Michael Adcock (p), Nik Lomax , Stephen Clark

Jean Dube

Consumer vulnerability has recently attracted increased media attention due to a number of high profile cases. While consumer vulnerability has been discussed at length for over a decade and guidelines have been produced, there has not been a comprehensive geographical assessment of consumer vulnerability in the United Kingdom. This work creates a geodemographic classification of consumer vulnerability at the geography of output area using data from the 2011 census. This classification was achieved using a k-means approach. There were found to be six distinct clusters of varying levels of vulnerability. Two clusters in particular can be considered to contain high levels of vulnerable people, identified as “Vulnerable Communities” and “Vulnerable Pensioners”. To aid interpretability of the results, pen portraits are provided for each cluster along with an interactive map showing cluster assignment for every output area in the United Kingdom. To further explore the characteristics of the clusters, sub-clusters were created using the same k-means methods on each individual cluster dataset. Comparison of these clusters and sub-clusters with data obtained from a commercial partner provided validation for the character of the clusters and provided further characterisation of the clusters by using variables that were not available in the census data.

Itzhak Benenson (p), Eran Ben-Elia , Bella Akkerman

Michael Adcock

We present a unique methodology for understanding travel behavior patterns of Israeli PT users based on the tap-on only smartcard system. Currently, we have analyzed around 100M boarding records from 2M Israeli travelers over two-month period. To understand travelers behavior we grouped travel patterns based on the ID and direction of travelers’ initial line in each day and then noted consecutive boardings. Based on this description, we established daily travel patterns, grouped together travelers of similar travel patterns and classified and interpreted the obtained pattern clusters.
The first unexpected outcome from the data analysis is that a high percent of travelers use PT service only for one leg of their daily trip. Assuming most travelers return to their origin on the same day, the data shows that around one-quarter of the PT users use multiple travel modes on their daily activity. In addition to the high volume of one leg trips, a week by week analysis shows that majority of travelers' travel two or less times a week. In addition, around 40% of travelers switch modes on a daily basis opting for public transport one day and selecting a different mode the next. These and additional rather surprising result suggest that the national PT network is used by essential fraction of travelers as a non-exclusive travel mode. That is, the public transportation system that was established and optimized for the everyday commuting is exploited by these users according to Mobility as a Service (MaaS) concept and half of its users integrate, possibly seamlessly, several types of mobility services. MaaS comes as the next necessary step in the evolution of smart transportation and is strongly endorsed by various transport providers who present it as the next stage of the smart city transportation. While larger database will be investigated towards the conference, our findings suggest that Israeli public transport users are already ready to switch to the MaaS of a future.

Jean Dube (p)

Itzhak Benenson

Spatial econometrics is now well recognized in economic geography and regional science. All researchers working with data having geographical coordinates are now aware that not taking into account the spatial dimension of the data can lead to serious bias in the analysis. After the seminal work of LeSage and Pace (2009), it is also clearer how spatial econometrics models may affect the calculation of marginal effects by taking into account spatial (local and global) spillover effects. This is what Elhorst (2010) has referred to as the moment of “raising the bar.”
The recent availability of big data raises other methodological challenges for spatial econometrics, particularly when one estimates the models with the maximum likelihood approach. Even taking into account the sparse structure of the weights matrix (Pace, 2004), the estimation of the models results in high memory consumption when dealing with big data. One possible way to deal with such challenges is to proceed to spatial sampling, reducing the sample size and memory requirements. However, a spatial sampling scheme necessarily affects the structure of the (spatial) weights matrix if the selection of individuals is purely random (neighbours are not the same). Consequently, the “true” spatial relations are not all reported in the weights matrix. Thus, all the spatial statistics built on the weights matrix can introduce bias. This is true for spatial dependence measures (Moran’s I), but also for spatial spillover effect, as isolated in spatial econometric models through autoregressive coefficients.
The aims of the talk are to adopt different sampling procedures and test whether some procedures are better than others. It also targets proposing a “spatial bootstrap” method that can be applied to spatial big databases and check if such an approach can be proposed for empirical applications. The demonstration is based on a Monte Carlo experiment.