Methods for Dealing with Missing Data

From Displayr
Jump to: navigation, search

Missing Completely At Random methods

Using mean values

This method replaces the missing data with the average for that variable.

Listwise deletion

Pairwise deletion

At an intuitive level, this method seems superior to listwise deletion: as each correlation is estimated using all the available data, it seems obvious that a larger effective sample size is used in the analysis. However, analytic and simulation studies of linear regression reveal that this is only true when the correlations between the variables are low. Further, the statistical tests used in many commercial programs are wrong. [1]. For this reason, pairwise deletion should not be consider a better method than pairwise deletion, and is perhaps best only used in situations where the missing data makes listwise deletion impractical.


Imputation

Stochastic imputation

Multiple imputation

Maximum likelihood estimation

Nonignorable methods

Dummy variable adjustment

Extra category

  1. Allison, Paul D. (2001). Missing Data (Quantitative Applications in the Social Sciences) (Page 8). SAGE Publications. Kindle Edition.