General election head-to-head polls appear biased against Trump

Dozens of head-to-head general election polls between Donald Trump and Hillary Clinton have been released over the past several months.  Almost all of them purport to show Clinton ahead of Trump by a significant margin.

As noted by Thomas Lifson, during the past week, Rasmussen Reports has released a couple polls showing either a tie or a slight advantage to Trump.  The Rasmussen poll sits in a sea of other recent polling data suggesting that the edge remains with Clinton.

But are these other polls biased?  A look at their methodology suggests that many of them may be.  We'll consider a few representative examples that should give Trump's team the keys to start looking for as he turns to pivot and fire the vast ammunition at his disposal in Hillary Clinton's direction during the next six months.

For context, we need to understand the general ratio of Democratic to Republican voters for the presidential ticket among the public.  The best way to assess this is just to examine POTUS election results.  In 2012, the ratio was 1.08, shifting to 1.16 in 2008, 0.95 in 2004, and 1.01 in 2000.

In other words, the ratio is approximately equal to one.  At most, in recent times, perhaps a ratio of up to 1.1 is reasonable to assume, but that would be pushing the boundaries and introducing bias by way of erroneously assuming some permanent leftward shift in the electorate.

On to the problematic polling data.

CNN/ORC released a major national poll yesterday showing Clinton out in front of Trump by 13%.  This is apparently "based on 878 registered voters and 12 individuals who plan to register to vote, for a total of 890 registered voters."  The exact political composition of all 890 respondents is difficult to determine by looking at the methodology, but if the following information from the poll is any indication, there are some issues:

Based on 267 registered voters who describe themselves as Republicans and 139 who describe themselves as Independents who lean Republican, for a total of 406 Republicans ...

Based on 287 registered voters who describe themselves as Democrats and 118 who describe themselves as Independents who lean Democratic, for a total of 405 Democrats.

Looks balanced, until you wonder who the remaining 79 (or 9%) are, and until you look at the ratio of independents to Republicans or Democrats in each division.  More than one third (34%) of the "Republicans" are independents, while just 29% of the "Democrats" are Independents.  This modest difference is important.

"Independents who lean Republican" are far more likely to lean toward Clinton over Trump – even if they still dominantly support Trump – than are the more "pure" Republicans.  On the Democratic side, "Independents who lean Democratic" are often those Reagan Democrats who hold a much higher chance of choosing Trump over Clinton than do more "pure" Democrats – even if they still dominantly support Clinton.  Thus, the polling composition is potentially biased against Trump by having a higher proportion of Independents in the Republican group than in the Democratic group.

Then there is a Suffolk University/USA TODAY poll released April 25 showing Clinton ahead of Trump by 11%.  When asked, "Do you think of yourself as a Democrat, Republican, or Independent?," 35.7% said Democrat, 30.5% said Republican, and 29.3% said independent.  That is a ratio of 1.17 Democrats:Republicans, which is almost certainly too high.  Among the respondents, the ratio of those who voted in the Democratic primaries to those who voted in the Republican primaries was a whopping 1.24.  Chalk this poll up as one biased toward the Democrats and against Trump.

Moving along to a GWU/Battleground poll from a couple weeks back having Clinton ahead of Trump by just 3%.  When respondents were asked their political preferences, just 21% indicated "strong Republican" versus 27% "strong Democrat," and an overall 42%-to-40% advantage for those leaning Democrat over Republican.  Once again, the poll is skewed toward hardcore Democrats, who are least likely to vote Trump.  The ratio of strong Democrats to strong Republicans (1.3) is probably far too high to be representative.  An unbiased sampling of voter preferences would likely have come out with an overall tie between Trump and Clinton, or perhaps a modest Trump lead.

In a NBC News/Wall Street Journal poll released during mid-April, Clinton beat Trump by 11% head to head.  And then we look at the polling details.  When survey participants were asked whom they voted for in the 2012 election, 43% said Barack Obama, and just 31% indicated Mitt Romney.  That is a polling ratio of Obama:Romney voters at 1.39 compared to the general election ratio of just 1.08.  Nothing to see here, move along...along to the other data in the methodology showing a several-point advantage to those leaning Democratic over Republican.

A McClatchy-Marist poll from late March had Clinton over Trump by 9%, and – once again – the tables are tilted against Trump.  There is a 5% bias toward those leaning Democratic in the survey composition.

The goal for every poll should be to accurately represent the public, and if the public has undergone some seismic shift into Democratic territory, then the polling data on Trump vs. Clinton may be correct.  But if history is any lesson, data biased toward Democratic voters will severely underestimate the actual mood of the voters and GOP presidential candidate performance come election time.  And we see this apparent bias in the historical data.

Take the historical Pew Research Center data on voter "leanings."  If this data actually represented the way the public leaned, and individuals voted according to their leanings, the 2000 election would have been won handily by Al Gore (rather than the popular vote tie that occurred), George W. Bush would have lost the 2004 election to John Kerry in a massive landslide (instead, a solid Bush win), and the 2008 and 2012 elections would have been truly staggering victories for Barack Obama well beyond the margins that we saw in practice.

Left-leaning media organizations keep coming up with polling data claiming to show a dominantly and continuously left-leaning public, yet right-of-center politicians win their fair share of elections.  What gives?  Undoubtedly bad (read: biased) polling data.

The likely state of the Trump vs. Clinton head-to-head is a historically small Clinton lead – on the order of a few percentage points, but not the 10+% claimed – that has evaporated in the past couple weeks to effective parity, or even a small Trump lead.

Dozens of head-to-head general election polls between Donald Trump and Hillary Clinton have been released over the past several months.  Almost all of them purport to show Clinton ahead of Trump by a significant margin.

As noted by Thomas Lifson, during the past week, Rasmussen Reports has released a couple polls showing either a tie or a slight advantage to Trump.  The Rasmussen poll sits in a sea of other recent polling data suggesting that the edge remains with Clinton.

But are these other polls biased?  A look at their methodology suggests that many of them may be.  We'll consider a few representative examples that should give Trump's team the keys to start looking for as he turns to pivot and fire the vast ammunition at his disposal in Hillary Clinton's direction during the next six months.

For context, we need to understand the general ratio of Democratic to Republican voters for the presidential ticket among the public.  The best way to assess this is just to examine POTUS election results.  In 2012, the ratio was 1.08, shifting to 1.16 in 2008, 0.95 in 2004, and 1.01 in 2000.

In other words, the ratio is approximately equal to one.  At most, in recent times, perhaps a ratio of up to 1.1 is reasonable to assume, but that would be pushing the boundaries and introducing bias by way of erroneously assuming some permanent leftward shift in the electorate.

On to the problematic polling data.

CNN/ORC released a major national poll yesterday showing Clinton out in front of Trump by 13%.  This is apparently "based on 878 registered voters and 12 individuals who plan to register to vote, for a total of 890 registered voters."  The exact political composition of all 890 respondents is difficult to determine by looking at the methodology, but if the following information from the poll is any indication, there are some issues:

Based on 267 registered voters who describe themselves as Republicans and 139 who describe themselves as Independents who lean Republican, for a total of 406 Republicans ...

Based on 287 registered voters who describe themselves as Democrats and 118 who describe themselves as Independents who lean Democratic, for a total of 405 Democrats.

Looks balanced, until you wonder who the remaining 79 (or 9%) are, and until you look at the ratio of independents to Republicans or Democrats in each division.  More than one third (34%) of the "Republicans" are independents, while just 29% of the "Democrats" are Independents.  This modest difference is important.

"Independents who lean Republican" are far more likely to lean toward Clinton over Trump – even if they still dominantly support Trump – than are the more "pure" Republicans.  On the Democratic side, "Independents who lean Democratic" are often those Reagan Democrats who hold a much higher chance of choosing Trump over Clinton than do more "pure" Democrats – even if they still dominantly support Clinton.  Thus, the polling composition is potentially biased against Trump by having a higher proportion of Independents in the Republican group than in the Democratic group.

Then there is a Suffolk University/USA TODAY poll released April 25 showing Clinton ahead of Trump by 11%.  When asked, "Do you think of yourself as a Democrat, Republican, or Independent?," 35.7% said Democrat, 30.5% said Republican, and 29.3% said independent.  That is a ratio of 1.17 Democrats:Republicans, which is almost certainly too high.  Among the respondents, the ratio of those who voted in the Democratic primaries to those who voted in the Republican primaries was a whopping 1.24.  Chalk this poll up as one biased toward the Democrats and against Trump.

Moving along to a GWU/Battleground poll from a couple weeks back having Clinton ahead of Trump by just 3%.  When respondents were asked their political preferences, just 21% indicated "strong Republican" versus 27% "strong Democrat," and an overall 42%-to-40% advantage for those leaning Democrat over Republican.  Once again, the poll is skewed toward hardcore Democrats, who are least likely to vote Trump.  The ratio of strong Democrats to strong Republicans (1.3) is probably far too high to be representative.  An unbiased sampling of voter preferences would likely have come out with an overall tie between Trump and Clinton, or perhaps a modest Trump lead.

In a NBC News/Wall Street Journal poll released during mid-April, Clinton beat Trump by 11% head to head.  And then we look at the polling details.  When survey participants were asked whom they voted for in the 2012 election, 43% said Barack Obama, and just 31% indicated Mitt Romney.  That is a polling ratio of Obama:Romney voters at 1.39 compared to the general election ratio of just 1.08.  Nothing to see here, move along...along to the other data in the methodology showing a several-point advantage to those leaning Democratic over Republican.

A McClatchy-Marist poll from late March had Clinton over Trump by 9%, and – once again – the tables are tilted against Trump.  There is a 5% bias toward those leaning Democratic in the survey composition.

The goal for every poll should be to accurately represent the public, and if the public has undergone some seismic shift into Democratic territory, then the polling data on Trump vs. Clinton may be correct.  But if history is any lesson, data biased toward Democratic voters will severely underestimate the actual mood of the voters and GOP presidential candidate performance come election time.  And we see this apparent bias in the historical data.

Take the historical Pew Research Center data on voter "leanings."  If this data actually represented the way the public leaned, and individuals voted according to their leanings, the 2000 election would have been won handily by Al Gore (rather than the popular vote tie that occurred), George W. Bush would have lost the 2004 election to John Kerry in a massive landslide (instead, a solid Bush win), and the 2008 and 2012 elections would have been truly staggering victories for Barack Obama well beyond the margins that we saw in practice.

Left-leaning media organizations keep coming up with polling data claiming to show a dominantly and continuously left-leaning public, yet right-of-center politicians win their fair share of elections.  What gives?  Undoubtedly bad (read: biased) polling data.

The likely state of the Trump vs. Clinton head-to-head is a historically small Clinton lead – on the order of a few percentage points, but not the 10+% claimed – that has evaporated in the past couple weeks to effective parity, or even a small Trump lead.