Insufficient-effort/careless responding (IER) on self-report background questionnaires attached to large-scale assessments (e.g., PIRLS, PISA, and TIMSS) is routinely screened with indicator-based methods (e.g., longstring, response variability, Mahalanobis distance, and psychometric synonym/antonym consistency). These indices are computed from a respondent’s own response pattern and implicitly treat any inconsistency as inattention. We show this assumption is not fairness-neutral: lower-ability respondents may produce more inconsistent-looking patterns simply because background-questionnaire items are harder for them to process, independent of genuine effort. We introduce ierfair, an R package that benchmarks standard IER indices against ability-stratified (or local-regression) reference bands built from an external achievement score, flagging a respondent only when their index is extreme relative to others at the same ability level. Applying ierfair to the PIRLS 2011 Indonesia Student Background Questionnaire ( N = 4,720), we confirm a significant ability confound in three raw indices (r = −.15 to .25) that per-index conditioning removes but find that a natural rule for combining several conditioned indices (“flag if ≥ k of m fire”) can reintroduce a significant residual ability trend even when no individual index shows one. Requiring unanimity or switching to local-regression conditioning reduces this trend, and pooling indices into a single score before thresholding, implemented as iec_composite (), removes it most cleanly, keeping the flag rate within 1-2 percentage points of the sample mean across every stratum. The package requires no response-time data, an important constraint for archival large-scale-assessment data, and produces an audit-trail report summarizing flag rates across ability strata and demographic subgroups.
The Graded Response Model (GRM) is widely used for analyzing polytomous response data from attitude, personality, and character measurement instruments. However, when response data are collected from a range-restricted sample such as selected applicants, screened participants, or high-performing students, GRM discrimination parameters are systematically underestimated. No existing R package implements range-restriction bias correction specifically for GRM discrimination parameters. This paper introduces grmcor , an R package that implements the truncated normal standard deviation ratio correction for GRM discrimination parameters under direct range restriction. The package provides functions for estimating GRM parameters from restricted samples, applying the correction, comparing restricted and corrected estimates, visualizing bias patterns, and validating the correction procedure via Monte Carlo simulation. Monte Carlo results demonstrate that grmcor substantially reduces bias in discrimination parameter estimates, with mean bias reduced from −0.64 to near zero and RMSE reduced by approximately 52% under 50% range restriction. The package is freely available via GitHub.
Assessments calibrated under item response theory are often expected to support diagnostic reporting, though their parameters differ from diagnostic classification models. This study introduces a projection procedure that summarizes continuous response functions into diagnostic metrics based on explicit reference distributions and mastery thresholds. These metrics—the projected nonmaster success summary, projected master success summary, and projected separation index—are conditional averages over defined mastery regions rather than portable parameters. Simulations and an empirical application show strong item-level agreement with recalibrated diagnostic models, while downstream classification showed varying sensitivity across calibration contrasts. Under a fixed reference distribution and threshold system, the projected separation index provided a conditional item-screening criterion and supported targeted warning-flag review.
facomplex R package implements procedures for estimating complexity and simplicity derived from factor-analytic results of multidimensional scales. In this type of analysis, factor loadings are estimated to represent the relationship between the items and their latent factors or constructs, as well as with the other factors in the model where cross-factor loadings are estimated. An evaluation of these cross-factorial loadings helps determine whether the result is factorially simple or complex. There are indicators for assessing the complexity and simplicity of factorial results, and facomplex calculates all known indicators from various approaches focused on the direct estimation of factorial simplicity and complexity at the item, dimension, and total scale levels: the loading simplicity index, the factor simplicity index, Hofmann’s complexity coefficient (for items and factors), the Bentler Simplicity Index, and the proportion of explained variance derived from the squared sum of target versus non-target factor loadings.
Fairness in automated scoring is typically evaluated with a single global statistic contrasting a focal and reference group—an approach that can either mask a disparity that changes sign across the ability range, or overstate one by conflating it with genuine ability differences between groups (impact). We introduce condfair, an R package that adapts differential item functioning (DIF) logic to automated scoring: it estimates a conditional disparity function across ability levels, tests it with a wild-bootstrap omnibus procedure, decomposes bias into uniform and non-uniform components, and identifies candidate feature-level sources of a detected disparity via conditional SHAP disparity testing. Using the PERSUADE 2.0 essay corpus, we show a global measure can conceal a large, ability-concentrated gender disparity (marginal gap = 0.003; peak conditional disparity = 0.276, p = .001) while overstating an English Language Learner disparity by conflating it with impact (marginal gap = 0.280; conditional bias = 0.045).
