c**********e 发帖数: 2007 | 1 I have a dataset, which as variable X, Y, Z, W, all continuous.
The two variables Z and W are "comparable" -- they
have the same range, similar mean, standard deviation, etc.
I would like to compare parameters of Z and of W
in the following 2 models
Y = X Z
Y = X W
Does anybody have any idea on this?
Thanks a lot! |
D******n 发帖数: 2836 | 2 dont understand. elaborate it more please.
【在 c**********e 的大作中提到】 : I have a dataset, which as variable X, Y, Z, W, all continuous. : The two variables Z and W are "comparable" -- they : have the same range, similar mean, standard deviation, etc. : I would like to compare parameters of Z and of W : in the following 2 models : Y = X Z : Y = X W : Does anybody have any idea on this? : Thanks a lot!
|
c**********e 发帖数: 2007 | 3 The fitted models are
Y = a1 + b1*X + c1*Z + error
and
Y = a2 + b2*X + c2*W + error
I want to test H0: c1=c2.
【在 D******n 的大作中提到】 : dont understand. elaborate it more please.
|
s*r 发帖数: 2757 | 4 it will be easier to you get c1 from one part of the sample
and get c2 from another part.
things will be also easier if you know z and w are independent
【在 c**********e 的大作中提到】 : The fitted models are : Y = a1 + b1*X + c1*Z + error : and : Y = a2 + b2*X + c2*W + error : I want to test H0: c1=c2.
|
D******n 发帖数: 2836 | 5 i think this question was asked in this board before, if the Ys are differen
t and Z,W are not dependent, it is easy.
【在 s*r 的大作中提到】 : it will be easier to you get c1 from one part of the sample : and get c2 from another part. : things will be also easier if you know z and w are independent
|
s*r 发帖数: 2757 | 6 or we can do some permutation test
differen
【在 D******n 的大作中提到】 : i think this question was asked in this board before, if the Ys are differen : t and Z,W are not dependent, it is easy.
|
c**********e 发帖数: 2007 | 7 If two part of sample, it would be a straight-forward F-test.
Z and W are highly correlated. The test is to gain insight how
much difference between Z and W in predicting Y.
Yes, it is a hard question. The trouble is that c1_hat and c2_hat
are from 2 different models. So we do not know which to use to
calculate their expectations and variances.
【在 s*r 的大作中提到】 : it will be easier to you get c1 from one part of the sample : and get c2 from another part. : things will be also easier if you know z and w are independent
|
c**********e 发帖数: 2007 | 8 Yes, I am asking the hard part of the question.
Thanks for your input.
differen
【在 D******n 的大作中提到】 : i think this question was asked in this board before, if the Ys are differen : t and Z,W are not dependent, it is easy.
|
D******n 发帖数: 2836 | 9 btw, why do u wanna test c1=c2, if u wanna know which predict Y better?
【在 c**********e 的大作中提到】 : If two part of sample, it would be a straight-forward F-test. : Z and W are highly correlated. The test is to gain insight how : much difference between Z and W in predicting Y. : Yes, it is a hard question. The trouble is that c1_hat and c2_hat : are from 2 different models. So we do not know which to use to : calculate their expectations and variances.
|
c**********e 发帖数: 2007 | 10 The two R^2's are close, with one slightly better than the other.
It comes up a generic question to test c1=c2.
【在 D******n 的大作中提到】 : btw, why do u wanna test c1=c2, if u wanna know which predict Y better?
|
|
|
c**********e 发帖数: 2007 | 11 A question is: whether these two R^2's statistically equal.
It also boils down to the problem of F-test without the "full model".
【在 c**********e 的大作中提到】 : The two R^2's are close, with one slightly better than the other. : It comes up a generic question to test c1=c2.
|
r******r 发帖数: 138 | 12 is orthogonalizing Z & W the first step you want? and then check if the
orthogonal term significant?
【在 c**********e 的大作中提到】 : A question is: whether these two R^2's statistically equal. : It also boils down to the problem of F-test without the "full model".
|
s*r 发帖数: 2757 | 13 f test is usually for nested model
i guess you can use bayes factor to compare one fitting is better than the
other
【在 c**********e 的大作中提到】 : A question is: whether these two R^2's statistically equal. : It also boils down to the problem of F-test without the "full model".
|
c**********e 发帖数: 2007 | 14 Orthogonalizing means that to check the additional effect of Z after
including W first, or the additional effect of W after including Z first.
Either way it is not a head to head comparison.
【在 r******r 的大作中提到】 : is orthogonalizing Z & W the first step you want? and then check if the : orthogonal term significant?
|
r******r 发帖数: 138 | 15 then what exactly are you comparing here?
【在 c**********e 的大作中提到】 : Orthogonalizing means that to check the additional effect of Z after : including W first, or the additional effect of W after including Z first. : Either way it is not a head to head comparison.
|
D******n 发帖数: 2836 | 16 it actually doesnt quite make sense to me to compare the coefficients of two
unrelated variables. It is more seen to compare the coefficents of the same
varialbe under different treatments.
y=b0+b1x1+b2x2+b3x1x2+e
x2 is class varialbe = 1 if in class 1, =0 in class 2
test if b3=0
【在 r******r 的大作中提到】 : then what exactly are you comparing here?
|
g******7 发帖数: 19 | 17 one possible solution is using the simutaneous equation modeling, or
structural equation modeling (the so-called SEM)
but first you will need to establish a theoretical frame/model. if you can,
it is very straitforward to test: coefficient of Z = coefficient of W by
model comparison (something similar to GLH test in GLM models).
In SAS, you use proc Calis
the other software to do it include: LISREL, AMOS etc.
【在 c**********e 的大作中提到】 : I have a dataset, which as variable X, Y, Z, W, all continuous. : The two variables Z and W are "comparable" -- they : have the same range, similar mean, standard deviation, etc. : I would like to compare parameters of Z and of W : in the following 2 models : Y = X Z : Y = X W : Does anybody have any idea on this? : Thanks a lot!
|