Enoch O. Oladunmoye

The 3PL and 4PL Item Response Theory Models

The three-parameter logistic (3PL) and four-parameter logistic (4PL) models extend the two-parameter logistic (2PL) Item Response Theory (IRT) framework by introducing additional parameters intended to capture response behaviour that discrimination and item location alone cannot represent. The 3PL model introduces a lower asymptote, conventionally interpreted as pseudo-guessing, while the 4PL model further incorporates an upper asymptote representing the probability that a respondent answers incorrectly despite possessing sufficient latent ability, commonly described as slipping or careless responding. This paper reviews the conceptual, mathematical, statistical and practical foundations of the 3PL and 4PL models, explaining what each parameter does and does not represent, and examining the consequences for item information, ability estimation, identifiability and sample-size requirements. Applications in educational, psychological, online and adaptive testing are discussed, alongside model comparison, cross-validation, and differential item functioning. Particular care is taken to distinguish genuine guessing from pseudo-guessing, and careless responding from the statistical construct of the upper asymptote. The paper concludes that the 3PL and 4PL should be regarded as theoretically motivated extensions rather than automatically superior alternatives to the 1PL or 2PL, and sets out functional requirements for a PsychtrixWeb implementation supporting model comparison, diagnostic transparency, fairness analysis and reproducible reporting.