Statistics 102H: FPP (or FPPA) Freedman's book on Statistics

Administrivia

• Start next project. You and your group will
  1. design an experiment
  2. collect the data
  3. analyse it
  4. write it up
• Each of the above 4 steps will be individually turned in.
• First step due next week.

Summary of course

• Studied so far: prediction
• focusing on soon: causation and explanation
• Business often deals with prediction, but science requires causation

Chapter 1: controlled experiments

Comparisons

• Contrasts are easy to understand (men are taller than women)
• Contrasts are useful for decisions (college educationed people make more money than non-college educated people.)
• But are the comparisons fair?

Ideal experiment

• controled
• randomized
• double blind

Why all that rigamarole? The story of the Polio vacine

• Case study: Polio vacine study
• History:
  • 50,000 per year
  • cripeling disease
  • Several proposed vacines
  • Salk won. Why?
• Study designs
  • NFIP: National Foundation for Infantile Paralysis
  • Historical controls:
    • vacinate all children and compare to last year
    • Oops: 1952=60k, 1953=30k, no vacine
    • Problem with epidemics
  • Treat all those who are willing
    • Oops: willingness is correlated with income
    • Income is correlated with desease
    • Double oops: poor are healther in this case!
  • Treat grade 2, use grades 1 and 3 for controls (we'll meet this again soon)
  • Best: compare to randomized population
• Implemented
  • both 2 grade treatment and randomzied study
  • See table from book for results
  • Notice: the effect is VERY different
• Placebo and double blinding
  • Diagnosis is subjective (I've got a bad knee--polio or not?)
  • patient reporting is subjective and may change actual behavior

Portacaval shunt

So obviously wrong that why do we need to discuss it? See the number of studies that didn't do it!

Draw picture of flow of patients

Other desease:

• Coronary bypass surgery: What does placebo look like?
• DES: "to prevent spontaneous abortion" actually caused cancer. Looked good on non-randomized trials. Banned eventually.

Ethics

Is doing a bad study ethical?

Is misleading doctors and patients ethical?

Homework

• Read chapter 1 of FPP(A)
• Start thinking about your second project (you can work in a group). I'll put up some details soon after class.
• First step:
  • Pick a group of 2 or three people
  • Decide on something to study
  • Can you do a controlled experiment? If so do so. If not, be ready to defend your reasoning.
  • Examples
    • Does wording of a question effect the outcome?
    • Does order of question effect the outcome?