Is Joe Really Up by 10 Points over Trump?

It has been said that numbers never lie, but statistics do.  A political poll is a relative handful of numbers statistically analyzed and adjusted to yield, hopefully, simple results.  On their best days, polls are educated guesses.  On bad days, polls are outright lies.  The Real Clear Politics National Poll Average has both.

For a poll to be reasonably accurate, it needs a relatively large sample size, a good model reflective of the electorate, and it should include only those people likely to vote, AKA likely voters.  The voter model is much more important than either of the other two issues.  Get this wrong, and you can get a really skewed result.

One of the basic elements of a voter model is party affiliation.  This Gallup party affiliation poll shows that most voters claim to be independents, and those claiming to be either Republicans or Democrats are pretty evenly split.  If you average the numbers for 2020, you find that 28.5% claim to be Republicans and 29.7% are Democrats, a gap of just over 1 percentage point.  In June 2020, a Pew Research survey found party affiliation at 33% Democrat, 29% Republican, with 34% identifying as independents.  There are many other things involved in a voter model, such as age, sex, ethnicity, turnout, etc., but the point is, no national voter model should have a gap in party affiliation between Republicans and Democrats greater than about 3–4% and probably less.

Now let's look at the oft cited Real Clear Politics National Poll Average, shown below, which has Joe Biden at 7.1 percentage points over Donald Trump.  Is this an accurate reflection of the presidential race?  Probably not.

The RCP National Poll Average has several problems.  One issue is that the Democracy Institute/Daily Express poll, taken over August 26–28, and showing Trump with a 3-point lead over Biden, is left out.  Another problem is that the average includes several polls of registered voters, which are less accurate than polls of likely voters.  A third problem is that several polls do not mention the number of Democrats, Republicans, and independents included in their surveys.  A fourth issue is that the most recent CBS News/YouGov poll of likely voters has a party affiliation split that heavily favors Democrats.  This poll has Biden up by 10 points over Trump.

If you look at the data, you will see that the CBS News/YouGov poll has a Biden bias.  Note the table below.  The section shown is taken from the survey question that concerns Presidential Vote Intention.  If I read the unweighted numbers correctly from the red box, which represent Party ID, there are 974 Democrats, 700 independents, and 730 Republicans.  That is a total of 2,404 people, apparently with another 29 unspecified people included, which brings the total to 2,433.  If you do the math, that works out to about 40% Democrats, 30% Republicans, and 29% independents, with another 1% in there somewhere.  Remember, a reasonable real-world split between Republicans and Democrats in 2020 is about 2–3% not 10%.

Even if you use the 2016 election model taken from this Pew Research article, of 35% Democrats, 31% Republicans, and 34% independents, the split is just 4%, and there are more Republicans and independents.  This would still result in a higher vote for Donald Trump.  Obviously, the survey has been juiced by loading it up with Democrats.  It is no wonder Biden is up by ten.

Let's look at the other polls in the RCP National Poll Average.  You can review any of them simply by going to the web page above and clicking on them.

The USC Dornsife poll at Biden +10 seems remarkably detailed and complex at first glance, and it has a reputation of having predicted Trump ahead in 2016, although with a fairly wide miss in the popular vote.  You can find all sorts of statistical formulae used for analysis, raking, and sorting, and crosstabs galore.  They claim that their poll is reflective of age, sex, ethnicity, party affiliation, etc. of the nation as a whole.  However, the one thing not readily apparent is the party affiliation split in its group of likely voters.

This is not uncommon.  The Harvard/Harris poll of likely voters has no party affiliation split I can find.  Neither does the Quinnipiac poll or Grinnell/Selzer.  Same thing for the IBD/TIPP poll of registered voters and the Hill/Harris-X poll.  Are they hiding something?

Even Rasmussen at Biden +4 does not declare a party affiliation split, instead saying it uses a weighting program that reflects the overall population.  While this is similar to the methodology claimed by USC Dornsife, you will notice there is a six-point difference in their predictions, so they both cannot be right.

The Emerson poll at Biden +2 claims to be using the 2016 election model.  If so, that is not a problem.

The USA Today poll of registered voters may have party affiliation information, but I cannot get access to the data without a subscription.

Surprisingly, CNN at Biden +8 has a reasonable party affiliation split at 30% Democrats, 27% Republicans, and the rest independents.  However, this survey uses fewer than 1,000 registered voters, and the margin of error is plus or minus 4%.

The Reuters/Ipsos poll at Biden +7 has a voter model, but it is difficult to figure out.  The party affiliation percentages I calculated do not match the numbers they provide.  Remarkably, there are only 11% independents, 13% according to them, included in the poll, and 9% of the participants claim to be none of the above.  Keep in mind that the number of independent voters in the electorate is considered somewhere between 35 and 40%.  Is this poll trustworthy?

I have no idea why the DI/Daily Express poll is left out of the national average, other than that it makes Biden look bad at Trump +3.  If I do something simple, like remove the CBS News/YouGov poll and replace it with the DI/Daily Express poll, the poll average drops to Biden +6.1.

There is one other issue weighing on the polls.  What about the rumored shy Trump vote?  Is it real?

I tried to make the case for a hidden Trump vote in another post on American Thinker.  Since that piece was published, a poll done by CloudResearch appears to have discovered a truthfulness gap that may confirm the existence of shy Trump voters.  According to the poll, 11.7% of Republicans and 10.5% of independents were reluctant to share their true opinions on presidential candidates over the phone, versus 5.4% of Democrats.  Also, 10.1% of Trump voters were likely to be untruthful on phone surveys versus 5.1% of Biden voters.  If you assume a net five percent of Trump voters are lying to the polls and apply this to 45% of the electorate, then the shy Trump vote could be north of two percent.

Expect the polls to tighten in the runup to Election Day as we see more polls of likely voters and more people make up their minds.  Is Joe really up by 10 points over Trump?  You make the call.

Image: Marc Nozell via Flickr, CC BY 2.0.

It has been said that numbers never lie, but statistics do.  A political poll is a relative handful of numbers statistically analyzed and adjusted to yield, hopefully, simple results.  On their best days, polls are educated guesses.  On bad days, polls are outright lies.  The Real Clear Politics National Poll Average has both.

For a poll to be reasonably accurate, it needs a relatively large sample size, a good model reflective of the electorate, and it should include only those people likely to vote, AKA likely voters.  The voter model is much more important than either of the other two issues.  Get this wrong, and you can get a really skewed result.

One of the basic elements of a voter model is party affiliation.  This Gallup party affiliation poll shows that most voters claim to be independents, and those claiming to be either Republicans or Democrats are pretty evenly split.  If you average the numbers for 2020, you find that 28.5% claim to be Republicans and 29.7% are Democrats, a gap of just over 1 percentage point.  In June 2020, a Pew Research survey found party affiliation at 33% Democrat, 29% Republican, with 34% identifying as independents.  There are many other things involved in a voter model, such as age, sex, ethnicity, turnout, etc., but the point is, no national voter model should have a gap in party affiliation between Republicans and Democrats greater than about 3–4% and probably less.

Now let's look at the oft cited Real Clear Politics National Poll Average, shown below, which has Joe Biden at 7.1 percentage points over Donald Trump.  Is this an accurate reflection of the presidential race?  Probably not.

The RCP National Poll Average has several problems.  One issue is that the Democracy Institute/Daily Express poll, taken over August 26–28, and showing Trump with a 3-point lead over Biden, is left out.  Another problem is that the average includes several polls of registered voters, which are less accurate than polls of likely voters.  A third problem is that several polls do not mention the number of Democrats, Republicans, and independents included in their surveys.  A fourth issue is that the most recent CBS News/YouGov poll of likely voters has a party affiliation split that heavily favors Democrats.  This poll has Biden up by 10 points over Trump.

If you look at the data, you will see that the CBS News/YouGov poll has a Biden bias.  Note the table below.  The section shown is taken from the survey question that concerns Presidential Vote Intention.  If I read the unweighted numbers correctly from the red box, which represent Party ID, there are 974 Democrats, 700 independents, and 730 Republicans.  That is a total of 2,404 people, apparently with another 29 unspecified people included, which brings the total to 2,433.  If you do the math, that works out to about 40% Democrats, 30% Republicans, and 29% independents, with another 1% in there somewhere.  Remember, a reasonable real-world split between Republicans and Democrats in 2020 is about 2–3% not 10%.

Even if you use the 2016 election model taken from this Pew Research article, of 35% Democrats, 31% Republicans, and 34% independents, the split is just 4%, and there are more Republicans and independents.  This would still result in a higher vote for Donald Trump.  Obviously, the survey has been juiced by loading it up with Democrats.  It is no wonder Biden is up by ten.

Let's look at the other polls in the RCP National Poll Average.  You can review any of them simply by going to the web page above and clicking on them.

The USC Dornsife poll at Biden +10 seems remarkably detailed and complex at first glance, and it has a reputation of having predicted Trump ahead in 2016, although with a fairly wide miss in the popular vote.  You can find all sorts of statistical formulae used for analysis, raking, and sorting, and crosstabs galore.  They claim that their poll is reflective of age, sex, ethnicity, party affiliation, etc. of the nation as a whole.  However, the one thing not readily apparent is the party affiliation split in its group of likely voters.

This is not uncommon.  The Harvard/Harris poll of likely voters has no party affiliation split I can find.  Neither does the Quinnipiac poll or Grinnell/Selzer.  Same thing for the IBD/TIPP poll of registered voters and the Hill/Harris-X poll.  Are they hiding something?

Even Rasmussen at Biden +4 does not declare a party affiliation split, instead saying it uses a weighting program that reflects the overall population.  While this is similar to the methodology claimed by USC Dornsife, you will notice there is a six-point difference in their predictions, so they both cannot be right.

The Emerson poll at Biden +2 claims to be using the 2016 election model.  If so, that is not a problem.

The USA Today poll of registered voters may have party affiliation information, but I cannot get access to the data without a subscription.

Surprisingly, CNN at Biden +8 has a reasonable party affiliation split at 30% Democrats, 27% Republicans, and the rest independents.  However, this survey uses fewer than 1,000 registered voters, and the margin of error is plus or minus 4%.

The Reuters/Ipsos poll at Biden +7 has a voter model, but it is difficult to figure out.  The party affiliation percentages I calculated do not match the numbers they provide.  Remarkably, there are only 11% independents, 13% according to them, included in the poll, and 9% of the participants claim to be none of the above.  Keep in mind that the number of independent voters in the electorate is considered somewhere between 35 and 40%.  Is this poll trustworthy?

I have no idea why the DI/Daily Express poll is left out of the national average, other than that it makes Biden look bad at Trump +3.  If I do something simple, like remove the CBS News/YouGov poll and replace it with the DI/Daily Express poll, the poll average drops to Biden +6.1.

There is one other issue weighing on the polls.  What about the rumored shy Trump vote?  Is it real?

I tried to make the case for a hidden Trump vote in another post on American Thinker.  Since that piece was published, a poll done by CloudResearch appears to have discovered a truthfulness gap that may confirm the existence of shy Trump voters.  According to the poll, 11.7% of Republicans and 10.5% of independents were reluctant to share their true opinions on presidential candidates over the phone, versus 5.4% of Democrats.  Also, 10.1% of Trump voters were likely to be untruthful on phone surveys versus 5.1% of Biden voters.  If you assume a net five percent of Trump voters are lying to the polls and apply this to 45% of the electorate, then the shy Trump vote could be north of two percent.

Expect the polls to tighten in the runup to Election Day as we see more polls of likely voters and more people make up their minds.  Is Joe really up by 10 points over Trump?  You make the call.

Image: Marc Nozell via Flickr, CC BY 2.0.