Mark Blumenthal figured out in April 2012 that Gallup's tracking polls were overestimating Mitt Romney's support because they were underestimating how many nonwhite voters would turn out. That turned out to be correct, and Gallup's consistently less favorable results turned out to be wrong. This year, Gallup decided not to do a daily tracking poll.

On Wednesday, Blumenthal (now at SurveyMonkey) pointed out another apparent anomaly. After walking through the current poll numbers (including Libertarian Gary Johnson and the Green Party's Jill Stein), he made a simple point.

Historically, summertime polls that prompted for third party candidates grossly overstated their ultimate share of the vote. Polls that included Johnson and Stein in September 2012, for example, estimated their combined vote totals at levels varying between 4 and 7 percent; actual support for those candidates barely exceeded a single percentage point when all the votes were counted. Polls conducted in the summers of 2008, 2004 and 2000 that included third party choices exhibited very similar patterns, finding support that reached the high single digits yet melted away to just a percentage point or two on Election Day.

He also pointed to a recap of research showing how third party candidates can skew polls. If they are included as an "other" option, without being named, support is underestimated. If they are asked about by name, it's overestimated.

Anyway, I pulled numbers to see what it looked like. For consistency's sake, I grabbed two polls that were by CNN (or CNN and partners), one from mid-August to mid-September and one as close to Election Day as possible. This isn't a poll average, it's just one poll, so keep that in mind. But with that warning, the point largely holds up.

The two exceptions in the numbers I pulled came in 1992 and 1996, when Ross Perot ended up doing a bit better than his late-summer polling. In 1992 Perot also made the debate stage, which seems like it could play a role -- but he didn't make the debate four years later. That's particularly important for Gary Johnson as he tries to scratch his way to the 15-percent-support level he needs to hit to get into the debate.

This correlates to the fact that people commit to the major-party candidates later in the cycle. We plotted this in July when we saw the percent of the electorate picking Hillary Clinton or Donald Trump at that point was far higher than in previous cycles.

Note that the sample size here is very small, both in terms of the polls I pulled and in terms of the number of independent candidates that were lucky enough to be included in major polling at all. And with that small sample size, big grains of salt.

But note, too, that others have spent more than an afternoon looking at this and they expect the final margins for Johnson and Stein to be below where we see them now. Which means -- as suggested by the second graph above -- that support for the major-party candidates will be that much higher. And that two-way polling, often reviled by fans of Johnson and Stein, still holds quite a lot of value.