The Role of Social Networks and Personal Characteristics in Shaping Fertility Intentions: A Multi-Method Machine Learning Perspective

Abstract

This study aims to explore how individual and social network characteristics influence women’s desire to have children. For this purpose, individual and network attributes were used for predicting fertility intentions. The data was acquired from the "Social Networks and Fertility Research" at the LISS panel. The sample of the study consists of 738 Dutch women aged 18 to 40, and their individual and social network characteristics. In total, over 18,000 relationships were collected from respondents. First, a Graph Neural Network (GNN) was applied to grasp the effect of the network variables. Since the accuracy of the GNN performed low (with 0.30- 0.40 accuracy) and it was able to produce prediction only for one class the other machine learning methods were used. The methods, Histogram-Based Gradient Boosting Decision Tree (HGBT), Support Vector Machine, and Random Forest were trained and tested with 5-fold cross-validation also Grid Search CV hyperparameter tuning was implemented for reaching the best parameters. HGBT performed the best among all, so this model was used in further steps to describe the relations. It was found that individual characteristics, especially age and family pressure, had a more significant impact compared to network variables. On the other hand, while not as influential as individual attributes, network variables also demonstrated a significant role in explaining fertility intentions. The results showed that the network variables, such as the total number of children in women's networks, the number of people who want to have children, their connections with these individuals, and the frequency of their contact also have an influence.