Class notes 9-23-2004
Review of ANOVA post-hocs
EPSY
6301
23 Sept 2004
Review of some symbols for different tests, one-way ANOVA
- we will review some of the highlights from Chapter 5 of the Kirk book (we will not get a copy)
after finding a significant F, we do multiple comparison procedures for
- we have discussed:
-- Tukey
-- Tukey-Kramer
-- SNK
-- Fisher-Hayter
-- LISD
-- Scheffe
pairwise or complex comparisons
- wanted to test Mew1 - Mew2 = 0, that is one individual comparison we could test
- these are pairwise comparisons (the Tabachnick book calls these CONTRASTS)
-- some experimental design books say for every contrast, we will call it psi (see hand notes)
sum of coefficients of contrasts should be a zero
- these are they symbols we use (psi)
- these are a-posteriori contrasts
If you do all of the contrasts, you are "going fishing" rather than targeting (tailoring the way you question certain things)
# of pairwise comparisons = [J (j-1)] / 2
Orthangonal contrasts can be used for interesting pairwise contrasts
- a zero gives me orthagonality (they are unrelated)
you don't want to do pairwise comparisons because of all the overlap
- so we want to customize the way we ask the question
- like just answer 2 pairwise comparisons / hypotheses, and look at orthagonal relationships to do that
orthagonality: there is no relationship
repeated measures design involve pretest and posttest
so you can ask more customized questions than doing all the pairwise comparisons
- this is called tailored hypothesis testing
- this gives you more power because you are hypothesis testing just a few
- still trying to identify where the significance is located
you have to custom program the syntax into the computer when you want to ask customized questions
pairwise comparisons equate to just a fishing expedition
whatever is interesting to you is what you are going to test
- the resarch interests drive the questions that you are investigations
- the statistics should not drive what you do
to test for orthagonality, it is mainly a process of playing with the coefficients
you never test the obvious: test what you don't know
- ask important questions with your data
experimental design is basically about finding or creating a methodology that can be replicated
- target specific groups
no one in biology at a high level is doing pairwise comparisons
- I can't spend more time on this because you don't have chapter 5 from the Kirk book
From Kirk book chapter 4, Table 4.1-2
- shows recommended procedures when ssumptionis are tenable
We will discuss the following now:
- Tukey
- Newman-Kewls
- Fisher-Hayter
Scheffe must be used for a more complex hypothesis that goes beyond just pairwise comparisons
- you can group your data in interesting ways to identify new things / trends
-- you will always use the AVERAGE of your variable results for comparison, not just the sum
Fisher-Hayter
- each contrast (phi) is going to have a unique standard error (psi hat: a pair of means)
posthoc power comes from using the pooled variance: the mean of squares between
23 Sept 2004
Review of some symbols for different tests, one-way ANOVA
- we will review some of the highlights from Chapter 5 of the Kirk book (we will not get a copy)
after finding a significant F, we do multiple comparison procedures for
- we have discussed:
-- Tukey
-- Tukey-Kramer
-- SNK
-- Fisher-Hayter
-- LISD
-- Scheffe
pairwise or complex comparisons
- wanted to test Mew1 - Mew2 = 0, that is one individual comparison we could test
- these are pairwise comparisons (the Tabachnick book calls these CONTRASTS)
-- some experimental design books say for every contrast, we will call it psi (see hand notes)
sum of coefficients of contrasts should be a zero
- these are they symbols we use (psi)
- these are a-posteriori contrasts
If you do all of the contrasts, you are "going fishing" rather than targeting (tailoring the way you question certain things)
# of pairwise comparisons = [J (j-1)] / 2
Orthangonal contrasts can be used for interesting pairwise contrasts
- a zero gives me orthagonality (they are unrelated)
you don't want to do pairwise comparisons because of all the overlap
- so we want to customize the way we ask the question
- like just answer 2 pairwise comparisons / hypotheses, and look at orthagonal relationships to do that
orthagonality: there is no relationship
repeated measures design involve pretest and posttest
so you can ask more customized questions than doing all the pairwise comparisons
- this is called tailored hypothesis testing
- this gives you more power because you are hypothesis testing just a few
- still trying to identify where the significance is located
you have to custom program the syntax into the computer when you want to ask customized questions
pairwise comparisons equate to just a fishing expedition
whatever is interesting to you is what you are going to test
- the resarch interests drive the questions that you are investigations
- the statistics should not drive what you do
to test for orthagonality, it is mainly a process of playing with the coefficients
you never test the obvious: test what you don't know
- ask important questions with your data
experimental design is basically about finding or creating a methodology that can be replicated
- target specific groups
no one in biology at a high level is doing pairwise comparisons
- I can't spend more time on this because you don't have chapter 5 from the Kirk book
From Kirk book chapter 4, Table 4.1-2
- shows recommended procedures when ssumptionis are tenable
We will discuss the following now:
- Tukey
- Newman-Kewls
- Fisher-Hayter
Scheffe must be used for a more complex hypothesis that goes beyond just pairwise comparisons
- you can group your data in interesting ways to identify new things / trends
-- you will always use the AVERAGE of your variable results for comparison, not just the sum
Fisher-Hayter
- each contrast (phi) is going to have a unique standard error (psi hat: a pair of means)
posthoc power comes from using the pooled variance: the mean of squares between
Posted: Fri - September 24, 2004 at 12:02 AM
Quick Links
Calendar
Sun | Mon | Tue | Wed | Thu | Fri | Sat |
Categories
Archives
XML/RSS Feed
Statistics
Total entries in this blog:
Total entries in this category: 3
Published On: Sep 24, 2004 12:02 AM
Total entries in this category: 3
Published On: Sep 24, 2004 12:02 AM
