More Power

1
More Power
2
Factors affecting power

• the size of the difference you wish to be able to
  detect
• the variability in your data
• your sample size
• (alpha, but this is cheating)
• in general, power f(mHa, s, n)

3
Making power complicated

• define effect size, d, to entangle the difference
  between groups and the variability
• d f(mHa, s)
• define delta, d, as d times some function of
  sample size to further abstract things
• d df(n)
• create a magic table to relate power, d, and a.

4
Calculating (simple) power

• Sketch sampling distribution given mHa, s, n,
  then look up the corresponding area.
• Use the d, d, magic table method
• Use software (e.g. GPower)

5
Calculating complicated power

• Use software, Use software, Use software
• (this doesnt actually mean it is easy - SPSS
  only gives post-hoc power, for example)
• If you do complicated (inelegant?) experiments,
  you will have to do complicated power
  calculations or approximations (with no help from
  me).

6
Post-hoc Power
alpha
your data
7
The most important thing about calculating power
It forces you to think about the theoretical or
practical meaning of possible experimental
outcomes.
8
Additional bonuses

• A negative result with high power is both
  informative and publishable
• We dont waste time doing futile experiments
• We can operate intelligently with limited
  resources (medicine, industry, etc.)
• (academia is actually the only place it is
  possible to get away without computing power)

9
about this exam

• 3 essay questions
• direct (not a great deal of creativity required)
• cover large, encompassing concepts
• 8 short answer
• most are direct and cover basic concepts
• a couple might be more abstract curve balls
• no need for calculator or crib sheet

10
about exams writing in general

• read the question
• stop and think about your answer
• organize your answer paragraph by paragraph
• illustrate your answers
• budget your time (for me, this means quickly
  dispatching of the ones I know)
• write neatly (happy graders are a good thing)