The Turnout Model Doesn't Always Turn Out Like That

With Labor Day weekend approaching, we're starting to see pollsters switch from registered voters to likely voters, that is raw numbers of voters to weighing those numbers based upon who pollsters think will actually turn out to vote in November.

Traditionally this switch greatly favors the Republican candidate, as polling outfits eliminate more of the younger, more liberal, Democratic-leaning voters from their likely voter models as they are less likely to actually vote than older, more conservative Republican voters.  In presidential election years, this switch usually happens around Labor Day, where the campaign season's home stretch begins.

But a lot of those turnout models haven't done so well recently.  Remember four years ago when Gallup was predicting a Romney win, and Obama won by 5 points instead?  Many likely voter models are heavily weighted against black and Latino turnout, and yet black voters came out in record numbers in 2008 and 2012.

It looks like the pollsters are making the same mistake this year as well.  Let's start with the latest IBD/TIPP poll, showing Trump now tied with Clinton at 39% in a 4-way race.

In a sharp turnaround in an already volatile election season, support for Hillary Clinton tumbled as Donald Trump made gains over the past month, leaving the race a virtual tie.

The latest IBD/TIPP Poll shows that Clinton is now ahead of Trump by just one percentage point, 44% to 43% among likely voters. Last month, Clinton had a seven-point lead over Trump — 46% to 39% -- among registered voters.

Clinton and Trump are tied at 39% each in a four-way matchup that includes Libertarian candidate Gary Johnson, who gets 12% support, and Green Party candidate Jill Stein, who gets 3%.

As the election nears, IBD/TIPP is narrowing the horse-race results from registered to likely voters. This month's survey included a total of 934 respondents, 887 of whom were registered voters and 861 were deemed likely voters. The margin of error for the horse-race results is +/‐3.4 percentage points. The IBD/TIPP Poll has been cited as the most accurate in the past three presidential elections.

Now IBD/TIPP has traditionally been pretty accurate when it comes to the final poll of the presidential campaign, they called 2008 right on the nose.  But the jump from registered to likely is always jarring and is almost always the likely voter model from four years previous, without any adjustments.  We're seeing that now.

Reuters/Ipsos too has made the jump to their first real likely voter model, and it now finds Trump leading.

Republican presidential candidate Donald Trump has pulled into an effective tie with Democratic rival Hillary Clinton, erasing a substantial deficit as he consolidated support among his party’s likely voters in recent weeks, according to the latest Reuters/Ipsos national tracking poll released Friday.

The poll showed 40 percent of likely voters supporting Trump and 39 percent backing Clinton for the week of Aug. 26 to Sept. 1. Clinton's support has dropped steadily in the weekly tracking poll since Aug. 25, eliminating what had been a eight-point lead for her.

Trump's gains came as Republican support for their party’s candidate jumped by six percentage points over the past two weeks, to about 78 percent. That is still below the 85 percent support Republican nominee Mitt Romney enjoyed in the summer of 2012, but the improvement helps explain Trump’s rise in the poll.

The Reuters/Ipsos poll is conducted online in English in all 50 states. The latest poll surveyed 1,804 likely voters over the course of the week; it had a credibility interval, a measure of accuracy, of three percent.

Different polls have produced widely different results over the course of the campaign. In part that's because some, like Reuters/Ipsos, have attempted to measure the preferences of who's likely to vote, while others have surveyed the larger pool of all registered voters. And even those that survey likely voters have different ways of estimating who is likely to cast a ballot.

Again, a very, very similar outcome to IBD/TIPP.  Rasmussen too has made the jump to the likely voter model this week.

Hillary Clinton’s post-convention lead has disappeared, putting her behind Donald Trump for the first time nationally since mid-July.

The latest weekly Rasmussen Reports White House Watch national telephone and online survey shows Trump with 40% support to Clinton’s 39% among Likely U.S. Voters, after Clinton led 42% to 38% a week ago. Libertarian candidate Gary Johnson now earns seven percent (7%) of the vote, down from nine percent (9%) the previous two weeks, while Green Party candidate Jill Stein picks up three percent (3%) support. Three percent (3%) like some other candidate, and seven percent (7%) are undecided. (To see survey question wording,click here.)

Clinton's support has been trending down from a high of 44% in early August just after the Democratic National Convention. This is her lowest level of support since mid-July. Trump's support has been eroding, too, from his high of 44% at that time. A one-point lead is statistically insignificant in a survey with a +/- 3 percentage point of margin of error. It highlights, however, that this remains a very close race.

Again, a nearly identical outcome.  All three polls show Hillary Clinton at just 39% among likely voters in a 4-way race, tied with Trump, and frankly I don't believe that for a second. Once again these likely voter models almost always underestimate black and Latino voters, and those are the groups that Hillary is polling the best with.

So expect to see adjustments in these likely voter models that favor Clinton as we move ahead as these models start accounting for the changes in the electorate from four years ago (some will do this better than others, which is why Gallup was so badly off four years ago in late October, having Romney up by 7 two weeks before the election.)

Shorter article: relax.  Then go vote.