9.9 Number of imputations revisited

In Section 9.4.6 you were instructed to simply used 20 imputations. That approach was taken because the author did not want to complicate the presentation of the other aspects of multiple imputation. However, the correct number of imputations needed is actually a function of information that is obtained after fitting the regression model. von Hippel (2020) suggests to first fit the imputation model 20 times, then fit your regression model, and then recompute the number of imputations using the how_many_imputations() function in the howManyImputations library. If that number is larger than 20, then re-fit the imputation model with the larger number of imputations.

For example:

imp <- mice(nhanes,
            seed  = 3,
            m     = 20,
            print = F)

# Temporary fit, just to get the number of imputations
fit <- with(imp,
            lm(LBDGLUSI_trans ~ BMXWAIST + smoker + RIDAGEYR +
                                RIAGENDR + race_eth + income))

## [1] 7

Since we already fit the model with m = 20 imputations, and more imputations is better than fewer, there is no need to refit the imputation and regression models. However, if how_many_imputations() had returned a number larger than 20, then refit the imputation model with that larger number and then refit the regression model with the updated mids object.


———. 2020. “How Many Imputations Do You Need? A Two-Stage Calculation Using a Quadratic Rule.” Sociological Methods & Research 49 (3): 699–718. https://doi.org/10.1177/0049124117747303.