This preview shows pages 1–12. Sign up to view the full content.
This preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full Document
Unformatted text preview: IE 410 Notes For Lecture 5 Comparing Individual treatment Means The general ANOVA test tests: H : all treatment means are equal Suppose H is rejected. We will want to know more about why it was rejected. We have 4 General methods for making comparisons between individual treatment means: 1. Orthogonal Contrasts (in lecture 4) 2. Scheffe's test on all possible contrasts 3. Pairwise comparison tests 4. Comparisons with a control (Dunnett's test) 3. Pairwise comparison tests i) Studentized range methods Tukey's test NewmanKeuls test ii) Duncan's test iii) Least significant difference (LSD) method iv) Fisher significant difference (FSD) or Bonferroni method Variance of a contrast estimator Let L be a general contrast: L = i c i i We will test H : L = 0 Consider an estimate of L, L L = i i i y c We can show that E[ L ] = L and that V[ L ] = 2 i c i 2 /n i Recall the following based on our Linear Model : . i y ~ NID( i , 2 / n i ) ] E( L ) = E[ i c i . i y ] = i E[c i . i y ] = i c i E[ . i y ] = i c i i V( L ) =V[ i c i . i y ] = i c i 2 V[ . i y ] = i c i 2 2 /n i Estimating 2 with MS E , we can estimate V( L ) = MS E i c i 2 /n i This is a key component in most of the methods we will discuss Let us use S( L ) to denote the standard deviation of the contrast estimate: S( L ) = i i i E n C MS 2 Generalized Confidence Interval on a Contrast Let: L CutOff( )*S( L ) be a (1 )*100% CI on the true value of L Recall we are interested in H : L=0 Thus if the CI contains zero, do not reject. This CI is based on our best (unbiased) estimate of L: ( L ) L (some multiple of)S( L ) This CI should make intuitive sense....
View
Full
Document
This note was uploaded on 08/06/2008 for the course IE 410 taught by Professor Storer during the Fall '04 term at Lehigh University .
 Fall '04
 Storer

Click to edit the document details