THE CROSS-CULTURAL EFFECTS OF RESCALING VERBAL AND NUMERIC RATING SCALES USING CORRESPONDENCE ANALYSIS

It is common in cross-cultural research to treat variables as if they have interval scale properties irrespective of how these scales were constructed. The purpose of this research is to explore the differences in and consequences of how respondents from different countries use the same scales over the same scale items. Previously collected data from four countries using the same 6-point verbal and 10-point numeric rating scales were used. The data was rescaled using the delta chi-square and correspondence analysis techniques (Bendixen & Yurova, 2012). The differences in means of the variables were compared in pairs for raw and rescaled data. Bootstrapping was used to estimate confidence intervals of the differences between raw and rescaled values. Furthermore, a method of partitioning the differences in means into rescaling and cultural components was devised. In 83.3% and 94.9% of the cases, the differences in raw versus rescaled means were significant at the 5% level for verbal and numeric variables respectively. The results of partitioning indicate that by not rescaling the data, the differences in raw means consistently exaggerate the true cultural differences: the extracted cultural component was underestimated by 12.8% on average for verbal variables and by 5.3% for numeric variables. Therefore, reporting the differences in raw means as a true reflection of cultural differences is in error. Finally, the effect of rescaling in pancultural research was investigated by comparing the factor structure of doubly standardized raw and rescaled data. Pancultural research attempts to identify etic, or universal, dimensions of human culture and employs double standardization to remove cultural and individual biases inherent to cross-cultural data. While no differences in latent factors extracted were found for raw and rescaled data, considerable differences in the variance explained and slight differences in factor structure were found for double standardized rescaled and double standardized raw data. The results of this research indicate that researchers and practitioners in the field of cross-cultural research should choose their scales very carefully. Furthermore, to extract true cultural differences it is probably necessary to rescale and partition differences in means. Further research on the impact of rescaling is proposed