October 20, 2009
Midterm exam
The midterm exam will be held during class on Tuesday, October 27th. It will cover the material in the first set of notes "Background and review." You are not responsible for anything beyond that.
The best way to prepare is to review the problem sets you have done so far this semester and their solutions. The short problem set due this week is based on the background/review material, so that is relevant practice for the exam (you'll have the solutions for it on Friday).
I also posted last year's midterm and a set of practice problems on the course web site. Last year's midterm has solutions, but the practice problems do not. I will be happy to discuss the practice problems, but I will not be preparing solutions for them. Note that a few topics have changed from last year -- you do not need to know about the "delta method."
You cannot use notes or books during the exam. You may use a simple calculator (not a PDA, laptop, etc.). I will trust everyone to only use calculators for checking arithmetic.
This Thursday (10/22) during class we will do some review and I will take questions. I will not lecture on new material until after the exam.
Things you should know:
* Definition and calculations of probabilities and quantiles.
* Definitions, calculations, and properties of expected values, variances, standard deviations, covariances, and correlation coefficients.
* The definition of power and coverage probability.
* The variance of the sample mean for independent and for dependent data.
* The double expectation theorem and law of total variation.
* The test statistic for the two-sample Z-test.
* How to construct confidence intervals.
* Definitions of conditional, marginal, and joint probabilities.
Things you do not need to know:
* Anything having to do with degrees of freedom or the t distribution.
* The formula for sample variance, sample covariance, or sample correlation coefficient.
You should be able to do power problems similar to what appeared on last year's exam. I do not recommend memorizing the power formulas on slides 63-65. You can do these sorts of calculations by standardizing the test statistic, as long as you know the properties of means and variances.
Posted by kshedden at 04:45 PM | Comments (0)
October 13, 2009
Calculating correlations (ps 5 #2)
Here is some advice for calculating correlation coefficients.
* There is very little you can do directly with a correlation coefficient. You need to convert the correlation coefficients into expressions involving variances and covariances (slide 101).
* You should use properties of covariances (slides 96/97) and variances (slide 38) to simplify the expressions. It is not helpful to use the definition of the variance (slide 36) or the definition of covariance (slide 95). Using these definitions to convert things into expectations would be a more complicated way to solve the problems (not advised).
* Properties of variances and covariances do not apply to standard deviations and correlations. For example, if X and Y are independent we know that var(X+Y) is equal to var(X)+var(Y), but SD(X+Y) is not equal to SD(X)+SD(Y). Similarly, we know that cov(A+B,C) = cov(A,C) + cov(B,C), but it is not true in general that cor(A+B,C) = cor(A,C) + cor(B,C) (note carefully whether you have a "cor" or a "cov").
Posted by kshedden at 11:45 AM | Comments (0)
September 22, 2009
Problem set 2, number 4
For this question you don't need to answer part (ii) about the standard deviation. You can just answer part (i) about the mean. We haven't gotten far enough yet to do the standard deviation part. I will put that on a future homework.
Posted by kshedden at 11:22 PM | Comments (0)
