Scaling unobservable constructs in social science

SUMMARY The use of the logit–probit model for constructing social measurement scales is explored by using two data sets. The first was specifically designed for scaling and consists of answers by 1490 individuals to five questions on economic self-determination. The second is from the Workplace Industrial Relations Survey and relates to answers to six questions addressed to 1005 firms about innovations. It is shown that although scales can be constructed their precision (reliability) is likely to be low. Further, for examples like these, latent class models may give an equally satisfactory account of the data.