AM is a statistical software package for analyzing data from complex samples, especially large-scale assessments such as the National Assessment of Educational Progress (NAEP) and the Trends in International Mathematics and Science Study (TIMSS).
From its origin as a specialized tool for analyzing large-scale assessment data, AM has evolved into a more generalized and growing tool for analyzing data from complex samples in general. Originally, AM was developed to estimate regression models through marginal maximum likelihood (MML). Because large-scale assessments are often low-stakes assessments for students, students are usually asked to respond to only a few items; each student sees only part of the whole test. Otherwise, they would be unlikely to expend real effort on any items. As a result, individual test scores are subject to substantial measurement error, which would bias many statistical estimates. Rather than assign each student an error-filled score, MML procedures represent each student’s proficiency as a probability distribution over all possible scores. MML procedures use these probability distributions in the estimation process.
Another characteristic of large-scale assessments has led to wider applicability of AM—they almost always draw a sample from a complex design. AM automatically provides appropriate standard errors for complex samples using a Taylor-series approximation. This happens automatically even when new procedures are added to the software. Over time, the software has grown to offer a set of non-MML statistics, including regression, probit, logit, cross-tabs, and other statistics that are useful for survey data in general.
The American Institutes for Research is committed to keeping AM available as a free and growing tool for the research community. Visit this website for further information, updates, and technical support.