More on polls and the American voter

As they do every election year about this time, the polls strongly favor the Democrats, which is very disheartening.  Right around now I find it especially worrisome, because I think that Obama, emboldened by media adulation, is being more and more forthright about his political agenda, and it’s one that ordinarily would should have the American people running for the hills:  He’s for higher, way higher taxes; gutting the military; embracing the UN as our political guide; and, most significantly, keeping fuel prices up, up, up.

McCain, for all his weaknesses as a conservative, isn’t for any of those things.  Indeed, as to the last issue, I think he’s regurgitating the green Kool-Aid he drank, which turned him into a raving environmentalist.  Instead, he’s starting to use rising fuel prices as a face-saving way to edge back to common sense.  McCain is infinitely more the centrist American candidate and, traditionally, he’s the candidate who should win.

And yet, over and over, poll after poll, Obama has the edge (although sometimes so insignificant that it’s a statistical dead heat).  What gets interesting is when one drills into the polls.  It turns out that Democrats are consistently oversampled, as are African-Americans.  Then, after all that oversampling, some magical jiggery pokery goes on in a back room, and they reveal the oracular numbers — favoring Obama.  In other words, the polls don’t seem honest.  Rather, to the uninitiated, they seem more along the lines of Twain’s lies, damn lies and statistics.

Given these delicately created polls, Paul Geary makes the interesting argument that, because the polls use a very small sampling and then taint that sampling by selecting disproportionate numbers of respondents likely to choose Obama, an enormous, unscientific AOL poll might actually be more accurate:

For the second consecutive week, Political Machine at AOL is conducting an open straw poll. And for the second consecutive week, John McCain beats Barack Obama about 56-44 with about half a million votes tallied, total. This week Obama leads only in California, Washington, New York, Maryland, and DC.

But this is not a random sample survey, you say?

True. But as Tom Elia has pointed out, recent polls have not been using an accurate cross-section of the American electorate in their surveys. The skew is almost always favorable to Democrats. The proof is in consistent election results that are several points better for the Republicans than polls predicted. (And the requisite delusional whining about vote manipulation.)

Why could a non-scientific poll be more accurate than the “random sample” polls? There are several reasons.

I’m not going to steal Geary’s thunder.  Instead, you can read all the reasons he gives here.

As I’ve repeatedly said, I’m not a numbers person, so I cannot approach Geary’s argument with much analytical fire power.  What I can say is that it is a very clear argument and that, within it’s parameters, it makes a lot of sense and is hearteningly compelling.