Daniel Larison, who knows better, throws caution to the wind and decides to defend the credibility of a Dick Morris column. Has anyone ever made money on that bet? In the Morris column in question, Tricky Dick puts forward the thesis that the presidential race is in fact tied, and times it perfectly to coincide with the late week dip last week in Obama’s numbers in one prominent daily tracking poll, but not to coincide with those numbers predictably reverting to the mean of a modest but unequivocal and stable Obama lead. Having completed that sleight of hand — have you spotted how he knew which card was yours? he’s resting every word of his column on a single poll from which, at the time of his writing, drawing any strong conclusions was literally unintelligible, and which was contradicted by nearly simultaneous polls; and did I mention that the solitary result upon which the whole charade hangs happens to conform perfectly to P. Dicky’s pre-existing biases?; really now, I’ve seen birthday magicians whose tricks showed fewer seams and took longer to crack — the Dickster then projects his own biases onto the electorate as a whole, under the pseudo-objective guise of listing Obama’s flip-flops on salient issues in the campaign, thereby (since he presents neither evidence nor argument for their salience) just gruesomely begging the question a second time (the first time being his mendacious assertion of an Obama ex-lead). In other words, bullshit from top to bottom, an effort tha not only fails to add any value, but destroys existing value. In yet other words, exactly what one expects from a Dick Morris column.

Daniel, who (not that I don’t have my crippling biases) has a tendency to latch onto polls showing sub-par results for Obama, defends Morris from Andrew Sullivan’s charge of ignoring the sustained Obama lead in national polling averages thusly:

First of all, the averages for the presidential race, whether at Pollster or RCP, incorporate polls (such as the methodologically suspect June Newsweek survey and the LAT/Bloomberg poll) that are entirely unlike all others that skew the average in a certain direction.  Second, the “average” acquires a strange, almost canonical authority because it includes a number of polling results rather than being subject to the flaws of any particular poll, and people then invoke the average as if it were more authoritative rather than being correctly understood as less precise.  For the purposes of measuring overall momentum, averages can be useful, but when you include everything from typically reliable Rasmussen and Gallup numbers to the volatile Zogby results to the ridiculous PPP you are going to have an average that splits the difference between meaningful and useless.

So, a couple of things: this only begins to make prima facie sense if one assumes that individual polls do not inherently “split the difference between meaningful and useless,” and thus that polling averages stand at a relative disadvantage to individual polls (at least if the individual polls carry a strong enough brand name). But that assumption is, as they say in France, crazy. Also, the invocation of the concept of precision is confused. It’s unclear whether Daniel is using the concept technically, or in a lay way. If the usage is technical, then he’s technically correct: individual polls are more (technically) precise than averages. But that doesn’t imply a whole lot that Daniel needs it to imply for the rest of this to follow. More (non-technically) precisely, it doesn’t imply any of the many things it would need to for the rest to follow. It just highlights the familiar distinction between precision and accuracy, which are frequently traded off against one another in alternative methodologies to suit various purposes. Sometimes precision is more important than accuracy, sometimes the opposite. In the case of averaging polls, the whole idea is to trade precision for accuracy, and that’s what gives the averages a greater claim to authority than individual polls. (Though Daniel is right to object to averages acquiring canonical status. If canonical status depends on either perfect precision or accuracy, then it’s a chimera, but if being canonical just means being the best of the available epistemic lenses, then averages have forfeited their claim to being canonical to Nate and Sean of FiveThirtyEight, a site I’m going to assume everyone reading here is already well acquainted with, and thus only say by way of describing it that media companies no longer have any fig leaf of an economic rationale for maintaining the employment of almost any staff political reporter, and possibly all of them. It’s not the case that their salaries didn’t used to be more usefully spent gambling on greyhound races but suddenly are; rather, that’s been the case for ages but any remaining pretense for denying it no longer exists.)

Now accuracy and precision are important in data analysis, but from the perspective of your average news consumer, all that matters (and all that counts towards establishing the canonical authority of a method of data collection or presentation) is its accuracy — i.e., its proximity to the truth, i.e., in this case, the confidence one can have that the candidate the average says is in the lead is really the lead. Precision, for these purposes, can basically just piss off. 99.99999% of the people who follow this stuff don’t know what a tendency to replicate results is, and if they did know, wouldn’t care. And even those of us who do care would probably be willing to say that if we somehow found a totally imprecise — i.e. utterly non-replicating, i.e. utterly unscientific — method of measuring voter preferences that was somehow 100% accurate at the same time, we wouldn’t hesitate to use the 100% accurate, unscientific method. (Metaphysical claim about the nature of the world: given the laws of nature, that scenario is impossible.)

If, on the other hand, Daniel is just using the term “precise” according to its ordinary lay usage as basically equivalent to “accurate,” and thus that polling averages tend to wind up further away from the truth than ordinary (credible) polls if the former include crappy polls in their averages, he’s simply wrong. First, crappy polls are less accurate than uncrappy polls, but their accuracy is still >0, so they offer some information, just not as much as good polls do. Second, if you’re going to countenance junking particular polls and not including them in the average, you’ll need criteria for when to junk polls. And on the assumption that you’ll be junking polls on grounds significantly less obvious than that they were written up in crayon or feces, I wish you all the best luck in your endeavor. It’s much less work just to stop taking polling averages altogether, and therefore leaves you with far more time to watch reruns of Coach.

Third, this is gonna blow your mind, especially you libertarians, but single authorities and experts on a subject cannot know more than an entire mass public (a crowd, if you will) knows collectively. Provided there is lots of data, the odds that a single poll (no matter how ridiculously dope the pollster who conducted it) is more accurate than an average of many polls (even if many of them are teh suck) are on their asymptotic roller-coaster ride towards zero. Sorry mate, but insisting on the accuracy of individual polls over the average is the last refuge of a scoundrel who doesn’t want to acknowledge the reality of an election.

Still some more epistemological and methodological issues to work through. Daniel disparages the LAT/Bloomberg and Newsweek polls as “methodologically suspect,” but this is true of them as opposed to other polls only in the sense that the McCain campaign publicly aired suspicions of the methodology when it didn’t like the results. The truth is that, even more than usual, we are shooting in the dark this year on questions of voter ID breakdown, turnout, and factors on individual voting likelihood. Every poll is suspect or none is, which come to the same thing. Which underscores an important motivation for taking polling averages, namely that even if you find a particular poll is suspicious to you, you have to hedge against the possibility that that poll is right and you’ve been working from a bad assumption somewhere along the way. Daniel continues:

This is not really to defend Dick Morris, but when he says that Obama’s lead had virtually disappeared in the Rasmussen tracking poll, he was stating something that is demonstrably true.  In the days since he wrote that, a small Obama lead has reappeared.

Sort of resolves itself, doesn’t it? A brief late week outlying slump in one daily tracking poll — which given the nature of daily tracking polls is very likely the result of a single sample skewing away from the true numbers — only counts as good grounds for writing op-eds to innumerate hacks. On the basis of the evidence Dickmorris presents, there is nothing whatsoever that can be inferred. Not that that stops our Big Dick from inferring lots and lots of things he oughtn’t. (But really, how thoroughly awful is he? The Gallup tracking poll also showed a bit of a skid around the same time and has also now reverted back to the mean. That means there was a hit of a twinkle of a glimmer of a germ of a seed of an acorn of a reason for writing a >1 sentence blog post about the sub-sub-sub-sub-sub embryonic trend. But pulling in the Gallup numbers wouldn’t have worked with the “race is tied” headline, and you don’t have to see the end result how the Clown Prince of the Potomac is going to come down on that choice.)

Now Morris Dick (as they’d call him in Hungary), having artfully dodged, through cunning use of deadlines, the predictable reversion of a brief skid from the mean right back to it, could without fear of logical contradiction insist that the fluctuating numbers reflected a real, albeit brief shift in public opinion. But both common sense and Occam’s razor (if you’re not a fan of either one, just rely on the other; trouble awaits, though, if you think Occam’s razor is common sense) militate for concluding that the poll captured nothing but statistical noise. Not every demonstrably true statement is interesting or informative. In fact, the vast majority of demonstrably true statements, countably infinitely many of them and perhaps as many as uncountably infinitely many of them, are not the least bit interesting or informative, and thus better left unsaid to conserve the calories required to say them. There now, any residual thought that you might want to say anything that might compel you to issue the disclaimer that “this is not really to defend Bigus Dickus the Morris Dancer” still lingering about? No? Don’t mention it. Yup, anytime. I’m glad to have been of help.

Having risked so much (i.e. anything) so recklessly for so worthless a human specimen as you know who, Daniel concludes, Psalmist-style, that “Dick Morris is wrong about enough things that we don’t need to impute more errors to him than he already makes on his own.” He is counseling us to invite massive systematic error into our own epistemology. It is never irrational to proceed from the assumption that for any substantive claim Dick Morris makes that requires more than elementary logic or an elementary school education to confirm, the probability function for the correctness of the evaluation of the Morris clam that “it is steaming bullshit left on fire at our doors for the sake of some repugnant political cause” converges on 100 percent. So let’s really not defend Dick Morris. If he’s the polling and elections expert he claims to be, he’s deliberately lying and shouldn’t be getting paid in a currency that is legal tender for debts public and private to type lies and email them to stupid, uneducated, credulous editors. If — and this is my hunch — he’s not an expert at anything, but a shaved orangutan of below average intellectual capacities swapped at birth with a human child in a tragic mishap, who has the basic sub-mammalian instinct of latching onto a benefactor, Slick Willie in his case, and spending the rest of his life subsisting at minimal energy cost to himself of eating the regurgitations of what said benefactor couldn’t hold down, in his case the laugh out loud hilarious idea that it took a political genius for Bill Clinton to beat Bob Dole (let alone, with Ross Perot in the race, let alone with a staggering financial advantage, let alone with Dole’s campaign not being able to pay the bills for several months and having the lights and phone turned off) , then it takes the sort of person I want lots more of at my poker games to pay Morris real rather than play money to write things. (More productive use of one’s resources: paying Morris not to write things. More productive still: paying Morris to unwrite things he’s already written.)

Ordinarily, at this point, I’d launch into the current edition of my unabridged gripe about innumeracy in the media and what to do about it, but I’ve gone on too long as it is. So let’s just say that reporters who write about statistical and mathematical concepts without the aptitude and either formal training or freakish prodigy ability to be qualified to do so, can now without hesitation add to their no doubt already impressive resumes that for at least one extended temporal part of their lives, they were as bad as Dick Morris. Congratulations chaps, I suggest you throw a theme party where everybody gets to contribute an idea ball for the next Tom Friedman column.

Tags: innumeracy, journamalism, statistics 101