First: Understand WHY Data Is Missing

MCAR (Missing Completely at Random): No pattern to missingness

• Example: Random sensor failures
• Safe to delete rows or impute

MAR (Missing at Random): Missingness related to other variables

• Example: Younger people skip income questions more
• Can use advanced imputation methods

MNAR (Missing Not at Random): Missingness related to the missing value itself

• Example: High earners refuse to report income
• Hardest case - deletion causes bias!

Handling Strategies

1. Deletion

Listwise deletion: Remove entire row if any value missing

When to use: Less than 5% missing, MCAR data, large dataset

Risk: Reduces sample size, introduces bias if not MCAR

2. Simple Imputation

Mean/median/mode: Replace missing values with average

• ✅ Quick and simple
• ❌ Reduces variance
• ❌ Ignores relationships between variables

3. Advanced Imputation

Multiple imputation: Generate several plausible values

Regression imputation: Predict missing values from other variables

When to use: More than 5% missing, important analysis

4. Flag as Missing

Create indicator variable: "income_missing" (0/1)

When to use: Missingness itself is informative

Common Mistakes

❌ Replacing missing with 0 - Zero is a value, not "unknown"

❌ Imputing without investigating why - Pattern might be important

❌ Using mean for skewed data - Use median instead