Missing data recovery in environmental time series
Main Article Content
Abstract
As many environmental data are increasingly recorded on a long term basis, it is unfortunately frequent that they show missing data (MD). In addition to information losses, MD also prevent the use of time series analysis and present the researcher the dilemma of either apply sophisticated methods of analysis or attempt to fill those MD gaps in order to apply conventional methods. In any case, further statistical treatment usually needs complete time series and hence MD must be estimated. The main statistical methods to tackle this problem are briefly outlined here, and available software is reported as well. A case of time series reconstruction of Spanish rainfall and water quality to exemplify these methods is also described, using the maximum likelihood approach of the Expectation-Maximization- Bayesian (EMB) algorithm and the AMELIA-II free software.
Article Details
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.
References
Allison, P. D. 2001. Missing Data. Sage University, Paper 136,London.
Benítez-Gilabert, M. y Alvarez-Cobelas, M. 2008. Stream waterquality is influenced by climatic change and teleconnectionsin Southwestern Europe. Climatic Change (en prensa).
Coenders, G., Batista-Folget, J. M. y Willem, E. S. 2005. Temasavanzados en modelos de ecuaciones estructurales. LaMuralla S.A., Madrid.
Cohen, J. y Cohen, P. 1985. Applied multiple Regression andCorrelation Analysis for the behavioural Sciences. 2ndedition. Eribaum Press. Hillsdale, New Jersey.
Dempster, A. B., Laird, N. M. y Rubin, D. B. 1997. Maximumlikelihood estimation from incomplete data via the EMalgorithm. Journal of the Royal Statistical Society, Series B:39: 1-38.
Enders, C. K. 2001. The impact of non-normality on fullinformation maximum likelihood estimation for structuralequation models with missing data. Psychological Methods6: 352-370.
Finkbeiner, C. 1979. Estimation for the multiple factor modelwhen data are missing. Psychometrika 44: 409-420.
Ford, B. L. 1983. An overview of Hot-Deck procedures. InMadow, W. G., Olkin, I. y Rubin, D. B. (Eds.), IncompleteData in multiple Surveys, pp 185-207. Academic Press,New York.
Graham, J. W. y Schafer, J. L. 1999. On the performance ofmultiple imputation for multivariate data with small samplesize. In Hoyle, R. H. (Ed.), Statistical Strategies for smallSample Research, pp 1-29. Sage Publications, London.
Graham, J. W, Hofer, S. M. y Piccinin, M.,1994. Analysis withmissing data in drug prevention research. In Collins, I. M. ySeitz, L. (Eds.), Advances in Data Analysis for PreventionIntervention Research. National Institute on Drugs Abuse.Monography series 142. Washington DC.
Homaker, J., King, G. y Blackwell, M. 2006. Amelia II: Aprogram for missing data.(http://gking.harvard.edu/amelia/).
Jones, M. P. 1996. Indicator and stratification methods formissing explanatory variables in multiple linear regressions.Journal of the American Statistical Association 91: 222-230.
Kim, J.O. y Curry, J. 1977. The treatment of missing data inmultivariate analysis. Sociological Methods and Research 6:215-240.
King, G. J., Honaker, A. J. y Kenneth, S. 2001. Analyzingincomplete Political Science Data. An alternative Algorithmfor multiple imputation. Political Science Review 95: 49-69.(http://gking.harvard,edu/files/abs/evil-abs.shtml).
Little, R. J. A. y Rubin, D. B. 1987. Statistical Analysis withmissing Data. Wiley, New York.
McLachlan, G. J. y Krishnan, T. 1997. The EM Algorithm andExtension. Wiley, New York.
Meng, X. 1994. Multiple imputation inference with uncongenialsources of input. Statistical Science 9: 538-573.
Plaia, A. y Bondi, A. L. 2006. Single imputation method ofmissing values in environmental pollution data sets.Atmospheric Environment 40: 7316–7330.
Rao, J. N. K. y Shao, J. 1992. Jacknife variance estimation withsurvey data under Hot-Deck imputation. Biometrika 70:811-822.
Reales, J. M. 2007. Valores perdidos en el análisis de datos enPsicología. Apuntes en preparación. UNED, Facultad dePsicología. Dpto. de Metodología de las Ciencias delComportamiento. Madrid.
Schafer, J. L. 1997. Analysis of incomplete multivariate Data.Chapman and Hall, London.
Van Buuren, S. y Oudshoorn, K. 1999. Flexible multipleimputation by MICE. Report TNO-PG 99054, TNOPrevention and Health, Leiden. (http://www.multipleimputation.com).