Bayesian regression entry methods and the difference between the predictor matrix and formulas

[Max-Lovell]

Hi,

I have two questions regarding the entry methods of the bayesian regression (norm) option.
How is this handled when using a predictor matrix - is it forced entry down the list of variables or simultaneous? If so, could regression entry order by be specified using formulas? I could then calculate stepwise entry orders earlier.

A final P.S. if anyone if anyone has any opinions or relevant papers to hand, I'm stuck between handling the items of my scales as single scale blocks or doing each item individually? I have missing data at post-test and am predicting this from pre-test. At this point I am handling it item by item - predictors in my regression equations will be the same, although outcomes will differ.

Any help much appreciated! thanks.

[stefvanbuuren]

There is currently no stepwise modelling in mice.impute.norm(), but there is one in preparation using lasso #438.

[Max-Lovell]

Thanks for your reply Stef! Could I ask how entry is handled in the impute.norm method, and if this changes when specifying blocks and formulas vs. a predictor matrix?

[thomvolker]

The default order of imputation specified in mice is from left to right. However, you can easily change the sequence by specifying the visitSequence parameter within the mice::mice function. You can do this once, specifying the same sequence for every variable that is to be imputed, or for every variable separately.

Hope this helps.

[stefvanbuuren]

Sorry, I do not quite understand your question(s). Could you perhaps illustrate the problem with a small example?