So, how would these patterns of missingness, or data inaccuracy alter the results from models using network data? One example comes from literature on partnership concurrency, a concept that measures when multiple sexual partnerships overlap in time. From a series of mathematical models, concurrency was proposed as a potential population-level accelerant for HIV transmission (Morris and Kretzschmar 1997), by increasing the rates of potential contact between serodiscordant partners. These ideas have also been supported with available empirical evidence (Helleringer, Kohler, and Kalilani-Phiri 2009). Researchers have shown that modest failures of corroboration in the reporting of partnerships, and the timing of when they begin or end, can substantially alter our ability to estimate concurrency rates among populations in the US (Brewer et al. 2006) and Malawi (Helleringer et al. 2011). Helleringer, Mkandawire, and Kohler (2014) proposed a method for including sensitivity adjustments that incorporate these levels of partnership uncertainty into estimates of concurrency prevalence. Such adjustments may account for the limited number of studies that have countered the concurrency hypotheses.
Marcel Salathé and colleagues deployed remote sensors to observe the interactions between members of a high school (Salathé et al. 2010), which they then used to develop a series of models of influenza outbreak and vaccination strategies. Interaction distance and duration of interaction are each factors directly associated with the remote sensors’ ability to record the actual interactions, and therefore are potential sources of measurement error in this sort of work. They show in a subsequent paper when modeling behavioral interactions and disease spread from these data, that different—temporal and distance—thresholds altered estimates of the roles that key network factors (including homophily, clustering, and formal organizational roles) played in the extent of modeled epidemics, or the efficacy of various vaccination intervention scenarios (Salathé and Jones 2010; Barclay et al. 2014).