Applications of item response theory to analysis of attitude scale translations.

Abstract

Methods of detecting item bias developed from item response theory (IRT) are generalized to analyze the fidelity of translations of psychological scales into foreign languages. These IRT methods are considered as alternatives to classical methods. Item characteristic curves obtained from responses to the original and target language versions of the instrument are examined for significance of differences. Data from a Spanish translation of the Job Descriptive Index administered to 203 Spanish-speaking employees are used to illustrate the method. Significance tests indicate that three items on the 72-item instrument are biased. Subsequent inspection of these three items by a language consultant revealed inadequate translation of one item. Because the number of items determined to be biased by significance testing is quite close to the expected number of Type I errors, it is concluded that the overall quality of the translation is quite good. Finally, it is argued that equivalent item characteristic curves across the original and translated items of a scale produce equivalent measurements in both languages, and nonequivalent item characteristic curves pinpoint differences between the two versions of the scale.Two increasingly important areas of applied psychological research are cross-cultural social psychology and cross-national industrial-organizational psychology. The need to assess and make statements about differences between cultures as well as within cultures is an integral part of both areas. Herein lies the problem addressed in this article: Cross-cultural research depends on cross-cultural comparisons, which, in turn, usually depend upon the meaningfulness of measuring instruments and scale scores both within and between the cultures in question. Thus, cross-cultural research and attitude survey This research was supported by Contract N000-14-75-C-0904 from the Office of Naval Research, Charles L. Hulin, principal investigator. The authors would like to thank Frank J. Smith for providing the sample of bilingual subjects and Robert Linn, Neil Dorans, Malcolm Ree, and Harry Triandis for reading and commenting on previous drafts. We would also like to thank Michael V. Levine for suggesting the logit transformation.

Analysis of Differential Item Functioning in Translated Assessment Instruments

“…Within the IRT framework, Drasgow has defined measurement equivalence as a property that exists &dquo;... when the relations between observed test scores and the latent attribute measured by the test are identical across subpopulations&dquo; (Drasgow, 1984, p. 134 (Candell & Hulin, 1987;Candell & Roznowski, 1984;de Vera, 1985;Drasgow & Hulin, 1989;Drasgow & Lissak, 1983;Ellis, 1989Ellis, , 1991Ellis, Minsel, & Becker, 1989;Hulin, Drasgow, & Komocar, 1982;Hulin & Mayer, 1986;Osberg, Scott, & Raju, 1985). Some of these IRT analyses were based on small samples and, in others, there was only limited control over the linguistic abilities of the samples used.…”

International Journal of Business and Management, Vol. 5, No. 3, March 2010, all in one PDF file

“…It has also been used in P-E fit studies to measure job satisfaction (Smart, Elton and McLaughlin, 1986). It has been shown to be reliable and valid not only in America but also in Spain (Hulin, Drasgow and Komocar, 1982), Saudi Arabia (Maghrabi and Johnson, 1995), Hong Kong (Lam, 1995) and Singapore (Goh, Koh and Low, 1991).…”

“…This aspect was tested by Hanisch (1992) and shown to be correct. Total job satisfaction was obtained by simply summing up the results of the five facet scales to obtain an acceptable measure of global job satisfaction as was done by Hulin, Drasgow, and Komocar (1982).…”

The job characteristics model in Hong Kong.

“…This back translation procedure helped insure an accurate prose translation that was decentered from a literal English language translation (Werner & Campbell, 1970). It did not, however, involve a psychometric analysis (Hulin, Drasgow, & Komocar, 1982). The decentered Chinese version was used to collect data from the 13 schools, whereas the original English language version was used to collect data from the banks, telephone company, and Labour Department.…”