COMPSTAT 2022: Start Registration
View Submission - COMPSTAT2022
Title: Two-stage target rotation with computational efficiency by asymmetric least squares criterion Authors:  Naoto Yamashita - Kansai University (Japan) [presenting]
Abstract: In factor analysis, a factor loading matrix is often rotated to a simple target matrix for its simplicity. As such rotational procedures, Promax and Simplimax are commonly used. A well-known limitation of Simplimax rotation is the computational inefficiency in estimating the sparse target matrix, which yields a considerable number of local minima. The target rotation procedures approximate the non-zeros in the loading matrix to zeros or non-zeros in the target matrix, but the existing procedures equally treat the two types of approximation, while the former is of importance for simplifying the loading matrix. The research proposes a new rotation procedure that consists of the following two stages. The first stage estimates a sparse target matrix with lesser computational cost by a regularization technique. In the second stage, a loading matrix is rotated to the target, emphasising the approximation of non-zeros to zeros in the target by the asymmetric least squares criterion. The simulation study and real data examples showed that the proposed method simplifies loading matrices, and its performance is superior to the existing procedures.