Modeling Mind and Matter: Reductionism and Psychological Measurement in Cognitive Neuroscience

Click to increase image sizeClick to decrease image size

Acknowledgments

This article is original. None of the materials have been published elsewhere. We thank Angelique Cramer, Petry Kievit-Tyson, and Anne-Laura van Harmelen for valuable comments on earlier versions of this article.

Notes

¹Although both Burnston et al. (this issue) and Vul (this issue) point out that one may also consider computational and algorithmic levels as different reductionist levels, for now we focus on the questions on, roughly speaking, the granularity or types of brain properties we should study, and why (although our approach is compatible with the algorithmic or computational level, more on the flexibility of N-indicators in a later section).

²Although we agree that there are much more sophisticated and predictively better neurological indicators with respect to intelligence, we disagree with Vul's assertion that intelligence is “not well predicted by coarse neural measures” (p. 139): Despite the coarse granularity, these measures together explain 25.1% of the variance in g, which is the higher end of explained variance in most research papers on similar topics.

³Haig (this issue) specifically addresses the importance of parsimony in comparing structural equation models. Although quantitative assessments are not the only interpretation of parsimony, the fit indices we use, (the AIC, the BIC, and the RMSEA) all take into account parsimony in some different ways (Claeskens & Hjort, 2008).

⁴Although the suggestion by Bagozzi to test the deviation of the correlation from zero instead of the full reflective model is interesting (and his comments concerning method factors valid), we prefer a model with a single factor. By representing the latent construct as one latent variable, we also imply certain interpretations in terms of possible interventions that are different from having two separate, albeit highly correlated, latent factors.