Rasch Dichotomous Model vs. One-parameter Logistic Model

Aspect Rasch Dichotomous Model Item Response Theory:
One-Parameter Logistic Model
Abbreviation Rasch 1-PL IRT, also 1PL
Motivation Prescriptive: Distribution-free person ability estimates and distribution-free item difficulty estimates on a linear latent variable Descriptive: Computationally simpler approximation to the Normal Ogive Model of L.L. Thurstone, D.N. Lawley, F.M. Lord
Persons, objects, subjects, cases, etc. Person n of ability Bn, or
Person ν (Greek nu) of ability βn in logits
Normally-distributed person sample of ability distribution θ, conceptualized as N(0,1), in probits: incidental parameters
Items, agents, prompts, probes, multiple-choice questions, etc.: structural parameters Item i of difficulty Di, or
Item ι (Greek iota) of difficulty δi in logits
Itemi of difficulty bi (the "one parameter") in probits
Nature of binary data1 = "success" - presence of property
0 = "failure" - absence of property
1 = "success" - presence of property
0 = "failure" - absence of property
Probability of binary data Pni = probability that person n is observed to have the requisite property, "succeeds", when encountering item i Pi(θ) = overall probability of "success" by person distribution θ on item i
Formulation: exponential form
e = 2.71828
Formulation: logit-linear form
loge = natural logarithm
Local origin of scale: zero of parameter estimatesAverage item difficulty, or difficulty of specified item. (Criterion-referenced) Average person ability. (Norm-referenced)
Item discriminationItem characteristic curves (ICCs) modeled to be parallel with a slope of 1 (the natural logistic ogive)ICCs modeled to be parallel with a slope of 1.7 (approximating the slope of the cumulative normal ogive)
Missing data allowedYes, depending on estimation methodYes, depending on estimation method
Fixed (anchored) parameter values for persons and itemsYes, depending on softwareItems: depending on software. Persons: only for distributional form.
Fit evaluationFit of the data to the model
Local, one parameter at a time
Fit of the model to the data
Global, accept or reject the model
Data-model mismatchDefective data do not support parameter separability in a linear framework. Consider editing the data.Defective model does not adequately describe the data. Consider adding discrimination (2-PL), lower asymptote (guessability, 3-PL) parameters.
Differential item functioning (DIF) detectionYes, in secondary analysisYes, in secondary analysis
First conspicuous appearanceRasch, Georg. (1960) Probabilistic models for some intelligence and attainment tests. Copenhagen: Danish Institute for Educational Research.Birnbaum, Allan. (1968). Some latent trait models. In F.M. Lord & M.R. Novick, (Eds.), Statistical theories of mental test scores. Reading, MA: Addison-Wesley.
First conspicuous advocateBenjamin D. Wright, University of ChicagoFrederic M. Lord, Educational Testing Service
Widely-authoritative currently-active proponentDavid Andrich, Murdoch Univ., Perth, AustraliaRonald Hambleton, University of Massachusetts
Introductory textbookApplying The Rasch Model.T.G. Bond and C.M. FoxFundamentals of Item Response Theory.R.K. Hambleton, H. Swaminathan, and H.J. Rogers.
Widely used softwareWinsteps, RUMM, ConQuestLogist, BILOG

Rasch dichotomous model vs. One-parameter Logistic Model … Rasch Measurement Transactions, 2005, 19:3 p. 1032

The URL of this page is www.rasch.org/rmt/rmt193h.htm

Website: www.rasch.org/rmt/contents.htm