Dear All,
The REVISED HW6 (v2) has now been posted. PLEASE DO NOT USE any version that you may have downloaded prior to 2:25 pm today. We took down the older assignment at 2:25pm, and have changed several questions. Please be sure to use the version that says “v2” in it. Based on recent f/b and realizing unintended hiccups occurred from the last 2 HW assignments, we wished to review/update this current HW for further clarity. Thank you for your patience.
As mentioned in class, this module has 7 ppt parts -- 5 recorded videos; one REQUIRED but NOT RECORDED slide deck, Part 3, which is central to the material and will be needed for the HW; Part 3 is mostly definitions and formulas; and one Appendix of extra/optional material.
This is a new set of videos, so I can’t say for sure which is the best order to do things in, based on any experience. However, in class, I suggested that you may want to view the slides/videos first (with the additional caveat about Parts 2a and 2b below), then go to the assigned readings, starting with Vanderweele, then RGL, then the Turner article. As further background, Part 1 is general/introductions, explaining terms and draws both from RGL and Vanderweele; Part 2 is mainly stats and doesn’t draw on any reading in particular, but rather several examples provided by biostats faculty and Dr. Glymour; Parts 3-4 are drawn largely from Vanderweele, sections 1.1-1.7; and Part 5 is RGL (p. 71-76; or Hernan 5.1-5.3).
For Parts 2a and 2b – these are all about interpreting coefficients from linear regression equations that have an interaction term (cross-product term) in them. There are 3 examples – first when there are 2 binary exposures; next, 1 binary/1 con’t exposure; and lastly, 2 con’t exposures. The module is about defining each beta coefficient in each example, to get a sense of what it really means when we include an interaction term in a regression model (when testing for multiplicative interaction). This is done commonly and is easy to do with statistical software programs, but sometimes we lose a feel for how this affects the other terms in the model. For this section, there’s a lot of math and algebra. People learn things like this differently. If you’ve had modeling of statistical interaction in biostats recently, you may want to look at the slides/examples (without the video); work through some math on your own with pencil/paper, write down your answers for the interpretations of all the beta’s (in full sentences), then compare them to the answers on the slides… or you may benefit from viewing the video, where I try to use pen/ink/color to motivate what math/algebra need to be done and a few “tricks” about how to re-scale variables. The goal is to see the patterns and recognize how inserting a cross-product term into a model changes the interpretation of the coefficients for the other variables. For some, the math may be helpful, for others the words may be more helpful. I would be interested in folks' f/b. thanks.
Good luck!
JMC
PS - For those not in class today, we apologized for the hiccup on the QBA HW regarding q4a, where we were not sure of how Dr. Mayeda intended folks to solve, given that she omitted one of the slides from last year that went with that problem. We tried to contact her, but didn't hear back, and so I eventually posted the slide from last year. I heard that some of you may have been spending a lot of time on what was not supposed to be a tricky problem and we are sorry for that. Also, Francois went over the p-value problem from last week, DAG HW. We recognize that that question could have been clearer with regards to the overall big-picture objective, or graded differently; and as mentioned previously, that question is now counted only as bonus. I wish you smoother sailing on this coming problem set and we took extra time this evening to review it again, in light of these events.
PPS - Also as mentioned in class - Head’s up for the PhD students, I think that over the course of your program, you will be recommended to read all of Vanderweele (parts 1 and 2 and the appendix; and Chapters 4 and 5 in Hernan and Robins).