Date of Award

Fall 2021

Document Type

Open Access Thesis



First Advisor

Alexander C. McLain


The current duration approach to modeling time-to-pregnancy (TTP) models the length of pregnancy attempt for women that are currently attempting pregnancy. There is a scarcity of studies, let alone TTP studies, that account for measurement error in the outcome. Previously, the benefits of a piecewise constant model with regards to bias in estimates of the survival function with measurement error and the parametric modelling of TTP was shown. In this thesis, correcting for measurement error in the outcome with the current duration approach is explored through piecewise constant models with log-normal measurement error. Five different methods are compared to determine the optimal method in reducing the bias associated with the estimating the survival function. Specifically, we compared three naïve approaches (untransformed, shifted, and transformed), and two different simulation and extrapolation, or SIMEX, methods. The SIMEX method uses increasing measurement error variance to generate data with increasing measurement error variance that is fit to a piecewise constant model to approximate a quadratic model that is extrapolated for the approximate true coefficient value. The SIMEX method was tested with log-transformed and untransformed hazard rates from the parametric piecewise model, which we refer to as log SIMEX and regular SIMEX methods, respectively. The methods are further compared through their confidence intervals to determine the extent of precision when correcting for measurement error in the outcome. The log SIMEX method has somewhat high variance and instances of the highest and lowest bias depending on the Weibull parameters. Nevertheless, it is shown to be an inconsistent method in correcting. For measurement error in the outcome. The transformed method performs the worst with regards to bias. Overall, there is no consistently best method to correct for measurement error in the outcome as the extent of bias and MSE depend on the measurement error variance and Weibull parameters.

Included in

Biostatistics Commons