RealClearPolitics’ polling arm, RealClear Polling (RCP), updated their general election poll average by dropping off all of the polls from before March 5 (makes sense) but then adding one small-sample poll that showed Trump ahead by seven percentage points (Grinnell/Selzer poll of 715 likely voters). This maintained Trump’s lead in the RCP Average of about 2 percent over Biden.
Something occurred to me, though. How, exactly, has RealClear Polling been deriving their average? Have they just been adding up the percentages of each poll and dividing that by the number of polls? So that the Morning Consult’s 5,777-person poll and Reuters 3,356 person poll counts the same as Grinnell’s 715-person poll? To be fair, Grinnell is upfront about their margin of error (3.7 percentage points, compared to 1.0 for Morning Consult and 1.8 for Reuters).
The answer: Yes, RealClear Polling has, in fact, been adding up the percentages of each poll and dividing that by the number of polls. They are upfront about this, however – anyone can go to their website and check their numbers to confirm.
So what does the polling average look like if the polls are instead weighted by number of respondents per poll? The easiest way to do this is to apply the percentages each candidate got in the polls and multiply by the total number of respondents. Round your results to whole numbers. Then you can convert each poll into sample counts – that is, instead of saying Trump got 45% of the Grinnell poll, you say he got 322 out of 715 respondents.
Doing this for all the polls and calculating a weighted average brings Biden and Trump within a point of each other … at least using the nine specific polls RealClear polling used to calculate their published unweighted average.
I believe they are averaging the results of the latest polls of each pollster. Rasmussen does NOT get more weighting because they poll daily with, in the course of a week, a large sample size.
Re weighting by sample size, this would be a mistake because the relationship between sample size, and ultimate post-election accuracy, is questionable. The justification for averaging is less to increase sample size, and more to average results from different methodologies. Selzer, which you mention as having a small sample size, could be the best pollster (not that their polls should be overweighted).
If 538 still has the staff to do what they did last cycle, their averages, once they start publishing them, will be more sophisticated. But I don’t think they are historically much different from the simpler RealClearPolitics system.
If you haven’t seen it yet, it gives some ideas of similarities and differences between pollsters, and generally how polls work:
There are still voters refusing to go for Biden or Trump in a poll. Their random/libertarian choice gets too much weight. They will probably not vote at all.
Really refusing, or just expressing a third party preference or indecision before saying who they lean towards?
We would have to look at detailed procedures in each poll where undecided and/or third party respondents were asked if they leaned towards Biden or Trump should those be the only choices — and refused to say. I have been impressed by how, among those willing to substantially participate in a given survey, more than 99 percent are often willing to express a lean towards Trump or Biden. Example:
Biden Gains Ground Against Trump in Six Key States, Poll Shows
The move in the president’s direction comes after five months of mostly consistent Trump leads.
Paywalled, but beginning of article is given free (below). Poll results can be seen here under the ‘Tuesday, March 26’ banner.
President Joe Biden has gained ground against Republican Donald Trump in six of seven key swing states, and significantly so in at least two of them. The results make for the Democrat’s strongest position yet in a monthly Bloomberg News/Morning Consult poll.
The move in the president’s direction comes after five months of mostly consistent Trump leads, and follows a State of the Union address that rallied Democrats and seemed to mitigate concerns about Biden’s age.
Per Bloomberg/Morning Consult, Trump & Biden have just polled to tie in both Pennsylvania and Michigan, two of the three most populous swing states.
Re recent upticks in Biden polling, even in a relatively static race, rolling polling averages move around a bit. Trump ticked up a tiny bit in September and a bit more in October. Now Biden is ticking up. No pundit whose intensity of interest in upticks and downticks is correlated with who they want to win the election should be trusted.
I’m also skeptical of people who claim that campaigning changes the race. The state of the union didn’t cause any polling average change.
However, I will make a polling prediction. If a replacement Baltimore bridge can be substantially opened in the next seven months, that will create a Biden uptick bigger than any seen in the past year. Campaigning on infrastructure under construction does not impress, but completion would be something else. If the seven month time-frame sounds miraculous, well, uh, yes, the way I read the 2020 - 2024 polling comparisons to date, Joe does need a miraculous national-pride-inducing accomplishment. Pontoon bridge?
If a temporary bridge counts, then that particular bridge crossing in Baltimore can probably be back online in a few months. I’m not sure how comparable the Key Bridge is to the I-10 Twin Span bridge in metro New Orleans, but here’s the story of the Twin Spans reconstruction after Hurricane Katrina damage:
Before
Temporary Bridge (opened six weeks after Katrina, couldn’t find good overhead shot)
The I35W replacement bridge in Minneapolis took 14 1/2 months in 2007, using “Design Build” to expedite the bidding process. This was a full replacement, not a temporary bridge. It was also a lot smaller (504 feet-long main spans).
I bet there a lot of civil engineers across the country looking at neq bridge possibilities.
It’s like something has changed since the Democrat’s prior campaign just over a year ago. Wait, it’ll come to me…
Seriously, thanks for sharing the good news. I’ll still spend all my time worrying (and drinking) until we’re done with the election cycle, because polling is dependent upon the law of large numbers, as well as efforts by the poll makers to do good faith methodology, both of which leave far too many unknowns. But I will be happy if all the MAGA “winning” on their extreme ideology tanks their final results.
For what it’s worth, the recent Biden uptick in swing states is getting some traction in the media. This morning (at about 9:30 EDT), CNN had a guest on talking about it – I’ll look for the video and post it later. Sometimes, the simple fact that something get said on the air helps that something actualize.
Flip side: last night, CNN aired yet another person-on-the-street interview of a purported Democratic voter, once again dutifully spouting out about how Biden “has to earn our vote, he hasn’t shown us anything”. Well, OK then.
I’m also skeptical of people who claim that campaigning changes the race. The state of the union didn’t cause any polling average change.
Also for what it’s worth: This morning’s CNN guest did credit Biden’s increase in swing states in part to his State of the Union address.
538 chat on the subject. Not much data yet, but it seems Biden’s approval rating is improving, but his head-to-head matchup numbers might not be. Perhaps some people who were probably going to vote for him anyway, are now starting to actually feel okay about that, and not just be resigned to the “lesser of two evils” thing.
I live in the middle out here, but as my son is in Marland, we went across the bay. In the tunnel. It will be necessary to take the trucks to yet a third crossing, as that tunnel is really small and would not take an increase in truck traffic. So in practical terms they DO need a bridge quick.
So … the CNN guest I was posting about above, I have learned, was Congressman Ro Khanna (D-CA). Since I can’t post the video, I’ll post the next best thing:
REP. RO KHANNA (D-CA): … But let me tell you, the president as some momentum. The latest polls, John, in Bloomberg/Morning Consult shows that he has closed the gap in a lot of the battleground states. He’s up in Wisconsin, tied in Pennsylvania, tied in Michigan. I actually think it’s after his strong State of the Union speech where he laid out the clear economic differences between him and Donald Trump. The country’s responding.
CNN HOST JOHN BERMAN: Why do you think or why did you think he had work to do among younger voters?
KHANNA: Well, I think he – we needed to make the clear contrast with Donald Trump. And the president has started to do that. That he has made a contrast when it comes to being for unions, when it comes to being for housing, affordable housing, of being for increasing the minimum wage. He said, look, Donald Trump is just going to give more tax cuts to the very, very wealthy. That’s what he did. And he’s also taken a position on getting humanitarian aid into Gaza. He didn’t veto – the administration didn’t veto a critical resolution, that he called for a ceasefire.
So, you see in the State of the Union, I think, him laying out an agenda for the next term. That it seems to be resonating.
Now, there’s no doubt that a Democratic Congressman is going to shill for a Democratic incumbent president. Even as we’re more or less forced to take Khanna’s statements with a grain of salt … he did make reasonable points.
Campaigning – especially in this particular Biden-Trump general election – may not flip Trump voters to Biden or Biden voters to Trump. But surely it can:
Get at least some people off of the sidelines – you don’t need a lot; and
Coalesce the loose Democratic coalition of likely voters more tightly around Biden. Again, not needing every last person to “come home” – but as many as can be gotten. Why write off Democratic-leaning likely voters already in March?
