Seems odd to me that Marshall only gains 2% on Kobach's #s, but with Marshall involved, suddenly Bollier loses *8%* of her total. We go from 87% decided to somehow, less decided (81%) in a Marshall-Bollier race?
Hard to speculate without a full data set, but I would suspect one of two things:
1) We're seeing a post-messaging ballot on the Marshall/Bollier race
2) There's a relatively big group of voters who know they will vote for Bollier against Kobach, but are undecided in a Bollier vs. Marshall race
1 is more likely than 2.