Gravity models do not explain, and cannot predict, international migration dynamics

Robert M Beyer,Hermann Lotze-Campen,Jacob Schewe

doi:10.1057/s41599-022-01067-x

Abstract

The major social and economic impacts of international migration have led to a strong interest in better understanding the drivers of cross-border movement. Quantitative models have sought to explain global migration patterns in terms of economic, social, climatic, and other variables, and future projections of these variables are increasingly being used to forecast international migration flows. An important implicit assumption in the most widely used class of these approaches, so-called gravity models, is that their parameterisation based on panel data enables them to describe the effects of predictor variables on migration flows across both space and time, i.e., that they explain flow variation both across country pairs at a given time and across time for a given country pair. Here we show that this assumption does not hold. Whilst gravity models describe spatial patterns of international migration very well, they fail to capture even basic temporal dynamics, indeed, often worse than even the time-invariant average of the historical flows. We show that standard validation techniques have been unable to detect this important limitation of gravity models due to the different orders of magnitude of migration flows across spatial corridors, on the one hand, and over time, on the other hand. Our analysis suggests that gravity-model-based inferences about the effects that certain variables have had, or will have, on international migration over time may in reality represent statistical artefacts rather than true mechanisms. We argue that future predictions based on gravity models lack statistical support and that, in its current form, this class of models is not suited for informing policy makers about migration trajectories in the coming years and decades.

Highlights

International migration has shaped global socio-political discourses in recent decades unlike any other demographic process and has had lasting effects on domestic political landscapes (Kapur, 2014; Trauner and Turton, 2017; Blinder and Allen, 2020) and international relationships, both between countries of origin and destination and between different countries of destination (Papademetriou and Banulescu-Bogdan, 2016; Martin, 2017)
Our results demonstrate that a successful calibration of gravity models on the basis of spatio-temporally pooled observed migration flows does not imply, or obviate the need to validate, a model’s ability to describe flows between specific pairs of countries in general and changes in migration over time in particular
The fact that the example gravity models considered in our analysis represented temporal dynamics overall worse than even the timeinvariant mean of the historical observations provides strong evidence that predictions of international migration flows based on these models would most likely be highly unreliable

Summary

Introduction

International migration has shaped global socio-political discourses in recent decades unlike any other demographic process and has had lasting effects on domestic political landscapes (Kapur, 2014; Trauner and Turton, 2017; Blinder and Allen, 2020) and international relationships, both between countries of origin and destination and between different countries of destination (Papademetriou and Banulescu-Bogdan, 2016; Martin, 2017). A strong interest in better understanding the drivers of cross-border movements, coupled with an increasing availability of global bilateral migration estimates (Özden et al, 2011; Abel, 2018), has led to the development of quantitative models in which migration flows are estimated as a function of predictors such as country- or country-pair-specific geographic, demographic, political, economic, environmental, linguistic, or cultural variables (e.g., Belot and Ederveen, 2012; Backhaus et al, 2015; Adserà and Pytliková, 2015; Poot et al, 2016; Cai et al, 2016; Cattaneo and Peri, 2016). Gravity models have been used to infer that migration from poor countries will increase as national income levels rise in the coming decades (Docquier, 2018; Rikani and Schewe, 2021), and, in the context of future climatic changes, that migration increases as the result of environmental degradation (Afifi and Warner, 2008; Reuveny and Moore, 2009), more frequent natural disasters (Afifi and Warner, 2008; Ragazzi, 2012; Gröschl and Steinwachs, 2017), higher temperatures (Backhaus et al, 2015; Maurel and Tuccio, 2016; Cai et al, 2016; Cattaneo and Peri, 2016; Helbling and Meierrieks, 2021), or exposure and vulnerability to climate change (Benveniste et al, 2020)

Methods

Results

Discussion

Conclusion