whitney test

use

  • non-parm
  • for non-normal dist
  • test if 2 samples from same dist
  • analogous to t-test
  • calculated from ranks

calc

  • rank both data sets, in one array, keeping labels (groups)
  • sum ranks for each sample
  • higher sum is sample 1
  • calc the U value
  • U=n_1n_2 + (n_1(n_1+1)/2)-R_1
  • R_1 is sum of sample 1 ranks

significance

  • use table “Critical Vals of Mann-Whit U”
  • two-tailed testing
  • signif level of 0.05
  • < than table val indics significant diff between samples

look at “pig weight T-test”

Excel

  • Sort
  • Filter aggregates

Stat Review

t-test - MW U test

- Margin of error

significance 5% level (5% prob that could be chance)

t-test

  • diff between means of 2 sampls
  • assumes normal dist
  • in excel, use 2-tailed and type-2 test

t-test signif

mann-whitney U

  • non-parametric
  • for not-normal

mw sig

  • table
  • 2-tailed
  • smaller value indicates sig diff

if margins overlap, % vals not sig different

correlation - is one factor related to another? on pairs of data -1 perf neg correl 0 no corr 1 perf corr

types of study

  • randomized
  • prospective
    • two samples, exposed and not (eg down and upwind from hanford)
    • effect is correlated?
  • retrospective
    • group by presence of effect