Skip to main content icon/video/no-internet

flexMIRT

flexMIRT is statistical software, authored by Li Cai and commercially distributed by Vector Psychometric Group, LLC, for item analysis and test scoring. Item analysis includes the estimation of item response theory (IRT) models and diagnostic classification models, both of which are widely used in educational research and measurement. The development of ever-more complex modeling frameworks and IRT models requires an adaptable and regularly updated software program capable of keeping pace with advancements in both computing and statistical/educational measurement theory; flexMIRT seeks to fulfill this need. This entry provides a broad overview of the capabilities of flexMIRT and briefly details licensing information.

First released in 2012, flexMIRT was initially published as a statistical software primarily for multidimensional, multiple group, multiple-level IRT model estimation, evaluation, and scoring within a confirmatory modeling framework using marginal maximum likelihood (or, optionally, modal Bayes) estimation. From its initial release, flexMIRT has also been able to simulate data from any model it is able to estimate. As of Version 3.0, released in the summer of 2015, updates to the program have included an alternate estimation routine better suited for truly high-dimensional models, intuitive syntax for the estimation of diagnostic classification models, expanded capabilities to estimate exploratory factor analysis models with analytic rotations, and an allowance for models that include covariates predicting the latent variables.

flexMIRT is a syntax-driven program written in C++, meaning that any system that is able to compile the language is capable of running the flexMIRT statistical engine. A graphical user interface has been created for computers running Windows, which allows for some point-and-click functionality. For operational and research situations, such as testing companies scoring thousands of respondents in the real-time or simulation studies that require a large of number of repeated analyses, flexMIRT is also able to call through the command-line interface (e.g., Windows Command Prompt), either for an individual analysis or to run a batch file that automates the running of an unlimited number of existing syntax files.

Capabilities and Features

Whether being used by a large-scale testing company or an individual user running a single analysis, all versions of flexMIRT are able to fit a wide variety of models, provide a large number of item-level and model-level fit statistics and diagnostics, produce a number of different IRT score types, and simulate data. There is no limit to the number of groups, individual observations, items, or item response categories that may be submitted for analysis, outside of the constraints of available memory and processing power of the computer on which flexMIRT is run.

The default estimation method of flexMIRT is marginal maximum likelihood via the Bock–Aiken expectation–maximum algorithm, which is the estimation method typically available in IRT software. Somewhat unique to flexMIRT is a generalized dimension reduction algorithm, which allows the program to estimate a certain subset of multidimensional IRT (MIRT) models with increased efficiency. Within the scope of models that contain more than a single dimension, flexMIRT is able to accommodate multilevel (sometimes called hierarchical) models. These models are often seen in educational research, as they allow researchers to properly account for nesting in data, such as students within the same classroom or teachers within a school; as of Version 3.0, flexMIRT is limited to models with two levels of nesting. Additionally, flexMIRT also has an alternate estimation routine called a Metropolis–Hastings Robbins–Monro algorithm, related to Markov chain Monte Carlo techniques, that is able to provide estimation for truly high-dimensional models that, historically, could not be estimated due to known computational issues associated with Bock–Aiken expectation–maximum estimation and high-dimensional models.

...

  • Loading...
locked icon

Sign in to access this content

Get a 30 day FREE TRIAL

  • Watch videos from a variety of sources bringing classroom topics to life
  • Read modern, diverse business cases
  • Explore hundreds of books and reference titles

Sage Recommends

We found other relevant content for you on other Sage platforms.

Loading