The NEO, Eysenck and IPIP inventories were especially conducive to being linked. Linking was slightly less successful for TCI and MPQ with the NEO, Eysenck and IPIP inventories. The mapping of Harm Avoidance onto Neuroticism, despite theoretical differences between the concepts, was found to be relatively good. However, the mapping of Reward Dependence to Extraversion was less feasible, as was suspected. Such imperfect linking results in bias when individuals are ranked, which is very important in for example educational settings (e.g. pass/fail decisions on a test or determining the final class rank). However, when scientific interest is in population effects, like a correlation in twins or between the phenotype and a SNP, results are highly satisfactory. When dealing with non-identical but correlated traits, an alternative could be the use of multidimensional IRT models (van den Berg and Service 2012), because such models allow for relatively low correlations between multiple latent construct, but still enable borrowing statistical information from the respective sets of items, which leads to more precise estimation of latent scores.
