A0301
Title: Analysis of parameter and partial parameter impacts
Authors: Serhat Guenay - Competence Center for Clinical Trials Bremen (Germany) [presenting]
Abstract: Building on work on counterfactual distributions, we consider the problem of defining and estimating the influence of small changes in a given statistical parameter (called the independent parameter) on another statistical parameter (called the target parameter) under the assumption that a set of statistical control parameters remains constant. Small changes in statistical parameters are realised by small changes in the distribution, using the theory of influence functions. Examples of the parameters are expected values, quantiles and regression coefficients. In the absence of control parameters, the influence of the independent parameter on the target parameter (called parameter impact) is defined by the ratio between the change in the target parameter and the change in the independent parameter when the distribution is perturbed along the independent parameter. We show that with a set of control parameters, orthogonalising the influence functions allows these parameters to be kept constant. This leads to a quantity we call the partial parameter impact. Point estimation and resampling-based statistical inference for influence and partial parameter impact are discussed. The method is illustrated with an observational study aimed at investigating factors that may affect the quality of inpatient geriatric care.
