A Field Test of Optional Unrelated Question Randomized Response Models
Tracy Spears Gill, Anna Tuck, Sat Gupta, Mary Crowe, Jennifer Figueroa,
University of North Carolina, Greensboro
Optional unrelated question randomized response models for both the binary and quantitative response to sensitive questions were recently introduced by Gupta et al (2013). In this field test, the binary and quantitative response models are utilized in surveys of sensitive behaviors to establish the validity of these tests in vivo. The two sensitive questions used were "Have you ever been told that you have a sexually transmitted disease?" and "How many sexual partners have you had in the last 12 months?" The target population was undergraduate students enrolled at UNC Greensboro during the 2012-2013 academic year. Subjects were asked the questions by optional unrelated question RRT, check-box survey method, and direct face-to-face interview. The results of these methods are compared to each other, as well as to existing published information on these sensitive behaviors. Estimates provided by the optional unrelated question randomized response models are in line with the mathematical and computer simulation results in Gupta et al (2013). This study also provides the first estimate of sensitivity level through a fieldwork study.
Generalized Functional Concurrent Model
Janet S. Kim, Ana-Maria Staicu, and Arnab Maity North Carolina State University
We consider the generalized functional model, where both the response and the covariate are functional data and are observed on the same domain. In contrast to typical functional linear concurrent models, we allow the relationship between the response and covariate to be nonlinear, depending on both the value of the covariate at a specific time point as well as the time point itself. In this framework we develop methodology for estimation of the unknown relationship and construction of point-wise confidence bands, allowing for correlated error structure as well as sparse and/or irregular design. We investigate this approach in finite sample size through simulations and a real data application.
Interactive Q-learning for Generalized Outcomes
Kristin A. Linn, North Carolina State University
A dynamic treatment regime (DTR) consists of decision rules that operationalize clinical decision-making by recommending treatments over time. These rules take into account patient information such as medical history, past treatments, and response to treatment. Each decision rule is a function that inputs patient history and outputs a recommended treatment, hence personalizing treatment to the individual. Thus far in the DTR literature, methods for estimating optimal sequential treatment rules have focused on maximizing the expected value of a continuous response. However, the mean may not be the best summary of the outcome distribution, or the primary clinical response may not be continuous, for example, whether or not a patient achieved remission. Little attention has been given to the problem of estimating a policy that maximizes a binary outcome, such as remission, or a particular quantile of the response distribution. Using an approximate dynamic programming framework, we derive the form of the optimal policy for a two-stage randomized trial and propose estimators for the optimal decision rules at each stage when the primary outcome is a binary indicator or pre-specified quantile of a continuous distribution. We study the small sample performance of our estimators across a broad range of generative models.
