Modelling population flows using spatial interaction models

Abstract

Background Spatial Interaction Models have been used for decades to explain and predict flows (of migrants, capital, traffic, trade etc.) between geographic locations.Aims This paper will guide users through the process of fitting and calibrating spatial interaction models in order to understand, explain and predict internal migration flows in Australia. Data and methods Internal migration data from the Australian 2011 Census of Population and Housing, which records people who have moved between Greater Capital City Statistical Areas over 5-year periods, is used to exemplify the modelling process. The R statistical software is used to process and visualise the data as well as run the models. Results The full suite of Wilson’s family of spatial interaction models is fitted to the internal migration data, revealing that distance and origin/destination populations are some of the most important influencing factors affecting internal migration flows. We see whether constraining the model to known flows about origins and/or destinations will improve the fits and model estimates. Conclusions Spatial interaction modelling has been a tool in the box of some population geographers for a number of decades. However, recent advances in more forgiving programming languages such as R and Python now mean that this powerful modelling methodology is no longer only available to those who also possess advanced computer programming skills. This guide has exemplified the process of fitting and calibrating spatial interaction models on Australian internal migration data, but the methods could easily be applied to other flow data sets in other contexts.

Accessibility and allocation of public parks and gardens in England and Wales: A COVID-19 social distancing perspective

“…where population weights ( W ij ) are defined as the probability that people living at postcode i visits park j , A j is the area of park j , d ij is the distance between postcode i and park j , and α is the distance decay parameter, which based on previous studies was assumed equal to 2.0 [ 30 , 31 ]. We used the per-capita space measure as an indicator of the possible crowdedness of the park or garden, which is also useful in evaluating the feasibility of urban parks and gardens to facilitate social distancing.…”

An Open‐Source Model for Projecting Small Area Demographic and Land‐Use Change

“…The idea that areas that are closer to one another are more attractive than those which are further away from a human migration perspective was developed by Zipf (1946) as “gravity models.” Subsequent work by Wilson (1971) introduced constraints at origin, destination, or both (termed doubly constrained models) which meant that limits could be imposed on the total outflow or inflow. A useful guide for the implementation of SIMs can be found in Dennett (2018). In their SMILE model of the Irish population, Ballas et al (2005) note that the treatment of internal migration could be better handled using a SIM.…”

“…Subsequent work by Wilson (1971) introduced constraints at origin, destination, or both (termed doubly constrained models) which meant that limits could be imposed on the total outflow or inflow. A useful guide for the implementation of SIMs can be found in Dennett (2018). In their SMILE model of the Irish population, note that the treatment of internal migration could be better handled using a SIM.…”

Internal migration and housing costs—A panel analysis for Germany

“…By adopting a modelling approach that was originally proposed by the UN and WTO (2012) for the study of international trade flows, we go beyond cross‐sectional gravity models or spatial interaction models that have conventionally been applied to interregional migration flows. The conventional Poisson regression models have been used to estimate the size of interregional migration flows based on population size and distance between origin and destination, as well as other contextual factors such as income and GDP (e.g., Boyle, Flowerdew, & Shen, 1998; Dennett, 2018; Flowerdew, 2000; Flowerdew & Lovett, 1988). In these models, the focus lies on the effect of regional variation in the size of population and the level of GDP, rather than the effect of regional variation in the change over time in these variables.…”

“…Given that the vigorous public and political debate has been focused on the increase in housing costs over time, the modelling strategy should focus on the effect of increasing housing costs on migration. This cannot be achieved with the cross‐sectional gravity models that have conventionally been used to study migration flows (see, e.g., Boyle, Flowerdew, & Shen, 1998; Dennett, 2018; Metulini, 2013). In this study, we overcome these challenges by applying a fixed effects Poisson panel regression model, which enables us to determine the association between increasing housing costs and shifts of migration flows, and thus, as some may argue, shed some light on the direction of causality (Allison, 2009; Brüderl & Ludwig, 2015).…”