The Gallup generic ballot has provided plenty of fodder with election analysts this summer. No clearer was the importance of the Gallup generic ballot to the news cycle than it was on Monday when Gallup showed a 10% Republican advantage. As noted, a 10% Republican lead on Gallup's generic ballot is unprecedented, and it will likely get worse once Gallup switches over to a likely voter model. Congressmen and political analysts alike have mentioned that Republicans could possibly do 4% better on a likely voter model. Upon further examination, however, I think it could be worse for Democrats. Why? History.
Gallup has a relatively famous likely voter model that has been in place since 1950. Therefore, we can compare past differences in the generic ballot between registered and likely voter models to give us an idea of how different they will be this year. Below, I have gathered the relevant Gallup data from every final midterm poll since 1994 and contrasted the registered and likely voter model. I have also placed the difference in net enthusiasm (percentage of voters saying they are more enthusiastic about voting and less enthusiastic) between Republicans and Democrats in the table. You should notice two important pieces of data.
First, Republicans have for the past four midterms always done better on the final Gallup likely voter poll than registered voter poll by at least 4%. This deviation is to be expected as midterm electorates tend to be older and whiter than presidential year ones.
Second, the gap between the likely and registered models benefited Republicans greatest in years where they had large leads in enthusiasm. In both 1994 and 2002 (where Republicans held at least a 8%+ edge in Gallup's final measure of enthusiasm), the Republicans margin was 7% and 11% higher respectively on the likely voter model. In 1998 and 2002 when Democrats had a lead in enthusiasm, they "only" picked up 5% and 4%. The Republicans edge on net enthusiasm was 28% a month ago, which means that voters this year are even more enthusiastic than in 1994 or 2002. So what does that mean for the registered to likely voter transition for this year?
It is important to keep in mind that the correlation between enthusiasm and differences between the registered and likely voter model has only been about 68% (not statistically significant at 95% confidence). Still, there seems to be some relationship. And keeping this connection in mind, the general Republican advantage on the likely voter model, and the large Republican lead in net enthusiasm this year, I believe that it is quite possible that at least on the final Gallup generic ballot (prior ones may differ) the Republican margin on the likely voter model could be 5-10% greater than on the registered voter model.