Class Notes 11 Nov 2004
Another ANCOVA example, start Tabaschnik book
(a little)
EPSY
6301
11 Nov 2004
Arturo needs the problem and the data and the writeup first
- then the calculations
- SPSS last
- staple the work
- it should look like a top notch graduate project
- do it in folders if possible (as neat as possible)
-- no more points are given for doing sigma locations by hand
-- learn this not to learn math type
- do some reminders about the model you are using, what type of methodology you are using
- don't just give lots of tables without explanation, that is what the writeup is about
Dr O needs to know if we are understanding what is going on
- the more you can elaborate the better
Now you don't have another chance if you get the project wrong, you just get one shot at turning it in
Future meetings
- we meet Nov 11 and Nov 18
- we don't meet Nov 25 for Tgiving
- we meet Dec 2, Dec 9, and not Dec 16
- projects are due Dec 16th
With tonight then, we have 4 more meetings (we still have not dealt with Tabachnik book)
PROJECT #4: Chap 15 Problem #9
Tonight we will do another example of ANCOVA
ANCOVA one way is basically an adjusted ANOVA
- adjusting for initial differences
- our Brunin example today talks about: "In order to make a decision concerning future equipment needs..."
a lot of times we do this type of assignment because we cannot do random assignment in terms of people assigned to a treatment, we just assign a treatment to the classroom
- there are several experimental problems here
- using a covariate is a way
- you don't want to use the same test because of the carryover effects (that is why they didn't do a spelling test as the pretest)
hypothesis is about the Y value: dependent score
question is are the differences in spelling test post) averages due to the different treatment levels
can see from the initial vocab test that the groups are not starting out the same
do the same information for the covariate
- have to get some crossproducts between x and y, because they are related
- first thing is to get a straight correlation between x and y, ignoring treatment levels
Fact that ANOVA table for entire dataset on y (posttest) is signficant tells us this may have been caused from
can do a linear regression instead of a covariate, but that could not sepearate things out by treatment method / group
covariate analysis lets you separate out the impact from the covariate versus the impact of the treatment
if you don't have random assignment or random selection, many people will try to use covariance
in ANCOVA we need to do comparison on adjusted means, not straight means
- we use Fischer Hayter, it is called teh Brian Haulson method in this case
- it uses adjusted means
covariates are almost always random effects, not fixed
My comment: I guess this weekend I need to buy the Kirk book, I should have done this at the start of term
3.61 is the critical value from Kirk table E-17 for project 4 (we need to use 24 instead of 26)
SPSS calls it the marginal means, in this example it is Sxy and Sxx
These post-hoc mean comparisons help us compare apples to apples, we are able to discount the initial differences which are observable in the initial vocab test scores we had before the treatment
If the interaction is NOT significant then you CAN do ANCOVA
- this would mean the lines are parallel
- if the interaction is significant then you can do ANCOVA
Usually when you enter the computer data, put the covariate in FIRST, then the method
- has to do with the concept of partialing
Now we are done with the Kirk book
Tabachnick p 275
ANCOVA is an adjusted ANOVA
blocking is used for a discrete/group variable (like gender) not for a continuous variable! (with a continuous variable you can do a covariance)
- blocking is a way to homogenize even further
- in a survey, remember it is always best to collect data at the highest level possible (continuous if you can), so you can use the most powerful statistical methodologies available (you can always categorize data later)
Results sections in Tabaschnik book are great examples of what Dr O wants us to have in our discussion and results sections
next week: will review regression and multiple regression
- we will have at least 1 example but no project on multiple regression
We WILL have a project on Tabaschnik Chap 9 on Multivariate Analysis of Variance and Covariance
we dropped 3 way project, we may drop multiple var
we will have 2 more projects (6 total)
- no MANCOVA
- we will do MANOVA
no #7 or #3
11 Nov 2004
Arturo needs the problem and the data and the writeup first
- then the calculations
- SPSS last
- staple the work
- it should look like a top notch graduate project
- do it in folders if possible (as neat as possible)
-- no more points are given for doing sigma locations by hand
-- learn this not to learn math type
- do some reminders about the model you are using, what type of methodology you are using
- don't just give lots of tables without explanation, that is what the writeup is about
Dr O needs to know if we are understanding what is going on
- the more you can elaborate the better
Now you don't have another chance if you get the project wrong, you just get one shot at turning it in
Future meetings
- we meet Nov 11 and Nov 18
- we don't meet Nov 25 for Tgiving
- we meet Dec 2, Dec 9, and not Dec 16
- projects are due Dec 16th
With tonight then, we have 4 more meetings (we still have not dealt with Tabachnik book)
PROJECT #4: Chap 15 Problem #9
Tonight we will do another example of ANCOVA
ANCOVA one way is basically an adjusted ANOVA
- adjusting for initial differences
- our Brunin example today talks about: "In order to make a decision concerning future equipment needs..."
a lot of times we do this type of assignment because we cannot do random assignment in terms of people assigned to a treatment, we just assign a treatment to the classroom
- there are several experimental problems here
- using a covariate is a way
- you don't want to use the same test because of the carryover effects (that is why they didn't do a spelling test as the pretest)
hypothesis is about the Y value: dependent score
question is are the differences in spelling test post) averages due to the different treatment levels
can see from the initial vocab test that the groups are not starting out the same
do the same information for the covariate
- have to get some crossproducts between x and y, because they are related
- first thing is to get a straight correlation between x and y, ignoring treatment levels
Fact that ANOVA table for entire dataset on y (posttest) is signficant tells us this may have been caused from
can do a linear regression instead of a covariate, but that could not sepearate things out by treatment method / group
covariate analysis lets you separate out the impact from the covariate versus the impact of the treatment
if you don't have random assignment or random selection, many people will try to use covariance
in ANCOVA we need to do comparison on adjusted means, not straight means
- we use Fischer Hayter, it is called teh Brian Haulson method in this case
- it uses adjusted means
covariates are almost always random effects, not fixed
My comment: I guess this weekend I need to buy the Kirk book, I should have done this at the start of term
3.61 is the critical value from Kirk table E-17 for project 4 (we need to use 24 instead of 26)
SPSS calls it the marginal means, in this example it is Sxy and Sxx
These post-hoc mean comparisons help us compare apples to apples, we are able to discount the initial differences which are observable in the initial vocab test scores we had before the treatment
If the interaction is NOT significant then you CAN do ANCOVA
- this would mean the lines are parallel
- if the interaction is significant then you can do ANCOVA
Usually when you enter the computer data, put the covariate in FIRST, then the method
- has to do with the concept of partialing
Now we are done with the Kirk book
Tabachnick p 275
ANCOVA is an adjusted ANOVA
blocking is used for a discrete/group variable (like gender) not for a continuous variable! (with a continuous variable you can do a covariance)
- blocking is a way to homogenize even further
- in a survey, remember it is always best to collect data at the highest level possible (continuous if you can), so you can use the most powerful statistical methodologies available (you can always categorize data later)
Results sections in Tabaschnik book are great examples of what Dr O wants us to have in our discussion and results sections
next week: will review regression and multiple regression
- we will have at least 1 example but no project on multiple regression
We WILL have a project on Tabaschnik Chap 9 on Multivariate Analysis of Variance and Covariance
we dropped 3 way project, we may drop multiple var
we will have 2 more projects (6 total)
- no MANCOVA
- we will do MANOVA
no #7 or #3
Posted: Thu - November 11, 2004 at 10:51 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: 9
Published On: Nov 16, 2004 12:43 PM
Total entries in this category: 9
Published On: Nov 16, 2004 12:43 PM
