Estimating Multidimensional Item Parameters in Exploratory Graph Analysis with Directional Minimum-Distance Calibration
Year: 2026
Type: preprint
Abstract: Exploratory graph analysis (EGA) estimates the dimensions of psychological data, and revised network loadings provide item-by-dimension coordinates for its communities, but EGA lacks an item response calibration. This paper presents Directional Minimum-Distance (DMD) calibration, a limited-information estimator that fixes the item directions identified by revised network loadings and estimates the only quantity these loadings leave unidentified, one communality per item, by reproducing the observed off-diagonal correlations. Discrimination vectors and intercepts follow in closed form, and the factor carrying the uniqueness cancels from directional difficulty, which reduces to the marginal normal threshold over the square root of the communality. Identification, coordinate invariance, and exact population recovery are proved, and under exact fit, direction-adjusted correlations factor into products of item magnitudes whose ratios are invariant across reference items. A reciprocal Monte Carlo study of 12,000 datasets compared fully empirical EGA-DMD with correctly dimensioned exploratory multidimensional item response theory (MIRT). EGA-DMD recovered item directions (mean row cosine .995 vs. .996) and difficulties (mean RMSE .066 vs. .063) nearly as accurately as MIRT, was robust to the logistic link, and was several hundred to several thousand times faster. In data from 322 students, EGA-DMD difficulties agreed almost perfectly with confirmatory item factor analysis.
Source: PsyArXiv (OSF Preprints)
Language: English
Field: Decision Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7234766354
