James Heckman and colleagues have argued that social and emotional abilities are key to children’s success in life. Canada’s National Longitudinal Survey of Children and Youth (NLSCY) included numerous markers of social/emotional development. It has been a decade since data of that kind was collected, but there is now appetite and resources to collect new data on child outcomes. The data on social and emotional skills collected in the NLSCY were primarily centred on mental health and children “at-risk”. Using Item Response Theory (IRT) as a guide, we show that NLSCY items typically targeted a narrow range of social/emotional abilities, at the extremes of the latent construct. We argue that new data collection should focus on social skills and abilities rather than risk and should measure the range of skills of children from diverse backgrounds in the population. The NLSCY provided little guidance about appropriate interpretation of derived social and emotional scales. Scales designed and analysed with IRT are more interpretable; resulting scale scores are on an interval scale, and equating techniques allow for construction of measures of intra- and inter-personal change without relying on a single fixed set of items (the set of items can be expanded, improved, or better targeted to individual sub-populations). IRT models, using plausible values and latent regression models, have the added benefit of being appropriate for recovering estimates of population parameters (including correlations amongst latent constructs), where traditional (point estimate) methods have been shown to produce biased estimates of variance components.