To the Editor:

The jangle fallacy is the assumption that two identical or highly similar things are different because they are named or labeled differently [1]. In their examination of factors associated with risk for Alzheimer’s disease (AD), the major conclusion of Hu et al. [2] is that “cognitive performance may protect against AD independently of intelligence” and that “intelligence may protect against AD dependently of cognitive performance”. Consideration of the jangle fallacy, however, demonstrates that this conclusion is not meaningful. The jangle fallacy here refers to the constructs of intelligence and cognitive performance. Hu et al. also refer to cognitive performance as general cognitive function, and we therefore use these two terms interchangeably.

In our view, Hu et al.’s conceptualization of intelligence and general cognitive function is problematic. It is that conceptualization that leads, in part, to the jangle fallacy, i.e., the erroneous assumption that these are distinct constructs. Hu et al. state that in addition to intelligence, “general cognitive function estimates the overall cognitive performance for a person, and is a prominent and relatively stable human phenotype compared with specific cognitive abilities, such as intelligence.” However, intelligence is not a specific cognitive ability; it represents general cognitive function. Indeed, in neuropsychology, intelligence/general cognitive function is always differentiated from specific cognitive abilities such as memory, executive function, or visual-spatial ability. Hu et al. also state that intelligence is “largely fixed early in life.” While correlational results suggest stability in adulthood, other results indicate that intelligence may not be so fixed across the lifespan. Stability of intelligence from early in life is rather hard to measure because the cognitive tests used to evaluate young children and adults usually need to be very different. One exception is the Lothian Birth Cohorts in which the same intelligence test was given to children at age 11 and then to the same individuals in their 70 s [3]. The correlation was 0.63 over this very long interval, but there was a mean score increase of 1.12–1.51 SDs [3]. In contrast, when considering intelligence from young adulthood to older adulthood, mean level change tends to be in the opposite direction. For example, the same Wechsler IQ is achieved with lower raw scores for a 75-year old compared with a 25-year old [4]. Thus, there is conceptual confusion in the basic arguments of the paper.

Regarding the jangle fallacy specifically, examination of measures of intelligence and cognitive performance used by Hu et al. shows considerable overlap. As Hu et al. explain (their Supplementary Table 5), their index of cognitive performance is based on data from the COGENT consortium and UK Biobank. COGENT used the first principal component (g-score) of at least 3 cognitive tests, many of which are intelligence test subscales. The UK Biobank measure was a verbal-numeric reasoning task, considered a measure of fluid score. However, there are many different measures of fluid intelligence, and intelligence tests also typically include multiple fluid intelligence measures. The Wechsler Adult Intelligence Test-IV, for example, includes tests of verbal reasoning (Similarities), matrix reasoning, and block design [4]. The Hu et al. intelligence GWAS dataset (their Supplementary Table 6) includes several samples that used a g-score, i.e., the first principal component calculated as it is in the COGENT measures which are considered to be cognitive performance by Hu et al. Two measures from UK Biobank in Table 6 are labeled as fluid intelligence, which is also how the UK Biobank measure for cognitive performance (in their Supplementary Table 5) is labeled. Others are scores from well-known intelligence tests. Taking the first principal component of the intelligence test subscales would yield what neuropsychologists and cognitive scientists would conceptualize as a g-score. Thus, g does not represent a different construct; it is simply a summary measure of intelligence, typically accounting for ~40% of the variance in the cognitive tests that contribute to it. Thus, the considerable theoretical and measurement overlap between Hu et al.’s intelligence and cognitive performance measures seriously challenges the idea that they are distinct constructs.

Despite considerable overlap, the intelligence and cognitive performance measures may still appear to be different simply because they are derived from varied measures from different samples, with additional variability introduced by the fact that the measures for some individuals were based on a single brief test rather than the first principal component based on multiple different tests. Such differences are most likely largely attributable to measurement error rather than to meaningful differences between intelligence and cognitive performance. Thus, the data support the view that cognitive performance (general cognitive function) and intelligence are not measuring different constructs or different cognitive abilities. To further emphasize this point, two major GWAS meta-analyses cited by Hu et al. include many of the same samples and use the same principal component as their measure in the samples that contribute to their respective analyses. Yet in the GWAS of Savage et al. [5] the phenotype being assessed is called intelligence” whereas in the GWAS of Davies et al. [6] the same phenotype is called general cognitive function.

In the end, the idea that Hu et al. are capturing something different by intelligence and cognitive performance cannot be considered a valid inference. Differences in the two are no more than differences we might find between any two different sets of data measuring the same phenotype with varying tests. Their separate analyses comparing the association of education with either their intelligence or their cognitive performance measures constitute valid contributions to the literature. However, their analyses that include intelligence controlling for cognitive performance or cognitive performance controlling for intelligence cannot be considered meaningful because there is no meaningful distinction between these two terms as defined by Hu et al.