5.9 Final model
Unfortunately, there is no single index of model fit that we can optimise. Modelling is more like a balancing act among multiple competing objectives, such as
- preserving as many items as possible that fit the model;
- finding high-quality active equate groups that span many cohorts and instruments;
- picking active equate groups for which we have enough information;
- providing reasonable age-conditional distributions of the D-score;
- representing various developmental domains in a fair way;
- preserving well-fitting historical models as new data become available;
- maintaining a reasonable calculation time.
This section showed various modelling techniques and ways to assess the validity of the model. In real life, we fitted a total number of 140 models on the data and made many choices that weigh the above objectives. The final model for the GCDG data consists of 565 items (originating from 14 instruments) that fit the Rasch model and that connect through 18 equate groups. Due to the sparseness of data at the very young ages, the quality of the model is best for ages between 4-36 months.
Model 565_18
formed the basis of the publication by Weber et al. (2019). Additional detail on model 565_18
is available through the dmodel
shiny app.