Likelihood Inference in Small-Area Estimation Using P-Spline and Time Series Models

Farhad Shokoohi: Likelihood Inference in Small-Area Estimation Using P-Spline and Time Series Models. Joint Statistical Meeting 2013. August 3-8, 2013, at the Palais des congres de Montreal, Montreal, QC, Canada, 2013, (Contributed Talk).

Abstract

Nonparametric methods, despite their advantages, have not been often used in Small Area Estimation (SAE) due to methodological difficulties. Recently, a nonparametric linear model using cross-sectional data was introduced in SAE. However, there are many real applications in SAE which are time-related as well. In this talk, we introduce non-parametric models for Normal and non-Normal data situations with incorporating cross-sectional and time-series data. Frequentist analysis of these models is computationally difficult. Recent method of data cloning has overcome computational difficulties of mixed models from frequentist perspective. We demonstrate how the data cloning approach can be used to perform frequentist analysis of these complex models for both continuous and discrete data. One advantage of the data cloning approach is that both prediction and prediction intervals are easily obtained. The performance of the proposed approach is evaluated through several simulation studies and also by a real application.

BibTeX (Download)

@conference{Shokoohi2013jsmb,
title = {Likelihood Inference in Small-Area Estimation Using P-Spline and Time Series Models},
author = {Farhad Shokoohi},
url = {https://ww2.amstat.org/meetings/jsm/2013/onlineprogram/AbstractDetails.cfm?abstractid=309172},
year  = {2013},
date = {2013-08-03},
address = {Joint Statistical Meeting 2013. August 3-8, 2013,  at the Palais des congres de Montreal, Montreal, QC, Canada},
abstract = {Nonparametric methods, despite their advantages, have not been often used in Small Area Estimation (SAE) due to methodological difficulties. Recently, a nonparametric linear model using cross-sectional data was introduced in SAE. However, there are many real applications in SAE which are time-related as well. In this talk, we introduce non-parametric models for Normal and non-Normal data situations with incorporating cross-sectional and time-series data. Frequentist analysis of these models is computationally difficult. Recent method of data cloning has overcome computational difficulties of mixed models from frequentist perspective. We demonstrate how the data cloning approach can be used to perform frequentist analysis of these complex models for both continuous and discrete data. One advantage of the data cloning approach is that both prediction and prediction intervals are easily obtained. The performance of the proposed approach is evaluated through several simulation studies and also by a real application.},
note = {Contributed Talk},
keywords = {Likelihood based estimation, P-Splines, Small-area estimation, Time series},
pubstate = {published},
tppubtype = {conference}
}