QSG this week features two talks by Alex Brodersen and Minami Hattori, respectively. The title of the first talk is "Parallel analysis with large amounts of planned missingness" and the title of the second talk is "Local solutions of Geomin rotation". The abstracts are given below.
QSG meets on Thursday from 3:30-4:45pm in Haggar 212. All are welcome, and we hope to see you there.
For a complete list of speakers for Fall 2016, please visit http://nd.psychstat.org/classes/qsg2016f/index.
Thanks,
Haiyan Liu
Johnny Zhang
================================================
Parallel analysis with large amounts of planned missingness
================================================
Parallel analysis is often suggested as an effective and accurate
method for determining the number of factors to retain in exploratory
factor analysis. However, in a design with large amount of planned
missingness, standard parallel analysis methods perform very poorly,
often suggesting extracting an unreasonably high number of factors. To
address this issue a modification to parallel analysis is suggested to
account for the additional imprecision in the correlation matrix
estimation due to the missing data. We present preliminary results
from a simulation conducted on both continuous and ordinal data by
applying the proposed method earlier developments of parallel
analysis. Preliminary results suggest a substantial improvement in
identifying the number of underlying factors in the context of large
planned missingness.
============================
Local solutions of Geomin rotation
============================
In exploratory factor analysis, factor loading matrices are rotated to
improve interpretability. A promising factor rotation criterion is
the geomin criterion. However, the geomin rotation frequently
produces multiple local solutions. Unawareness of this issue may lead
to its suboptimal use. We conducted a simulation study that explores
the number, frequencies, and the nature of local solutions. In this
walk, we report the findings and discuss the implications.