STRONG TRUE SCORE THEORY- IRT LECTURE 12 EPSY 625
Strong True Score Theory Equivalent to g-theory: subject ability item difficulty Extension of true score theory Uses form of logistic regression: e Dag( - bg ) Pr(1) = 1 + e Dag( - bg )
Strong True Score Theory Equivalent to g-theory: subject ability item difficulty Extension of true score theory Uses form of logistic regression: e Dag( - bg ) Pr(1) = 1 + e Dag( - bg )
Pg()Pg() ABILITY Difficulty b g Probability of Correct Answer Item Response Model Discrimination a g Difficulty: the ability score needed for a 50% probability of getting the item right Discrimination: slope of the IRT curve at the 50% probability intersection Assumptions:.local independence of items.single ability true score.logistic model for items: e Dag( - bg ) Pr(1) = 1 + e Dag( - bg )
MODELS One parameter model- only b g varies across items Two parameter model- both a g and b g vary across items
1-PARAMETER ESTIMATION MPLUS: TITLE:this is an example of a one –parameter logistic item response theory (IRT) model DATA:FILE IS ex5.5.dat; VARIABLE:NAMES ARE u1-u5; CATEGORICAL ARE u1-u5; ANALYSIS:ESTIMATOR = MLR; MODEL:f BY u1 (1) u2 (1) u3 (1) u4 (1) u5 (1); OUTPUT:TECH1 TECH8;
MPLUS 5.5 OUTPUT Thresholds Estimates S.E. Est./S.E. F BY U U U U U Thresholds U1$ U2$ U3$ U4$ U5$ Fixed slopes Item difficulties
2-PARAMETER ESTIMATION MPLUS: TITLE:this is an example of a two- parameter logistic item response theory (IRT) model DATA:FILE IS ex5.5.dat; VARIABLE:NAMES ARE u1-u20; CATEGORICAL ARE u1-u20; ANALYSIS:ESTIMATOR = MLR; MODEL:f BY u1-u20; OUTPUT:TECH1 TECH8;
MPLUS 5.5 OUTPUT MODEL RESULTS Estimates S.E. Est./S.E. F BY U U U U U U U U U U U U U U U U U U U U Thresholds U1$ U2$ U3$ U4$ U5$ U6$ U7$ U8$ U9$ U10$ U11$ U12$ U13$ U14$ U15$ U16$ U17$ U18$ U19$ U20$ Slopes (a parameters) difficulties (b parameters)
Three parameter model a g and b g vary across items parameter c g for guessing is added: Empirical studies indicate c g is usually lower than guessing rate Requires 5, ,000 cases for stable estimation (ETS, ACT or NAEP samples)
Pg()Pg() ABILITY Probability of Correct Answer agag bgbg cgcg
Pg()Pg() 11.5 (1,2)(1,2) Pg()Pg() MULTIDIMENSIONAL IRT - CONCEPTS AND ISSUES - Difficulty in getting estimates - Inconsistent with factor model analysis