4.4 Roadmap

4.4.1 Rearview Mirror

  • Statisticians create a population model to represent the world.
  • \(E[X], V[X], Cov[X,Y]\) are “simple” summaries of complex joint distributions, which are hooks for our analyses.
  • They also have useful properties – for example, \(E[X + Y] = E[X] + E[Y]\).

4.4.2 This week

  • We look at situations with one or more “input” random variables, and one “output.”
  • Conditional expectation summarizes the output, given values for the inputs.
  • The conditional expectation function (CEF) is a predictor – a function that yields a value for the output, give values for the inputs.
  • The best linear predictor (BLP) summarizes a relationship using a line / linear function.

4.4.3 Coming Attractions

  • OLS regression is a workhorse of modern statistics, causal analysis, etc
    • It is also the basis for many other models in classical stats and machine learning
  • The target that OLS estimates is exactly the BLP, which we’re learning about this week.