Suitably Flip

In New Hampshire, Life Imitates Must See TV

"There is an 80% chance in the next election that I will tell all my friends that I'm voting for Barack Obama but I will secretly vote for John McCain."

Liz Lemon (Tina Fey), "30 Rock", April 5, 2007

Well, last night's primary results were certainly a surprise.

Yesterday morning, it looked like McCain and Romney were neck and neck, while Obama was set to mercilessly trounce Clinton.  Early in the evening, it looked like John McCain would win by double-digits, while the Clinton-Obama race suddenly became a squeaker.  Once all the dust settled, Clinton had unexpectedly nabbed the win (39-37) and, despite a gradually shrinking spread, McCain held onto his victory (37-32).

The magnitude of the surprise on the Democrat side had some Obama supporters quickly screaming "Diebold!" and darkly fantasizing about the Clinton machine's dastardly deeds.  While I'm loathe to argue with the characterization of the Clintons as dastardly, simpler explanations seem more plausible.  There are lots and lots of theories, but I think the one that makes the most sense is the one variously described as the Wilder effect, the Bradley effect, or hereafter, I suspect, the Obama effect (or possibly the Liz Lemon effect).

Simply put, the idea is that people are damn dirty liars.  Specifically, people will lie to pollsters (and perhaps to family, friends, etc.) about their intent to vote for a minority, whether for the simple sake of doing so, for bandwagon fashionability, or to ward off the glancing-askance of people who might think them to be no good racsists who simply aren't ready for a black President.  If this is true, we can expect a significant polling error in favor of a minority candidate and significant ballot box underperformance versus those polls.  The effect (if it exists) may also apply to female candidates, but in all likelihood, race would trump gender, especially when the female candidate has long enough been ensconced in the most powerful political circles as to need no psychic validation from a dishonest poll respondent.

I heard some cable news commentator offer this possibility up a couple months back (I want to say it was Bill Sammon, but I don't remember; it could've been Charles Krauthammer) and I watched the Iowa returns half-expecting Clinton to blow away expectations.  Of course, in Iowa, we saw the opposite effect, with Obama exceeding expectations by 6 points.  It's important to remember, though, that unlike New Hampshire, which was a typical electoral scene (with typical voting booths and typical opaque curtains), Iowa was an open outcry political marketplace with a very public, visual, physical representation of one's preference.  It therefore probably shouldn't surprise us to see an exacerbation of the false pro-Obama shift when voters are made to stand in gymnasiums, gather in their candidates' corners, then hoot and holler at each other to come over to their respective sides.

One reason I put more stock in this effect than some of the other theories is that it explains the result vs. polling variance on both sides of the aisle.  The conventional wisdom going into yesterday's primary was that the perceived Obama surge would swing a large fraction of New Hampshire's huge independent voting bloc away from McCain and toward Obama.  This, of course, would have been good for Obama and Romney, and bad for McCain and Clinton.  The fact that both variances went the other way (Clinton overcame a huge polling disadvantage, while McCain expanded his margin of victory) is explained by this single theory.  If the perceived Obama surge was an illusion of (or at least grossly overstated by) polling, those independents were simply never really there for Obama; Clinton never had to worry; and Romney never had a shot of dethroning New Hampshire's favorite Republican.

Of course, the variance was significantly higher on the Democrat side, so yesterday's surprises aren't fully explained by independents faking toward Obama, then defecting back to McCain.  But we'd expect to see the results lopsided in this direction assuming at least some of the independents "for Obama" (and some of the registered Democrats "supporting Obama") actually voted for Clinton when they were done proving their lack of racism for pollsters.  Clinton's out-out-performance, relative to McCain's, could also be due in part to some fraction of registered Democrat voters making false Obama pledges.

While I think this is the most plausible theory and does the best job explaining what we saw last night, there are so many countervailing forces that it's hard (and usually foolish) to try to pin down precisely what accounted for what.  The one theory I simply don't buy though (other than the "Diebold!" theory) is the fact that New Hampshire voters are just so fiercely independent that they're unpollable.  Just as Hannibal Lecter is so brilliant as to be impervious to psychoanalysis, so to are Granite Staters opaque to researchers, we're led to believe.

Pundits slinging that idea have tossed around one statistic incessantly in the hours since the polls closed - namely that nearly half (!) of New Hampshire voters wait until just before the primary before deciding whom to vote for and that's what makes them so poll-immune.  This is a cynical read on my part, but the half million people who vote in the New Hampshire primaries have a tremendous incentive to convince the other 302.7 million people (and the candidates who hope to serve them) that their votes are swayable right up until the end.  It keeps the candidates around longer, it dumps rocket fuel on tiny hamlet economies, and it preserves the perversely disproportionate influence that the state has on Presidential elections.  If half of the primary-voting population truly couldn't make up their mind about a contest that was launched nearly a year ago and covered in depth every day since, we really need to think hard about how best to normalize their special electoral status.

The good news is that the accuracy of the Lemon effect theory should be testable.  In future Democratic primaries (but not open outcry cauci), we should expect to see, not Clinton necessarily winning, but Clinton outperforming the freshest polling data, relative to Obama.  No good prospective remedy in polling methodology jumps to mind when the problem rests with the filthy liars on the other end of the call, so the error will likely persist (and newstalkers will find more politically palatable explanations for the ongoing polling error).  And if New Hampshire (not to mention the reverse effect seen in Iowa) was any indication, the size of this error is massive - possibly double-digit.

My guess would be that states with smaller minority populations might see a larger Lemon effect (yes, I'm trying to make that term stick), which may mean that Iowa (91% white) and New Hampshire (97% white) showed us somewhat more dramatic variances than we'll see in the upcoming Democratic races in South Carolina (69% white), where Obama's up by 13% or Florida (81% white), where Clinton is up by 20%.

But next stop Nevada is a caucus state, so it will be interesting to see how results diverge from polling in that matchup.  Clinton is currently ahead by 20 points in that contest as well, but we may see another Iowa-style "reverse Lemon" that at least makes it close for Obama.

Comments

In New Hampshire, Life Imitates Must See TV

"There is an 80% chance in the next election that I will tell all my friends that I'm voting for Barack Obama but I will secretly vote for John McCain."

Liz Lemon (Tina Fey), "30 Rock", April 5, 2007

Well, last night's primary results were certainly a surprise.

Yesterday morning, it looked like McCain and Romney were neck and neck, while Obama was set to mercilessly trounce Clinton.  Early in the evening, it looked like John McCain would win by double-digits, while the Clinton-Obama race suddenly became a squeaker.  Once all the dust settled, Clinton had unexpectedly nabbed the win (39-37) and, despite a gradually shrinking spread, McCain held onto his victory (37-32).

The magnitude of the surprise on the Democrat side had some Obama supporters quickly screaming "Diebold!" and darkly fantasizing about the Clinton machine's dastardly deeds.  While I'm loathe to argue with the characterization of the Clintons as dastardly, simpler explanations seem more plausible.  There are lots and lots of theories, but I think the one that makes the most sense is the one variously described as the Wilder effect, the Bradley effect, or hereafter, I suspect, the Obama effect (or possibly the Liz Lemon effect).

Simply put, the idea is that people are damn dirty liars.  Specifically, people will lie to pollsters (and perhaps to family, friends, etc.) about their intent to vote for a minority, whether for the simple sake of doing so, for bandwagon fashionability, or to ward off the glancing-askance of people who might think them to be no good racsists who simply aren't ready for a black President.  If this is true, we can expect a significant polling error in favor of a minority candidate and significant ballot box underperformance versus those polls.  The effect (if it exists) may also apply to female candidates, but in all likelihood, race would trump gender, especially when the female candidate has long enough been ensconced in the most powerful political circles as to need no psychic validation from a dishonest poll respondent.

I heard some cable news commentator offer this possibility up a couple months back (I want to say it was Bill Sammon, but I don't remember; it could've been Charles Krauthammer) and I watched the Iowa returns half-expecting Clinton to blow away expectations.  Of course, in Iowa, we saw the opposite effect, with Obama exceeding expectations by 6 points.  It's important to remember, though, that unlike New Hampshire, which was a typical electoral scene (with typical voting booths and typical opaque curtains), Iowa was an open outcry political marketplace with a very public, visual, physical representation of one's preference.  It therefore probably shouldn't surprise us to see an exacerbation of the false pro-Obama shift when voters are made to stand in gymnasiums, gather in their candidates' corners, then hoot and holler at each other to come over to their respective sides.

One reason I put more stock in this effect than some of the other theories is that it explains the result vs. polling variance on both sides of the aisle.  The conventional wisdom going into yesterday's primary was that the perceived Obama surge would swing a large fraction of New Hampshire's huge independent voting bloc away from McCain and toward Obama.  This, of course, would have been good for Obama and Romney, and bad for McCain and Clinton.  The fact that both variances went the other way (Clinton overcame a huge polling disadvantage, while McCain expanded his margin of victory) is explained by this single theory.  If the perceived Obama surge was an illusion of (or at least grossly overstated by) polling, those independents were simply never really there for Obama; Clinton never had to worry; and Romney never had a shot of dethroning New Hampshire's favorite Republican.

Of course, the variance was significantly higher on the Democrat side, so yesterday's surprises aren't fully explained by independents faking toward Obama, then defecting back to McCain.  But we'd expect to see the results lopsided in this direction assuming at least some of the independents "for Obama" (and some of the registered Democrats "supporting Obama") actually voted for Clinton when they were done proving their lack of racism for pollsters.  Clinton's out-out-performance, relative to McCain's, could also be due in part to some fraction of registered Democrat voters making false Obama pledges.

While I think this is the most plausible theory and does the best job explaining what we saw last night, there are so many countervailing forces that it's hard (and usually foolish) to try to pin down precisely what accounted for what.  The one theory I simply don't buy though (other than the "Diebold!" theory) is the fact that New Hampshire voters are just so fiercely independent that they're unpollable.  Just as Hannibal Lecter is so brilliant as to be impervious to psychoanalysis, so to are Granite Staters opaque to researchers, we're led to believe.

Pundits slinging that idea have tossed around one statistic incessantly in the hours since the polls closed - namely that nearly half (!) of New Hampshire voters wait until just before the primary before deciding whom to vote for and that's what makes them so poll-immune.  This is a cynical read on my part, but the half million people who vote in the New Hampshire primaries have a tremendous incentive to convince the other 302.7 million people (and the candidates who hope to serve them) that their votes are swayable right up until the end.  It keeps the candidates around longer, it dumps rocket fuel on tiny hamlet economies, and it preserves the perversely disproportionate influence that the state has on Presidential elections.  If half of the primary-voting population truly couldn't make up their mind about a contest that was launched nearly a year ago and covered in depth every day since, we really need to think hard about how best to normalize their special electoral status.

The good news is that the accuracy of the Lemon effect theory should be testable.  In future Democratic primaries (but not open outcry cauci), we should expect to see, not Clinton necessarily winning, but Clinton outperforming the freshest polling data, relative to Obama.  No good prospective remedy in polling methodology jumps to mind when the problem rests with the filthy liars on the other end of the call, so the error will likely persist (and newstalkers will find more politically palatable explanations for the ongoing polling error).  And if New Hampshire (not to mention the reverse effect seen in Iowa) was any indication, the size of this error is massive - possibly double-digit.

My guess would be that states with smaller minority populations might see a larger Lemon effect (yes, I'm trying to make that term stick), which may mean that Iowa (91% white) and New Hampshire (97% white) showed us somewhat more dramatic variances than we'll see in the upcoming Democratic races in South Carolina (69% white), where Obama's up by 13% or Florida (81% white), where Clinton is up by 20%.

But next stop Nevada is a caucus state, so it will be interesting to see how results diverge from polling in that matchup.  Clinton is currently ahead by 20 points in that contest as well, but we may see another Iowa-style "reverse Lemon" that at least makes it close for Obama.