Unidimensional factor models imply weaker partial correlations than zero-order correlations
Authors:
Riet van Bork,
Raoul P. P. P. Grasman,
Lourens J. Waldorp
Abstract:
In a unidimensional factor model it is assumed that the set of indicators that loads on this factor are conditionally independent given the latent factor. Two indicators are, however, never conditionally independent given (a set of) other indicators that load on this factor, as this would require one of the indicators that is conditioned on to correlate one with the latent factor. Although partial…
▽ More
In a unidimensional factor model it is assumed that the set of indicators that loads on this factor are conditionally independent given the latent factor. Two indicators are, however, never conditionally independent given (a set of) other indicators that load on this factor, as this would require one of the indicators that is conditioned on to correlate one with the latent factor. Although partial correlations between two indicators given the other indicators can thus never equal zero (Holland and Rosenbaum, 1986), we show in this paper that the partial correlations do need to be always weaker than the zero-order correlations. More precisely, we prove that the partial correlation between two observed variables that load on one factor given all other observed variables that load on this factor, is always closer to zero than the zero-order correlation between these two variables.
△ Less
Submitted 12 October, 2016; v1 submitted 11 October, 2016;
originally announced October 2016.
Structure estimation for mixed graphical models in high-dimensional data
Authors:
Jonas M. B. Haslbeck,
Lourens J. Waldorp
Abstract:
Undirected graphical models are a key component in the analysis of complex observational data in a large variety of disciplines. In many of these applications one is interested in estimating the undirected graphical model underlying a distribution over variables with different domains. Despite the pervasive need for such an estimation method, to date there is no such method that models all variabl…
▽ More
Undirected graphical models are a key component in the analysis of complex observational data in a large variety of disciplines. In many of these applications one is interested in estimating the undirected graphical model underlying a distribution over variables with different domains. Despite the pervasive need for such an estimation method, to date there is no such method that models all variables on their proper domain. We close this methodological gap by combining a new class of mixed graphical models with a structure estimation approach based on generalized covariance matrices. We report the performance of our methods using simulations, illustrate the method with a dataset on Autism Spectrum Disorder (ASD) and provide an implementation as an R-package.
△ Less
Submitted 19 October, 2015;
originally announced October 2015.
