Alison Gibbs' STA 302 Page

Back to Alison Gibbs' Home Page

STA 302 / 1001 -- Regression Analysis -- Fall 2003

Marks on term tests and assignments (including Assignment 2)

Exam information:

The only aid allowed is a calculator.
The exam covers Chapters 1 through 7 and 11.
From Chapter 1 you can skip "Alternative version of regression model" on pages 13-14 and How to find maximum likelihood estimates on pages 30-35.
In Chapter 2 you can skip the material on power on pages 55-56, "Prediction for mean of m New Observations for given X" on pages 66-67, "Expected mean squares" on pages 74-76, and Section 2.8.
In Chapter 3 we covered the following sections: 3.1, 3.2, 3.3, 3.8, 3.9 (excluding Box-Cox transformations).
In Chapter 4 we covered the following sections: 4.1, 4.2, 4.6.
We covered all of Chapter 5.
In Chapter 6 you can skip: "Confidence region for Regression Surfaces", "Simultaneous CI for several mean responses", "Prediction of mean of m new observations at X_h", and "Prediction of g new observations" from Section 6.7; "Three dimensional scatter plots", the formal diagnostic tests and "Box-Cox transformations" from Section 6.8.
In Chapter 7 skip section 7.5.
In Chapter 11 skip "Time Series" (page 479), "Discontinuity" (pages 477-478), "Other coding" (pages 481-482). The case study in Section 11.7 includes material we didn't cover.
Last page of exam (formula sheet)
There is a typo on the formula sheet: It should say Cov(X)=E(XX')-(EX)(EX)'. (On the formula sheet the plus sign should be a minus sign.)
Note that you are not responsible for the formulae for Cp and PRESSp that are on the formula sheet. Also, the variance formulae needed to construct prediction intervals and confidence intervals for the mean value of Y for the multiple regression case are not on the formula sheet. That's because you'd have to do matrix multiplication with, at smallest, a 3x3 matrix in order to use these formulae, and I'm not going to make you multiply any matrices that big on the exam.

Office hours before the exam:
Wedneday, December 10, 2-5 p.m. Zheng in SS 2101
Thursday, December 11, 10 a.m. - 12 p.m. Gibbs in SS 3103
Friday, December 12, 2-4 p.m. Zheng in SS 1080
Monday, December 15, 10 a.m. - 12 p.m. Gibbs in SS 3103
Tuesday, December 16, 9 a.m. - 12 p.m. Mazumder in SS 1080

Assignment 2 in pdf
Data for question 1 (text)
Data for question 2 (text) Note that there is an additional variable that was omitted from the list on the question sheet. coll is the percentage completing college education. Add this variable to your initial model.
Key to data for question 2, including list of Florida counties (text)
Data for question 3 (text)
Solutions (pdf)
SAS output for question 1 (text)
SAS output for question 2 using square root of Y only (text)
SAS output for question 2 using square root of both X and Y (text)
SAS output for question 2 using log of both X and Y (text)
SAS output for question 3 (text)
(I won't be posting the plots -- there are too many.)

Assignment 1 in pdf
Typo: In Question 1(b) Everson should be Emerson.
Data for question 1 of Assignment 1 (text)
Data for question 2 of Assignment 1 (text)
Data for question 3 of Assignment 1 (text)
ASSIGNMENTS HANDED IN (UNDER MY OFFICE DOOR) BEFORE 2:00 p.m. FRIDAY, OCTOBER 24 WILL NOT BE CONSIDERED LATE. Assignments received on Monday will be considered two days late (20% deduction).
Solutions (excluding SAS output) for Assignment 1 in pdf

Here are the test 2 solutions in pdf. (corrected at 3:10 on Nov. 17 to put in a missing square root)
Note that for question 3(a), giving the confidence interval for the expected square root of strength was sufficient for full marks. To convert back to untransformed units, you would square the final answer. It is incorrect to square the predicted value of Y before calculating the CI -- the s.e. of Y and the s.e. of the square root of Y do not have a simple relationship.
Details on the marking scheme for question 2(b)
The class average on the test was 17.3/30. I haven't decided yet whether there will be any adjustment. I may wait until the end of the course. The final exam will cover the entire course. If you do better on it then you did on your tests, I will substitute your exam mark for your test marks.

Here are the test 1 solutions in pdf.
Test 1 adjustment: your mark will be considered as a mark out of 25 rather than 30.

Alison Weir's practice tests are available here.

Course outline in pdf

Photocopies of the overheads used in lecture are available for 2 hour loan in Gerstein library. Two copies of the Instructor's Solutions Manual are also in Gerstein on short-term loan. They are number 203 in the short-term loan section.

SAS examples from lecture

Using SAS in batch mode on CQUEST:

If you have secure shell software on your local machine (a free version for windows can be downloaded; see the CQUEST web site for more details) you can ssh into login.cquest.utoronto.ca. Save your SAS program file into filename.sas and type sas filename to execute the commands. This will produce the text files filename.log and, if there were no fatal errors, filename.lst. Your output is in filename.lst. Graphics (plots produced using proc gplot or within proc reg) will need to be saved separately in a file (details to come). Currently (there is a problem with CQUEST) you must have your SAS program in your home directory for SAS to find it. For an easy to use UNIX text editor, try pico.

E-mail course instructor: alison.gibbs@utstat.utoronto.ca