Calibrating a Household Relocation Model for Leicestershire
E Revill, D Simmonds, David Simmonds Consultancy, UK
This paper reports on the calibration of the household relocation model within the land-use component of the Leicester and Leicestershire Integrated Transport Model (LLITM), using data from a local household survey carried out for the project.
The land-use component of LLITM is a new implementation of the DELTA land-use modelling package. The treatment of residential location is very similar to that in previous DELTA applications, in that
- households are classified by composition/age/economic status (10 types) and by socio-economic level (4 levels based on occupation)
- households change over time (eg from young singles to young couples, or older couple to retired couple)
- only a proportion of households consider moving in each year - those that have changed type, and a proportion of others - giving the effect (consistent with observation) that younger households are generally more mobile
- households moves are a mixture of local moves, in which distance has a strong deterrent effect, dominated by housing issues, and longer-distance moves more related to employment issues.
The work here related to the choice of coefficients for the local moves model. Previous applications of this DELTA component have relied on limited, indirect calibration, drawing on previous research by others (including hedonic price analyses), matching the broad patterns of moves (eg average distance) reported in other surveys, and an important element of professional judgement informed by subjecting the resulting model to sensitivity testing. For LLITM we set out to supplement this with some direct calibration of a model of household location choice for moving households. A separate survey for this purpose was not affordable, so we attempted to obtain suitable data from additional questions in the Leicester and Leicestershire Travel Census, the major household survey conducted as part of the model development project.
The LLITM household survey was a mail-out, mail-back self-completion survey sent to 30,000 thousand households across the county, of whom 1880 sent back usable forms. The core of the data obtained was about household composition, car ownership and travel, including a one-day travel diary for each household member. The additional questions for modelling relocation asked how long households had been at their current address, and - if they had been at their current address for less than 5 years - where they had moved from. Some 223 households reported having moved within Leicestershire in the last 5 years, and these provided the sample on which the calibration of the relocation model was attempted.
The observed data provided a sample of moves by "from" and "to" zones. The main focus of the calibration was to explain these moves in a multinomial logit model with the independent variables being our modelled values for accessibility, cost of location, floorspace per household, and distance from old to new location. Since only the chosen "new" location was reported, and since it would have been highly impractical to identify all of the other (>900) zones of the model as "rejected" locations, we resorted to the standard technique of specifying a random (or semi-random) sample of other zones as the rejected alternatives for each observation. Despite the small number of observations we succeeded in obtaining reasonably significant coefficients on each of the independent variables. These informed the final choice of coefficients for the working LLITM model.
Apart from its contribution to the development of LLITM, the exercise demonstrated the feasilibity of calibrating a model of household relocation on data about such relocations, avoiding the limitations and pitfalls of building static models on cross-sectional data. Some lessons for future exercises of this type will be identified.
Association for European Transport