One thing that can NOT be concluded from this paper, is that statistical/positive economics is using tools that have sketchy underpinnings… because it does not matter one iota if probability is taxonomically ‘objective’ or not – it only matters whether or not the inferences drawn from sample data are (a) representative [the core 'thrust' of statistics and probability]; and (b) useful.

In a sense, the subjective nature of probability is also kind of obvious: we use probability (and its cousin, statistics) when the costs of obtaining a complete characterisation of a system are prohibitive.

Example: airframe MTTF and MTBF (mean time to, and between, failure[s] respectively). Rather than run every airplane to failure, we accept less-than-complete information (i.e., more uncertainty) and instead run a SAMPLE of KEY COMPONENTS to failure, and then extrapolate.

In effect, we DELIBERATELY CONSTRAIN our information set – sacrificing certainty for empirically-robust sample-based data that (1) enables sensible inference; (2) is replicable; and (3) is operationally relevant.

And so it is with economic data: we cannot completely-characterise every process that makes up an economy, but we can make inferences from a subset of observations that are generated by economic interaction. To the extent that the analysis is performed without a ‘strong prior’ (i.e., a bias – i.e., corruption-of-process), the results will be robust. And if there IS bias in the selection of data or method (say, in Friendman and Schwarz’ ‘masterpiece’ of dodgy stats), the results can be scrutinised by well-trained individuals and the fabrication or operational inadequacy revealed.

TL;DR: it doesn’t matter if probability is subjective or not. It matters that the tools that probability gives us are useful, neutral (under certain conditions) and amenable to scrutiny.