More accurate asymptotic standard error formulas for IRT ability estimators

Most-known IRT ability estimators under dichotomous scoring (MLE, BME, WLE and robust) have associated, simple and fancy formulas to derive their associated (asymptotic) standard errors (ASEs). Such ASEs are of primary interest for determining the degree of precision of the ability estimates but also in more specific contexts, such as e.g., CAT stopping rules. However, some of these ASEs were derived under spurious assumptions or only recently and are therefore not yet widespread. The purpose of ths talk is to present a general and unified approach to derive ASE formulas for a broad cass of dichotomous IRT ability estimators that encompass the most-known ones. Using mathematical derovations for asymptotic convergence of Taylor series expansion, a general ASE formula is derived and can be immediately applied to any classical IRT estimator. Some surprising results are encountered and discussed. Eventually, their potential usefulness in e.g., CAT context, is outlined.