Im not sure if Im the only one, but Im not quite sure how to do problem
3. any tips? Thanks!
Cheers,
Dave
Show replies by date
Dave,
We dissected the formula for sampling variance of multiple regression
(see Fox 217), which is the V(b) = sigma^2(X'X)^ -1
into its component parts, and solved to get a big messy-looking
variance-covariance matrix. Then showed the elements on the diagonal
were equal to V(A) and V(B), respectively.
The following isn't as clear as I'd hoped, but here is a description of
what we did.
Using the form given on Fox 204-5, where X = row vector of regressors
(with 1 as a constant) and Yi = (1, x1, x2.xik)
So, in matrix form, you have a column vector of Ys = (matrix with 1s
in the first column, xs in the other columns)
So we can write X as a 2xn matrix, with 1s in the first column and the
row of x's in the second. Therefore X' is a row vector with first row
of 1s and second row of x's.
Plugging these in, you can solve for X'X, (X'X)^1, etc.
This gives the variance-covariance matrix. You'll can then work with
the formulas on the diagonal and show that they are equal to the
formulas V(A) (show it's equal to first element in var-covar matrix) and
V(B) (show that it's equal to the bottom-right element in var-covar
matrix). (see Alison's handout for the two equations you have to show
are equal).
Brodi
-----Original Message-----
From: gov1000-list-bounces(a)lists.fas.harvard.edu
[mailto:gov1000-list-bounces@lists.fas.harvard.edu] On Behalf Of David
Ni
Sent: Sunday, November 28, 2004 2:52 PM
To: gov1000-list(a)lists.fas.harvard.edu
Subject: [gov1000-list] problem 3?
I'm not sure if I'm the only one, but I'm not quite sure how to do
problem 3. any tips? Thanks!
Cheers,
Dave