When to use BFinc across all models
We've had a discussion within our researchgroup on when one should calculate effects across all models and when one should calculate effects across matched models.
My general view is that if one is interested in the suport for an interaction in itself (disregarding the support for associated main effects and thus the principle of marginality) one should calculate effects across matched models. However, if one wants to uphold the principle of marginality and e.g. investigate the effects of an interaction together with the effects of associated main-effects one should calculate effects across all models.
Are there other general reason to calculate across all models rather than across mathed models? What is your general argument for calculating effects across all models rather than across matched models. When do you use across all models, when do you use across matched models?