|
Abstract:
|
Multiple imputation has entered mainstream statistics over the past decade. Now there are several software options that allow different models for multiple imputation. Past software required joint modeling of all variables causing problems of data dimensionality and convergence, especially for categorical ontinuous outcomes. The next generation of software have adopted a conditional, sequential regression approach, implemented in IVEWare, MICE, SAS and Stata. These allow for flexible modeling and more efficient estimation. However, it is still hard to estimate many models with large datasets and small sample sizes. MI analysis routines are also limited. We will discuss specific applications of large-scale MI, limitations we face and possible ways to address these.
|
