Predicting verbal reasoning from virtual community membership in a sample of Russian young adults.

Predicting personality traits from social networking site profiles can help to assess individual differences in verbal reasoning without using long questionnaires. Inspired by earlier studies, which investigated whether abstract-thinking ability are predictable by social networking sites data, we used supervised machine learning to predict verbal-reasoning ability based on a proposed set of features extracted from virtual community membership. A large sample (N = 3,646) of Russian young adults aged 18-22 years approved access to the data from their social networking accounts and completed an online test on verbal reasoning. We experimented with binary classification machine-learning models for verbal-reasoning prediction. Prediction performance was tested on isolated control subsamples for men and women. The results of prediction on AUC-ROC metrics for control subsamples over 0.7 indicated reasonably good performance on predicting verbal-reasoning level. We also investigated the contribution of virtual community's genres to verbal reasoning level prediction for male and female participants. Theoretical interpretations of results stemming from both Vygotsky's sociocultural theory and behavioural genomics are discussed, including the implication that virtual communities make up a non-shared environment that can cause variance in verbal reasoning. We intend to conduct studies to explore the implications of the results further.