However, that w would lead to fairly serious tension with PanTHEON and similar supernova samples, and non-negligible tension with BAOs as well, so lower w is not really a panacea but just shunts the tension somewhere else.

There is an interesting paper by Aylor et al, arXiv:1811.00537, which makes the case that if the H0 tension is “real”, the solution is more likely to be in high-z changes (such as extra neutrino-like species and N_eff ) which lead to a shorter sound horizon length, rather than purely low-z modifications e.g. dark energy, curvature etc.

]]>Please elaborate as if, as it were, “you were trying to explain it to a small child, or a golden retriever” (Jeremy Irons in ‘Margin Call’).

Thanks. ]]>

The details give me the heebie-jeebies: pairwise the test are based on the Bayes factor comparing the joint fit versus the completely independent fit which is a set up intrinsically geared in favour of rejecting the independent fit; ditto, for the all-but-one checks; also the Bayes factors seem like they might be computed from the BIC (certainly it is used elsewhere, so we’re talking an order 1 approximation). If you had a suspicion that there might be systematics would failing to reject the null hypothesis in this way give you comfort? OR would you try to be Bayesian and specify some models for what the systematic errors might look like in your measurements?

All that to say, I’m not dumping on the whole idea of calculating between-dataset or between-experiment Bayes factors as some measure of discrepancy (though certainly not an absolute or perfect one). But I think this is nuts as a test of systematics, which is really another way of saying model misspecification.

PS. Do you know your colleagues in climate change (sea level history reconstruction)? They’re doing cool things with integrated Gaussian processes etc.

]]>