Prevalence-Aware Networks of Social Activity and Community Facility Use Among Older Adults in the Republic of Korea: Comparisons by Age, Residence, and Sex

Background/Objectives: Older adults engage in diverse activities and use community facilities, but how these forms of participation connect is unclear. We examined these connections and whether the findings depended on how common each activity or facility was. Methods: Data from 10,078 participants in the 2023 National Survey of Older Koreans were used to estimate 13-node domain–facility and 20-node specific activity–facility networks with survey-weighted φ coefficients and 1000 stratified cluster-bootstrap replicates. Leisure–leisure pairs were excluded from the specific network because only two main leisure activities could be reported. Sensitivity analyses used sign-specific attainable φ limits, normalized φ, covariate-adjusted partial correlations, semantic-overlap exclusions, and subgroup log-odds-ratio contrasts. Results: The domain network retained 27 positive and one negative edge; the specific network retained 34 positive and one negative edge among 162 estimable pairs. Original and φ-limit-normalized strength rankings correlated at ρ = 0.852 (domain) and 0.968 (specific), but public cultural facilities moved from first to fourth and third, respectively, with socializing groups ranking first in both normalized networks. Eighteen of 28 domain edges and 22 of 35 specific edges were retained with the covariate-adjusted partial-correlation criterion. Public cultural facilities had the largest strength point estimates in both primary networks, although several leading nodes were not statistically distinguishable; the prominence of education/learning was also sensitive to broad removal of definitionally overlapping links. Subgroup strength rankings were similar, whereas several differential φ edges were not confirmed by log-odds-ratio contrasts. Conclusions: The networks identify disproportionate co-participation patterns rather than prevalence-free or causal relationships. Overall rank agreement under normalization was higher in the specific than the domain network, but neither preserved the leading node. Marginal constraints, item overlap, and the association measure affected interpretation.