Clarification on chained equations methodology #556
VictorPorcelli
started this conversation in
Missing data methodology
Replies: 1 comment 1 reply
-
Section https://stefvanbuuren.name/fimd/sec-FCS.html contains the pseudo code of the MICE algorithm. I leave it up you to spot the differences with your version. |
Beta Was this translation helpful? Give feedback.
1 reply
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
Hi there, I'm trying to better understand the chained equations methodology used by mice and am particularly confused about when previously imputed values are used in subsequent imputations/iterations.
From my understanding, the process is as follows:
Step 1: Missing values in predictors for a variable being imputed are replaced with random draws from observed values.
Step 2: The variable is imputed using whatever method is specified (e.g., logistic regression).
Step 3: Steps 1 and 2 are repeated for each variable being imputed, except imputations from Step 2 replace random draws from Step 1. The end result is one fully imputed dataset (end of imputation 1).
Step 4: Steps 1-3 are repeated for a set number of imputations (defined by parameter m), the result of which is one iteration.
Step 5: Steps 1-4 are repeated for a set number of iterations (defined by parameter maxit).
My first question is am I understanding the process correctly?
Assuming I am, what I'm slightly unclear on but have attempted to ascertain using context clues from various readings, is whether:
(a) Subsequent imputations use imputed values rather than random draws (i.e., skip step 1) within a single iteration. For example, after Steps 1-3 are completed producing a single fully imputed dataset, and mice moves on to imputation number 2, is the process reset, or does imputation 2 use imputed values from imputation 1 in place of random draws from observed values?
(b) Subsequent iterations use imputed values rather than random draws. That is, after mice produces m fully imputed datasets and completes iteration 1, does imputation 1 of iteration 2 use random draws from observed values to fill in missing values in predictors, or does it use imputed values from iteration 1?
Beta Was this translation helpful? Give feedback.
All reactions