Top, Lars (2023) How well can personal network characteristics predict women’s ideal family size using machine learning techniques? Master thesis, Sociology.
|
Text
Eindversie scriptie_LT_S3862003_2.pdf Download (1MB) | Preview |
Abstract
Since the dawn of network analysis, several social influence mechanisms fostered by social networks have been identified to impact women’s fertility behavior. Earlier studies found that the content and structure of social networks can be a source of helping to raise children (social support), spread the thoughts and ideas of having children (social contagion), spread information about having children (social learning), and enforce pro-natal norms (social pressure). These studies often lack a comprehensive understanding, however, since they usually only test a limited number of network characteristics and often produce poorly generalizable results. This study aimed to overcome these issues with a more holistic and data-driven approach by applying LASSO regression. This Machine Learning technique is a relatively novel method for social science that performs well when the number of variables in statistical models is high relative to the sample size. Furthermore, it identifies which variables are most influential in prediction novel, out-of-sample cases. For the analysis, a unique egocentric dataset is used which included 758 Dutch women, who reported on more than 18.500 relations and information about these relations. The nine models of this study were able to explain between 2% to 11% of the out-of-sample variation. The results indicate that the proportion of kin in a network and the number of strong ties strongly increase the ideal family size of women. Furthermore, whether network members want children or not and the strength of the relationship with these people influenced women’s ideal family size. Network members who prefer to have children or close relations with these people, increase the ideal family size. The opposite is true for network members that do not want children. Network density hardly had any impact, which is at odds with earlier studies. The results indicate that machine learning techniques like LASSO regression can provide promising new insights into social science.
| Item Type: | Thesis (Master) |
|---|---|
| Supervisor name: | Stulp, G. |
| Degree programme: | Sociology |
| Differentiation route: | Social Networks In A Sustainable Society [Master Sociology] |
| Date Deposited: | 20 Jul 2023 08:58 |
| Last Modified: | 20 Jul 2023 08:58 |
| URI: | http://gmwpublic.studenttheses.ub.rug.nl/id/eprint/2498 |
Actions (login required)
 |
View Item |