Measurement bias in cross-cultural surveys can seriously threaten the validity of hypothesis tests. Direct comparisons of means depend on the assumption that differences in observed variables reflect differences in the underlying constructs, and not an additive bias that may be caused by cultural differences in the understanding of item wording or response category labels. However, experience suggests that additive bias can be found more often than not. Based on the concept of partial measurement invariance (Byrne, Shavelson and Muthén 1989), the present paper develops a procedure for eliminating additive bias from cross-cultural data. The procedure involves four steps: (1) embed a potentially biased item in a factor-analytic measurement model, (2) test for the existence of additive bias between populations, (3) use the factor-analytic model to estimate the magnitude of the bias, and (4) replace differences between observed item means by unbiased estimates of the difference between "true" population means. Sampling moments are derived for the purpose of hypothesis testing. The procedure is demonstrated in a numerical example. Potential applications include comparisons of means from different cultures, and correction of raw data before use in cross-cultural segmentation studies.
Journal of Business Research, 2005, Vol 58, Issue 1, p. 72-78