Review of Yates’ order stuff
Review of fill in the blank ANOVA
Please feel free to comment here on the blog to discuss the notes. Oh and the University web site was down all weekend, so that explains why this was not posted until right now.
That title should be self explanatory…
Here are the answers to the fill in the blank questions. Feel free to discuss it here.
Dave
Some regression stuff:
Model R2 C(P)
Y=X1 .39 6.2
Y=X2 .10 2.1
Y=X3 .29 5.7
Y=X1X2 .49 3.1
Y=X2X3 .39 2.2
Y=X1X3 .58 6.1
Y=X1X2X3.68 4.0
Note, the expected value of C(P) is p-1.
Which of the following is the best model and why?
OH yeah, and if C(p) = p-1 then you have an unbiased model, it overpredicts as much as it underpredicts.
Could you please post the yates order answer?
Thank you
@Kellie yeah I will do that today, in the mean time, here are the solutions for assignment 4:
Sv df
B 1
S(B) 8
A 1
AB 1
AS(B) 8
Sv df
B 1
S(B) 8
A 1
AB 1
AS(B) 8
C 1
CB 1
CS(B) 8
CA 1
CAB 1
CAS(B) 8
Assuming you went back to the original design
Sv df
B 3
S(B) 16
A 1
AB 3
AS(B) 16
OK the review questions
B 1
S(B) 8
A 1
AB 1
AS(B) 8
The second one is really hard actually. The key is recognizing that A(C) is a between factor.
B 1
C 1
BC 1
A(C) 1
BA(C) 1
S(ABC) 40
Dave can you please post how you got the df. I don’t understand the last yate’s question.
(n-1)abc
I’m a little confused. Should it not be
B =(b-1) =1
C =(c-1) = 1
BC =(c-1) (b-1) = 1
A = (a-1) = 1
A(c) = (a-1)c = 2
BA (C) = (b-1)(a-1)c = 2
S (abc) = (n-1)abc = 40
total = 47
N-1
I do understand that part, but for A(c), i did: (a-1)c so, 1×2… sould I be using c=1 instead? I am guessing c=1 is right, but I don’t understand why. Help Please 🙂
I agree with you andrea i re did my math and it still comes out different. total should be 47 but the way i did it , it came out to 48?
Ya I also got 47 when I did mine.
confused.
@Kellie You can’t have A and A(C), if you take that out it is 47, and alas, you are right and I am a weenie…..
@Andrea it would be 2, I made a mistake copying my notes.. That said, the point is moot
Most comments ever….
It should come out to 47, N-1 = 47
so it should be
B 1
C 1
BC 1
A(C) 2
BA(C) 2
S(ABC) 40
ya? no?
Bobby for the win… Yes.
Dave can you please explain the regression stuff you posted. 🙂
@Andrea, on the regression stuff, check those models, which one is the best do you think? You want a combo of a high R2 and a C(p) that is close to the number of terms in the model. Oh and don’t forget about complexity….
Cool, can you post the answer(s) for the regression stuff, with explanation(s)?
Well, there are a couple of good choices there. If I had a choice (and, well I guess I do…) I would take the X1 X2 model. It has a decent R2, it is not that complex, the C(p) is close to p-1, and there is no correlation between X1 and X2. An argument could be made for the three variable model too, but having no multicolinearity would make me pretty happy if these were my data.
Okay Dave, tell me if I’m right here: X1X2 is good because, it has a pretty high R2 (Highest would be 1?). Also there is no correlation between X1 and X2 (meaning no overlap?–because X1 is .39 and X2 is .10 = .49). But I don’t understand what C(P) is. If there are three X’s should we want C(P) to be 3? and why?
Andrea, yeah you are right. C(p) is a measure of bias, if it is roughly the same as the number of predictors your model underpredict as much as it overpredicts.
Got it! Thanks Dave!