| title: | Prognosing multivariate missing data using additional information |
|---|---|
| reg no: | ETF5521 |
| project type: | Estonian Science Foundation research grant |
| subject: |
1.1-1.5. Exact Sciences 5. Social Sciences |
| status: | completed |
| institution: | TU Faculty of Mathematics |
| head of project: | Ene-Margit Tiit |
| duration: | 01.01.2003 - 31.12.2005 |
| description: | Missing values in data is one of the most important problems in several areas of applied statistics, especially in social sciences, where usually very high-dimensional data are used and, in general, the assumption about randomness of missing values does not hold. Due to the complicity of the problem, the most sophisticated methods of mathematical statistics are used to solve it, as, e.g. MCMC (Monte Carlo Markov Chains), based on Bayes approach. Nevertheless, a series of tasks, including time-dependent measurements (e.g. panel data and repeated data) are still waiting new ideas and consistent solutions. In the framework of given grant the solving of following tasks is planned: 1. To analyse the theoretical basis of recently proposed methods for handling missing data (EM, "hot deck", methods using MCMC-approach), to investigate their robustness in the case of different underlying distributions of data and missingness types. 2. To test the effectivity of different methods using specially constructed/ generated testing data-sets. 3. To make comparative analysis of the results of different missing-data-handling methods in the vase of different real data-sets. 4. To modify standard methods using the properties of multivariate extremal distributions that allow to estimate the allowed intervals and most probable intervals of missing data -- the so-called sensitivity approach that has since been used mainly for binary data. 5. To find optimal methods for handling with missing data in the case of different underlying distributions, especially in the case of time-dependent measurements and high-dimensional data of social surveys. To check the adequacy of the results of data handling methods. |
| project group | ||||
|---|---|---|---|---|
| no | name | institution | position | |
| 1. | Ene Käärik | Inst Math Stat, UT | doctoral student | |
| 2. | Ene-Margit Tiit | TU Faculty of Mathematics | prof emer | |
| 3. | Mare Vähi | Inst Math Stat, UT | lecturer | |