Measurement scales and multivariate analysis

✍ Scribed by G. J. Boris Allan

• Publisher: Springer Netherlands
• Year: 1979
• Tongue: English
• Weight: 346 KB
• Volume: 13
• Category: Article
• ISSN: 0033-5177
• DOI: 10.1007/bf00188016

✦ Synopsis

All, or nearly all, multivariate analysis is the multivariate analysis of ordinal-scaled variates; and, rather than being concerned with "... the consideration of cases in which the (Pearson) product-moment coefficient does not detect spuriousness when it should according to ordinal models ..." (Hawkes, 1976: p. 191), we should wonder how we can attempt to perform a multivariate analysis of intelval-scaled variates. The force of Labovitz's (1970) argument, as I construe it, is that all our measurement is in essence ordinal -"income", for example, is at first sight interval-scaled, but "income" is only an operational measurement of some concept such as "wealth" and the two need not be linearly related. I will argue that, in accepting the inherent ordinallity of the scale of measurement, we are not committed to what may be termed "ordinal" multivariate analysis.