Which States are Underpolled in the Presidential Race?

I mentioned in yesterday's electoral college post that I thought Nevada was a state that had been polled less often than than it should have been given how closely contested the Silver state appears to be. Of course I was called on to elaborate on that assessment.* Jack may have been asking for a simple gut reaction as to what I considered to be underpolled. However, I cannot help but over-analyze even the simplest of questions. Why provide a feeling when we can put the data we have to good use?

With that in mind, what is underpolled?

We can go about answering that question in a couple of ways. The simplest way is to take an average. With the release of Rasmussen's poll in North Carolina this morning, the grand total of polls in our data set (from Super Tuesday to now) is 553 polls. That's an average of just over 11 polls per state. States coming in under that line, then, are underpolled. Sure, that certainly isn't false, but that is rather a low bar to set in defining what "underpolled" means.

Another couple of layers can also be added to this. We would expect that the number of polls conducted in a state would vary based upon how close and how large the state was. We'll get to a state's size in a moment, but let's focus initially on the "how close" question. An easy way to extend the simple approach is to split the states into groups according to how close they are. Well, that's already been done for us. We can take an average of the toss up states, the lean states and the strong states with the expectation that toss up states would have more polls conducting in them on average than a lean state or a strong state (Likewise, lean states would have more polling than strong states.).

Average Number of Polls in States by Level of Competitiveness
	States	Poll Frequency	Average
Toss Ups (14)	AK CO FL IN MI MO MT NV NH NC ND OH PA VA	218	15.53
Leans (10)	DE GA IA NJ NM OR SC SD TX WI	116	11.6
Strongs (26)	AL AZ AR CA CT HI ID IL KY KS LA ME MD MA MN MS NE NY OK RI TN UT VT WA WV WY	219	8.42

And that is what we see in the table. So instead of saying the overall average of polls across all 50 states is 11 and there have only been 10 polls in Nevada. We can say that among toss up states, the average number of polls is 15.5 and Nevada has had only 10 polls conducted since February. That gives us a better definition of underpolled.

It gives us a better definition, but perhaps not a very efficient one. What about state size? We'll get to that in a minute. First, we can take a page out of FiveThirtyEight's book and run a regression with the number of polls conducted so far in a state as the dependent variable and the competitiveness that state (as measured by our weighted average) as our explanatory variable. In other words, we would expect that as the spread between the two candidates increases, the number of polls in that state decreases. That's exactly what the graph below depicts.

Predicted Polling Frequency

[Click Graph to Enlarge]

And with that handy regression line, we can predict where a state's frequency of polling should be given its level of competitiveness. So, Nevada, with ten polls thus far is about six polls under what we would expect in light of how close the race appears in the Silver state. But right there in that lower left quadrant of the graph are several toss up states clustered together. Alaska, Indiana, Montana, New Hampshire and North Dakota all come in under that prediction line. Even lean states like New Mexico, South Carolina and South Dakota are underpolled.

And what about a state's size? The number of electoral votes at stake in a state -- a reasonable proxy for size in this context -- could affect the frequency of polling in a state as well. When we add that into the regression how are the things we see above affected? Again, that would add to our understanding of what is causing polling frequency to vary across states and ultimately increases the efficiency of our prediction. Competitiveness alone explains about a quarter of the variation in polling frequency and competitiveness and state size bumps that up to just over half. If we focus our attention on the 14 toss up states -- six of which were underpolled when compared to the original prediction -- only four were significantly underpolled: Indiana, Montana, Nevada and North Dakota. Alaska, Michigan and New Hampshire were about on par with where they would be predicted to be with 10, 17 and 13 polls, respectively. The remaining seven states could be considered "overpolled" based on competitiveness and state size. You cannot over poll in my opinion, but in a world of finite resources and comparatively speaking, that's the reality.

So, long story short, it is that small group of toss up (and some) lean states that are underpolled at the moment.


*Our readers and commenters here are great. I certainly have my own ideas of what to post here, but it is in my conversations both here in the comments section and with colleagues here at UGA that spur some of the great ideas that ultimately appear in this space. I don't say it often enough, but thank you all for your support of the site and for your contributions.


Recent Posts:
The Electoral College Map (8/14/08)

2008 vs. 2004, Part II: What Happened in the Final 100 Days in 2004 and What That May Mean for the Rest of This Campaign

2008 vs. 2004, Part I: What Things Would Have Looked Like 4 Years Ago This Time