Annie's Chpt 14 Notes

[anniecollins]

• Introduction: put “All models are wrong, some are useful” quote in here somewhere? Or are you intentionally avoiding using it?
• Line 102: I get why you included “some of whom were eugenicists”, but it feels very out of place in this sentence. My first thought would be to remove it but is there a way to contextualize why you’re mentioning this detail more clearly?
• Simple Linear Regression section: I think it would be useful to connect your explanation of t-statistics to the output of summary(sim_run_data_first_model) and summary(sim_run_data_centered_model), also because this could help emphasize the different interpretations of the intercept term between the centred and uncentered models. Love the bit on p-values. Can you put a Reid citation in here? The “Shoulders of giants” call out kind of comes out of no where (for those who are unfamiliar with her work).
• Line 566: This seems like a weird place to introduce test_that() and test_file() given that it was not introduced when testthat was introduced in chapter 11. I think it would make more sense to introduce these briefly in chapter 11 and then re-emphasize their utility with modelling here.
• After code @ line 702: Add interpretation/explanation of table 14.3
• Line 836: is cache: true a code chunk option in Quarto? If so, do you think it is worth mentioning alongside saveRDS()?
• Line 902: What does “the comparison between individuals” mean here? Pls clarify (I have changed this in the file already but I’m not sure if there’s something I’m missing)
• Inference section: It might be useful to include a brief discussion of Bayes’ theorem at the beginning of this section when talking about priors, and then revisit it later when posteriors come up. I think this might help better explain the implications of Figure 14.5 in terms of why the coefficients were unexpected
• Line 1003: can you explain slightly more explicitly what it means to auto-scale and what the function is generally doing when autoscale = TRUE is used? Just saying that it “adjusts the scale” is not super informative, especially given the different between the specified and adjusted priors in the example prior summary.
• Line 1075: define posterior distribution, how it is calculated, and how it is different from an autoscaled prior (because I think this could be confusing to first time Bayesian modelers)
• Line 1916: the references to Figure 14.14 in the text are showing up as Figure 14.18, but clicking the links still leads to the correct image (see image below)

• Figure 14.15: This doesn’t appear to match up with the procedure outlined in the first paragraph of this section. Should the y-axis not be “Number of e/E letters in line”? The simulated data also doesn’t seem realistic for either scenario (first paragraph or current axes)
• Negative Binomial section: This section feels too fast and not detailed enough in comparison with the other sections in this chapter. Some suggestions:
  • Add some theory about what the negative binomial distribution means (i.e. relate “every year each individual either dies or does not” to a bernoulli trial with probability $p$ as in the previous section, also how it is the probability distribution of # of trials before first success)
  • Introduce more clearly what we’re trying to predict with modelling at the beginning of the example section (i.e. predicting number of deaths using cause)
  • It seems weird to include Figure 14.19 and then not use year as a predictor of total deaths in the models, especially given how clear the upwards trend is for some causes, so maybe either remove this figure or justify why year isn’t included in the model
  • It might be good to include something at the end explaining when to use Poisson versus neg binomial regression, as it currently seems like Poisson is a more restricted version of neg binomial.
• Table 14.12: Is this supposed to be by cause as the caption indicates, or is the caption wrong?
• Line 2497 & 2503: is “compute” intentional here or a typo? I’ve never seen it used as a noun before
• Amazon Web Services section: Would be nice to include something that connects this section to modelling specifically, even just a few lines at the end about how you can then run your model the same as locally and explaining how others can access your model/data using the cloud after you’re set up
• Plumber and model APIs, Local model section: do we need to do anything with the “server.R” file in this section?
• Plumber and model APIs, Cloud model section: Mention how someone else may access the model via the API that you’ve set up
• Exercises, Q6: what do you mean by “imperfectly” here?
• Exercises, Q16: I don’t think this question is necessary after Q15
• Exercises, Q21: what does small vs. large world mean here?