arch M 7, 2011
We can possibly resubmit exam the take home part. Due F riday.
Take Home Exam
General
If there is output but not verbiage then there is no points.
We have to wri te down. Less output is better. Only give output that is very
i mportant.
P roblem 1
Be sure to specify your n ull and alte rnative hypothesis explicitly.
P ai red test should have null and alternative should be different. Everything
should have a separate hypothesis.
B e careful t o report what you do.
D ependencies: Paired
I ndependent : Pooled
P roblem 2
Dont give a test with a p-value.
Give a confidence internal and say what it means.
You can do the p-test but dont have to give i t, just do i t for yourself.
USE the actual 95% confidence inte rnal.
U SE the inte rval to do the test.
I f we are giving the p-value we are doing the w rong thing.
M ajority of people talked more.
A lpha & do p-value not test statistical significance.
Confidence In terval 2 S IDE D
Dont give 2-sided p-values for a one-sided test.
C lasswork
When Y is t ransformed, it:
Changes equal variance
Changes distribution
Changes shape
We can look at measure of variability vs measure of mean.
I n essence, equal variance and distribution are tied together.
W hen X is t ransformed, it only effects:
t he shape of the curve, nothing else.
I f red line sits on top of blue line, it is not statistically significant.
I f there is curvature on the dotted line, i t means we h ave to t ransform the
v ariable .
Parameter estimates are for predicting (weights).
S ince there is log on Y even though there isnt log in all of the Xs, the answer
to predict will still be in terms of L og Y.
I n a r esidual plot, we should see randomness and if there is a pattern then
something is not r ight.