Predictive power of surveys across rounds.

Table shows which round(s) of data were used to train the network (i.e. identify which students are connected by a single link and not), and which round of data this network was tested against. ‘No link’ and ‘link’ indicates how many students are connected by a single link in the training data. When two training sets are used, students are sometimes linked in one and not linked in the other; these are classed as ‘ambiguous’. Potential links gives the total number of possible single links that could exist between nodes in that network. The ability of a training dataset to predict links in a test dataset is assessed using precision, recall, F-score and accuracy (see Methods).