A Comparison of Missing-Data Procedures for Arima Time-Series Analysis
Authors:
- 1 August 2005
- journal article
- Published by SAGE Publications in Educational and Psychological Measurement
- Vol. 65 (4) , 596-615
Abstract
Missing data are a common practical problem for longitudinal designs. Time-series analysis is a longitudinal method that involves a large number of observations on a single unit. Four different missing-data methods (deletion, mean substitution, mean of adjacent observations, and maximum likelihood estimation) were evaluated. Computer-generated time-series data of length 100 were generated for 50 different conditions representing five levels of autocorrelation, two levels of slope, and five levels of proportion of missing data. Methods were compared with respect to the accuracy of estimation for four parameters (level, error variance, degree of autocorrelation, and slope). The choice of method had a major impact on the analysis. The maximum likelihood very accurately estimated all four parameters under all conditions tested. The mean of the series was the least accurate approach. Statistical methods such as the maximum likelihood procedure represent a superior approach to missing data.Keywords
This publication has 25 references indexed in Scilit:
- Time Series Analysis2003
- Estimation, Prediction, and Interpolation for ARIMA Models with Missing DataJournal of the American Statistical Association, 1986
- Estimating Missing Observations in Economic Time SeriesJournal of the American Statistical Association, 1984
- The Reliability and Accuracy of Time Series Model IdentificationEvaluation Review, 1983
- Maximum Likelihood Fitting of ARMA Models to Time Series With Missing ObservationsTechnometrics, 1980
- Alternatives to Simonton's analyses of the interrupted and multiple-group time-series designs.Psychological Bulletin, 1979
- Inference and missing dataBiometrika, 1976
- A Maximization Technique Occurring in the Statistical Analysis of Probabilistic Functions of Markov ChainsThe Annals of Mathematical Statistics, 1970
- A New Approach to Linear Filtering and Prediction ProblemsJournal of Basic Engineering, 1960
- Maximum Likelihood Estimation from Incomplete DataBiometrics, 1958