Two Days: And So It Begins

For those whom the election matters the least:

There are 57 hours and 17 minutes until the 2012 election, as I write these words at 6:42p EST.

The most important data I’m working with is a daily average of the last five polls taken in eight swing states. Today I’m introducing another dimension: margin of error. But first, here’s how things have changed in the last 24 hours.

Barack Obama increased his lead in Colorado from 1 point yesterday to 2.2 points today, in Ohio from 2.4 points to 3, and Nevada from 3.6 points to 4. His lead slipped in Iowa from 3.2 points to 2.4, New Hampshire from 3.6 to 3.2, and Virginia from 2 points to 1.6. Mitt Romney took back the lead in Florida to 0.6 points, from a 0.2 point deficit, largely on the back of an outlier poll from Mason Dixon showing Romney up by 6 points.

Here’s the last ten polls of Florida:

10/25 – 10/28: Romney +1 (CNN)
10/26 – 10/28: Obama +1 (PPP)
10/28 – 10/29: Obama +1 (Grove Insight (D))
10/28 – 10/30: Obama +1 (JZ Analytics/Newsmax)
10/30 – 10/30: Romney +1 (We Ask America)
10/30 – 10/30: Romney +3 (Gravis)
10/27 – 10/31: Tie (Ipsos)
10/30 – 11/01: Romney +6 (Mason Dixon)

10/31 – 11/01: Obama +2 (NBC/WSJ/Marist)
11/01 – 11/03: Tie (Ipsos)

Only right-leaning Garvis has had Mitt Romney leading by more than one point lately and it looks like Mason, for whatever reason, simply blew that poll. Not sure why.

Romney maintained his lead in North Carolina as there hasn’t been any new polling there.

Here are the swing states from largest Obama advantage to largest Romney advantage:

Nevada: Obama +4
New Hampshire: Obama +3.2
Ohio: Obama +3
Iowa: Obama +2.4
Colorado: Obama +2.2
Virginia: Obama +1.6
Florida: Romney +0.6
North Carolina: Romney +1.4

Each candidate starts with their safe states, Obama has 247 electoral votes and Romney has 191. Nevada (6), New Hampshire (4), and Ohio get Obama to 275 with Iowa (6), Colorado (9), and Virginia (13) acting as firewall states. The 28 combined electoral votes in those firewall states are enough that Obama could lose Nevada, New Hampshire, and Ohio, and still win. Meaning between those two groups, Obama can lose either the top half (Nevada, NH, Ohio) or the bottom half (Iowa, Colorado, Virginia), and win the election. Romney needs Florida (29), Virginia (13), Colorado (9), Iowa (6), and Ohio (18) to get to 281 (without Ohio he only has 263). There are no groups of states where Romney is leading that he can afford to lose. He has to win everything between North Carolina and Ohio.

This might make it more clear:

Tipping point states

This is similar to what Nate Silver has been doing and I think it’s a nice way of showing where each candidate has to be, and where they are right now. Obama’s electoral total starts at the top with his safe states, and his column “Obama Total” increases for each state he wins going down the list. “Romney Total” begins at 191 and increases with each state he wins, from the bottom up. And the lead column shows you who is leading each state, from Obama’s best lead at the top to Romney’s best lead at the bottom.

Green cells are electoral vote totals that represent at least an election win for each candidate.

Ohio is literally the must-win state for both candidates, this year more than ever. And now you can see why Romney isn’t the favorite to win the election even though national polls are so close. Obama doesn’t have to win every battleground state where he currently has to lead in order to win. Romney has to flip several of them just to have a shot at breaking 250, much less 270.

* * *

Now, about that new dimension. It goes something like this. A population of 312 million people with a sample size of 1,000 will give you a margin of error of 3.1% about 95% of the time. A poll showing Obama leading in Ohio by 2 points is within the margin of error for a poll of that sample size and thus is of limited value. That’s why we average results. Doing so will increase the sample size and reduce the margin of error. A population of 312 million and a sample size of 17,000 will give you a MoE of 0.75%.

To put this in practice, here are the last five polls of Colorado:

Obama +4: PPP, 825 likely voters.
Obama +2: Denver Post/SurveyUSA, 695 LV.
Obama +2: CNN, 764 LV.
Obama +3: We Ask America, 1,246 LV.
Romney +3: Rasmussen, 750 LV.

I discarded a poll from Ipsos which I’m guessing was a national survey that broke out results for battleground states, which means their sample size for those states could literally be in the dozens. That’s no good for what I’m trying to do here, so I dumped it for this example (it’s in *all* my other figures).

Now, these are not national polls. So you don’t need to sample a thousand people here. Colorado’s estimated population as of 2011 was 5,116,796 . PPP’s sample size of 825 likely voters gave them a margin of error of 3.41%. But across those five polls, we actually have a sample size of 4,280. That gives us a MoE of just 1.5% about 95% of the time.

That folks, is why an average lead of 1.6 points (the five above polls) for Barack Obama in Colorado is statistically significant, and why the state is *not* a tossup. 1.6 points exceeds the 1.5% margin of error, which gives you (I guess) a 95% confidence interval. Even with the maximum accountable error which is the worst case scenario, Obama is still leading Colorado by 0.1 points with 95% confidence.

With that in mind, here are today’s battleground averages and their margin of error:

MoE states

That’s why Nate Silver has Barack Obama as an 83%+ favorite to win.

Update – I just realized today is a scheduled day to update the electoral college projections, and the day where I begin updating them daily. Stand by for that…

Update II – TPM has Obama up to 303, 538/Silver up to 305, PEC up to 323. It looks like the two latter sites finally found their courage and started calling the race like it actually is. These changes have boosted both candidates in the projection from 271-202 to 279-209 in the average, and 277-206 to 290-206 in the median. This represents a shift in the middle, not at the extremes. Otherwise you’d have seen the average move more, and then median not move at all.

The full photo below, squished, so click on it to enlarge:

ec projection

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.