Thursday, October 25, 2012

There is Only 1 Swing State

The discussion of “swing states” in the media had left a lot to be desired in my opinion.  There are various definitions but most election observers seem to count any state that would likely be close if the overall national vote ends up being close as a swing state.  This is unsatisfying.  There is insight to be gained by analyzing the correlation of outcomes across states.  If Pennsylvania goes Republican, you can be nearly %100 sure Ohio will go Republican as well.  Likewise with Virginia and North Carolina, if Virginia goes GOP, so does North Carolina.

One particular use of swing state polling data that I think is misleading is the “swing state average” poll.  A newspaper, website, or even a polling firm will average the results of several swing states together.  The resulting number is meaningful, in the same way a national poll is meaningful, but not helpful in determining who might win a very close election,

To demonstrate cross-state correlation, let’s look at the case of Ohio and Pennsylvania.  Below is the Republican candidate’s share of the vote in each presidential election since 1948.  The tight correlation is obvious, as is the fact that, in each and every election since 1952, the share in Ohio is higher.  If Ohio falls for the Democrat in 2012, we can be sure Pennsylvania goes Democrat as well.  Likewise, if Pennsylvania goes for the Republican, we know Ohio is going GOP.

This conclusion is also supported by the polling data from 2012.  I used polls tracked on to create an unbalanced two-week moving average of polls for each state.  The numbers for each state are the average Republican share of all polls concluded within 2 weeks of the date.  Polling data can be messy, especially early polling data, but it is obvious the fact still holds that Pennsylvania will not go Republican without Ohio going first.

Not all the correlations are as clean as this one, but still more insight can be gained by looking at these same numbers for a broader set of states.  Different sets of swing states are used, and I started with a broad set.  This set of 11 states incorporates most states defined as a swing state by the major newspapers or websites: Colorado, Florida, Iowa, Michigan, Nevada, New Hampshire, North Carolina, Ohio, Pennsylvania, Virginia and Wisconsin.

Using these 11 states I calculated the poll moving average for each state.  After looking at the numbers, it became clear that Ohio is in the middle of the distribution.  I grouped them into 2 sets, with the 5 leaning more Democratic than Ohio in one set, the D6, and the 5 leaning more Republican than Ohio in the other set, the R6.  Ohio is included in both.

The next chart shows the poll moving average for the R6.  It is clear that Ohio is consistently the lowest among the 6.  Using the correlation logic I have explained, the conclusion should be: if Ohio goes Republican, all 6 should go Republican.  Now, if you add the electoral votes of these 6 to the “safe” Republican electoral votes (states which surely go GOP if the election is close), you get 281, enough to win.  It’s possible for Ohio to go Democrat, while all the others go the other way.  This leaves the Republican with 263 electoral votes, not enough to win.  The conclusion is if Ohio goes for the Republicans, so will the rest of the R6, leading to a Republican win.

Now let’s look at the D6.  Here it is not clear from the poll moving average that Ohio is the most Republican leaning state of this group.  It looks like New Hampshire, Iowa and Wisconsin are all close to Ohio.

Here the historical data can help lead to a conclusion.   If Iowa or Wisconsin does lean more Republican than Ohio in 2012, it would buck a 40 year historical trend.  If New Hampshire leans more Republican than Ohio, it would buck a 20 year trend.   Below are the Republican shares for each of these states in presidential elections since 1948.  The key is that, since 1992, Ohio has leaned more Republican than the other 3 in every election.  Furthermore, since 1972, only New Hampshire has leaned more Republican than Ohio.  This is important because NH is not an essential state for the Democrats.   I conclude that it’s very unlikely that Wisconsin or Iowa leans more Republican than Ohio.

So if Ohio goes Democrat, so will the rest of the D6.  Adding these to the safe Democratic states gives Democrats 275 electoral votes, enough to win the election.  Even if New Hampshire breaks the other way, that still leaves the Democrats enough votes to win with 271.

Now here are a couple disclaimers.  First, do not read too much into the absolute magnitudes on the poll moving average charts, since these polls always include a healthy amount of undecided voters, it’s the relative magnitudes that matter here.  Second, Ohio does not cause other states to vote a certain way, these are correlations arising from many factors, including the campaign efforts of each candidate in every close state.  So you should not conclude that only campaigning in Ohio is a reasonable strategy.  What you can conclude is, as an observer, you can identify the outcome of the election, with a very high degree of probability, by observing just 1 state.

There you have it.  The winner of Ohio wins the election.  If a swing state is a state that can independently swing the election outcome, there are not 11 swing states, or 9, or 6.  There is only 1 swing state.