Class notes 4 Nov 2004
More on designs that are not projects for this
class but we should be familiar with
EPSY
6301
4 Nov 2004
2 nuisance variables is called Latin Science design
- assuming additivity
3 way ANOVA is also called a 2 x 2 x 2
- between subjects design
- looking at a design that is 3 dimenstional like a cube
- you have 7 hypotheses with this: 3 main effects, 3 first order interactions, 1 2nd order interaction
Designs addressed by Kirk:
- CR-P 1 hypothesis
- RB-P 1 hypothesis
- CRF-pq 3 hypotheses
- LS-pq
CRF-per 7 hypotheses
CRE-pqrt 15 hypotheses
Chap 10 of Kirk now
- hierarchical design, blocking
- not deal with within subject design
This book also does not deal with repeated measures design
My comment: Here it is Nov 4th and we are still working on the Kirk book which I did not buy because I was told not to: I bought Tabaschnik......
we are going to use the pretest as the covariate (do this when you know there are differences at the onset)
the altimeter experience with pilots was a bad example for blocking because it would have been better to use hours of flight time as a covariate
the horse is the question, don't put the cart before the horse
- you must have your question first!
Covariance is just ANOVA adjusted (because of the covariate)
- first term accounts for the covariance
If I can randomly select classrooms but not randomly assign, then covariance can be good
- collect more info on the group: like IQ (it tends to affect school performance of any type)
- covariate can be used to adjust for initial differences
- lets you say that the difference in your treatment variable is due to the control variable
- design is still a t-test, but we control for initial differences by using this statistical control versus a design control
covariates should not be highly related: so first you will look for correlation, should not be .9, can be something like .4 and .5
Covariates also have to be going in the same direction
statistical control and experimental control are not mutually exclusive
- Dr O's favorite is doing both
- should attempt to use experimental control whenever possible
can use absenteeism as a covariate for performance
for covariate want at least interval level, should be continuous variables
Kirk p 720 Covariance
- to do this we must have 2 values for each subject
-
ANCOVA = analysis of variance adjusted
significance of the covariate is bogus: doesn't matter if that shows significance in the table, we are looking for significance in the groups
to change the y axis origin:
- right click, choose SPSS chart object, enter the enditor, click Y and and on scale tab change minimum to zero
4 Nov 2004
2 nuisance variables is called Latin Science design
- assuming additivity
3 way ANOVA is also called a 2 x 2 x 2
- between subjects design
- looking at a design that is 3 dimenstional like a cube
- you have 7 hypotheses with this: 3 main effects, 3 first order interactions, 1 2nd order interaction
Designs addressed by Kirk:
- CR-P 1 hypothesis
- RB-P 1 hypothesis
- CRF-pq 3 hypotheses
- LS-pq
CRF-per 7 hypotheses
CRE-pqrt 15 hypotheses
Chap 10 of Kirk now
- hierarchical design, blocking
- not deal with within subject design
This book also does not deal with repeated measures design
My comment: Here it is Nov 4th and we are still working on the Kirk book which I did not buy because I was told not to: I bought Tabaschnik......
we are going to use the pretest as the covariate (do this when you know there are differences at the onset)
the altimeter experience with pilots was a bad example for blocking because it would have been better to use hours of flight time as a covariate
the horse is the question, don't put the cart before the horse
- you must have your question first!
Covariance is just ANOVA adjusted (because of the covariate)
- first term accounts for the covariance
If I can randomly select classrooms but not randomly assign, then covariance can be good
- collect more info on the group: like IQ (it tends to affect school performance of any type)
- covariate can be used to adjust for initial differences
- lets you say that the difference in your treatment variable is due to the control variable
- design is still a t-test, but we control for initial differences by using this statistical control versus a design control
covariates should not be highly related: so first you will look for correlation, should not be .9, can be something like .4 and .5
Covariates also have to be going in the same direction
statistical control and experimental control are not mutually exclusive
- Dr O's favorite is doing both
- should attempt to use experimental control whenever possible
can use absenteeism as a covariate for performance
for covariate want at least interval level, should be continuous variables
Kirk p 720 Covariance
- to do this we must have 2 values for each subject
-
ANCOVA = analysis of variance adjusted
significance of the covariate is bogus: doesn't matter if that shows significance in the table, we are looking for significance in the groups
to change the y axis origin:
- right click, choose SPSS chart object, enter the enditor, click Y and and on scale tab change minimum to zero
Posted: Mon - November 8, 2004 at 08:25 PM
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: 8
Published On: Nov 08, 2004 08:25 PM
Total entries in this category: 8
Published On: Nov 08, 2004 08:25 PM
