Structural Exposure and Subsequent Neighborhood Stability in Temporal Networks: Evidence and Limits of Generalizability

Abstract

Can temporal path based structural exposure serve as a reliable indicator of subsequent local neighborhood stability across different temporal networks? We investigate this question using the completed observational results from 12 networks spanning face to face nteraction, communication, online activity, and urban mobility. Exposure aggregates source node structural mass along time respecting paths; the outcome is Jaccard neighborhood turnover between later, nonoverlapping windows, with greater turnover indicating lower stability. Six complementary analyses examine the direction and consistency of exposure-turnover associations, proximity to high mass nodes, functional response shapes, distance decay kernels, alternative mass definitions, and temporally ordered changes in source mass. The evidence is heterogeneous rather than uniformly negative or positive. In the primary association analysis, two networks retained positive coefficients after within question false discovery rate adjustment, indicating lower subsequent stability at higher exposure, whereas no negative association survived that adjustment. The pooled proximity coefficient was not distinguishable from zero, and significant mass change associations appeared in opposite directions across networks. More flexible functional, kernel, and mass choices yielded only modest cross validated improvements over simpler comparators; available temporal randomizations also gave model dependent contrasts. The study therefore does not establish structural exposure as a directionally consistent, network independent stability indicator. It also cannot establish a lack of predictive value relative to a no exposure baseline, because that specific comparison and independent cross network validation were unavailable. Incomplete null simulations, unrecorded covariance estimation fallbacks constrain inference.