However, I think you're incorrect in contending first vote preference should always be used for comparisons.
In fact, final ballot in a caucus is a perfectly good measure to use against first ballot in primaries.
The luxury of not spoiling your ballot within a RBV system alters voting strategies. People think, work out their own odds and adapt to this by voting for "riskier" contenders while keeping their "safer" as immediate backup. Single Transferable Vote is the universal voting method in Ireland, so I have some experience of it.
A pre-caucus ballot like this: 1. Gabbard / 2. Warren / 3. Sanders
is not going anywhere but to Bernie in a runoff sytem without Gabbard & Warren.
Your claim it should be entirely discounted is odd and appears borne from a lack of forethought about the psychology of ballot spoiling. These voters were motivated enough to waste their afternoon in casting a marginally insignificant ballot. It's time to give voters some credit - Democratic ones at least.
I understand the ideas behind ballot spoiling/ranked choice voting/etc. and I definitely think that there is merit to using final-alignment voting as a cross state indicator. If we're trying to see how voters in Iowa might compare to, say, voters in New Hampshire, final-alignment might be the better way to look at the two sets of numbers. However, there is a big difference between the process of casting votes in Iowa vs. New Hampshire. In New Hampshire, you go in and cast your secret ballot, perhaps with some strategic understanding of who may and may not be viable. In Iowa, you go to your precinct caucus and if your preferred candidate is viable in just that one location you stay with your person, even if that's their only precinct in the whole state. Again, final-alignment might still be the better comparison, but first preference is a much more similar to the process of going and voting. As with everything else, it's a messy process when it comes to caucusing.
However, the point I was making in my original comment had to deal with what is important for entrance/exit polls. In trying to get a sense of what groups supported who, the conductors of the entrance polls use first-alignment numbers to try to match their sample with the actual electorate. It still muddles the results even if you were using final-alignment numbers for the statewide outcome and first-alignment numbers for the performance of a candidate by demographic group.
Also, even though it wasn't my original point, I think that this fact, that first-alignment votes are what are used for the entrance polls, makes it so first-alignment is the better cross state measure. If I want to look at how Sanders is doing across the country among non-white voters, I can't do that if one of my measures is showing something else. This might just be a specific thing for cross state demographic use, but I think first-alignment vote if a perfectly reasonable measure, even if final-alignment is too.
