irtDFI: An R Package for Dynamic Fit Index Cutoffs in Polytomous Item Response Models

Limited-information fit indices such as the RMSEA and SRMSR are routinely reported for polytomous item response theory (IRT) models, yet they are judged against fixed rules of thumb borrowed from structural equation modeling. Dynamic fit index (DFI) cutoffs, derived by simulating from the researcher’s own model, address this problem for factor analysis, but to our knowledge no implementation exists for the IRT indices computed from models fitted with mirt. This paper introduces irtDFI, an R package that, given a fitted graded response, generalized partial credit, or partial credit model, simulates data from the model and from versions with increasing local dependence between item pairs, refits the model, and returns cutoffs for the RMSEA, SRMSR, CFI, and TLI based on the C2 statistic. In a Monte Carlo study of nine conditions (200 replications per level), the Level 0 RMSEA cutoff ranged from .019 to .050 depending on sample size, test length, and number of categories. With 12 items and 3 categories, the conventional RMSEA <.05 rule flagged 0% to 5% of data sets with pronounced local dependence, whereas the tailored cutoff flagged 35% to 94%. The package is freely available via GitHub.