Well, I put out a call and people responded. Unfortunately, it seems that nobody really knows what happened, because that sort of thing happened all the time.
From SABR member Pete Palmer:
Although RBI were first officially recorded in 1920, there were no rules about them until 1930. The bases loaded walk was specifically noted as an RBI then. In the 1920s, some scorers gave RBI for a bases loaded walk and some didn’t. Since there were no rules, you can’t really fault the scorers for what they did. After all, in the 1880s a walk was counted as an error for the pitcher. The modern encyclopedias subtracted these as well as the assists on a strikeout for pitchers.
It would make sense to go back and dig these bases loaded walks up and credit an RBI for them. Of course, we don’t have a complete set of play-by-plays, but we do have a lot. The whole compilation of RBI especially in the 20s, had hundreds if not thousands of errors besides this type, so it is better to assume the posted number is an approximation.
Elias recently changed some stats, although we don’t know which ones except for the league leaders shown in their record book. In 1921, George Kelly had 122 RBI. Elias changed him to 127 which led the league, but Retrosheet found 131.
It’s as good an explanation as I’ve yet heard. In the course of reporting his book 1921, Steve Steinberg reports that he uncovered a number of statistical discrepancies, essentially confirming Palmer’s account.
It might not be entirely satisfying, but it is an answer.