Multiple testing

• Administrivia:
  • Bring your surveys next time if you can collect data from 112 students

Estimating contrasts

• Consider our managers again:

  coefficient	estimate	SE
  Intercept	20	5
  Manager[A-C]	10	20
  Manager[B-C]	5	20

• Write the equation for each manager
• What is the difference between Manager A and manager C?
  • Suppose manager C was "new" and had only done 10 jobs, but manager A and B had each done 1000 jobs.
  • Then [A-C] has a large SE, and [B-C] has a large SE
  • But, [A-B] has a very small SE say SE = 1.
  • Is [A-B] significant? We can't tell.
  • What is the standard error of [A-B]? We can't tell.

• We COULD change the name of manager B to by ZZZ. Then JMP would tell us the answer.
• Not such a bad idea when there are three--but what if there were 10?

Broken glassware

• We are moving our appartment from NJ to Philly this week
• Hence, breakage of glassware is on the mind
• There are many different ways you could pack glassware:
  • A: pack in peanuts by owner
  • B: pack in paper by owner
  • C: Pack in clothes or other soft stuff by owner
  • D: Pack in orginal shipping box by owner
  • E: pack in paper by shipper
  • F: pack in foam by shipper

• List contrasts: A-B, A-C, A-D, A-E, A-F, B-C, B-D, B-E, B-F, C-D, C-E, C-F, D-E, D-F, E-F.
• Why don't we need B-A? We can compute it by looking at minus A-B
• number of contrasts: p(p+1)/2
• Even more comparisons than in regression!
• Tukey solved this problem also (same Tukey of the bulging rule)

Computer repair

• see Handout

Last modified: Thu Apr 6 10:48:09 2000