Response style bias has been shown to seriously contaminate the substantive results drawn from
survey data; particularly those conducted using cross-cultural samples. As a consequence. identification of
response formats that suffer least from responst style bias has been called for. Previous studies show that
respondents' personal characteristics, such as age, education level and culture, are connected with response
Differences in the way respondents interpret and utilise researcher-defined fixed rating-scales (e.g. Likert
formats), poses a problem for survey researchers. Techniques that are currently used to remove response
bias from survey data are inadequate as they cannot accurately determine the level of contamination present
and frequently blur true score variance. Inappropriate rating-scales can impact on the level of response style
bias manifested, insofar as they may not represent respondents' cognitions. Rating-scale lengths that are too
long present respondents with some response categories that are not 'meaningful', whereas rating-scales that
are too short force respondents into compressing their cognitive rating-scales into the number of response
categories provided (this can cause ERS contamination - extreme responding). We are therefore not able to
guard against two respondents, who share the same cognitive position on a continuum, reporting their stance
using different numbers on the rating-scale provided. This is especially problematic where a standard fixed
rating-scale is used in cross-cultural surveys.
This paper details the development of the Individualised Rating-Scale Procedure (IRSP), a means of
extracting a respondent's 'ideal' rating-scale length, and as such 'designing out' response bias, for use as the
measurement instrument in a survey. Whilst the fundamental ideas for self-anchoring rating-scales have been
posited in the literature, the IRSP was developed using a series of qualitative interviews with participants.
Finally, we discuss how the IRSP's reliability and validity can be quantitatively assessed and compared to
typical fixed researcher-defined rating-scales, such as the Likert format.