Value Imputation

I’m a substitute for another guy

I look pretty tall but my heels are high

The simple things you see are all complicated

I look pretty young, but I’m just back-dated, yeah

–Pete Townsend

Data can be missing or untrusted in a variety of ways, and for a variety of reasons. These ways are discussed especially in Chapter 4, Anomaly Detection, and Chapter 5, Data Quality. Sometimes your best option for dealing with bad data is simply to discard it. However, many times it is more useful to impute values in some manner, in order to retain the rest of the features within an observation. From the perspective of this chapter, let us assume that all data values identified as untrusted—even if initially present with bad values—have already been explicitly marked as missing.

When imputing data, it is important to keep a good record of the difference between values you have invented (imputed) and those...