問題文
A column of household income has values missing for 12 percent of the rows. Investigation shows the value is missing precisely when the respondent declined to answer, and those respondents skew toward higher incomes. What is the most defensible treatment?
選択肢
- Add a flag column recording that the value was missing, and impute the value with a method chosen for the distribution, so the refusal itself becomes usable information.
- Fill the missing values with the mean of the observed incomes, because the mean is the least biased single value to use.
- Drop the rows with the missing value, because 12 percent is a small enough share that the remaining sample stays representative.
- Replace the missing values with zero, because zero is unambiguous and the model can learn from the columns that a zero in this column stands for an unanswered question.