bertvanderveen / st2304 Goto Github PK

Might benefit of a more intuitive explanation why we need a constrained and unconstrained scale: why do we need to model things on the link scale in an unconstrained fashion, and what are the benefits from having it constrained on the response scale. E.g., we want to model the probability of land and this ranges between 0 and 1: connect to that concept.

I changed some exercises + they already needed updating. Also check for correct Likelihood notation if applicable.

For the lm function the exercise phrases things as if there are x and y arguments, rather than making clear that the function requires a formula input that represents y and x in the regression formula. Many students get confused here.

• Students find the explanation of "pnorm" and "qnorm" confusing, as well as the combination of distributions for the response and for the estimators.
• at least pnorm-qnorm can be explained easier; pnorm gets p(X<x) while providing x qnorm does the reverse; it gives X while providing p(X<x). Visualising this it's "just" swapping the x-y axes of the plot.

Students found it a bit long. Might be good to shorten.

The course could use a video on interactions, explaining in detail how to calculate different outcomes, because the students seem to struggle with that

To prevent students from copying answers from the previous year.