Monthly Archives: October 2018

Fall 2018 video solutions posted

I finished up the video solutions for the Fall 2018 exam today.  You can find these in the sample exams tab, just below sample exam 4.

These are free samples until the next exam sitting, so you don’t have to be logged into the site to view them.

I counted about the same number of problems that I would think of as type II or type III on this exam as I counted on the spring exam.  So, the overall difficulty level seems to me to be fairly similar.  It felt harder for me when I went through it.  But I’ve never been able to tell how to connect what I think of the difficulty to the eventual pass mark anyway.    I will say that more of the type II questions seemed to crop up in different places than usual.  I found 4 in the first 10 problems, and that tends to rattle folks, which can cause overall performance to go down.   Also, there were less type II’s on the ELM material, but that’s also the material that people struggle with most, so even though that section was easier, people may still make more mistakes on that part.   I definitely had the feeling that they were trying to make it trickier on more problems than I usually do.
Also, I think that #34 is defective. If it were to get thrown out, that would lower the pass mark, and for folks either skipped that or didn’t choose D, that will help.
I hope you passed!  If you didn’t pass, then you might be eligible for a free extension.  Our extension policy can be found here:

Time series “Intercept” clarification, and new problems this week

Thanks to Tiffany for bringing my attention to the fact that the solution to the second example in the review lesson for E.2 was not consistent with the solutions to somewhat similar problems from old exams and sample exams.

I think that this lesson problem was not being solved the way that you would be expected to solve it on an exam, so I’ve removed that problem.

The source text we are using utilizes these two routines in R:
The ar routine in R subtracts the mean before estimating and does not report a mean or an estimate. It is up to us to compute the mean separately and mean adjust as necessary.
The arima routine spits out an intercept whenever d = 0, and that “intercept” is to be understood as an estimate for the mean, and not as a traditional GLM style intercept that we just add into the right-hand side of the model.

This issue has come up one time on previous exams, on Exam S Fall 2016 #43.
On this exam problem, they provided an intercept in the output, but they didn’t say what routine was used to produce it. If you assumed that it was the arima, so that you mean-adjusted, then you got the preliminary answer key result. However, they eventually also accepted the answer that you get by using the intercept as an intercept instead of mean adjusting, probably because they didn’t say where the intercept was coming from or what it should have represented. I think there is a high likelihood that when they get back around to testing this, they will try to make it more clear what is expected.

Additionally, this week problems were added to D.6.3, D.4.4, D.4.2, and D.4.1.