European Academic Research ISSN 2286-4822
ISSN-L 2286-4822
Impact Factor: 3.4546 (UIF)
DRJI Value : 5.9 (B+)
Article Details :
Article Name :
Bootstrap and multiple imputation under missing data in AR(1) models
Author Name :
ELJONA MILO, LORENA MARGO
Publisher :
Bridge Center
Article URL :
Abstract :
Missing data is a phenomenon detected in many scientific investigations due to the bias often caused and inefficient analysis of the data. Determining the appropriate analytic approach in the presence of incomplete observations is a major question for data analysts. Recently many scientists have developed several statistical methods to address missingness. Since the introduction by Efron (1979), bootstrap has resulted to be an important method for estimating the distribution of an estimator by applying the resampling of the data. This bootstrap method resulted efficient in the case of independent and identically distributed observations, but in the case of dependent data like time series, classic bootstrap gives incorrect answers. For this reason, to adapt the bootstrap in the case of time series Kunsch (1989) presented a bootstrap method with blocks compounded by a fixed number of observations. Block bootstrap methods developed by researchers resulted suitable in the case of time series and give good results under specific assumptions. In this paper we will realize simulations with intention to compare the results obtained using block bootstrap combined with missing data mechanisms. We are interested in estimating the parameter in the AR(1) time series model using block bootstrap procedure after filling in the missing values using multiple imputation. We compare the results using several block length also different missing data mechanisms and packages for completing the missing values.
Keywords :
Bootstrap, time series, missing data, imputation, autoregressive
Announcements
New Launched Project
onlineresearch
Recommend & Share
Copyright © 2013: www.euacademic.org , Best Viewed in IE7+, Chrome, Mozilla Firefox.