Combining longitudinal data analysis with networks to examine spatio-temporal variation

Spatio-temporal networks are a useful tool for examining systems such as transport networks. However, it is relatively difficult to examine continuous temporal change in the properties of network connections. Spatially correlated time series analysis often uses only distance to estimate correlations between time series, which is not necessarily applicable to a network system.
Longitudinal data analysis methods that are common in epidemiology and psychology may be combined with spatio-temporal network data to capture complex temporal patterns at the level of individual observation-units, for example individual network objects. These can be used to distil complex temporal information into specific easily interpretable variables that represent a specific part, or feature, of a temporal pattern, such as the timing of maxima.
This paper illustrates how these methods could be combined with geographical methods to generate meaningful and interpretable results describing spatial variation in temporal patterns of temperature in a rail network. Longitudinal methods considered were: multilevel modelling and functional data analysis.
Results show differences across the longitudinal methods used that are likely down to necessary differences in model specification. The appropriate parameterisation of each method is one of several factors that will affect the utility of these methods to accurately capture temporal pattern features in a meaningful way. There is considerable scope for further investigation of the utility of these methods through simulation.