Color for exclusion-restriction variable in plot_pred() is coded wrong

[hanneoberman]

pred <- mice::quickpred(mice::nhanes)
pred[,1] <- -3
ggmice::plot_pred(pred)

^{Created on 2023-10-04 with reprex v2.0.2}

[stefvanbuuren]

Hmm, interesting. Does 1 stand for "incomplete predictor" and -3 for "complete predictor"?

[hanneoberman]

There is currently no difference between complete and incomplete predictors. I have taken this from the vignette:

"In the predictor matrix, -2 denotes the class variable, a value 1 indicates a fixed effect and a value 2 indicates a random effect"

Instead of this from the mitml package reference:

"1: target variables containing missing data
2: predictors with fixed effect on all targets (completely observed)
3: predictors with random effect on all targets (completely observed)
-1: grouping variable within which the imputation is run separately
-2: cluster indicator variable
0: variables not featured in the model"

The value -3 would denote a new imputation model predictor type, implemented in the Heckman method in micemd.

[hanneoberman]

pred <- mice::quickpred(mice::nhanes)
pred[,1] <- -3
ggmice::plot_pred(pred)

^{Created on 2023-10-04 with reprex v2.0.2}

[stefvanbuuren]

Thanks. I wasn't aware of the -3 code. Perhaps we should start a code registry?

[hanneoberman]

Good idea! Maybe the Wiki feature on Github or a reference page on amices.org would be nice.

[stefvanbuuren]

I made a previous inventory. Perhaps we could that into a table of code by method

• Univariate imputation methods for two-level data use other codes than 0 and 1
• 2l.bin, 2l.lmer, 2l.norm, 2l.pan, 2lonly.mean, 2lonly.norm and 2lonly.pmm use code -2 to indicate the class variable
• 2l.bin, 2l.lmer, 2l.norm and 2l.pan use code 2 to indicate the random effects
• 2l.pan uses codes 3 and 4 to add class means to codes 1 and 2 respectively