mi (version 0.10-2)

noise.control: Auxiliary for Adding Priors to Missing Data Imputation

Description

Auxiliary function as user interface for adding noise for mi procedure. Typically only used when calling mi.

Usage

noise.control(method=c("reshuffling", "fading"), pct.aug=10, K=1, post.run.iter=20)

Arguments

method

two methods are implemented: reshuffling and fading.

pct.aug

percent of N being add into the exisitng data, where N is the number of observation of the completed data set, default is 10.

K

the cooling parameter, default is 1.

post.run.iter

number of iterations after a imputation, default is 20. This is to mitigate the influence of a imputation with the noise.

Details

If reshuffling method is used, the imputation will randomly switch between randomly imputing data from marginal (imputing data from the observed values) and drawing from the modeled based values. And with each iteration (s), the probability of cooling (q) decreases by number of iterations, such that q = K/s. If fading method is used, the imputation will augment

References

Yu-Sung Su, Andrew Gelman, Jennifer Hill, Masanao Yajima. (2011). Multiple Imputation with Diagnostics (mi) in R: Opening Windows into the Black Box. Journal of Statistical Software 45(2).

See Also

mi

Examples

### NOT RUN
#===========================================================
# add fadding empirical noise by augmenting 10% of the data
#===========================================================
## data(CHAIN)
## IMP <- mi(CHAIN, add.noise=noise.control(method="fading", pct.aug=10, post.run.iter=20))
#=================================================
# add noise by randomly drawing from the marginal
#=================================================
## This is the default setting
## IMP <- mi(CHAIN, add.noise=noise.control(method="reshuffling", K=1, post.run.iter=20))
#=================================================
# add no noise
#=================================================
## IMP <- mi(CHAIN, add.noise=FALSE)
#=================================================
# add noise but no post.run
#=================================================
## IMP <- mi(CHAIN, add.noise=noise.control(post.run.iter=0))