Individual alpha peak frequency is related to latent factors of general cognitive abilities

Highlights

• The association of EEG α frequency to cognition is a long standing question.

• We investigated this relationship using structural equation modeling (SEM).

• SEM allowed estimation of this association unbiased by measurement error.

• α frequency and g intelligence are significantly correlated at the latent level.

• This parsimoniously explains correlations with diverse tasks reported up to now.

Abstract

Some eighty years after the discovery of the human electroencephalogram (EEG) and its dominant rhythm, alpha (~ 10 Hz), the neurophysiological functions and behavioral correlates of alpha oscillations are still under debate. Similarly, the biological mechanisms contributing to the general factor of intelligence, or g, have been under scrutiny for decades. Individual alpha frequency (IAF), a trait-like parameter of the EEG, has been found to correlate with individual differences in cognitive performance and cognitive abilities. Informed by large-scale theories of neural organization emphasizing the general functional significance of oscillatory activity, the present study replicates and extends these findings by testing the hypothesis that IAF is related to intelligence at the level of g, rather than at the level of specific cognitive abilities. Structural equation modeling allowed us to statistically control for measurement error when estimating the association between IAF and intellectual functioning. In line with our hypothesis, we found a statistically reliable and substantial correlation between IAF and g (r = .40). The magnitude of this correlation did not differ significantly between younger and older adults, and captured all of the covariation between IAF and the cognitive abilities of reasoning, memory, and perceptual speed. The observed association between IAF and g provides a parsimonious explanation for the commonly observed diffuse pattern of correlations between IAF and cognitive performance. We conclude that IAF is a marker of global architectural and functional properties of the human brain.

Introduction

Understanding the mapping of anatomical and functional properties of the brain onto individual differences in cognitive functioning is a major goal of cognitive neuroscience and central to attempts at delineating the neural mechanisms associated with intelligent behavior (Deary et al., 2010, Jung and Haier, 2007, Kanai and Rees, 2011, Narr et al., 2007, Toga and Thompson, 2005). The notion that characteristics of the human electroencephalogram (EEG) are related to measures of intelligence has a long history going back almost as far as the discovery of the EEG by Hans Berger, 1929, Berger, 1930, with the alpha rhythm being one of the first targets for studying the relationship between EEG markers and intelligence. It is easily discerned in the raw EEG traces at parieto-occipital electrodes when persons are awake and relaxed with their eyes closed (Adrian and Matthews, 1934, Adrian and Yamagiwa, 1935). The average of alpha frequency is around 10 Hz, with a range of approximately 8–12 Hz in healthy adults (cf. Aurlien et al., 2004, Chiang et al., 2011, Niedermeyer and Lopes da Silva, 1999). One of the earliest observations relating the alpha frequency (AF) to individual differences in cognitive functioning was the finding of ‘mentally retarded’ patients (Berger, 1933) exhibiting systematically slower alpha waves. Since then it has remained an appealing notion that ‘smarter brains are running faster’ (Posthuma et al., 2001), with a long history of findings supporting this idea but also with inconsistent reports (see below).

From a psychometric perspective, AF is a promising neurophysiological trait marker for investigating the association between brain and cognitive functioning (Grandy et al., in press). The AF is characterized by a remarkably high relative and mean test–retest stability in samples of healthy individuals (Deakin and Exley, 1979, Gasser et al., 1985, Kondacs and Szabó, 1999, Salinsky et al., 1991), with test–retest intervals ranging up to several years (Kondacs and Szabó, 1999). Stability coefficients of AF derived from EEG recordings with closed eyes were found to be typically around .75 to .90, with higher correlations for shorter test–retest intervals. It follows that AF shows trait-like characteristics, with substantial and stable differences between individuals that justify use of the term, individual alpha frequency (IAF), to denote the AF of a given person (cf. Doppelmayr et al., 1998, Klimesch, 1996, Klimesch, 1997). In full agreement with this line of reasoning, IAF has been found to show considerable heritability (Lykken et al., 1974, Posthuma et al., 2001, Smit et al., 2006, van Baal et al., 2001, van Beijsterveldt and Boomsma, 1994, van Beijsterveldt and van Baal, 2002, Vogel, 1970).

The magnitude and consistency of the association between IAF and individual differences in intellectual functioning are a matter of ongoing debate (cf. Anokhin and Vogel, 1996, Ellingson, 1966, Posthuma et al., 2001, Vogel and Broverman, 1964). Several studies have reported significant correlations between IAF and verbal abilities (Angelakis et al., 2004a, Anokhin and Vogel, 1996, Mundy-Castle, 1958), memory performance (Klimesch et al., 1990, Klimesch et al., 1993, Lebedev, 1994, Saletu and Grunberger, 1985), Digit Span performance (Angelakis et al., 2004b, Clark et al., 2004), performance on the Raven matrices (Anokhin and Vogel, 1996), response control (Angelakis et al., 2004a), speed of information processing (Klimesch et al., 1996, Mundy-Castle, 1958), reaction times (Surwillo, 1961, Surwillo, 1963, Surwillo, 1964), and indicators of general intelligence (Giannitrapani, 1985, Mundy-Castle, 1958, Mundy-Castle and Nelson, 1960). Collectively, these reports indicate that IAF is functionally related to various types of cognitive performance. At the same time, some studies failed to detect any associations between IAF and cognitive performance (Ellingson, 1966, Posthuma et al., 2001, Vogel and Broverman, 1964). Moreover, some of the initial findings, such as correlations between IAF and verbal subtests of the Wechsler Intelligence Test (Angelakis et al., 2004a, Mundy-Castle, 1958), were not replicated in later work (Posthuma et al., 2001). However, it needs to be cautioned that many of the studies had small sample sizes. In addition, associations between IAF and intelligence were often computed at the level of individual cognitive tests. Generally, such tests are poor indicators of intelligence because their reliability and validity is far from perfect. In sum, it is conceivable that combinations of sampling error, lack of statistical power, measurement error as well as variance specific to individual cognitive tests have obscured, and potentially lowered, the observable association between IAF and intelligence.

Despite these limitations, the predominantly positive correlations between IAF and a variety of cognitive performance measures point to an association between IAF and the positively correlated manifold of intellectual abilities (cf. Deary et al., 2010). Given its high stability and heritability, it is plausible to assume that IAF is linked to cognitive performance in a general rather than in a specific manner. Specifically, we would like to propose that IAF is a physiological marker of Spearman's ‘general intelligence’ (Spearman, 1904). It has been shown that second-order factors extracted from a wide range of intelligence test batteries tend to be highly correlated (Johnson et al., 2004, Johnson et al., 2008), suggesting that the common variance extracted from different intelligence test batteries can be seen as different expressions of a common construct termed general intelligence, or g. Hence, we predicted that allowing for an association between IAF and intelligence at the level of g would obliterate the need to model the IAF–intelligence link at lower levels of the intelligence hierarchy.

We tested these predictions using data from the COGITO study (Schmiedek et al., 2010). The COGITO study was conducted to investigate day-to-day variability of cognitive performance and effects of extensive cognitive training. Within the COGITO study, a large sample of 145 younger adults (20–31 years) and 142 older adults (65–81 years) completed 27 subtests of the Berlin Intelligence Structure test (BIS; Jäger et al., 1997) from the cognitive domains perceptual speed, memory, and reasoning as part of a large battery of baseline assessments. Additionally, approximately 30% of the participants – 45 younger and 40 older adults – took part in two EEG sessions, one directly after the baseline assessments and the other on average 6.6 months later. Resting state recordings with eyes closed and eyes open were obtained at each of these two occasions.

The constructs in the BIS test broadly cover the range and type of cognitive tests that have been shown to correlate with IAF in previous work (see above). Importantly, by making use of the fairly large full COGITO parent sample (n = 287), we were able to establish a sound confirmatory factor model representing the structure of intelligence by three first-order factors and one second-order latent factor g. Following the confirmation of this factor structure, structural equation modeling (SEM) with full information maximum likelihood (FIML) estimation was used to project a latent IAF factor, which was derived from eyes closed and eyes open resting state EEG measurements, into the latent space of intellectual abilities. This procedure allowed us to estimate the association between IAF and g unbiased by measurement error at the latent level, and to test whether links of IAF to cognitive performance at lower levels of the hierarchy (i.e., first-order factors, specific tests) are needed after accounting for the link between IAF and g. If there were no need to estimate such residual associations, this would be consistent with the guiding hypothesis that IAF–intelligence associations reflect general properties of the human brain. To our knowledge, the association between IAF and intelligence has not been investigated using SEM techniques.