Ian Harris, Southern Methodist University
Lynne Stokes, Southern Methodist University
Abstract: NAEP (National Assessment of Educational Progress) data are collected via a complex sampling design. As a result some students have higher probabilities of selection into the sample than others. Many secondary users of NAEP data fit hierarchical or multi-level models, but are unsure of exactly how to use the sampling weights supplied in the NAEP data files. Some commercial software such as HLM implements a first-order weighting approach, and recent research in the literature has suggested the use of second-order weights, which are inversely proportional to joint selection of students. This poster gives some recommendations on using both first-order and second-order weights for NAEP type designs, primarily for multi-level linear models, with some information on the generalized linear models.