Trajectories of loneliness across adolescence: An empirical comparison of longitudinal clustering methods using R.

Introduction: In this study, we compare three different longitudinal clustering methods. As a case study, the comparison of the methods is conducted for the development of loneliness from middle childhood to young adulthood. The aim is to explore how two nonparametric longitudinal cluster methods compare with a model-based latent class mixture model approach.

Methods: The trajectories of loneliness of 130 young people between 9 and 21 years of age, were analyzed to find a set clusters within these trajectories. The data for this study were obtained from the Nijmegen Longitudinal Study on Infant and Child Development (The Netherlands). Loneliness was measured at four waves at the age of 9, 13, 16, and 21 years. The nonparametric methods are in the R-packages kml and traj, and the model-based in the lcmm package.

Results: All methods indicated that the optimal number of clusters to describe the heterogeneity across the trajectories was three. The kml and lcmm methods showed the most similarity in shape of all clusters and fitted the data relatively well, while the traj method yielded somewhat different shapes and didn't fit the data well.

Conclusions: All three methods corroborate the literature in this field by finding that the largest portion of subjects experience stable and low levels of loneliness. However, the clustering methods also reveal that there is a portion of subjects that experience changes in loneliness during adolescence. By comparing the results of nonparametric clustering methods to the latent class mixture model, this study equips researchers with an example of how to implement these models and thereby contributes to the literature on longitudinal clustering in the social sciences. Altogether the analyses show that it might be useful to investigate different algorithms to identify the most robust solution.