News   Any MI-related news?

Please let me know.
Statistica Neerlandica Special issue
Number 1 of volume 57 (2003) of Statistica Neerlandica is a special issue on Incomplete data: Multiple imputation and model-based analysis. The editorial says: "Significant computational advances over the last decade have helped to establish multiple imputation as a respectable and versatile approach to a broad variety of incomplete data problems. Model-based analysis refers to the situation where a specific model for the missing data is needed, i.e., if the missing data are not missing at random. Multiple imputation still works in this case, but the emphasis shifts to the specification of the model that created the missing data.". Contents (.doc). (Jan. 2003).
Books!
Several books (partly) devoted to MI:

MICE or elephant
The American Statistician of August 2001 features a software comparison by Horton and Lipsitz (pdf). They study four packages, three commercial ones (SOLAS V3.0, SAS 8.2, Missing data library S-Plus) and MICE, our public domain library for S-Plus and R. MICE may not be fast and flashy, but it can keep up with the big guys. It is the only package in which the user can add custom imputation functions, useful for sensitivity analysis in nonignorable nonresponse. In comparison, MICE has compact syntax, and extensive facilities for checking convergence. (SvB, Sept 2001)
Hype in Stat Med
The year 2001 has brought tremendous growth of MI applications appearing in Statistics in Medicine. Here's the list thus far:
  • Geskus RB. Methods for estimating the AIDS incubation time distribution when date of seroconversion is censored. Stat Med. 2001 Mar 15;20(5):795-812. Abstract.
  • Gmel G. Imputation of missing values in the case of a multiple item instrument measuring alcohol consumption. Stat Med 2001 Aug 15;20(15):2369-2381. Abstract.
  • Hunsberger S, Murray D, Davis CE, Fabsitz RR. Imputation strategies for missing data in a school-based multi-centre study: the Pathways study. Stat Med. 2001 Jan 30;20(2):305-16. Abstract.
  • Landrum MB, Becker MP. A multiple imputation strategy for incomplete longitudinal data. Stat Med 2001 Sep 15;20(17-18):2741-2760. Abstract.
  • Lynn HS. Maximum likelihood inference for left-censored HIV RNA data. Stat Med. 2001 Jan 15;20(1):33-45. Abstract.
  • Wei L, Shih WJ. Partial imputation approach to analysis of repeated measurements with dependent drop-outs. Stat Med. 2001 Apr 30;20(8):1197-214. Abstract.
  • Wu H, Wu L. A multiple imputation method for missing covariates in non-linear mixed-effects models with application to HIV dynamics. Stat Med. 2001 Jun 30;20(12):1755-69. Abstract.
  • Zhou XH, Eckert GJ, Tierney WM. Multiple imputation in public health research. Stat Med. 2001 May 15-30;20(9-10):1541-9. Abstract.
S-Plus missing data library
S-Plus Version 6 appeared in August 2001 and features a new missing data library. You can create and analyse multiple imputations under Gaussian, Logistic and Conditional Gaussian models. The new methods are inspired on Schafer's 1997 book and software. Documentation is about 160 pages! Check out www.insightful.com for details. (SvB, Sept 2001).
SAS joins in...
Multiple imputation will become a serious option for a wide audience. SAS recently announced two new procedures: PROC MI and PROC MIANALYZE. These procedures will do more or less what SOLAS does, and add Schafers's NORM method to that. The procedures are experimental in V8.1 (July 2000) and V8.2 (expected late 2000). Documentation has not yet been finished and will probably take a couple of months, but still, there is a SUGI paper that gives a preview (.pdf, 90Kb) (SvB, July 2000).
Cox and interval censoring
Pan (A multiple imputation approach to Cox regression with interval-censored data, Biometrics, ,56, 199-203) proposes a general semiparametric method for Cox regression using multiple imputation. The basic idea is to impute exact survival times from interval-censored data and to take advantage of standard methods for right-censored data (SvB, July 2000).
Conditional means
Schafer and Schenker (Inference with imputed conditional means, JASA, 95, 144-154, 2000) present a new variance estimator for means and proportions that uses just a single data set imputed by conditional means. The estimator is a combination of the naive variance of the imputed data, the variance around the regression line, and the variability of the line. The authors claim that their method requires less computational effort and can yield more precise estimates than those resulting from multiple imputation (SvB, May 2000).
   
New variance estimator  
Robins and Wang (2000) propose a new estimator of the asymptotic variance in both single and multiple imputation. The authors write: "Our variance estimator, in contrast to the estimator proposed by Rubin (1987), is consistent even when the imputation and analysis models are misspecified and incompatible with one another." The new estimator requires more information to be passed from imputer to user. (SvB, April 2000).
   
Interval censoring  
Two articles on the use of MI in interval censoring recently appeared in Statistics in Medicine, one by Pan, the other by Bebchuk and Betensky. (Svb, Mar. 2000)
   
Adjusted degrees of freedom  
Barnard and Rubin (1999) derive new pooling rules specifically designed for use in small samples. These rules adjust the degrees of freedom of the t-distribution such that it is always less than or equal to the complete-data degrees of freedom. The authors conclude: "We recommend that our modified repeated-imputation reference distribution in all analysies of multiply imputed data, especially with datasets having few primary sampling units." (SvB, Jan. 2000).