We are now roughly six and a half months from election day or election month, depending on your state’s election laws.   

Today, we’ll focus on Joe Biden’s recent polling surge, as shown in the Real Clear Politics Poll Average of Trump vs Biden.  You will note Trump is up over Biden by a whopping 0.4 percentage points.  If we remove the possible outlier Rasmussen survey from the average, showing Trump at +8, Joe Biden would be ahead by 0.4.  Republicans might be dismayed by this, and Democrats encouraged.

In January of this year, Trump was ahead of Biden by +4.  Since then, the poll average has tightened.  Is this real, or is there something else going on?

Before getting into some poll internals, let’s see how a few polls have changed since the beginning of 2024.  A good one to start with is Emerson.

1/26  Trump +1

2/13  Trump +1

3/05  Biden +2

4/02  Trump +1

4/16  Trump +3

Next, let’s look at the Economist/YouGov Poll.

1/07  Tie

1/14  Biden +1

1/21  Trump +1

1/28  Biden +1

2/04  Trump +1

2/11  Tie

2/18  Trump +1

2/25  Tie

3/03  Trump +2

3/10  Trump +2

Finally, let’s look at the Quinnipiac Poll.

1/25  Biden +6

2/15  Biden +4

3/21  Biden +3

Looking at the poll trend rather than the chatter, you might conclude that Trump was slowly gaining strength.  But the RCP polling average contains different surveys in different combinations, some of which shouldn’t be included.  For example, the Reuters poll is all over the map on results and highly inconsistent regarding the number of voters included from poll to poll.  Here are the numbers.

1/03  Tie            4677 registered voters

1/22  Trump +5 1028 registered voters

3/07  Biden +2  3356 registered voters

4/03  Biden +4    833 registered voters

This doesn’t make their top-line numbers wrong, just suspect.

Polling organizations are supposed to be ethical, unbiased, and consistent in their methodology.  The truth is the way a poll is conducted can significantly affect the results.

There are two general processes involved.  One is data collection, which depends on where and when you poll, who you poll, and how the questions are asked.  Another depends on how the data is weighted.

Polling organizations often try to fit their data into a voter model.  A model could be tied to U.S. Census data, the results of the previous Presidential election, or any combination.  Voter responses are usually weighted according to party affiliation, gender, age, race, income, etc.  It would be great if every polling organization used the same voter model.  But they all want you to think their polling kung-fu is the best, so I doubt if any are the same.

The Gallup Party-affiliation Poll is a good source to see how voters are aligned.  If you average the twelve monthly polls between April 2023 and March 2024, the split is 27.7% Republican, 27.3% Democrat, and 42.8% Independent.  When party leaners are included, the average is a split of 45% Republican and 43.5% Democrat.  This means your national voter model and the data you collect should be close to an even split on party affiliation.

The only problem is Democrats have a registration advantage, thanks to New York and California.  So even though registration is roughly 50-50 in the other 48 states, national polls often give Democrats an extra 3- 4% advantage on party affiliation to account for the two big blue states.

If you look at voting in terms of gender, the story may surprise you.  Women have been responsible for roughly 10 million more votes in the past five presidential elections than men.  So, based on the 155 million votes in the 2020 election, the number of women in a poll should be roughly 6% higher than men. 

So, looking at the specifics, let’s start with the New York Times Poll, which has Trump up by 1 point.  Since the NYT includes its voter model in the cross-tabs, we can see how the data is weighted.  Party affiliation in their likely voter model is 33% D, 28% R, and 31% I/Other.  Men vs women in the model is 47% to 53%, a six-point split.  The 5% party affiliation split in favor of Democrats is probably too high, but that is how the Times rolls.  Think Trump +2 here.

 The Emerson Poll has Trump up by 3 points.  Party affiliation was 37% D, 35% R, and 28% I/O.  Gender was roughly 47% men vs 52% women.  Here, the data collection seems reasonable.  Take the points.

The Yahoo/YouGov Poll has the race as a tie.  According to YouGov, party affiliation is weighted to November 1, 2022, which was 33% D, 27% R, a six-point split.  Party leaners also favored Biden by about 2%.  Gender was weighted to the 2020 election turnout, roughly 47% to 53% men to women.  If their party affiliation matched the 2024 reality, this poll should be Trump +1 or +2.

The Data for Progress Poll has Biden at +1.  As the title suggests, this is a poll that is Democrat-friendly.  Party affiliation in their likely voter model is 38% D, 37% R, and 25% I/O.  Gender was roughly 46.5% men vs 53.5% women, a 7-point split.  The weighted number of women is a bit high, but party affiliation seems reasonable.  Interestingly, this is the first time this poll has been included in the average in 2024.  I’ll accept this one as is.

Daily Kos/Civiqs has Trump at +1.  Party affiliation is 36% D, 31% R, and 33% I/O.  Gender was 47% men vs 53% women.  Again, party affiliation is probably too high for 2024.  It could be Trump +2.

The NPR/Marist Poll has Biden at +2.  Party affiliation is 38% D, 31% R, and 31% I/O.  Gender was roughly 49% men vs 51% women, which is a split of 2%.  With all due respect to NPR, a 7-point split on party affiliation makes this look more like a blue state poll instead of a national one.  Maybe the narrow difference between men and women counters part of the affiliation bias.  This poll should be Biden +1 at best, maybe Even.

It would be nice to see the cross-tabs of the other polls in the RCP average.  But some are not available, and others are hidden behind paywalls.

One observation concerns the various Harris polls.  From January through mid-March 2024, these polls had Trump up from +4 to +8.  The latest one from 3/25, which is not currently in the RCP average, shows the race tied.  Is this a trend or just an outlier in the data?  We will have to wait for the next one or two Harris polls to get confirmation.

In conclusion, the polls may be tightening up a bit.  But it is not as bad as it looks.  Any poll using the 2020 Presidential election as a voter model will likely be surprised after Election Day.  2024 is not going to be 2020.

Image: Negative Space