Worst Fraction of Missing Information

DESCRIPTION:

Estimates the worst fraction of missing information, and the direction of the worst linear function of parameters for a class missmodel object produced by an EM algorithm.

This function is generic (see Methods); method functions can be written to handle specific classes of data. Classes which already have methods for this function include: Loglin, Gauss and cgm.

USAGE:

worstFraction(object, ...) 

REQUIRED ARGUMENTS:

object
a class "missmodel" object produced by an EM algorithm.

OPTIONAL ARGUMENTS:

...
most methods have additional arguments method, tol, lim, and h which affect the algorithm used to find the worst fraction. Additional arguments are possible. See the specific function called for a list of all possible arguments.

VALUE:

a list with two components. The fraction component estimates the largest eigenvalue of the jacobian of the EM operator.

The direction component contains the corresponding eigenvector. The direction is normalized to have length one.

DETAILS:

For an EM algorithm, convergence is approximately linear, with rate governed by the largest eigenvalue of the jacobian of the EM operator.

REFERENCES:

C. Fraley, "On Computing the Largest Fraction of Missing Information for the EM Algorithm and the Worst Linear Function for Data Augmentation", to appear in Computational Statistics and Data Analysis.

Schafer, J. L. (1997), Analysis of Incomplete Multivariate Data, Chapman & Hall, London.

SEE ALSO:

, , , .

EXAMPLES:

fit <- emLoglin(crime, frequency= count) 
worstFraction(fit) 
worstFraction.Loglin(fit)   #same