Measurement Error and Complex Data Workshop: Titles and Abstracts

Semiparametric regression is a generalization of the general linear model where the effects of some predictors are modeled nonparametrically, for example, as splines or Gaussian processes. Powerful R packages are now available that allow many semiparametric regression models to be fit easily. For more complex semiparametric regression models, a Bayesian analysis using Stan, which can be run from R, is possible. This course will cover univariate smoothing, additive models, bivariate smoothing, spatial models, grouped data, interaction models, random effects, quantile regression, and non-Gaussian responses. Aurore Delaigle Tutorial Title: Nonparametric Curve and Density Estimation with Measurement Error Abstract: We will introduce nonparametric kernel–based density and regression estimators that are valid when the data are contaminated by additive random measurement errors. We will discuss the main principles for constructing consistent estimators in this context, and show that often a solution can be found more easily by working in the Fourier domain. The estimators involved the choice of one or several smoothing parameters, and we will discuss ways of choosing them in a data–driven way. We will introduce nonparametric kernel–based density and regression estimators that are valid when the data are contaminated by additive random measurement errors. We will discuss the main principles for constructing consistent estimators in this context, and show that often a solution can be found more easily by working in the Fourier domain. The estimators involved the choice of one or several smoothing parameters, and we will discuss ways of choosing them in a data–driven way. Speaker: Raymond J. Carroll Title: Building a Data-Based Healthy Eating Index Abstract: The Healthy Eating Index-2005 (HEI-2005) and the Healthy Eating Index-2010 (HEI-2010), are multi-component dietary quality indices involving ratios of interrelated dietary components to energy. They were built by nutritionists based on USDA dietary recommendations, and have applied across many cancers and mortality to show links to diet and health. We build a new dietary index by using health outcome data from the NIH-AARP Study of Diet and Health, including lung, colorectal, and prostate cancers in men and lung, colorectal, breast and ovarian cancers in women. Technically, the tools involve a unique single-index modeling framework. The actual data are based on a food frequency questionnaire, which is known to be biased for usual intake and have considerable measurement error. I will outline a possible approach to accounting for the bias an measurement error in these data. The Healthy Eating Index-2005 (HEI-2005) and the Healthy Eating Index-2010 (HEI-2010), are multi-component dietary quality indices involving ratios of interrelated dietary components to energy. They were built by nutritionists based on USDA dietary recommendations, and have applied across many cancers and mortality to show links to diet and health. We build a new dietary index by using health outcome data from the NIH-AARP Study of Diet and Health, including lung, colorectal, and prostate cancers in men and lung, colorectal, breast and ovarian cancers in women. Technically, the tools involve a unique single-index modeling framework. The actual data are based on a food frequency questionnaire, which is known to be biased for usual intake and have considerable measurement error. I will outline a possible approach to accounting for the bias an measurement error in these data.