Effect of multivariate outlier detection and imputation on poverty and inequality indicators

RAND Statistics Seminar Series

Effect of multivariate outlier detection and imputation on poverty and inequality indicators

Presented by Beat Hulliger, University of Northwestern Switzerland, School of Business

Date: Thursday, July 25th, 2012
Time: 12:00–1:00 p.m. Pacific / 3:00–4:00pm Eastern
Host Location: Santa Monica, Forum 1226
Other Locations: Pittsburgh, room 6202 & Washington, DC room 7126

Abstract

The Statistics on Income and Living Conditions of the European Union (EU-SILC) establishes a set of coordinated surveys across Europe and a set of social cohesion indicators. In particular, the monetary indicators are based on the equivalized disposable income, an aggregation of person- and household-specific income components (e.g., income from employment and capital; unemployment-, old-age-, survivor-, and disability benefits). The income components span a multidimensional space with the following characteristics: (1) the marginal distribution of each component is very skew and has a heavy point mass at zero. In addition the data contains missing values and outliers. The BACON-EEM, the epidemic and the GIMCD algorithms for multivariate outlier detection and subsequent imputation are investigated to determine whether the methods are capable to protect the social cohesion indicators from outliers in the components. The algorithms can cope with missing values and the first two take the sample design into account. Results from a large simulation study in the AMELI project show the performance of the methods.

Speaker Bio

Beat Hulliger is professor of economic and social research at University of Northwestern Switzerland (FHNW). He studied Mathematics at ETH Zürich and specialised in Statistics at Universidad Autónoma de Madrid and ETH Zürich. His Ph.D. at ETH Zürich in 1991 treated robust estimators of finite population means. Hulliger worked as an expert for statistical methods of sample surveys at the Swiss Federal Statistical Office and was deputy head of its Statistical Methods Unit. He has published reports and articles on survey methodology. His research interests are survey statistics, in particular multivariate outliers in incomplete survey data, data preparation for sample surveys and quality of surveys. Hulliger is president of the section Education and Research of the Swiss Statistical Society and Associate Editor of Survey Methodology and of Wirtschafts- und Sozialstatistisches Archiv

To Attend

Visitors to RAND’s Santa Monica and Pittsburgh locations are welcomed to attend & must RSVP at least one day prior to the seminar. To ensure attendance please, contact Catherine Cruz with your name, company or affiliation & national citizenship (for security purposes).

Sponsored by the RAND Statistics Group