Non-parallel lines

• Administrivia:
  • Turn in project questionnaires
  • Start collecting data as soon as possible!
  • The more data you collect, the easier your anaylsis will be

R-sq, F-test and doing multiple tests

• Question: "How often do you worry about what you will do after you graduate?"
  • plot vs year.
  • We expect people to htink more about it as they get closer to graduating
  • But what about Wharton? It is all they think about! (My introduction to business school consisted of overhearing two students discussing the importance of bullets vs. paragraphs on their vitas--this was before the fall semester started!)
  • Maybe engineers worry at first (hence majoring in engineering) but realize they don't have to worry later.
  • So we would get different slopes for these different groups.

• Production example
  • R-sq'ed = variability explained = (SST - SSE)/SST
  • It always increases when you add more variables
  • But did it increase more than chance alone would predict?
  • This is the goal of the F-test

• Why not just use individual t-statistics/p-values?
  • Each protects at 5%
  • you will have at least 7 variables and 21 "interactions" on your project
  • Expected number of variables included by chance alone = .05 * (7 + 21) = 1.4 variables
  • So you will include some "noise" variables = bummer!
  • The F-test protects against this somewhat.

Last modified: Fri Mar 31 15:52:20 2000