|
Abstract:
|
Motivated by a real nutritional epidemiologic data, we develop a semiparametric method for estimating disease-diet association in the presence of measurement error in the diet variable. Usually in these problems, the primary data consist of a binary disease variable (Y) and a response to Food frequency questionary (FFQ) which is considered as a surrogate variable (Q) for the true diet variable (X), and a smaller study which is called validation data, contain the surrogate variable (Q) and repeated measures of an unbiased instrumental variable of the true diet. At the first stage, using the validation data we estimate how Q depends on the X, and at the second stage we estimate how the disease risk is associated with X. The method is semiparametric in the sense that at both the stages we use P-splines technique to estimate the nonparametric smooth functions.
|
