Arguments for a model-dependent inference?

Forest inventories (FIs) are increasingly supported by auxiliary variables (X) with a formulated link to a target variable (Y). The tenet forwarded here is that with simple single-stage sampling designs and a census of X, an analyst can choose between a model-assisted and a model-dependent finite population prediction approach to inference without the risk of important numerical differences in resulting estimates of interest. A small simulation study largely confirmed the tenet. A suite of FI-related issues regarding the two paradigms and their application in FIs are brought to the attention and discussed, hopefully triggering a more balanced reflections on the choice of inferential paradigm.