psychotools: Psychometric Modeling Infrastructure

Achim Zeileis, Carolin Strobl, Florian Wickelmaier, Basil Komboz, Julia Kopf, Lennart Schneider, Rudolf DebelakOpenalex Unknown 2011 · DOI · OpenAlex

Abstract: Infrastructure for psychometric modeling such as data classes (for item response data and paired comparisons), basic model fitting functions (for Bradley-Terry, Rasch, parametric logistic IRT, generalized partial credit, rating scale, multinomial processing tree models), extractor functions for different types of parameters (item, person, threshold, discrimination, guessing, upper asymptotes), unified inference and visualizations, and various datasets for illustration. Intended as a common lightweight and efficient toolbox for psychometric modeling and a common building block for fitting psychometric mixture models in package "psychomix" and trees based on psychometric models in package "psychotree".

Introduction

Psychometric modeling has historically been a fragmented field within the R ecosystem. Researchers often find themselves switching between disparate packages to handle different data types—such as dichotomous versus polytomous responses—or different model families like Item Response Theory (IRT) and paired comparisons. This fragmentation leads to inconsistent data formats, varying syntax for parameter extraction, and a lack of unified visualization tools.

psychotools addresses these challenges by providing a standardized, lightweight, and efficient framework for psychometric analysis. It streamlines the entire research pipeline, from data handling to parameter extraction and visualization, acting as a foundational building block for advanced methods like psychometric trees and mixture models.

Core Infrastructure: Data Classes and Model Fitting

The package introduces two primary data structures designed to standardize psychometric inputs:

  • itemresp: Specifically for item response data (questionnaires).
  • paircomp: Designed for paired comparison data (preferences).

Beyond data handling, psychotools implements or wraps fitting functions for a wide array of models, including IRT models (Rasch, 2PL, 3PL, 4PL, RSM, and GPCM), preference models (Bradley-Terry), and cognitive models (Multinomial Processing Tree models).

Advanced DIF Anchoring Framework

A standout feature of the package is its sophisticated anchoring infrastructure for detecting Differential Item Functioning (DIF). The anchor() function allows researchers to build a common scale for comparing groups using various classes and selection strategies.

The `anchor` function allows for detailed specification of anchor classes and selection strategies, such as constant or iterative forward selection.

By supporting methods like Gini-based selection (Strobl et al., 2021) and iterative forward anchoring, the package moves beyond simple "all-other" anchoring, providing more robust ways to detect item bias across different demographics.

Unified Extractor API and Visualization

One of the most powerful aspects of psychotools is its set of unified extractor functions (itempar, personpar, threshpar, discrpar, guesspar, upperpar). These allow researchers to pull specific parameters from different model types using a consistent syntax, significantly reducing the learning curve for multi-model analyses.

Furthermore, the package provides a suite of specialized diagnostic plots:

  • curveplot: Visualizes item characteristic curves.
  • infoplot: Displays information functions.
  • piplot: Generates person-item maps.
  • regionplot: Offers a unique view of item thresholds.

Integration and Inference

The utility of psychotools is further demonstrated through its integration into the broader R ecosystem, serving as the engine for psychomix (mixture models) and psychotree (model-based recursive partitioning). For formal inference, the anchortest() function provides a rigorous Wald test for uniform DIF.

The `anchortest` function provides the interface for conducting Wald tests for DIF, building upon the anchoring strategies defined in the package.

Conclusion

While psychotools is intended as a lightweight infrastructure, it provides the essential tools for modern psychometricians. It focuses on standard parametric models and provides a robust foundation for more complex structural questions. Future work continues to expand its scope, particularly in its role as the foundational layer for psychometric trees and mixture models.

Summary updated

Abstract

Infrastructure for psychometric modeling such as data classes (for item response data and paired comparisons), basic model fitting functions (for Bradley-Terry, Rasch, parametric logistic IRT, generalized partial credit, rating scale, multinomial processing tree models), extractor functions for different types of parameters (item, person, threshold, discrimination, guessing, upper asymptotes), unified inference and visualizations, and various datasets for illustration. Intended as a common lightweight and efficient toolbox for psychometric modeling and a common building block for fitting psychometric mixture models in package "psychomix" and trees based on psychometric models in package "psychotree".
BibTeX
@inproceedings{zeileis2011psychotools,
  author = {Achim Zeileis and Carolin Strobl and Florian Wickelmaier and Basil Komboz and Julia Kopf and Lennart Schneider and Rudolf Debelak},
  title = {{psychotools: Psychometric Modeling Infrastructure}},
  booktitle = {Openalex Unknown},
  year = {2011},
  doi = {10.32614/cran.package.psychotools},
}
Figure