Including the gender dimension of migration is essential to avoid systematic bias in migration predictions

Abstract

This study examines the theoretical and methodological limitations of migration research in understanding gender-specific trends of migration. In particular, theory-driven methods suffer from the gender blindness of migration theories, while data-driven methods suffer from the scarcity of gender disaggregated migration data. This research aims to evaluate how these dual limitations affect the accuracy of commonly used migration prediction models. By analyzing migration flows disaggregated by gender, the study compares the performance of deterministic methods and probabilistic gravity-type models in predicting migrant flows with varying gender compositions. The findings reveal significant differences in the predictive performance of gravity-type models based on the gender composition of migration flows. Drawing on migration theories and case studies, the study contextualizes these findings, concluding that the lack of robust theoretical frameworks and the limited availability of gender-specific migration data have critically undermined the accuracy of current prediction and forecasting methods. The implications of this research highlight the urgent need for a critical reassessment of migration theories and methodologies through the lens of gender biases, paving the way for more inclusive and accurate migration predictions.

Keywords: gender; methods; migration; modeling; sex.

Fig. 1.

This heat map reports the mean absolute scaled error (MASE) for the comparison between the predicted values and the observed data for the number of migrants, by type of migration flow and gender, based on pseudo-Bayes estimates by Abel and Cohen (1). The flows are classified as female-predominant (FPD), gender balanced (GB), or male-predominant flows (MPD), according to the typology proposed by Donato and Gabaccia (5). For each row, representing a specific model, the accuracy is color coded, from dark red (for types of flows with low predictive accuracy, meaning high MASE values) to dark green (for types of flows with high predictive accuracy, meaning low MASE values). Across models, the predictive accuracy is generally lowest for female migrants in male-predominant flows and male migrants in female-predominant flows. For detailed descriptions of the applied methods and model specifications please see Materials and Methods.

Fig. 2.

Each panel represents a different migration corridor. Panel (A) shows the migration corridor from Mexico to the United States, panel (B) from Poland to Germany, panel (C) from Morocco to Spain, panel (D) from Venezuela to Colombia, panel (E) from Nepal to Malaysia, and panel (F) from Burkina Faso to Côte d’Ivoire. The bars indicate absolute migration flows by time period in each of the six migration corridors based on pseudo-Bayes estimates by Abel and Cohen (1). Black squares indicate the share of females in the migration flow in a given corridor and time period. We can observe different developments in the share of females over time while the gender composition in absolute migration flows can differ greatly.

Fig. 3.

Gender composition of migration flows by different characteristics of origin and destination countries. The color-coded values indicate the percentage of flows, within each category, that are female predominant, gender balanced, or male predominant, respectively. In Panels (A and B), the three categories for gender equality (GE) are determined by the Gender Inequality Index (GII) for 2022, as provided in the United Nations Human Development Reports (52) (low GE denotes countries that appear in the Top third of the GII ranking, medium GE denotes countries that appear in the Middle third of the GII ranking, and high GE countries that appear in the Bottom third of the GII ranking). In Panel (C), the categories Low and Middle Income Countries (LMIC) and High Income Countries (HIC) are based on the World Bank definition. In Panel (D), the share of educated women represents the share of females in the population that obtained at least a Bachelor’s degree or equivalent. N denotes the total observations used for generating the figures. The numbers of observations vary as certain country characteristics were not available for all countries.

Fig. 4.

Panel (A) describes the median distances covered by migrants based on pseudo-Bayes estimates of flows by Abel and Cohen (1). Error bars across the panels indicate the 80% CIs obtained via bootstrap. In Panel (B), the global migration intensity is calculated as the size of migration flows as a share of world population, for the respective time period, and by gender. The global emigrant spread (Panel C) measures the level at which global migration is dispersed across destination countries, while the global immigrant spread (Panel D) measures the extent to which global migration comes from different origin countries. Both indices range from 0 to 1, with values close to 1 indicating a larger number of origins or destinations, and values close to 0 indicating a smaller number of origins and destinations chosen by migrants. For details on the methods, see Materials and Methods.