Steven Ellis, PhD
Associate Professor of Clinical Neurobiology in Psychiatry
Steven P. Ellis, Ph.D. is Associate Prof. of Clinical Neurobiology (in Psychiatry). He is Director, Statistics and Computing Core, Conte Center: Neurobiological and Developmental Antecedents to Suicidal Behavior: The Neurobiology of Suicidal Behavior (CCNDASB). He is also PI on Project 8 of the CCNDASB, "Statistical methods in suicide research." He graduated cum laude, with Honors with Exceptional Distinction in Mathematics, and with a masters degree in mathematics awarded simultaneously with the bachelor's degree after four years at Yale. He also was awarded the DeForest Mathematical Prize at Yale. He was awarded an NSF Graduate Fellowship in mathematics.
Dr. Ellis was PI on the grant "Statistical Analysis of Neurochemical Maps" (R21 MH60995, NIMH). Dr. Ellis's interests include multivariate analysis and statistical computing.
Dr. Ellis is interested in statistical prediction of suicide attempts, statistical computing, and instability of multivariate statistical methods.
Currently, Dr. Ellis is developing new statistical measures of high-order statistical dependence. When there are many variables this is a very challenging problems. However, a method called "concurrence topology" developed by Ellis is often successful in detecting interesting forms of high-order dependence. A paper by Arno Klein and Ellis applying this method to resting state fMRI data has recently been accepted for publication.
In a different direction, Ellis has discovered that for many multivariate statistical methods there are many data sets at which the method fails catastrophically. This had already been known to a limited extent, i.e., failure due to outliers or, in particular cases, due to ill conditioning of certain matrices. But Ellis has shown the extent of this problem is far greater than previously suspected. He is investigating the breadth and depth of the problem.