Document Type

Journal Article


A team of classically psychometrically trained researchers wanted to measure the construct of a yak. They did not bother to carefully conceptually define what a yak consists of (see Rossiter, 2002). Instead, they went out and found a horse, which was much more convenient because plenty of previous researchers have published items to measure the horse construct. They factor-analyzed the horse items and found that the four legs items correlated well, formed a unidimensional factor, and provided the necessary four indicators to meet the requirements of the oracular veterinarian, Lisrel. They dropped the two single items measuring the enormous head and the shaggy coat. They did not load on the legs factor. Multitrait-multimethod (MTMM) analysis of the "four legs" scale revealed that it converged very nicely with all other measures of land mammals (with the conspicuous exception of kangaroos) and that it showed low discriminant correlations with most measures of reptiles and all measures of fish. Subsequent researchers appreciated the nice psychometric properties of the "four legs" measure of the yak and thus it became established as the measure of yaks in all the leading journals. Yet again, statistics defined the construct and saved the researchers a lot of bothersome rational thinking. They used the conventional Likert answer scales to measure how much the raters (whoever they were) agreed or disagreed that the object (whatever that was) has four legs. Fortunately, no raters "strongly disagreed" with any of the legs so the researchers did not have to ponder what "strongly disagreeing with a leg" might mean when interpreting the score on this item. If someone did check "strongly disagree", then the researchers would just call it a "1" and throw it into the statistical package with all the other numbers, hopefully with some "7’s" among them. A group mean sum score of 5.3 across the four legs? That is clear to any manager, isn’t it? We are observing most of a yak here, aren’t we? You decide.