GLMs (Anderson)

[ClashCityRocker]

This new GLM paper by Anderson is pretty dense. Does anyone have any ideas on likely topics for exam questions off of this?

[glad]

I opened this paper, took a quick look at it, and said, yikes! almost made me not want to take this exam. At this point I'm focusing on some of the other exam parts that (for me) are easier to understand, and offer me more % of the total exam (parts E and D which are each about 35% of the exam; also have gone through part B).

I think this reading may be a candidate for skipping (for me), or I may try and get the basics and get any easy questions.

[GefilteFish144]

Never a good idea to skip new papers. They tend to get a lot of points and easy questions. Look for list questions (e.g., Identify the 3 components of Linear Models), as shown on p.12. There's a good chance they'll take one of the examples on p.24-28 and plug in different numbers. And for any items where they give formulas you'd better be familiar with them.

[BassFreq]

More likely questions will be broad rather than specific to a small point made in the paper.

The numeric examples in this paper can be solved using methods in the minimum bias procedure paper.

Some possible questions:

What assumptions differentiate a GLM from a simple linear model?
What advantages does GLM have over a) minimum bias? b) one-way analysis?

Possible, but less likely:
Why use a log link function when determining rates?
Name and describe 3 types of aliasing.

[jk]

Yeah, I've struggled with visualizing how they're going to test this. A lot of the math is too involved to be solved in an exam question. Which leads to the paradox, that the more densely mathematical the paper, the more likely the questions are to be general and qualitative.

I'm trying to concentrate on definitional things, so if (for example) they give us an input vector and a link function, I'll understand what those things are. Chances are the math will be simple if you know what goes where.

And for some reason, aliasing strikes me as a likely "quick hitter".

[Basso]

More likely questions will be broad rather than specific to a small point made in the paper.

The numeric examples in this paper can be solved using methods in the minimum bias procedure paper.

Some possible questions:

What assumptions differentiate a GLM from a simple linear model?
What advantages does GLM have over a) minimum bias? b) one-way analysis?

Possible, but less likely:
Why use a log link function when determining rates?
Name and describe 3 types of aliasing.

I doubt you will have to cacluate any estimates.
Equating distributions with type of model also strikes me as likely. Knowing how a distributions variance function varies with the mean is important.
For some reason I see an examiner not familar with modeling asking about the form of the exponential family of distributions also.

[Pack Fan]

Learning Objectives:
Formularize and solve General Linear Models (GLMs).
Range of weight: 0-5 percent

Knowledge Statements:
a) GLM assumptions compared to one-way analysis, minimum bias procedures, and classical linear analysis
b) Components of a GLM formula
c) Aliasing and near-aliasing