The conclusion of this paper is that the 15% gap estimate is likely far too low (he estimates its actually 21-28%). It also states that when gaps outpage gains,

The conclusion of this paper is that the 15% gap estimate is likely far too low (he estimates its actually 21-28%). It also states that when gaps outpage gains, downward wage pressure on nonunion wages due to the "spillover effect" can be understood to be outpacing the upward pressure of "threat effects" (page 4, labeled 236, p.3), which pretty strongly suggests that if his conclusions are correct, unions may have the effect of reducing nonunion wages.