Diagnostics for empirical likelihood inference NICOLE LAZAR DEPARTMENT OF STATISTICS CARNEGIE MELLON UNIVERSITY  Empirical likelihood inference is based on confidence intervals derived from the chi-square approximation to the log-likelihood ratio statistic.  One of the oft-touted advantages of the approach is that it produces confidence intervals that reflect the symmetries and asymmetries in the data, that is, the orientation of the sample.  This is in opposition to, for example, normal-theory intervals, which are perforce symmetric.  However, this flexibility also means that empirical likelihood confidence intervals are potentially more sensitive to extreme or unusual data points.  By making changes in the sample, it is possible to assess the influence of individual observations, on the empirical likelihood itself, on the resultant confidence intervals, and on the conclusions that are drawn from the data.  In this talk, I will examine ways of evaluating influence of data points, and suggest diagnostics that can be used to summarize this information.  I will demonstrate these ideas on a number of data examples and for a variety of functionals of interest.