On Usefulness of Maximum Likelihood Estimator Using Incomplete Data
We often encounter with missing data which consist of a complete part and an incomplete part. In this case, there are two types of maximum likelihood estimator. One is the conventional maximum likelihood estimator based on all of the data and another is based on the complete part only, which neglects the incomplete part. Let n and n* be the sample sizes of the complete part and the incompleste part, respectively. It is well-known that the former is asymptotically better than the latter when n*n is constant. However, other cases have not been well-known. This paper shows that the former is better than the latter in other cases in view of higher-order. In addition, this paper illustrates some exact comparison.
KeywordsProbability Density Function Maximum Likelihood Estimator Likelihood Estimator Exponential Family Fisher Information Matrix
Unable to display preview. Download preview PDF.
- Fujisawa, H. (2000). On usefulness of incomplete data for maximum likelihood estimator. Hiroshima University Technical Report, 99–19.Google Scholar
- Kano, Y., Bentler, P.M., and Mooijaart, A. (1993). Additional information and precision of estimators in multivariate structural models. In Stat.Sci.and Data Anal., Matsusita, K. et al. (eds.), 187–196, VSP, Utrecht.Google Scholar
- Kariya, T., Krishnaiah, P.R., and Rao, C.R. (1983). Inference on parameters of multivariate normal populations when some data is missing. In Developments in Statistics, Vol.4“, Krishnaiah, P.R. (eds.), 137–184, Academic Press, New York.Google Scholar