Background: To better understand the development of mental disorders, dynamic networks have gained more attention in recent years. Most of these network models use a single indicator per node despite the fact that measurement error may bias parameter estimates. This can lead to incorrect conclusions about the presence or absence of edges, as well as the relative strength of edges in the network. In this study, we compared single-indicator dynamic networks to approaches using information on multiple indicators per node to account for measurement error.

Data and Methods: We conducted two simulation studies, using time series (Study 1, N = 1) and panel (Study 2, N > 1) data, to compare the estimation of network parameters in the presence of measurement error in models with single indicators versus models with multiple indicators, namely as latent variables, plausible values, factor scores, and average scores. Across conditions, we varied the variance of the measurement error and the number of observations (in time series: number of timepoints; and in panel data: number of persons and waves). We evaluated the performance of each model by examining the correlation between the estimated and true network edge weights, as well as the sensitivity, specificity, and precision.

Results: In both studies, measurement error decreased correlations between the true and estimated network as well as sensitivity among all approaches, while specificity and precision were mostly unaffected. The single-indicator approach was the most sensitive to measurement error and the number of observations compared to other approaches. In Study 1, the factor and average score approaches performed best for temporal networks, and the latent variable approach for contemporaneous networks. In Study 2, generally the best-performing approach was the plausible value score.

Discussion: Measurement error may substantially bias estimates in dynamic networks, and multiple-indicator approaches can mitigate this bias. Multiple-indicator approaches generally outperformed the single-indicator approach, but the choice between different multiple-indicator approaches depends on several factors that must be carefully considered before deciding the best method for each study.