NEW YORK TIMES
By Nate Silver
October 13, 2012
http://fivethirtyeight.blogs.nytimes.com/2012/10/13/oct-13-arizona-and-the-spanish-speaking-vote/
President Obama halted an 8-day winning streak for Mitt Romney in the FiveThirtyEight forecast on Saturday, with his chances of winning the Electoral College ticking up to 62.9 percent from 61.1 percent on Friday.
One should be careful about making too much of this: Mr. Romney has made very strong gains in the forecast over the past week-and-a-half to draw the race nearly even. It is unlikely that there will be a major change in the landscape until Tuesday’s debate in New York.
However, the polls have shifted so much toward Mr. Romney in recent days that a mediocre day of polling for Mr. Obama may look good to the model in comparison to a very poor one, as he had on Friday.
Three of the four national tracking polls on Saturday, from Gallup, Rasmussen Reports, and Investors’ Business Daily, showed no change in their results among likely voters, although Mr. Obama gained one point in the version of the Gallup poll among registered voters. Mr. Obama did improve his margin by roughly one point in the online poll published by the RAND Corporation.
In general, the tracking polls have been slightly more favorable to Mr. Obama than other national surveys since the debate. The RAND Corporation poll has him leading by roughly 3 points, while the Investors’ Business Daily has him up by less than a full percentage point.
In the Rasmussen Reports poll and the Gallup poll of registered voters, Mr. Obama’s numbers are not far from their long-term averages, although that partly reflects methodological changes in each survey.
A note on this topic: My preference would be that polling firms did not change their methods so close to the election. Even if the changes are helpful to their accuracy in the long run, they make it harder to measure the trend in the race against a constant baseline. There are some changes that I’d like to make to the FiveThirtyEight forecast — for instance, in making it more aggressive about responding to news events like debates — and I’ll tell you when I think it may be doing something wrong. But I’m going to save those for 2016. I think the value of an objective method is undermined if one changes the rules in midstream.
Still, things were at least not getting worse for Mr. Obama in the tracking polls. We’ll know soon enough whether this reflects a dead-cat bounce or the beginning of a rebound.
A Surprising Poll in Arizona
Apart from the tracking polls, the only survey published as of our deadline on Saturday was one in Arizona. It had a surprising result, showing Mr. Obama 2 points ahead among likely voters.
The survey has a couple of things in its favor. The poll comes from a firm, the Behavior Research Center, that has had good results in the past. And almost all of its interviews postdated the Denver debate.
But I would not be too worried about the topline numbers if I were Mr. Romney’s campaign — or too enthralled with them if I were Mr. Obama’s. The survey contacted relatively few respondents — about 500 voters — and even a good polling firm can and will produce an outlying result or two with a sample size like that.
It is plausible that Mr. Obama could win Arizona if he is running strongly nationwide — but it is much less likely that he will do so in the current national environment, where the race is almost tied.
Consider that, in 2008, Mr. Obama lost Arizona by 8 points despite winning nationally by 7 points.
Part of that is because Arizona was John McCain’s home state. Historically, the home-state advantage for a candidate is on the order of 7 points. In other words, had Mr. McCain been from another state, Arizona might have been a toss-up in 2008.
Even so, that was an in election in which Mr. Obama had a clear victory nationwide. If Mr. Obama recovers from his debate swoon and wins another clear victory this November, Arizona could fall into his column. But it is unlikely to prove decisive in a tight national race.
The FiveThirtyEight forecast model considers what we call “state fundamentals” along with the polls — with the most important fundamental factor simply being how the state voted in recent elections. (The model incorporates an adjustment for home-state effects in Arizona and other states that were home to one of the candidates.)
When there is abundant polling in a state, the polls predominate in the forecast. But in a sparsely-polled state, like Arizona, the fundamentals have more influence.
Our research has shown that this will usually produce a more accurate forecast, reducing the risk of banking too much on potential outlier polls. In 2004, a pair of late polls showed George W. Bush just ahead of John Kerry in Hawaii — an implausible-seeming result given that the national race was fairly close and that Hawaii is ordinarily very Democratic-leaning.
Indeed, Mr. Kerry eventually won Hawaii by 9 points, despite a late visit there by then-Vice President Dick Cheney.
Thus, the forecast still makes Mr. Romney the clear favorite in Arizona. We will need to see more polls of Arizona, or a rebound for Mr. Obama nationally, for that story to change much. (Arizona could certainly use more high-quality polling. Although there is only an outside chance that it will influence the presidential race, the senate race there is highly competitive and a loss in the state would make it very hard for Republicans to win control of the Senate.)
Are Polls Underestimating Obama’s Hispanic Vote?
However, there is a potentially troubling aspect of the poll for Mr. Romney. The Behavior Research Center conducted interviews in Spanish along with English. That may account for some of why it had better results for Mr. Obama than other surveys of the state.
In the past couple of elections, polls have underestimated Democrats’ standing in states with heavy Hispanic populations. (The two senate races that the FiveThirtyEight forecast called incorrectly in 2010 — Nevada and Colorado — are both states with a healthy number of Hispanic voters.)
This may be because many polling firms that conduct interviews only in English miss some Hispanic voters who are more comfortable speaking Spanish. According to Matt Barreto of the polling firm Latino Decisions, which conducts bilingual interviews, primarily Spanish-speaking Hispanic voters are more likely to vote Democratic than those who have more English fluency.
Polling firms such as Latino Decisions that have conducted interviews in Spanish have shown Mr. Obama with a larger advantage among Hispanic voters than those which interview in English only. The most recent Latino Decisions poll, for example, had Mr. Obama ahead 72-20 among Hispanic voters. This poll is not an outlier; other polling firms that have conducted Spanish-language interviews have found similar results.
The countervailing factor is that Hispanics who speak mostly Spanish are unlikely to be registered or likely to vote. Still, this factor could make some difference in states with heavy Hispanic populations, like Arizona.
How much of an impact might this make? On Saturday, I ran an alternate version of the FiveThirtyEight simulation in which I assumed that Mr. Obama would in fact win Hispanic voters by 50 percentage points, his edge in the Latino Decisions poll, as opposed to the roughly 35-point margin he’s had on average in polls that were conducted in English only.
I scaled this adjustment based on the share of Hispanic voters in 2008 exit polls. So, for example, Mr. Obama gained a net of 3 points in Texas because of the adjustment, but almost nothing in Kentucky.
Even with this adjustment, Mr. Obama was far from being favored in Arizona. Instead, the model gave him just an 8 percent chance of winning there, although this was better than the 4 percent chance from the standard FiveThirtyEight model.
However, the adjustment increased Mr. Obama’s win probability in Colorado to 57 percent from 44 percent, in Florida to 53 percent from 35 percent, and in Nevada to 77 percent from 62 percent. It even helped him slightly in Virginia, where about 5 percent of voters identified as Hispanic in 2008 exit polls.
Overall, Mr. Obama’s chances of winning the Electoral College rose to 69 percent from 63 percent.
More of the Hispanic population is concentrated in noncompetitive states like Texas, California and New York. Thus, the increase in Mr. Obama’s Electoral College win probability was slightly smaller than increase in his probability of winning the popular vote, which rose to 71 percent from 61 percent.
This adjustment is speculative — as I mentioned, it isn’t clear how many primarily Spanish-speaking Hispanics will actually turn out to vote. And a few surveys do conduct interviews in Spanish.
I recommend the regular FiveThirtyEight model as providing the best overall forecast of the state of the race. There might be ways in which the polls could be underestimating Mr. Obama’s vote — but there are potentially other ways in which the surveys might be biased against Mr. Romney.
Still it’s the possibility that the polls are underestimating Mr. Obama’s performance among Hispanics that should be the concern for Mr. Romney in the Arizona poll — not the unlikely possibility that Arizona has suddenly become a swing state.
No comments:
Post a Comment