Basic Assumptions for the MI Method

The MI method is provided under the following assumptions:

  1. The base model for the Summary Statistics version of MI is a multivariate normal distribution with parameters (μ, Σ)

    1. μ is a vector of means.

    2. Σ is a variance-covariance matrix.

  2. Prior distribution of μ is a conditionally-multivariate Gaussian given Σ with parameters μ0∈R7 and τ-1Σ, where τ is a positive constant. The variance-covariance matrix Σ follows the inverted-Wishart distribution for fixed parameters mp and a positive-definite matrix Λ.

  3. Data points are Missed At Random (MAR).

The strict definition of this and other mechanisms supporting missing values are available in [Rubin1987].

For more complete information about compiler optimizations, see our Optimization Notice.