CASOS Working PAPER

"Multicollinearity Robust QAP for Multiple-Regression"(PDF file)
Authors: Dekker, D., Krackhardt, D., Snijders, T.
Abstract
The quadratic assignment procedures for inference on multiple-regression coefficients(MRQAP) has become popular in social network analysis. These tests have been developed to assess the sizes of a set of multiple-regression coefficients. However, research practitioners often use these tests to assess the size of individual multiple-regression coefficients. Although this might be a harmless extension, our our concern focuses on this practice under conditions of multicollinearity. In this paper we show analytically that different MRQAP-tests for individual parameter estimates are biased under multicollinearity. Subsequently, we propose a new MRQAP-test, which we call "semi-partialing" that is robust against multicollinearity. Extensive simulation results, as well as re-analysis of the classic Laumann-Marsden-Galaskiewicz(1978)data show the added value of this new "semi-partialing" method over the existing methods.