12.3 Roadmap

Rearview Mirror

• Statisticians create a population model to represent the world.
• The BLP is a useful summary for a relationship among random variables.
• OLS regression is an estimator for the Best Linear Predictor (BLP).
• For a large sample, we only need two mild assumptions to work with OLS
  • To know coefficients are consistent
  • To have valid standard errors, hypothesis tests

Today

• The Classical Linear Model (CLM) allows us to apply regression to smaller samples.
• The CLM requires more to be true of the data generating process, to make coefficients, standard errors, and tests meaningful in small samples.
• Understanding if the data meets these requirements (often called assumptions) requires considerable care.

Looking Ahead

• The CLM – and the methods that we use to evaluate the CLM – are the basis of advanced models (inter alia time-series)
• (Week 13) In a regression studies (and other studies), false discovery is a widespread problem. Understanding its causes can make you a better member of the scientific community.