Back in 2000, Barbara Weber put all the then-current research on child car seat safety into one amazing article: "Crash Protection for Child Passengers: A Review of Best Practices."
From what I remember from that and other articles: The 12-month number is based on crash statistics.
THe progression of "child's spinal cord is likely to snap and leave child paralyzed" is not a straight line, if you graph it. The chances of serious spinal injury due to the whole "heavy head weak neck" thing don't start high and drop gradually as the child matures. When they ran the numbers on thousands of crash outcomes, they found a major line could be drawn through the numbers at the 12-month mark. Beneath it, the chances of serious injury were so high that it makes sense to legislate mandatory rearfacing. Above it, the child is still significantly safer riding backwards, but the risk/benefit of turning vs. not turning is not as wildly clearcut.
So the line is "arbitrary" to a point. The line had to be drawn somewhere. BUt, as Weber describes, it's very clear that at less than 12 months, there is a difference.
As Weber says: "Accident experience has shown that a young child’s skull can be separated from its spine by the force of a crash, the spinal cord can be severed, or the child may live but suffer paraplegia or tetraplegia due to the stretched and damaged cord. Eleven cases studied in depth were included in the two 1993 reports. All children with severe injuries were 12 months old or younger, whereas others who suffered less severe injuries,
such as C2 odontoid fractures, were over 18 months."