living in areas with hillier terrains and higher land-use mixes tend to have higher levels of cycling dissonance. By drawing these results together, we develop a new policy tool that spatially delineates the place-based factors that matter for cycling dissonance and in doing so provide a new evidence base with the capacity to better target place-specific cycling-supportive policy.
Acknowledgements
The China Scholarship Council and the University of Queensland International Scholarship provided financial support for this study. The authors thank the anonymous reviewers for their comments during the revision of this paper.
Disclosure statement
The authors declared that there is no conflict of interest.
Notes
1 It should be acknowledged that the term of dissonance also emerges from social psychology such as the Theory of Cognitive Dissonance by Festinger (Citation1957) which denotes mismatching relationships among cognitions. Cycling dissonance is distinguished from cognitive dissonance given that cycling dissonance places an emphasis on cycling behavior in commuting and does not place a lens on trying to unpack individual cognitions.
2 The Local Government Areas are Non-ABS Structure regions, and GB is one of the Greater Capital City Statistical Areas that are defined by ABS (2016b123). The geographical boundary of GB does not exactly cover the entire boundary of all the LGAs mentioned here, for example, only part of Regional Council of Lockyer Valley is included within the GB, but all the administrative centers are included within the spatial scope of GB.
3 SA2s reflect functional areas that work as a community which interacts together socially and economically; in major urban areas, SA2s represent one or more function-related suburbs (ABS, Citation2019). When developing a neighborhood plan, a Community Planning Team provides local knowledge and feedback on specific proposals for their neighborhood (Brisbane City Council, Citation2020). The examination of all commuting trips which are within the cyclable distance highlights that on average 22% of trips begin and end in the same SA2 and 78% end in a neighboring SA2. We consider that SA2s (together with the neighboring SA2s) provide a suitable scale that acknowledges their distinctive planning function along with reflecting the general distribution of cyclable commuting trips and cycling dissonance.
4 We adopt a conservative approach to help ensure that we do not overestimate potential cyclists—rather we define both an upper and lower distance bound that attempts to capture potential cyclists that make commuting trips well within the cyclable distances reported in existing studies. In similar, we do not define either the upper or lower bounds to exist at the maximum or minimum distances that are made by cyclists commuting to work in census data. Rather, these bounds are established well within the maximum or minimum distance. In sum, we argue that this constitutes a ‘conservative approach’ to determining the cyclable distance.
5 We also made trials using cycling rate as the dependent variable instead of cycling dissonance. The differences in the estimation results between the two model settings are subtle but the models using cycling dissonance as the dependent variable have better performance and more variables that are significant. The differences to some extent unpack the complexity of the relationship between cycling dissonance and environmental characteristics in particular ways.
6 For a Poisson distribution, the variance of the population (i.e., var[y]) is equal to the mean (i.e., μ), namely var[y] = μ. In practice, the variance of the observation could be lower than the mean, which is called under-dispersion (Heinzl & Mittlböck, Citation2003).
7 The quasi-Poisson method adds a positive dispersion parameter φ to the Poisson variance function var[y] = μ, namely, var[yi] = φμi (where φ < 1 denotes under-dispersion) (Dunn & Smyth, Citation2018).
8 The opposite of under-dispersion is over-dispersion which indicates that the variance of the observation is higher than the mean (Heinzl & Mittlböck, Citation2003).
9 Among the 236 SA2s in Greater Brisbane, four SA2s as places of work have no trips commuting to and eight SA2s as places of residence have no trips commuting from. These zero-trip SA2s are places of military camp, reservoir, port, and airport, almost have no commuters working or living there and thus we excluded them from the analysis.
10 For the WCD model, cycle network density, intersection density, gender, education level, and employment status are excluded. For the RCD model, the excluded variables include education level and population density.