Predictive Patterns Beyond Partnership: How Graph Neural Networks Reveal Associations in the Complex Social Dynamics of Fertility Intentions

Abstract

Traditional demographic theories typically conceptualize romantic partnerships as the primary context in which fertility intentions are formed. This study challenges this assumption using Graph Neural Networks (GNNs) with three complementary explainability tools to analyze how different relationship types are associated with fertility intentions among 706 Dutch women aged 18–40. The analysis reveals a striking “partner paradox”: despite 69.8% of women including partners in their core networks, these relationships demonstrated the weakest predictive power across all analytical methods, ranking last in attention analysis (weight = 0.14), 19th out of 23 in GNNExplainer importance (0.05), and showing only moderate edge importance (0.12). Conversely, peripheral relationships showed the strongest predictive associations, with internet-mediated connections receiving maximum attention (1.00), followed by ties through ex-partners (0.71) and other unspecified relationships (0.52). The model achieved robust performance (R² = 0.43), with network features contributing 25.0% of predictive power beyond individual characteristics. Individual characteristics moderated which relationships showed strongest associations: mothers of multiple children demonstrated substantially stronger overall network associations (0.86) compared to childless women (0.51); highly educated women showed strongest associations with institutional ties while less educated women showed stronger associations with child-mediated connections; and economic resources amplified network associations. These findings suggest that low partner predictive power may paradoxically reflect successful preference alignment through assortative mating rather than lack of real-world influence. The results indicate that fertility intentions are associated with complex social ecologies encompassing diverse relationship types, challenging couple-centered demographic theories and suggesting that contemporary fertility patterns require understanding the full spectrum of social connections, from intimate partnerships to digital ties, that correlate with reproductive decision-making.