The biggest survey of it's kind, Latino Decisions
http://www.latinodecisions.com/blog/2012/10/22/latino-voters-highly-engaged-in-presidential-election-support-for-obama-rises/ has been conducting a large scale tracking poll in both English and Spanish for Hispanic/Latino voters in America. In most national polls Hispanic's are a small subset of voters; at time's no more than 50-100 people polled. However these small sample sizes can impact on the reported poll results.
Latino Decisions as of today found that 71% said they would vote for Obama. 20% said they would vote for Romney. During the aftermath of first debate 'bounce' the figures were 67-23. The Pew Hispanic Center
http://www.pewhispanic.org/ in their slightly older poll had Obama at 69% to 21% for Romney. Over at Gallup their September 24th-October 14th 'super sample' that had Obama up 3 in Registered Voters overal had the Hispanic vote at 69-26 for Obama.
So in these three large comparable surveys Obama leads Romney by 51, 48 and 43 among Hispanics.
If we allocate 68% to Obama and 22% to Romney (the average of all three figures) just for fun then Obama leads by 46 amongst Hispanics. Exit polls in 2008 gave Obama a 36 lead. So Obama appears to be doing better this time around.
What happens therefore if we remove the small samples of Hispanic voters from the national polls and drop down the above figure? This doesn't for example tinker with likely voter screens or implied turnout. It just corrects for small errors in the data amongst Hispanic voters.
Let's have a look at the PPP poll from 18th-20th October which gave us a draw with both candidates at 48%. Obama lead Hispanics by 57-35, making up 9% of the Likely Voters in a sample of 1200 voters. So theoretically, Hispanics made up around 108 responses. Obama got 62 responses and Romney 38. If we adjust this to the figues I suggested above (which remember are based on larger samples and in the case of two reports; directly designed to gauge voting intention amongst a minority group) Obama 'gains' 11 responses and Romney 'looses' 14 responses. Doesn't seem like much. PPP rounded their figures so we have to work with 576-576. If we adjust these then Obama gets 587 and Romney 562.
That gives Obama 48.9% to 46.8%. That's 49-47 for Obama. Such a small change in the subsample but a bigger change in the headline figures.
What about the IDB/TIPP poll that gives Obama a 5.7% lead? Well they don't have figures for what % of the electorate Hispanics represent, but do give a voting breakdown 62-24. Obama's lead in that poll would be higher. What about the Politico poll today which has Romney ahead 49-47? Well they reported 44% vote for Romney amongst Hispanics (which would be the highest % vote a Republican president has ever got with Hispanics). If we re-weight that poll then we end up with an Obama
lead of 49-47
Now of course small errors in the other demographic subsets could cancel out these errors. This model is simply designed to highlight potential pitfalls in measuring support in one key groupIf the Hispanic vote is as decisive for Obama as Latino Decisions, Pew and even Gallup suggest, then Obama is in a stronger position that it perhaps appears. Given that white voters make up 70+% of voters, then the sample sizes are large enough to reduce potential error. What we don't have of course are wider surveys for African/American and other minority voters.
