11. Commonly Asked Questions II

What are some typical approaches to model based imputation?

What is the trade-off in using MICE imputation vs. simpler imputation methods?

Additional Thoughts

Although model-based imputations are most effective in reflecting the distribution of the population, in an application setting, it’s important to consider the tradeoff between maintainability of your model and the effectiveness of the imputation. In a “real-life” production setting, models are often refreshed (sometimes daily). Does it make sense to run a MICE-based imputation for a feature every day? Is that an expensive computation? If the recommendation will not fundamentally change, I recommend using simpler imputation techniques to maintain your data science model in perpetuity.