SYMPOSIUM ON INCOMPLETE DATA November 8th., 2001 Utrecht, The Netherlands http://www.vvs-ssp.nl/symposium2001.html Organized by the Statistical Software section of The Netherlands Society for Statistics and Operations Research http://www.vvs-ssp.nl The program committee is delighted to be able to present a selection of the top researchers on this topic. If you would like to learn more about incomplete data, here is the symposium you should not miss!! Registration Please register via email to [EMAIL PROTECTED] or via the web site: http://www.vvs-ssp.nl/symposium2001registration.html Program 9:30 welcome 10.00 opening 10.05 Joseph L. Schafer Pennsylvania State University Multiple imputation in multivariate problems when the imputation and analysis models differ 10.45 Ineke A.L. Stoop Social and Cultural Planning Office Getting to know the nonrespondents 11.15 coffee break 11:45 Geert Molenberghs Limburgs Universitair Centrum Sensitivity analysis for incomplete data 12:15 Mark Huisman University of Groningen Handling missing item responses due to item nonresponse and incomplete designs 12:45 lunch 14.00 Carl-Erik Särndal Statistics Sweden & Statistics Canada Weighting for Nonresponse: Some Theoretical and Computational Issues 14:40 Susanne Rässler University of Erlangen Nürnberg Alternative approaches to statistical matching 15:10 tea break Software caroussel 15:40 Donald B. Rubin Harvard University The role of direct likelihood methods in statistical software to avoid missing data problems. 16:05 Joop Hox University of Utrecht Software for direct estimation 16:15 Karin Oudshoorn TNO Prevention and Health Software for multivariate imputation 16:30 Drinks Missing Values Everybody has them, nobody wants them More often than not empirical researchers are confronted by an abundance of missing values. As the phenomenon is usually not seen as a possible threat to the validity of the research, the most common approach to this problem is simply to deny it. The 4% to 5% missing values in a few variables in a small part of the data under research do not seem that important. However, when one looks closer at the 'important' variables in the data set, percentages of missing values of 30% to 70% are not uncommon. Now this would not be a problem if Fate dropped these blind spots all over the responses. The adequate action then would be to simply enlarge the sample and use only the known values. All statistical packages provide this simple escape route. Alas, Fate is not as blind as the old Greek let us believe. Just as, in former days, whole families were struck by disasters, nowadays SES-classes or age-classes or other minorities are disproportionately struck by missing values. If within those classes Fate is turning a blind eye, truth can still be found using ML-methods (e.g. the EM-algorithm) or simulating a few complete data sets from the single incomplete set (Multiple Imputation). The last 25 years have seen a rapid development of algorithms, but these are only slowly being applied in statistical software. Sometimes, Fate is not striking blind at all. The highest income appears to be unknown. Populations high at risk refuse to be sampled for blood, and in longitudinal clinical research most missing values are found for patients in the worst health conditions. In agricultural data, highly productive cows no longer show milk yields due to an illness correlated with productivity. But laboratories have their share as well, the lowest values being undetectable, the highest putting the instruments on ERROR. In short, in this situation the missing value depends on the true, but unknown, value of the variable itself. Missing values can really be a threat to your research. This one-day conference may help you to deal effectively with this problem. Organization Section Statistical Software of The Netherlands Society for Statistics and Operations Research VVS-SSP Nieuwpoortkade 25 1055 RX Amsterdam The Netherlands T +31 (0)20 5608410 F +31 (0)20 5608448 E [EMAIL PROTECTED] U www.vvs-ssp.nl Our main sponsors - CANdiensten Your Partner in Mathematics and Statistics http://www.candiensten.nl/english/ - Genstat With Genstat You know you can http://www.vsn-intl.com/genstat/index.htm - Muthen & Muthen Mplus - Statistical Analysis With Latent Variables http://www.statmodel.com - SOLAS for missing data analysis http://www.statsol.ie/solas/solas.htm - Springer-Verlag the publishing company for books and journals in Statistics http://www.springer.de/statistic/ - StatSoft Benelux BV http://www.statsoft.nl