Despite phylogenetic comparative methods being around for a few decades now (see Harvey and Pagel, 1991 for a good introduction), there is still a very strong tendency in comparative studies to report both the phylogenetic and non-phylogenetic comparative analyses.
My take on this phenomenon is that researchers are interpreting these two statistical procedures as equally valid alternative approaches, similar to a case where you don't know if one phylogeny is correct over another so you report both sets of results.
However, these two examples are not directly comparable, and reporting results from non-phylogenetic comparative analyses alongside results from phylogenetic comparative analyses is wrong.
This is simply because non-phylogenetic analyses violate statistical non-independence when data show strong phylogenetic non-independence, while phylogenetic comparative analyses account for this non-independence.
It's exactly the same issue with analysing temporally correlated time series data; applying traditional statistical tests would violate the assumption of non-independence so it has to be accounted for by using special statistical procedures, in this case, time series analyses.
Thus, it is nonsensical to report results from non-phylogenetic analyses, when your data shows strong non-independence, which is more often than not the case with comparative data.
Phylogenetic comparative methods are a statistical necessity, and not an equally viable alternative to traditional methods.
My take on this phenomenon is that researchers are interpreting these two statistical procedures as equally valid alternative approaches, similar to a case where you don't know if one phylogeny is correct over another so you report both sets of results.
However, these two examples are not directly comparable, and reporting results from non-phylogenetic comparative analyses alongside results from phylogenetic comparative analyses is wrong.
This is simply because non-phylogenetic analyses violate statistical non-independence when data show strong phylogenetic non-independence, while phylogenetic comparative analyses account for this non-independence.
It's exactly the same issue with analysing temporally correlated time series data; applying traditional statistical tests would violate the assumption of non-independence so it has to be accounted for by using special statistical procedures, in this case, time series analyses.
Thus, it is nonsensical to report results from non-phylogenetic analyses, when your data shows strong non-independence, which is more often than not the case with comparative data.
Phylogenetic comparative methods are a statistical necessity, and not an equally viable alternative to traditional methods.
Comments