In recent years IRT models have focused on the estimation of unidimensional abilities for instruments with sub-tests. While both testlet and higher-order models address potential local item dependence, a generalized subdimension model (GSM) allows for the construction of a unidimensional ability estimate while assuming multidimensional latent variables. This paper extends the model beyond the use of dichotomous data in large-scale test batteries to include partial credit data in order to check whether it yields comparable results in small-scale assessment data with only constructed response items. The GSM results yield accurate estimates for partial credit data; the increase in measurement precision in comparison to the unidimensional model is significantly higher for more heterogeneous data with lower correlations for the underlying dimensions.