Premium
ATTRIBUTE AND RELIABILITY
Author(s) -
Ross John,
Lumsden James
Publication year - 1968
Publication title -
british journal of mathematical and statistical psychology
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 3.157
H-Index - 51
eISSN - 2044-8317
pISSN - 0007-1102
DOI - 10.1111/j.2044-8317.1968.tb00414.x
Subject(s) - generalizability theory , reliability (semiconductor) , ambiguity , test (biology) , dimension (graph theory) , mathematics , classical test theory , statistics , computer science , econometrics , reliability engineering , item response theory , psychometrics , combinatorics , paleontology , power (physics) , physics , quantum mechanics , engineering , biology , programming language
An analysis of classical test theory and of more recent developments shows that the concepts of true score, parallel test and reliability are poorly defined. Where reliability is defined as generalizability it is not so much poorly defined as detached from the essential interior properties of the test. It is proposed that tests as such be restrictively defined as devices which map an attribute in a single dimension. A logistic model which fits this conception is examined and it is shown that true score, parallel test and reliability can be defined without ambiguity, inconsistency or circularity. Deeper investigation reveals that the true score concept is not only unnecessary as a basis of reliability but also a distraction. Reliability is formulated as a characterization of the test's mapping. It is shown that the mapping can be characterized for the model and it is also shown that there is a distinction between information loss due to regression and error which probably lies behind the poorly conceived notion of true score.