Introduction

Comparisons of the results of research undertaken in varied cultural settings can be misleading unless comparable methods have been used to collect the data. Depending on the nature of the text, the differences in meaning may be easy or difficult to detect. For example, redundancy in describing a particular event may help to convey meaning and facts correctly in spite of a non-equivalence in meaning of particular passages, phrases or words in the translated version. In questionnaires and standardised assessment instruments, however, non-equivalence can seriously distort results. Prince and Mombour (1967), for example, assigned 80 bilingual subjects randomly to two groups. The first group was given half of the Langner 22 item questionnaire (at that time used for screening purposes) in English, followed by the other half in French. The second group was given the questionnaire with the French half first and the English second. In four of the 22 items the frequency of positive answers differed significantly between the two languages, even though they have similar structure and syntax, and share a significant number of words. In a more recent study, Pichot and colleagues (1991, personal communication) compared the results obtained when the QD2 instrument – a self-report questionnaire containing items usually found in depressive illness – was given in Arabic and French to two groups of bilingual subjects. There was a significant difference between the two language versions on 24 items, all but two of which were scored higher in French. This was true for both male and female respondents. It is obvious that the danger of distortion is much larger if the languages belong to different families and the subjects have very different cultural backgrounds.

Tests of equivalence are also important when an instrument is applied to groups belonging to culturally or socially very distinct groups, although speaking the same language. The risk of distortion is lower with semi-standardised instruments applied by well-trained interviewers conversant with the culture and the languages in which the instrument is applied, than when a fully standardised instrument is used. Nevertheless the risk exists and should be borne in mind in testing the instrument and interpreting the results.

Several aspects of equivalence should be considered when comparing different language versions of a clinical instrument.