## 6.8 Exercises

Exercise 6.1 (Worm plot for normal model) Repeat the imputations in Section 6.6.1 using the linear normal model for the numerical variables. Draw the worm plot.

1. Does the imputation model for wgt fit the observed data? If not, describe in which aspects they differ.

2. Does the imputation model for wgt fit the imputed data? If not, describe in which aspects they differ.

3. Are there striking differences between your worm plot and Figure 6.10? If so, describe.

4. Which imputation model do you prefer? Why?

Exercise 6.2 (Defaults) Select a real dataset that is familiar to you and that contains at least 20% missing data. Impute the data with mice() under all the default settings.

1. Inspect the streams of the MICE algorithm. Does the sampler appear to converge?

2. Extend the analysis with 20 extra iterations using mice.mids(). Does this affect your conclusion about convergence?

3. Inspect the data with diagnostic plots for univariate data. Are the univariate distributions of the observed and imputed data similar? Do you have an explanation why they do (or do not) differ?

4. Inspect the data with diagnostic plots for the most interesting bivariate relations. Are the relations similar in the observed and imputed data? Do you have an explanation why they do (or do not) differ?

5. Consider each of the seven default choices in turn. Do you think the default is appropriate for your data? Explain why.

6. Do you have particular suggestions for improved choices? Which?

7. Implement one of your suggestions. Do the results now look more plausible or realistic? Explain.