How does cross-sectional sampling bias our understanding of adolescent romantic relationships?: An agent-based simulation

Introduction: Adolescent romantic relationships are developmentally significant, but relatively brief and often disrupted by changes in context. Large individual differences and age-related change make sampling complex. Most adolescents have multiple romantic relationships. Which should we sample? To better understand the issues involved, this study used a simulation - an agent-based computational model - to generate model worlds, each following the relationships formed and dissolved over 5 years. Cross-sectional sample estimates of the number, duration, and type of relationships were compared to population parameters of all relationships formed within the 5 years. Computational models can provide useful insight into sampling bias because (1) the processes producing the results are explicit, (2) results can be replicated to reduce sample idiosyncrasies, and (3) sample statistics can be compared to known population parameters.

Methods: 1000 iterations were run of an agent-based model following 1000 individuals interacting for 60 "months." The model included three types of individuals differing in relationship duration. Two sets of 1000 cross-sectional samples were drawn from the 60,000 cross-sectional "months." Sample statistics were compared to the population parameters.

Results: Cross-sectional samples systematically over-represented longer relationships. The ability to detect individual differences in the duration and number of partners varied with time. These results suggest that cross-sectional survey and observational studies may be time sensitive and systematically distort our understanding of adolescent romantic relationships by oversampling longer-term relationships. Results also illustrate how computational models can provide insight into complex phenomena.