I once took a Labview Intermediate class (long time ago) where the instructor explained why it was good practice to use a shift register. Of course we all know that using a shift register for error in/out is desirable to carry any errors from one loop iteration to the next. But since the reference will not change, why bother.
One explanation given was in case of a zero iteration loop, as Ben has already pointed out. If the loop is skipped because of a 0 wired to N or because of an empty array wired to an indexed input, then the refernece will be NULL, and an attempt to close the reference after the loop will result in an error.
The other explanation was to just get in the habit of using shift registers. Its a good habit, and you don't have to think about it. "Hmmm... do I need a shift register on this wire or not? Will it change? If not then I don't need the shift register....." Just do it as a habit and you won't ever miss a situation where you actually needed it but didn't put it.
I always put a shift register where a wire comes into a loop, goes into a function, and a copy comes out of the function to the other side of the loop. There may be a few exceptions though, where you don't want the data propogated to the next iteration. But by doing it by habit, I never miss putting it on error wires or other data in/out situations. Bottom line is that you don't need it (as others have said), but it doesn't hurt to have it, and it builds good habits.
