The last SurveyUSA poll of the Kentucky Senate race showed Republican Randal Paul running away with it from Democrat Jack Conway. However the new one tells a completely different story in its top line.
When two polls by the same firm of the same race differ by that much, there has to be a story behind the story. Fortunately SurveyUSA's detailed public reports make it easy to dig in and find what that story is.
Back at the beginning of the month, Paul led 55-40 (MoE 4.2), Paul +15. He led among Republicans 82-13, Independents 56-34, and he made a healthy 32-64 showing among Democrats.
Now though he only leads 49-47 (MoE 4), Paul +2. Republicans he still wins 83-12, but Independents have dropped to a 50-45 margin, and Democrats also fell to a 25-72 advantage for Conway.
On top of that, the party ID ratios changed over the month. Before the split was 42 R-47 D-10 I, while now it's 36 R-51 D-12 I. That shift of the Republican total is in itself outside the MoE, but it's perfectly reasonable to imagine that the actual figures are somewhere in between these two and still see neither poll as an outlier in itself.
But what matters more, the shifts in Party ID or the shifts in support within each partisan group? If we take the party ID breakdown from the beginning of the month, and plug in the support levels from now, the result becomes Paul 52-Conway 43, or Paul +9, a swing of 6 from the big lead. If we take the party ID breakdown from now, and plug in the support levels from the beginning of the month, the result becomes Paul 53-Conway 41, Paul +12, a swing of 3.
Conclusion: Randal Paul lost some of his lead in the SurveyUSA poll from the partisan ID swings, but he lost more from the 17 point swing among Independents, and the 12 point swing among Democrats, with both swings going in Jack Conway's direction. This could be real movement and I can't throw out the shift in the race as a mere partisan ID discrepancy.