December 14th-16th 2013
Programme Leader Ian White and senior scientists Dan Jackson and Shaun Seaman from the MRC Biostatistics Unit will give invited talks at the 6th International Conference of the European Research Consortium for Informatics and Mathematics Working Group on Computational and Methodological Statistics (ERCIM 2013)
The prestigious conference will take place at the University of London, United Kingdom, 14-16 December 2013. This event is organised by the ERCIM Working Group on Computational and Methodological Statistics (CMStatistics), Birkbeck University of London and London School of Economics.
The ERCIM WG CMStatistics includes members in all areas of methodological statistics and those of computing that have an impact on statistical techniques. Applications of statistics in diverse disciplines are strongly represented. These areas include economics, medicine, epidemiology, biology, finance, physics, chemistry, climatology and communication. The range of topics addressed and the depth of coverage establish the WG as an essential research network in the interdisciplinary area of advanced computational and methodological statistics.
The ERCIM aims to foster collaborative work within the European research community and to increase co-operation with European industry. Leading European research institutes are members of ERCIM.
Session Methods for handling missing data (Monday 16.12.2013) Chair: Shaun Seaman
- R. Keogh, I. White. “Using multiple imputation to improve analysis of case-control studies nested within cohorts”.
- S. Seaman, J. Bartlett, I. White. “Multiple imputation of missing covariates with non-linear effects and interactions: An evaluation of statistical methods”.
See related material at: http://link.springer.com/article/10.1186%2F1471-2288-12-46
- J. Bartlett, S. Seaman, I. White, J. Carpenter. “Multiple imputation of covariates by fully conditional specification: Accommodating the substantive model”.
See related material at: http://missingdata.lshtm.ac.uk/preprints/main.pdf
- D. Jackson. “Using the number of contact attempts to model non-ignorable missing outcome data”.
Selection models are an established way of modelling missing outcome data. However using these models in practice is fraught with difficulties. In order to overcome these problems, Repeated Attempts Models (“RAM”s) have more recently been proposed. These computationally quite intensive models describe the probability that each contact attempt to obtain outcome data is successful. If only one contact attempt is made for each subject, then the RAM reduces to a conventional selection model.
In this talk Dan Jackson will describe the types of dataset that motivate the use of RAMs. He will also explain why we can generally expect RAMs to perform better than selection models, whilst emphasising the additional assumptions made by the RAM. He will however argue that each dataset with information about failed contact attempts can be expected to provide its own quirks and difficulties, and hence providing generic software for RAMs is a very challenging task. Despite this he will describe the software that is available and he will present some results for a dataset where a RAM was used.
To find out more about the ERCIM 2013 please visit http://www.cmstatistics.org/ERCIM2013/