Hybrid Monte-Carlo in Multiple Missing Data Imputations with Application to a Bone Fracture Data
In this chapter we introduce Hybrid Monte-Carlo (HMC) as an efficient method to sample from complex posterior distributions of many correlated parameters from a semi-parametric missing data model. The HMC enables a distribution-free likelihood-based approach to multiple imputation of missing values. We describe the modeling approach for modeling missing values that does not require assuming any specific distributional forms. We then describe the use of the HMC sampler to obtain inferences and generate multiple imputations under the model, and touch upon various implementation issues, such as choosing starting values, determining burn-in period, monitoring convergence , deciding stopping times. An R program is provided for analyzing missing data from a Bone Fracture study.
KeywordsMultiple Imputation Dirichlet Process Imputation Model Impute Dataset Full Conditional Distribution
- Barengolts, E., Karanouh, D., Kolodny, L., & Kukreja, S. (2001). Risk factors for hip fractures in predominantly African-American veteran male population. Journal of Bone and Mineral Research, 16, S170.Google Scholar
- Chen, H.Y., Xie, H., & Qian, Y. (2011). Multiple imputation for missing values through conditional semiparametric odds ratio models. Biometrics, 10–13.Google Scholar
- Gelman, A., & Raghunathan, T. E. (2001). Discussion on “conditionally specified distributions: An introduction”. Statistical Science, 15, 268–269.Google Scholar
- Geweke, J. (1992). Evaluating the accuracy of sampling–based approaches to calculating posterior moments. In J. M. Bernado, J. O. Berger, A. P. Dawid & A. F. M. Smith (Eds.), Bayesian Statistics 4. Oxford: Clarendon Press.Google Scholar
- Raghunathan, T. E., Lepkowski, J. M., van Hoewyk, J., & Solenberger, P. (2001). A multivariate technique for multiply imputing missing values using a sequence of regression models. Survey Methodology, 27, 85–95.Google Scholar