A longitudinal mixed logit model for estimation of push and pull effects in residential location choice

Steele, F., Washbrook, E., Charlton, C. and Browne, W.J. A longitudinal mixed logit model for estimation of push and pull effects in residential location choice.

Further details

Choice of residential location reflects the interaction between a household’s preferences for dwelling and neighbourhood attributes and the extent of constraints on their ability to realise those preferences. Different configurations of preferences and constraints lead to different spatial distributions of populations, a key factor in many areas of social science research. For example, a large body of work aims to understand the forces leading to the segregation of groups such as ethnic minorities, immigrants and the economically deprived. Evidence on the extent to which different neighbourhood characteristics attract or repel different types of household can help to discriminate between alternative theories of the sources of spatial segregation

We consider households’ choice of neighbourhood over time using data from the British Household Panel Survey linked to neighbourhood data. A type of longitudinal multinomial logit model – commonly referred to as a conditional logit model – is developed to study neighbourhood choice within commuting (travel-to-work) areas. Household panel data provide information on individual residential choices over time together with information on changing household characteristics. We show how such longitudinal data allow separation of the influence of neighbourhood characteristics on the decision to stay in the current area (‘push’ effects) and on the choice of destination among movers (‘pull’ effects).

A particular focus of the research is the extent to which the push and pull effects of area deprivation on location choice at year t depend on household characteristics such as family type and housing tenure at t-1, and on life course transitions (such as a birth) between t-1 and t. We also consider differential effects of the geographical distance between a potential destination area s and a household’s current area r on the probability of moving from r to s. For example, families with young children might be reluctant to move far from their current area, leading to a pull effect of the geographical proximity between areas r and s.

As well as exploring differential effects of observed household characteristics, our model includes household random effects to allow for unobserved heterogeneity between households in their propensity to move, the push and pull effects of deprivation, and the pull effect of distance.