I can believe it was probably a temporary policy created by management of the day, rather than something they did historically, but I'd be curious to know if people with direct experience of IBM in the mid-1980s can comment.
It's a term from much earlier than that.
Punch cards held a fixed number of characters, the first IBM punch cards were 80 characters for instance, and that was the maximum length of a "line". One line per punch card.
It doesn't matter if your punch card was 1 character or all 80, you still used up an entire card to store it.
So it made little sense to count characters, words, etc. You count by card, or line. Thus LOC (lines of code)
Assembly language was a "one statement per line" language.
You didn't have text editors, at best you might have written a draft first on paper.
So a line/card was a direct measure of the number of instructions in your program.
Things like comments were hand written in the margins. They didn't count as lines of their own.
Oh, and punch cards came in packs of 1000, wrapped separately within the box of cards you ordered. Thus KLOC (1000 lines of code)
I don't know if programmers were paid by the line, but there was certainly a fixed cost of cards that was directly comparable, since cards and lines were the same thing.
If management types were involved at all it would be more related to reordering your consumables, not unlike measuring pages printed per month to figure out how often you need to reorder printer paper.