Top Videos
Related Opinion
Related Topics
2008 Polls NationalIowaNew HampshireGeneral Election
GOP | DemGOP | DemGOP | DemHead-to-Head

Send to a Friend | Print Article


Does Media Have Reason to Predict a Blow-Out?

By Jay Cost

I have received some excellent comments, in the constructively critical vein, from people today about my column that operationalizes the Big 3 rankers.

A criticism that is worth discussing is whether I can come up with the averages that I did. The averages that I came up with are dependent upon the assumption that each House race is statistically independent of every other House race. In some instances, relatively few, this is a problematic assumption for my column. In others it is not.

First off - what does it mean for one observation to be statistically independent of another? It means that the observation of a "success" in one instance makes the observation of a "success" in another instance neither more nor less probable.

When would it be problematic? Take, for instance, the races in PA 06 and PA 07. These are in the same media market, and one race might be affecting another race. In that instance, we might not be able to consider them independent of one another. Democratic success in PA 06 could help induce Democratic success in PA 07.

But what about, say, FL 22 and IL 06? Is the race in Miami affecting the race in Chicago? Certainly not in a causal sense. To say that FL 22 causes IL 06 is to make a spurious causal claim. In fact, a set of causes is affecting both FL 22 and IL 06, which have no causal affect upon one another. Causally speaking, they are independent of one another. Provided that we have delineated and controlled for all of the causes that are affecting these races, they are also statistically independent of one another, too.

The proviso of the last sentence is the important point. Think of it this way. Suppose we had, in the first instance, perfect information about the probability that FL 22 would switch to the Democrats, and perfect information about the probability that IL 06 would switch to the Democrats. In the second instance - we find that FL 22 switches. Would we change our estimate of IL 06? No. We would not. Why? Because we already had perfect knowledge about everything that causes IL 06. FL 22 is not one of those causes, and therefore the result in FL 22 does not induce a change in our probability distribution.

On the other hand, if we had, in the first instance, imperfect information about both races, and we found that FL 22 had indeed switched, would we change our opinion about IL 06? We might. If we believe that IL 06 and FL 22 are caused by the same set of forces, a "success" in FL 22 might incline us to update our prediction of a "success" in IL 06. This is akin to what social scientists call Bayesian updating of prior beliefs.

In other words, the column came down to the perfection of information - or, more specifically, the amount of stock you are willing to put into the rankings of the rankers. The rankers have weighed all of the various causal forces affecting each race, and have grouped them based upon those causes. If they have fully and properly delineated and weighed those causal forces, then any variation within their categories should be random. Therefore how one race goes in a given column has no effect on how another race goes.

The "if" in the last sentence is the heart of the issue. And I punted on that one. I said:

Before we get into this - note that this should not be taken as an endorsement of any of these rankings. The idea here is that I am correcting the conventional wisdom based upon the data it most frequently uses. I am not making any comments about whether that data is valid.

This was not a cheap, sophistic move on my part. It was consistent with my overarching analytical question: given that the media uses these rankings to analyze these races, do they have reasons to predict a 25+ seat blow-out? The answer is no. To argue that we should expect something more dramatic, they must be presuming that these rankers have made many pro-Republican errors in their assignments of probabilities. Have they done that? Perhaps. But the point is that this assumption violates their other assumption: that Cook, Rothenberg and CQ are valid guides. To "expect" 25 seats, rather than 15 to 18 - is to disagree with them. You are, of course, free to do that. But don't use them as your evidence for that expectation! That was my point.


Email Friend | Print | RSS | Add to Del.icio.us | Add to Digg
Sponsored Links

Jay Cost
Author Archive