When estimating latent trait scores, we preferred linking inventories within cohorts, but not across cohorts. Arguably, linking across cohorts would be even better, scaling all individuals from all cohorts to one common metric. Although theoretically possible, it can be infeasible in practice. In our study, it would require analyzing hundreds of items in over 160,000 individuals in one analysis, which is computationally infeasible. This approach would also only be possible if all inventories could in fact be linked to one another. In our study, this was not the case; for instance, different versions with different answer categories of the same inventory were used in different cohorts.
