All simulators
Missing-Data Mechanisms
MCAR, MAR and MNAR on an education–wage scatter: complete-case bias, IPW repair and slope attenuation.
Day 14: Missing Data & Inverse Probability Weightingobservedmissing (hollow)full-data regressioncomplete-case regression
Share missing: 39%
True mean wage
30.32
Complete-case mean
27.43
bias -2.89
IPW mean (by education)
28.34
bias -1.98
Full-data slope
1.987
truth 2.0
Complete-case slope
1.756
Hollow points are missing wages. IPW re-weights respondents by the inverse of their education group's response rate. It repairs MAR, but no model on education can repair MNAR.