Abstract
Existing evidence suggests social isolation provokes perceived loneliness in older adults, however; inconsistent results have been reported such as loneliness in the crowd. In this research, we aimed to identify a social network position that is associated with loneliness and classify lonely older adults from non-lonely using the complete social network of two entire villages (n = 1,636; Mean age = 74) from Korean Social life and Health Aging Project. Three types of social network characteristics were introduced as predictors of loneliness: degree centrality (size), k-core (embeddedness), and constraint (brokerage). Also, feeling of closeness, communication frequency within an individual’s social connections, and demographic data (age, sex, education, marital status, and health status) were included as covariates. Loneliness was measured using a question (“I felt lonely”) from CES-D scale. To predict loneliness of an individual, we applied a machine learning algorithm of regularized regression to all social and demographic variables. Optimizing procedure selected the predictors in a training set (village A; n = 733) and we tested predictability of the model in a new data set (village B; n = 933). As results, older adults with highly constrained social network, which reflects less opportunity for brokerage, tend to feel lonely (OR: 1.73, p < 0.05). Also, the logistic regression model which includes constraint in social network accurately identified lonely individuals in novel individuals (AUC of the ROC curve: 0.81 and 0.74 in training and test set, respectively). Our results suggest that a constrained social network is an important predictor of loneliness.