Kan-van-der-maas-and-levine-2019 - no support for g via network psychometrics and mutualism theory

File: kan-van-der-maas-and-levine-2019 - annotated.pdf

Annotation summary:

--- Page 53 ---

Highlight (yellow):
Intelligence 73 (2019) 52–62

Highlight (yellow):
Extending psychometric network analysis: Empirical evidence against g in favor of mutualism? Kees-Jan Kana⁎ , Han L.J. van der Maasb, Stephen Z. Levine

Highlight (yellow):
A B S T R A C T

The current study implements psychometric network analysis within the framework of confirmatory (structural equation) modeling. Utility is demonstrated by three applications on independent data sets. The first application uses WAIS data and shows that the same kind of fit statistics can be produced for network models as for traditional confirmatory factor models. This can assist deciding between factor analytical and network theories of intelligence, e.g.g theory versus mutualism theory. The second application uses the 'Holzinger and Swineford data' and illustrates how to cross-validate a network. The third application concerns a multigroup analysis on scores on the Brief Test of Adult Cognition by Telephone (BCATC). It exemplifies how to test if network parameters have the same values across groups. Of theoretical interest is that in all applications psychometric network models outperformed previously established (g) factor models. Simulations showed that this was unlikely due to overparameterization. Thus the overall results were more consistent with mutualism theory than with mainstreamg theory. The presence of common (e.g. genetic) influences is not excluded, however.


--- Page 54 ---

Highlight (yellow):
We note that from a descriptive (statistical) point of view bi-factor models may tend to fit better, but also that from an explanatory (sub-stantive theoretical) perspective, a bifactor model of intelligence is considered unsatisfactory (e.g., Jensen, 1998; Hood, 2008). Decisions as to which model to adopt as a the best model should rely on both fit and theory, not on fit itself (Morgan, Hodge, Wells, & Watkins, 2015; Murray & Johnson, 2013). In other words, theory drives, fit assists

Underline (red):
theory drives, fit assists

Highlight (yellow):
Kovacs & Conway, 2016)

Note (yellow):
POT

Highlight (yellow):
Secondly, that the general factor represents a single, unitary source of variance is not a given, but a hypothesis, which – like any other scientific hypothesis – requires empirical scrutiny

Underline (red):
not a given, but a hypothesis

Highlight (yellow):
results of other (non-psychometric) lines of re-search are of importance.

Highlight (yellow):
In brief, the mutualism model of intelligence is a model of cognitive development that was inspired by research in ecosystem modeling, where the dynamics between variables are due to reciprocal causation. The key idea is that such reciprocal causation also occurs among cog-nitive abilities during their development.

Highlight (yellow):
Unlike in g theory, these capacities are (or can be) con-sidered statistically independent. Yet, because the growth of a given cognitive ability is not only limited by its own, specific limiting capa-city, but is also affected by the level of other cognitive abilities (through the dynamical interactions), and thus by their corresponding limiting capacities, the cognitive abilities themselves become positively corre-lated throughout the course of their development.

Highlight (yellow):
The essence of statistical modeling and model selection (Kline, 2015) is the combination of Popperian logic (Popper, 2005) and Oc-cam's razor or 'the law of parsimony'

Highlight (yellow):
Mutualism thus provides an alternative explanation of the positive manifold,


--- Page 55 ---

Highlight (yellow):
As has been noted (van der Maas et al., 2017), mutualism – the idea of dynamic coupling between cognitive abilities – aligns neatly with some of the latest and most rapid developments in psychometrics, namely psychometric network modeling (Borsboom, 2008; Epskamp, Cramer, Waldorp, Schmittmann, & Borsboom, 2012)

Highlight (yellow):
It shows that one may conceptualize cognitive abilities as being related to each other directly, rather than through common, unobserved variables on which they depend. Indeed, the connections between any pairs of cognitive variables can be modeled using (full or full partial) correlations only, hence without postulating any latent factors.

Highlight (yellow):
In other words, factor models are nested within network models.


--- Page 59 ---

Highlight (yellow):
As demonstrated, the implementation of network models within a confirmatory (structural equation) modeling framework (Boker et al., 2011; Epskamp et al., 2017) permits, for instance, (1) the comparisons among factor and networks models, which can assist in the comparison of a priori theo-retically driven models, (2) the comparison of networks over groups, and (3) the combination of these two.

Highlight (yellow):
From a descriptive viewpoint concerning individual differences in cognitive performance, the major finding of interest was that the psy-chometric networks provided better descriptions of the data than pre-viously established confirmatory factor analytic models. Additional si-mulations showed this is unlikely due to overparameterization. In view of substantive theory, our results imply that the hypothesis of an un-derlying general factor of intelligence is not required in order to explain the pattern of correlations between the different cognitive performance measures. More strongly, the current results provide an empirical ar-gument against g theory (e.g. Jensen, 1998) favoring the mutualism theory of intelligence (van der Maas et al., 2006). The latter posits that positive associations between cognitive abilities arise through re-ciprocal dynamical interaction between those abilities during devel-opment, and that this is a sufficient explanation

Underline (red):
our results imply that the hypothesis of an un-derlying general factor of intelligence is not required in order to explain the pattern of correlations between the different cognitive performance measures. More strongly, the current results provide an empirical ar-gument against g theory (e.g. Jensen, 1998) favoring the mutualism theory of intelligence (van der Maas et al., 2006).


--- Page 61 ---

Highlight (yellow):
Apart from the fact that psychometric network models outperformed traditional factor models, we obtained additional findings of theoretical interest.

Highlight (yellow):
Overall, the current study promotes confirmatory psychometric network analysis, in the field of cognition and intelligence in particular

Highlight (yellow):
and in differential psychology in general.

Highlight (yellow):
With respect to the debate concerning the theoretical status ofg, we conclude the following. We do not exclude the presence of common or general influences, e.g. of certain genetic variants or environmental variables like exposure to education. The question to be answered is more how such effects could have arisen: Are they the result of dyna-mical reciprocal interactions or are they due to a single mediating variable g which has never been found to exist? The evidence from the current series of studies argues clearly against the latter and therefore against mainstream g theory. They favor the mutualism theory of in-telligence.