Darwinians would not be satisfied if all life on Earth derived from the same large slab of rock. (p. 269)
The final chapter of the book (apart from the concluding summary) is about common ancestry and may be the most statistically orientated of the three last chapters. This is not to say that the chapter is without faults, including, in particular, a tendency to repeat the same arguments, but this is somehow the chapter that I appreciated the most. It starts with a detailed analysis of how the hypothesis of common ancestry should be set, the main distinction being between one organism and several, while pointing out the confusing effect of lateral gene transfer. Inference about phylogenetic trees and the use of genetic sequences rather than simplistic traits gets us closer to the true issues at stake. Another interesting feature of this chapter is the reference to Darwin's reflections on the common origin of life on Earth, through many quotes.
If those prior probabilities are obscure, the same will be true of the posterior probabilities. (p. 277)
Thus, the statistical issue is one of testing for a common ancestor versus separate ancestors for a set of organisms. The nature of the information contained in the data is never described precisely enough to understand whether this fits the principle of total evidence stressed throughout the book. The chapter also shows a more lenient disposition towards Bayesian solutions (relying on priors on p. 301) but Section 4.3 ends up with a damning statement, due to the impossibility of defining an objective prior because Sober wants prior probabilities that have some authority. This is a self-defeating constraint, leading to empirically well-grounded priors (p. 276).
Those propositions suffice for similarity to be evidence for common ancestry, and they have broad applicability. (p. 283)
The part about Reichenbach's [13] sufficient condition for a common trait to induce a likelihood ratio larger than one in favour of the common ancestor hypothesis needs to be discussed, as this is the point that I find the most puzzling in the chapter. Indeed, most of Reichenbach's nine assumptions connect both models under comparison, (ie common ancestry versus separate ancestry) by
and
where X and Y are the observed character traits for two species, and Z is the common ancestor trait, while Z1 and Z2 are the separate ancestor traits. These types of assumption are statistically and philosophically meaningless, in the sense that the models under comparison should not share any parameters. If the point is about determining which model is 'true', the 'wrong' model does not exist, there is no Z or (Z1, Z2), and hence the corresponding parameters do not have any substance either. For instance, when building a Bayesian model to compare single ancestor and separate ancestors models, there is a separate prior distribution on each group of parameters. The common parameter assumption is thus not compatible with selecting one of the two models. This unrealistic framework may be the result of a reluctance to handle true (ie unknown) parameters as happens in a regular statistical analysis (see, for example, the lament that 'until values for adjustable parameters are specified, we cannot talk about the probability of the data under different hypotheses'; p. 338). What is striking is the reliance of the whole chapter on this unnatural set of hypotheses, since it keeps resurfacing throughout the chapter. Sober writes that Propositions l - 9 are not consequences of the axioms of probability, nor are they necessary conditions for common ancestry to have a higher likelihood than separate ancestry (p. 283). Nonetheless, this is creating an unnecessary bias in the perception of the problem which may induce critics of evolution to reject the whole approach.
If there was no such common ancestor, what would alignment ever mean. (p. 29l)
The theme of the missing model that I have alluded to earlier in this review is also recurrent in this chapter. There are a lot of paragraphs about the choice of the representation of the differences between two species, from trait to gene sequence, and the author acknowledges that the difficulty in this choice has to do with a requirement for a more advanced theoretical representation (model) adapted to more complex data. This sounds rather obvious when stated that way, but the book wanders around this point for several pages. (An example is the above quote, which misses the point about sequence alignment; this is a perfectly welldefined measure of distance, common ancestor or not. An interesting discussion appears on page 29l about the bias induced by alignment, the conclusion being that 'aligning sequences is not loading the dice'. I think that alignment is akin to maximum likelihood plug--in and hence favours the null hypothesis. It should therefore be accounted for in the statistical procedure.) The overall conclusion is a vague call for the principle of total evidence (which is a rephrasing of the likelihood principle), after rightly dismissing the majority rule (p. 295). As illustrated by the section on multiple characters, the discussion is confusing without a proper model. It is only on page 300 that a completely defined model for the evolution of a dichotomous trait (ie the simplest possible case) appears. (A minor point of contention here is the use of bias for the Markovian model of chromosome transformation, where reversibility would have been more appropriate.) This model is a rather crude tool, as it depends on arbitrary calibration factors such as P(Z = 0) = 0.99 (instead of the absorbing 1) and, more importantly, on an unspecified time t (as in 'what time is it on the evolutionary clock?'). The corresponding likelihood ratio is then (under one of the selection schemes)
where the dependence on those calibration factors is obvious. This illustrates the impossibility of reaching a satisfactory conclusion without first going through a statistical analysis of the problem.
Although this is not the purpose of the book, I think the debate about causality is rather superficial. For instance, while, in Section 3.6, causality and correlation are differentiated (see footnote 22 on p. 224 and p. 233), Section 3.8 embarks upon testing for a causal connection, discussing Reichenbach [13] without mentioning the Humean thesis of the logical and statistical impossibility of such a test. Most of Chapter 4 is about testing 'whether there was a common cause' (p. 247). A notion of information is mentioned on page 305 without being defined, and I do not understand whether or not this relates to Fisher's information [14] or to the Kullback divergence, as, apparently, no parameter is involved.
It is possible for data to discriminate among a set of hypotheses without saying anything about a proposition that is common to all the alternatives considered. (p. 3l5)
The debate about the phylogenetic tree reconstruction versus the test for common ancestry (Sections 4.7 and 4.8) lacks appeals for the very reason explained above. The tree structure may be incorporated within the model(s) and integrated out in a Bayesian fashion to provide the marginal likelihood of the model(s). Although this seems to be an important issue, as illustrated by the controversy with Templeton,[15, 16] the opposition between likelihood inference and 'cladistic' parsimony is not properly conducted, in that, as a naïve reader, I cannot understand Sober's presentation of the latter. This section is much more open to Bayesian processing by abstaining from the usual criticism about the lack of objectivity of the prior selection, but it entirely misses the ability of the Bayesian approach to integrate out the nuisance parameters, whether they are the tree topology (standard marginalisation) or the model index (model averaging). The debate about the limited meaning of statistical consistency makes the valid point that consistency only puts light on the case when the hypothesised model is true, but extended consistency could have been considered as well, namely that the procedure will bring the hypothesised model as close as possible to the 'true' model within the hypothesised family of models. What I gather from this final section is that cladistic parsimony tries to do without models (if not without assumptions), which seems to relate to Templeton's views about Bayesian inference.
This is the most enjoyable chapter of the book from my point of view, even though the lack of real illustrations makes it less potent than it could be. It also shows the limitation of a philosophical debate on simplistic idealisations of the real model. The book rarely acknowledges (see pp. 236 and 334) that genealogical hypotheses are composite. An incorporation of the parameter estimation in the inferential process would have improved the depth of the debate.