Outfluence – The impact of missing values

Issue title: Statistical Inference with Missing Data

Guest editors: Ma del Mar Rueda

Article type: Research Article

Authors: Harel, Ofer

Affiliations: Department of Statistics, University of Connecticut, 215 Glenbrook Road Unit 4120 Storrs, CT 06269-4120, USA. Tel.: +1 860 486 6989; Fax: +1 860 486 4113; E-mail: [email protected]

Abstract: There are numerous measures that assess the effect of an observation, group of observations, a variable, or variables and observations on the regression estimation. Incomplete data is a common difficulty in data analysis. We introduce a new measure which assesses the effect of a missing observation, a group of missing observations, an incomplete variable or any combination of these, on the overall estimation. We call this measurement “outfluence”. The outfluence measure can be used in a regression analysis context or any other parametric settings. We illustrate the major benefits of outfluence using two examples. The first example demonstrates the use of outfluence in a marijuana pilot study. The second example records blood cholesterol levels of heart attack victims at three time points after the heart attack. We evaluate the cholesterol level mean at time 3 and the difference between time 1 and time 3 with incomplete data.

Keywords: Missing data, multiple imputation, rates of missing information, influence

DOI: 10.3233/MAS-2008-3209

Journal: Model Assisted Statistics and Applications, vol. 3, no. 2, pp. 161-168, 2008