IRT diagnostics and measurement plots¶

eyeprocesspy includes an extensive IRT and process-psychometrics surface rather than treating eye-tracking features only as predictors. This worked page focuses on diagnostic visualization: information, item fit and DIF.

IRT information curve — **Information profile.** Test information across the latent continuum.

IRT item fit — **Item fit.** Item-level fit statistic relative to the reference value.

IRT DIF curve — **DIF curve.** Signed focal-reference probability difference across theta.

The executable example is examples/irt_diagnostics.py.

Information and conditional SEM¶

ax = ep.plot_eye_irt_information_profile(information_profile)
ax_sem = ep.plot_eye_irt_information_profile(
    information_profile,
    show_sem=True,
)

Information is conditional on theta; a single global reliability number cannot substitute for an information profile when precision varies substantially across the latent continuum.

Item fit¶

ax = ep.plot_eye_irt_item_fit(item_fit, statistic="infit")

Use fit statistics as diagnostics rather than automatic item-deletion rules. Investigate content, local dependence, dimensionality and data quality before changing an assessment model.

Differential item functioning¶

ax = ep.plot_eye_irt_dif_curve(dif_curve)

DIF is a measurement-invariance diagnostic. Statistical DIF is not itself proof of unfairness; substantive interpretation requires the grouping variable, item content, model specification and potential impact on scores.

Wider IRT surface¶

The package also exposes score uncertainty, Q3/local-dependence diagnostics, adaptive traces, link stability, DTF, recovery and SBC evidence, bank coverage, process alignment, sparse-design audits, prior sensitivity and advanced process-informed models. See the IRT and psychometrics guide and the API reference.