Chapter 6

Back to notes from chapter 5

  • p. 173 I really shouldn't use O (uppercase "oh") or l (lowercase "ell") for any variable names — it's too easy to confuse them with 0 and 1. (Add to the list c, q, t, C, D, F, I, T, which are all functions or reserved words in R.)
  • p. 173: probably should have mentioned that this code leads to warnings like In dbinom(x, size, prob, log) : NaNs produced, which are harmless — see note on p. 248 about ignoring such warnings (which just mean that the optimizer has tried some impossible values, such as $p<0$ or $p>1$, on its way to the solution). (thanks to Stu Field)
  • p. 177 Figure 6.2 figure caption could be better
  • p.178 Fig 6.3 — n.b. MEAN of Beta(30,10) (scaled likelihood) is 0.75 but MODE is 0.763; mean of of Beta(31,11) (flat prior) is 0.738, mode is 0.75
  • p. 179 strengthen warnings of Gelman 2006 about vague Gamma priors …
  • 181 mean depends on scale, mode doesn't (transformation-independent)
  • ? n.b. integrate probabilities, not log-probabilities
  • p. 184 dashed line is ugly for stair-step
  • p. 185 prefer different breaks in code chunk 2
  • p. 188 not entirely happy with Fig 6.6b: is there a better way?
  • p. 190 cite Royall 1993
  • p. 198 mu, sigma not the best parameter choices for eqs 6.5.7 , 6.5.8 , because of confusion between sigma and sigma^2
  • in general, might have been better off moving straight to deviance (-2 log likelihood), rather than switching among likelihood ($\cal L$), log-likelihood (L), negative log likelihood (-L), and deviance (-2L)
  • p. 199 Figure 6.13 might be better with a dashed line (rather than relying on grayscales)
  • p. 205 I'm not entirely sure that I use consistent parameterizations for the Shepherd function across all chapters …
  • p. 209 Akaike pronunciation: Kullback says "Pronunciation is important; Akaike is pronounced with an accent on the. “ka” and the “i” is pronounced like an “e” – AKAeke." (Kullback, Solomon. 2008. Information Theory and Entropy. In Model Based Inference in the Life Sciences: A Primer on Evidence, 51-82 (Accessed May 14, 2008))
  • p. 212 would really be best to deemphasize/delete AIC columns from ICtab, to help convince people that absolute AIC values don't matter.
  • p. 214: the BUGS code for running the models to compute the marginal likelihood estimates in Figure 6.5 is stashed away in the chap6.R code on the web site (chunk 99), but the BUGS files aren't there (they will be included in the new version of the emdbook package — with version or higher of the package you will be able to locate those files by typing

but in the meantime here they are: firfec0.bug, firfec1.bug, firfec2.bug

On to notes from chapter 7

Unless otherwise stated, the content of this page is licensed under Creative Commons Attribution-ShareAlike 3.0 License