𝔖 Bobbio Scriptorium
✦   LIBER   ✦

Appropriate questions about causal inference from “direction of causation” analyses

✍ Scribed by J. J. McArdle


Publisher
John Wiley and Sons
Year
1994
Tongue
English
Weight
367 KB
Volume
11
Category
Article
ISSN
0741-0395

No coin nor oath required. For personal study only.

✦ Synopsis


THE DIRECTION OF CAUSATION MODEL

My comments here are stimulated by the interesting lead articles on the "Direction of Causation" (DOC) model by Neale et al. [I9921 and by Duffy and Martin [1992]. In this paradigm, MZ and DZ twin paris are measured on at least two variables (X and Y ) , and a variety of latent variable structural equation models are fitted.

Under specific model constraints it is possible to show that the X -Y model fits better than the X -Y model. Under still more restrictive assumptions it seems possible to show that the X -Y feedback model is needed.

The first paper by Neale et al. lays out a foundation for these structural models, and this model presentation is both clear and creative. This paper also provides an interesting example of possible feedback influences between parents and children [see Bell, 19681, and an effort is made to detail the substantive limitations of this approach. The second paper by Duffy and Martin [ 19921 highlights some critical technical issues in using this structural model. They present a useful power analysis of key alternative models, they point out the fundamental importance of measurement error in these models, and they present a variety of additional substantive examples. Both papers deal with key issues about structural modeling, and both papers reflect a high level of technical sophistication on the part of the authors.

This research demonstrates an important principle for structural analysesinformation about hiometr-ic group digerences can help resohe other important aspects of structural equation systems. I have advocated this same principle for the determination of latent scores in a multivariate system; e.g., for latent factors in a factor analysis [McArdle and Goldsmith, 19901 and for latent growth curves in lon-Address reprint requests to J.