You are not playing correctly according to the rules. You are supposed to go in order (followed by reverse order) of seating for setup, not dice rolls. Only the highest roller matters.
Implementing this as a house rule would have a minimal effect on the game. The important thing about the setup is that you go in the reverse order for the second settlement, which you are still doing. This house rule doesn't change who goes first. So, the largest possible difference is that the person who goes last in setup can go second in the game.
Going last for setup can be very strong. In my experience, there are typically only 5 or 6 decent settlement spots on a random board (this is just anecdotal; I don't have hard data). Thus, the person who goes last has the advantage of getting decent spots for both starting settlements. The person who goes last in setup has the additional advantage of getting to coordinate their starting settlements with no intervening action by other players. Thus, they can take two spots that work well together, either in having a 2:1 port and solid production of the resources of that port, or having a solid diversity of resources and/or numbers.
We also need to analyze what going second vs fourth in the main game means in terms of advantage. Given each person gets resources from everyone's rolls, the only thing that really matters is when people have opportunity to build. This is much harder to analyze. Where this matters the most is if two players are fighting for a road spot right out of setup. In my opinion, that the person who went last in setup also goes last in play is by design, as it gives every other player a chance to challenge one of their double settlements on road placement. If the player who went last in setup went second in the main game, it would, in my opinion, only strengthen their advantage in setup.
In conclusion, I think this house rule can give a slight advantage to the player who goes last in setup. I believe this player is already at an advantage, but if that's not true in your experience, you may want to use this rule. I don't see any other interesting consequences of this change.