1. Some common but biased or inefficient missing data practices:
(1)Make up numbers: e.g., changing Party ID “don’t knows” to
“independent”
(2)Listwise deletion: used by 94% pre-2000 in AJPS/APSR/BJPS
(3)Various other ad hoc approaches
2. Application-specific methods: efficient, but model-dependent and hard to develop and use
3. An easy-to-use and statistically appropriate alternative, Multiple imputation: (1) fill in five data sets with different imputations for missing values, (2) analyze each one as you would without missingness, and (3) use a special method to combine the results.
摘编人：章文娟
