Lecture 17 - Measurement & Misclassification

Rose / Thorn

Rose: feels extremely relevant

Thorn: weird log functions

Measurement Error

• many variables are proxies of the cause of interest
• don’t consider how things are measured
• measurement error can have many effects on estimates

Modeling Measurement

• divorce, marriage, and age statistics are measured with error and the amount of error varies by state

  • imbalance in evidence quality

  • potential confounding through measurement error

• states with larger populations have less uncertainty/higher quality data

• confounding because measurement is influenced by population size but then effects such as divorce rate can also be influenced by population size

Misclassification

• categorical version of measurement error

• related models: hurdle models and occupancy models