The importance of isomorphism for conclusions about homology: A Bayesian multilevel structural equation modeling approach with ordinal indicators

Guenole, Nigel. 2016. The importance of isomorphism for conclusions about homology: A Bayesian multilevel structural equation modeling approach with ordinal indicators. Frontiers in Psychology, 7, 289. ISSN 1664-1078 [Article]

[img]
Preview
Text
FrontiersIsomorphism.pdf - Published Version
Available under License Creative Commons Attribution.

Download (521kB) | Preview

Abstract or Description

We describe a Monte Carlo study examining the impact of assuming item isomorphism (i.e., equivalent construct meaning across levels of analysis) on conclusions about homology (i.e., equivalent structural relations across levels of analysis) under varying degrees of non-isomorphism in the context of ordinal indicator multilevel structural equation models (MSEMs). We focus on the condition where one or more loadings are higher on the between level than on the within level to show that while much past research on homology has ignored the issue of psychometric isomorphism, psychometric isomorphism is in fact critical to valid conclusions about homology. More specifically, when a measurement model with non-isomorphic items occupies an exogenous position in a multilevel structural model and the non-isomorphism of these items is not modeled, the within level exogenous latent variance is under-estimated leading to over-estimation of the within level structural coefficient, while the between level exogenous latent variance is overestimated leading to underestimation of the between structural coefficient. When a measurement model with non-isomorphic items occupies an endogenous position in a multilevel structural model and the non-isomorphism of these items is not modeled, the endogenous within level latent variance is under-estimated leading to under-estimation of the within level structural coefficient while the endogenous between level latent variance is over-estimated leading to over-estimation of the between level structural coefficient. The innovative aspect of this article is demonstrating that even minor violations of psychometric isomorphism render claims of homology untenable. We also show that posterior predictive p-values for ordinal indicator Bayesian MSEMs are insensitive to violations of isomorphism even when they lead to severely biased within and between level structural parameters. We highlight conditions where poor estimation of even correctly specified models rules out empirical examination of isomorphism and homology without taking precautions, for instance, larger Level-2 sample sizes or using informative priors.

Item Type:

Article

Identification Number (DOI):

https://doi.org/10.3389/fpsyg.2016.00289

Keywords:

isomorphism, homology, multilevel structural equation modeling, ordinal indicators

Departments, Centres and Research Units:

Psychology

Dates:

DateEvent
14 February 2016Accepted
2 March 2016Published Online

Item ID:

16827

Date Deposited:

15 Feb 2016 16:41

Last Modified:

03 Aug 2021 15:04

Peer Reviewed:

Yes, this version has been peer-reviewed.

URI:

https://research.gold.ac.uk/id/eprint/16827

View statistics for this item...

Edit Record Edit Record (login required)