Measuring Non-Cognitive Skills Exploiting Log-Files on Online Behaviour

Non-cognitive skills are important components of individuals' human capital and good predictors of educational and labour market outcomes. Conventional self-reported measures of non-cognitive skills suffer from measure-ment errors stemming from self-presentation and reference group effects, which can produce paradoxical results in cross-country comparisons. We propose a novel source of measures derived from computer-generated log files on the behaviour of individuals taking an online test or respondents taking an online survey. We analyse mea-sures of two desirable non-cognitive skills, perseverance and deep learning, constructed with log-file data from two large-scale educational surveys. Compared with the self-reported measures, our log-based behavioural measures have higher cross-country comparability, as they predict the performance of tests consistently at both individual and country levels. They also show high predictive validity in schooling and labour market outcomes, offering promise for a wide range of applications. We discuss the methodological implications of log-based behavioural measures and encourage researchers to apply them in combination with conventional self -reported measures.