Missing Observations and Imputation

Abstract

Statistical analysis with missing data is an important problem as the problem of missing observation is very common in many situations. During the last two decades different methods have been developed to tackle the situation. One possible way to handle missing values is to remove either all features or all objects that contain missing values. Another possibility is imputation where we fill in the missing values by inferring new values for them. The imputation method may not be applicable to some astronomical data sets as the missing value may arise from physical process and imputing missing values is misleading and can skew subsequent analysis of data. For example, the Lyman break technique (Giavalisco 2002) can identify high-redshift galaxies based on the absence of detectable emissions in bands corresponding to the FUV rest frame of the objects.