Week 5 Flashcards
How to calculate the 95% confidence interval for beta.j?
(beta hat.j - 2.09 sqrt(s^2v.jj), beta hat.j + 2.09sqrt(s^2v.jj) Note in this case n = 20, replace 2.09 with the correct students t-value.
How to describe confidence intervals?
We cannot state that beta.j has a probability of 95% to be in the interval, we can state that: “The 95% confidence interval is (…, …)”
How do we derive the 95% prediction interval?
If we have predictor y* hat = X beta* hat of y*, we know that y* hat - y has mean zero and variance: Var(y* hat - y*) = sigma2 (X*(X’X)^-1X*’ + I.m).
If we have a squared normal random variable it follows the chi-squared distribution, we get that (y* hat - y*)’ Var(y hat* - y*) (y* hat - y*)/sigma2 is chi-squared distributed.
If we replace sigma2 by s2 we get an F-distribution (where m denotes the compontents in y*). (y* hat - y*)’ Var(y hat* - y*) (y* hat - y*)/ms2 ~ F(m, n - k).
In the special case that m = 1: (y hat * - y*)/(s sqrt(x’*(X’X)-1x* + 1))
What do n - k usually mean in this course?
n: number of data points
k: number of regressors
How to test a linear restriction on the beta?
- First rewrite the restriction to a linear equation. w is the vector of the equation.
- Then we test the following:
tw = (w’ beta hat - r)/(sqrt(s2w’(X’X)-1w)) ~ t(n -k)
How is the linear restriction derived?
It’s known that w’beta hat ~ N(w’beta, sigma2w’(X’X)-1w), thus if we write:
(from now on d = ) (w’beta hat - w’ beta) / (sqrt(sigma2w’(X’X)-1w)) ~ N(0, 1)
Then we can state:
d/sqrt(s2/sigma2) = (w’beta hat - w’beta)/sqrt(s2w’(X’X)-1w)
Which follows a students t-distribution, we can test for the restriction using:
tw = (w’beta hat - r) / sqrt(s2w’(X’X)-1w) ~ t(n - k)
In what formula is that includes s2 is it chi-squared distributed?
(n - k)s2/sigma2 is kai-squared distributed to degree n -k
When is students t-distribution often found?
When you divide a normal distribution by an chi-squared distribution.
How to test several linear restriction at the same time?
Because we have: R beta = N(R beta, sigma2R(X’X)-1R’), we can rewrite this as:
(beta hat - beta)’R’(R(X’X)-1R’)-1R beta (beta hat - beta) / sigma 2 ~ chi squared(m).
Thus if we rewrite to include s2, we have:
(beta hat - beta)’R’(R(X’X)-1R’)-1R beta (beta hat - beta) / m s2 ~ F(m, n - k)
What is the formula of the adjusted R2?
R bar2 = 1 - (e’e/(n - k))/(SST/(n - 1)) = 1 - s2/(SST/(n - 1))
What is model averaging?
It joins two models together by assigning them both some weight.