From 'aisle' to 'labile': a hierarchical NART scale revealed by Mokken scaling

Decline in cognitive ability is a core diagnostic criterion for dementia. Knowing the extent of decline requires a baseline score from which change can be reckoned. In the absence of prior cognitive ability scores, vocabulary-based cognitive tests are used to estimate premorbid cognitive ability. It is important that such tests are short yet informative, to maximise information and practicability. The National Adult ReadingTest (NART) is commonly used to estimate premorbid intelligence. People are asked to pronounce 50 words ranging from easy to difficult but whether its words conform to a hierarchy is unknown. 587 healthy, community-dwelling older people with known age
11 IQ scores completed the NART as part of the Lothian Birth Cohort 1936 study.
Mokken analysis was used to explore item responses for unidimensional, ordinal and hierarchical scales. A strong hierarchical scale ('mini-NART') of 23 of the 50 items was identified. These items are invariantly ordered across all ability levels. The validity of the interpretation of this briefer scale's score as an estimate of premorbid ability was examined using the actual age 11 IQ score. The mini-NART accounted for a similar amount of the variance in age 11 IQ as the full NART (NART=46.5%, MiniNART=44.8%).
The mini-NART is proposed as a useful short clinical tool to estimate
prior cognitive ability. The mini-NART has clinical relevance, comprising highly
discriminatory, invariantly ordered items allowing for sensitive measurement, and adaptive testing, reducing test administration time and patient stress.