Comparing Different Variance Component Estimation Methods Used in Generalizability Theory Decision Studies


Ozberk E. H., GELBAL S.

JOURNAL OF MEASUREMENT AND EVALUATION IN EDUCATION AND PSYCHOLOGY-EPOD, cilt.5, sa.2, ss.91-103, 2014 (ESCI) identifier

Özet

The aim of this research is to compare various variance component estimations procedures with using signal noise ratio and error tolerance ratio which is offered with generalizabiity and Phi coefficients in non-normal distrubitions (Brennan, 2001; Kane, 1999). This research compares variance components estimations with using ANOVA and bootstrap procedures in non-normal disturbitions in one facet design G studies. Data were gathered with using two seperate procedures (a) data simulation and (b) sampling simulation. In data simulation part, it's been simulated a non-normal dichotomous data set which fits to unidimensional personitem matrix 60x5 which fits to (bxm) design. All the simulations replicated 25 times. In sampling simulation sections datas, gathered from data simulation sections has been bootstrapped 1000 times according to the each facet. Standart errors, variance components, relative and absolute error are estimated according to the each facets with using ANOVA and bootstrap procedures. The results also show that in non-normal dichotomously scored datas best signal-noise ratio has estimated in boot. b procedure, and best error-tolerance ratio has been estimated in boot-m procedure. Thus, boot-m procedures gives more valid estimations and boot-b procedure gives more reliable and precise estimations of universe scores in G studies rather than other procedures.