A.3.5 has been removed (Normal-Normal conjugate priors are no longer on the syllabus).

I’ve added a new lesson B.1.0 introducing LMM with an intercept only example. This isn’t related to syllabus changes; the goal is to help clarify why we use random effects as that is something a lot of people ask about.

C.2.5: Multicollinearity is no longer on the reading list and this lesson will be shortened. Target date 3/19

C.2.9 and C.2.10: Probably 2 new lessons will be added to include material on splines, cross-validation, and prior distribution simulations. Target date 3/26

C.3.2: The 2nd edition of Statistical Rethinking adds an example related to shortcomings of Metropolis-Hastings and Gibbs Sampling. Three slides of the lesson (including the exercise) will be updated to reflect this. Target date: 3/13

C.3.5: New lesson on trace rank plots. This is the first fully new lesson to be added as I expect at least one question related to this to appear on the exam this sitting. Target date: 3/13

C.4.1: Edits to reduce the emphasis on general link functions as it seems that only log link (for Poisson) and logit link functions (for binomial) are still on syllabus. Target date: 3/19

C.5.5: New lesson on using uncentered priors to get better convergence of chains. Target date: 3/31.

]]>In the hypothesis tests that we will be doing, we will always care about either the right tail (likelihood ratio test with chi-square distributions, F-tests) or 2-sided tests (t distribution). Because of that, I would recommend always using the .RT or .2T versions of those functions. That is, with chi-squares, I would use CHISQ.DIST.RT(x, dof), F-test would use F.DIST.RT(x, numerator dof, denominator dof), and t-test would use T.DIST.2T(x, dof). The advantages of those is a) we don’t need to subtract from 1 and/or divide things by 2, b) the syntax is all fairly similar, being <distribution name>.DIST.<RTor 2T>(x, degrees of freedom), and c) the most common error is probably selecting the wrong tail type, and plugging in x=0 into any of those 3 formulas will output 1 for the tail type we want, so you have an easy way to double check you are using the correct formula.

]]>The full announcement (https://www.casact.org/press/index.cfm?fa=viewArticle&articleID=4889) also includes the following:

“**Registration for spring 2021 exams will open on December 15, 2020**, and candidates who register early will have more options for selecting their preferred site, date, and time than candidates who register later in the registration window. While we understand that candidates who sat for a fall 2020 exam may not want to register for their spring 2021 exam before knowing their exam results, candidates whose plans for the spring sitting are more certain should take advantage of the opportunity to register and secure their Pearson VUE appointment when registration opens. The window for registration will **close on April 9, 2021, and the refund deadline is April 16, 2021**.”

I think this means that you should register for whichever exam you think that you are most likely to take in the Spring as soon as the registration window opens and schedule your exam with Pearson VUE as soon as possible, even though results aren’t out yet. If you are correct, then you will have first pick at an appointment slot, while if you have to change later you can still cancel your scheduled slot and get a new one.

The release time for Fall 2020 results is still listed as “during Q1 2021″.

]]>You must bring your own mask. Masks with valves are not allowed. I would work all practice problems in the final week with a mask on so that you don’t get distracted by it on exam day.

For doing work by hand, you get a 5 page laminated notebook that is somewhat larger than legal size, along with a non-erasable marker. It’s fairly similar to https://www.amazon.com/Manhattan-GMAT-Simulation-Booklet-Marker/dp/0979017580 This initially sounded insufficient to me, but in working the released exams I actually find it works out ok, largely because I’m doing most of my work on a scratch spreadsheet you get in the exam.

More specifically, the CBT format gives you a calculator simulator as well as a scratch spreadsheet you can use. You also can, and should, bring your own calculator as using a physical calculator saves precious screen space and is faster than the onscreen calculator. You also may find that you use the spreadsheet exclusively for computations.

The Pearson VUE scratch spreadsheet is not Excel, but seems to be based on it. The tl;dr is that most Excel functions work, but most shortcuts don’t. When practicing, use Excel and not Google Sheets as Pearson VUE prefers Excel formula syntax when they differ.

Things you will dislike about the Pearson VUE scratch sheet:

The spreadsheet resets the active cell to A1 each problem so you have to scroll back down to where your work is. It may be tempting to clear the spreadsheet so you don’t have to scroll — I think this is a terrible idea. I would label each problem clearly, and save everything for checking work later.

Most Excel shortcuts don’t work. E.g., hitting F4 doesn’t insert absolute references, you can’t smart fill down by double clicking, etc. Try not to use keyboard shortcuts when practicing.

The spreadsheet has a fixed size of 100 rows and 40 columns (A1:AN100). You can’t insert or delete rows and columns, which makes editing somewhat harder, but you can select an area and drag it to move things.

Some good things about the Pearson VUE CBT format:

With the big exception of the normal distribution table, the exam tables are searchable, a big upgrade over the Prometric format you had for your first couple of exams. If you need the normal table, instead of typing normal, you can remember that it’s on p. 4.

Within each problem, you can highlight and cross out text. It will save your annotations for when you review your work. I think this is really helpful for the triple true-false questions, as you can highlight the ones you know are true, cross out the ones you know are false, and use what is left to make your guesses (yes, if you can eliminate answer choices you should guess even with the adjustment for guessing)

As these changes make things quite different than what you have seen before, I would practice on the Pearson VUE demo (https://home.pearsonvue.com/cas) at least twice before exam day. Good luck!

]]>The questions I found most problematic are #38, where statement I should be about splits rather than trees to be true (I could see them going either way on this one, or maybe accepting 2 answers), #35, where the given information is contradictory but I still think there is a best answer, and #34, which I think should be thrown out but they may stick with C.

If you have any questions about these or others, posting on the discussion forum is easier for me to respond to than blog responses (which tend to hit spam filters). If you want help with an appeal, let me know.

]]>I’ve also added solutions to the CAS sample problems to the `Before You Begin’ section. Most of these problems and solutions exist elsewhere in the seminar, sorted by topic, but some people have asked for them all in a single place.

]]>In terms of new content, there are going to be 3 new / revised lessons by the end of this month. There will be two new lessons at the end of chapter C.5, and I think I’ve come up with a better way of explaining principal component analysis so will revise at least one of those lessons in D.2.

After those are done, I will start expanding the practice problems. This is going to be a continual process, with some problems coming in July, more in August, etc.

Finally, I’m making formula sheets which are going in a new section labeled `Before You Begin’. This currently contains the formula sheets that the CAS will give you on exam day, plus formula sheets for Section A. The Section B formula sheet should be up later this week, and I expect Section C and D to follow by the end of the month.

]]>