Brendan Nyhan and I share a dissertation/thesis advisor (he got the PhD, I got the MA); perhaps as a consequence of this I frequently agree with him on matters. We also share a generalized desire to see more punditry subjected to statistical rigor.  But this post bugged me a bit, and it is representative of a growing problem I see as professional political scientists increasingly make their way into the world of punditry.

Writing on Peggy Noonan's columns about a "Snakebit President," and an earlier column titled "The Sentence," he takes Noonan to task for suggesting that Obama's falling approval ratings were due to an increased perception of weakness on the part of the President.  Nyhan concludes (in part):

[T}his is silliness. If the economy was strong, public perceptions about "The Sentence" wouldn't be a political problem. What was Bill Clinton's "Sentence" in his second term? (Indeed, Noonan has argued that Reagan "knew, going in, the sentence he wanted, and he got it" and yet his approval ratings still declined substantially when the economy was bad in 1982.)

The underlying problem is that Noonan and other pundits have strong professional incentives to construct these ad hoc explanations, which emphasize their own expertise in narrative construction and dramatize politics for public consumption. Until more pundits recognize the potential advantages of incorporating political science into their work, mysticism and superstition will continue to dominate.

Yes and no.  Let me say first that, as general matter I have no idea whether the public really perceives Obama as "snakebit" or "getting a bad Sentence."  It's probably not how I would choose to explain Obama's approval ratings.  And let me reiterate my shared belief with Nyhan that pundits absolutely should do more to incorporate political science models into their work.

But political scientists posing as pundits also need to be more modest about their work, and up-front about the limitation of their models.  Non-quantitative punditry has a huge place in our discourse for many reasons, including one that is directly applicable here.  There are all sorts of problems with these statistical models (the data are usually nonlinear, badly heteroskedastic, and limited (eg small "n" problems), and the political scientists are frequently every bit as ad hoc in selecting variables as is Noonan, just in their own way), but the most applicable problem here is that there is always  a large portion of the data that have to be explained qualitatively.

For example, take the Presidential Approval models.  There are any number of them out there, but all of them have a significant portion of the variation in Presidential approval (or variance, in geekspeak) that the model just can't account for.  Even for models that make political scientists giggle with glee at the high r-square they've produced, there will still be about 10 to 20% of the data that the model won't explain.  Political scientists like to call this "error," but it isn't really "error."  It's just "other stuff we can't readily turn into data."

Ignoring this error leads to overly ambitious claims, such as "it's the economy stupid columnists!"  The literature is replete with proof that the economy is a major driver of Presidential approval.  But consider this chart from Jay Cost earlier this cycle, which was used to make a different point:

Obviously, at the beginning there is a nice, tight relationship between unemployment and Reagan's approval rate.  It *was* the economy, stupid, back then. But note that from about September of 1984 until late 1986, the unemployment rate is more-or-less stable, even as Reagan's approval ratings continues to climb.

Now, the political scientist will likely respond by turning to another economic variable, such as real disposable income, which continued to improve sharply during that time period.  And in truth, all of these measurements probably *are* a generalized stand-in for "how do voters think the economy is doing in general," which can't be directly measured (for more of my thoughts on this, see here).

But note what happens next.  The economy-approval relationship breaks down immediately in the late 1980s, immediately after the Iran-Contra scandal breaks.   The economy chugs along, we aren't involved in any major military action gone awry -- this really is just Iran-Contra at work. The effect is huge -- 40 net approval points or so, and in the short-run, completely overwhelms the economic data.  Reagan's approval is back to where it was when unemployment was ten percent.

Quite frankly, the "Reagan lost his sentence as a good, honest leader" explanation here works as well as any explanation.  Why didn't Clinton lose his approval rating in 1997 after he received his scandal?  The economy is no doubt part of the explanation, but we could just as easily say that "philanderer" was already part of Clinton's "Sentence," and so further proof of this attribute did little to change the public's perception of him.  And quite frankly, if the President in 1997 had been Bob Dole and he had suffered a similar scandal, I think that we would have seen a very different trend in Presidential approval (I'll pause for a second and allow you to wipe that image from your brain).

As for today?  I don't believe that Obama's supposed snakebitten-ness is why his approval has dropped about 20% or so in the last 16 months.  I actually think the number one reason for the change is that we left the Presidential honeymoon period (for more on this, see Alfred Cuzan's work here), and would place the economy after this -- though I should also note that many of the economic indicators like real disposable income and GDP growth would probably indicate a somewhat-higher approval rating.

But it is perfectly plausible -- and within the error term for most of these models -- to say that the reason that Obama is at 46% approval in Gallup instead of 51% or maybe even 55% is because of supposed snakebitten-ness.  There isn't much in the way of data on this attribute so we can't really *disprove* a relationship here.  All we know is that there is always going to be a large portion of data -- whether it be presidential approval, congressional midterm elections, or presidential election results -- that can't be easily explained quantitatively.  This is where qualitative analysts like Noonan will always be valuable.

--------------------------------------------
Follow the RCP Blog on Twitter.
Become a fan of RCP on Facebook.
--------------------------------------------