So what the hell is going on in Colorado?

[SPC]

Democrats close well in Colorado, even in 2012, Romney was up in some polling at this stage.

Throughout September, the only poll showing Romney ahead in Colorado was Rasmussen. True, many firms in October showed Romney statistically tied, but almost all of those were Republican-affiliated polls. Excluding Rasmussen and Republican firms, Obama was ahead by ~4 points in Colorado polls, not significantly far off from the final margin of ~5 points.

[SPC]

Democrats close well in Colorado, even in 2012, Romney was up in some polling at this stage.

PPP at this point gave Obama a 6 point lead in Colorado(late September).  

http://www.realclearpolitics.com/epolls/2012/president/co/colorado_romney_vs_obama-2023.html#polls

I know you think Udall will win, but you've got to admit that the PPP poll must make you a little bit nervous.

No, it doesn't make me nervous, because Buck lead in 2010 in the final poll and went on to lose, which is actually a better measure to compare this race to.

One error within the margin of error is enough for you to assume some unique polling phenomenon must occur in Colorado to the benefit of Democratic candidates? Buck's lead (1%) was within the margin of error of the poll, so his loss does not lend credence to this novel theory, given that such an event both a) would be expected to occur one out of four times and b) was not replicated in either 2008 or 2012.

[SPC]

Democrats close well in Colorado, even in 2012, Romney was up in some polling at this stage.

PPP at this point gave Obama a 6 point lead in Colorado(late September).  

http://www.realclearpolitics.com/epolls/2012/president/co/colorado_romney_vs_obama-2023.html#polls

I know you think Udall will win, but you've got to admit that the PPP poll must make you a little bit nervous.

No, it doesn't make me nervous, because Buck lead in 2010 in the final poll and went on to lose, which is actually a better measure to compare this race to.

One error within the margin of error is enough for you to assume some unique polling phenomenon must occur in Colorado to the benefit of Democratic candidates? Buck's lead (1%) was within the margin of error of the poll, so his loss does not lend credence to this novel theory, given that such an event both a) would be expected to occur one out of four times and b) was not replicated in either 2008 or 2012.

Calm down. My point is, Udall can still win this, even if the last poll showed him down two, since that is hardly out of reach. Independents are tied in the last poll and Gardner would have to do a lot better than that to win outright, especially when there are still undecided Democrats.

As someone who volunteers for campaigns, there are lots of Democrats who don't decide until the last minute and in blue leaning states, that can make the difference.

I was not objecting to the belief that Udall can win as much as the amount of certainty you have in such an outcome. 

[SPC]

Where does this whole "Democrats close well" thing come from? Buck? He was just as much of a loon as Angle, and you didn't see Heller or Sandoval suddenly lose out of nowhere, nor does that stigma exist AT ALL in Nevada. It is about the candidate, and Gardner is a fantastic candidate. Is he up 6? No. Is it conceivable that he is up about 2? Sure.

While Heller didn't lose, he did come a lot closer to losing than the polls said he would. As for Buck, while he is the most notable case of Colorado Democrats overperforming the polls (because it actually changed the eventual victor), it's a notable trend regardless. Here's the data:

2004 RCP: Bush +5.2
2004 final margin: Bush +4.7
Bias: R+0.5

2004 RCP: Salazar +4.5
2004 final margin: Salazar +3.9
Bias: D+0.6

2006 RCP: Ritter +18.7
Final margin: Ritter +16.8
Bias: D+1.9

So far, these seem like fairly typical variations.

You must be logged in to read this quote.

These at first glance seem to be fairly evenly distributed once more. However, try omitting notoriously bad pollsters (Rasmussen, ARG) from the RCP Average and adding in credible ones (PPP), which gives a final average of Obama +7.3 (R+1.7) and Udall +13 (D+2.7).

You must be logged in to read this quote.

Again applying the above methodology, the Rassy-free averages are Buck +1.5 (R+2.4) and Hickenlooper +8.3 (R+6.3). Buck's loss does not seem as extraordinary in this light, and Hickenlooper's larger than expected margin can be explained by many Tancredo voters voting for the Republican Maes in order to ensure Republican Party ballot status, rather than backing Tancredo's futile bid.

You must be logged in to read this quote.

Again, omitting Rassy, ARG, and WAA gives Obama +3.3 (R+2.1)

You must be logged in to read this quote.

After corrections: R+1.0
Excluding the 2010 gubernatorial election (for the above described reasons): R+0.2

You must be logged in to read this quote.

Based on my analysis, the alleged Republican bias in Colorado polling appears to be indistinguishable from a phenomenon resulting from disproportionate polling of the state by Rasmussen and other firms with known GOP house effect. Unless Gardner is leading by a percentage point or less in a Rassy-free polling average (he's currently at +0.2 if you trade Rassy for PPP and add the Marist poll in), I see little reason to believe that some polling theory of dubious validity will push Udall (or Hickenlooper, for that matter) over the top.

[SPC]

And here's Nevada, since you mentioned that:

2004 RCP: Bush +6.3
Final margin: Bush +2.6
Bias: R+3.7

2006 RCP: Gibbons +4.0
Final margin:Gibbons +4.0
Bias: None

R2K (which is now known to be a complete fraud) was 1/3 of the average in 2006. I'll let the Rasmussen inclusion slide since their House effect did not seem to become particularly prominent until 2008. This leaves an average of Gibbons +3.0 (D+1.0)

You must be logged in to read this quote.

Exchanging Rasmussen for PPP here did not significantly affect the average. I will give you this one.

You must be logged in to read this quote.

Rasmussen did not have a big affect here either. Another peculiarity.

You must be logged in to read this quote.

Here, however, excluding Rasmussen from the average reduces Sandoval's lead to 13.5 (R+3.2)

You must be logged in to read this quote.

Eliminating Rasmussen and Gravis make this Obama +3.7 (R+3.0)

You must be logged in to read this quote.

Rassy did not play a big role here either.

You must be logged in to read this quote.

Unlike Colorado, the polling does seem to support the idea that Democrats outperform the polls by ~4% in Nevada. However, given that this phenomenon seems to be unique to Nevada (the Colorado hypothesis seems dubious upon further inspection and I do not see similar analyses for California, Arizona, or Texas), I think that it is a reach to consider this a virtue of Nevada's Hispanic population, rather than some other factor not yet proposed.

I am not sure that this stands up to any greater scrutiny than the Hispanic theory, but could Nevada's "None of These Candidates" option by playing a role here? It is possible that disgruntled conservative voters may ultimately choose the NOTC at the last minute while having previously pledged to hold their nose for the often-flawed Republican candidate.

[SPC]

You make some good points regarding Colorado, but even if the polling theory doesn't hold up when you exclude certain firms, the fact remains that even trustworthy and reliable firms have had a slight Republican skew in Colorado in the past two elections.

Two elections out of the last eight does not seem like particularly strong evidence for polling bias, especially when such a bias was only ~2% for the Republican candidate once the quality of the firms was controlled for. Seems almost like an invocation of the Texas sharpshooter fallacy, IMHO.

You must be logged in to read this quote.

The fact that Oregon did not have a bias while Washington did would seem to suggest that vote-by-mail is not a definitive factor suggesting Democratic overperformance. Did Washington's polling bias become worse with time, as vote-by-mail became more prevalent and eventually universal?

You must be logged in to read this quote.

Pascal's Wager is not a very convincing argument.

You must be logged in to read this quote.

Recent history seems to indicate that anything from a ~3% overestimation of the Democrat (e.g. 2008 Senate) to a ~2% overestimation of the Republican (e.g. 2010 Senate) is within the realm of possibility. Thus, I see little reason for Democrats to think of consistent but small deficits in recent polling as anything but a cause for alarm.

You must be logged in to read this quote.

Yes, looking at the exit polls, it does appear that NOTC voters were disproportionately Democrats. Mea culpa.

[SPC]

The effect that vote-by-mail has had on Washington polls:

2012 President:
Average: Obama +14.0
Actual: Obama +14.9
Error: R+0.9

2012 Senate:
Average: Cantwell +23.0
Actual: Cantwell +20.9
Error: D+2.1

2012 Gubernatorial:
Average: Inslee +1.0
Actual: Inslee +3.0
Error: R+2.0

2010 Senate:
Average: Murray +1.3
Actual: Murray +3.8
Error: R+2.5

2008 President:
Average: Obama +15.7
Actual: Obama +17.2
Error: R+1.5

2008 Governor:
Average: Gregoire +6.7
Actual: Gregoire +6.4
Error: D+0.3

Average Error: R+0.8

If Colorado's switch to vote-by-mail is supposed to save Udall, he better hope that his average polling deficit is still within a percentage point on Election Day.

[SPC]

These were the numbers I got.

It seems the polling in Washington is pretty consistently more Republican leaning than the final results.

2004 RCP: Kerry +4.5
Final margin: Kerry +7.2
Bias: R+2.7

2006 RCP: Cantwell +13.3
Final margin: Cantwell +17.0
Bias: R+3.7

2008 RCP: Obama +13.0
Final margin: Obama +17.2
Bias: R+4.2

2008 RCP: Gregoire +3.4
Final margin: Gregoire +6.4
Bias: R+3.0

2010 RCP: Murray +0.3
Final margin: Murray +3.8
Bias: R+3.5

2012 RCP: Obama +10.5
Final margin: Obama +14.9
Bias: R+4.4

2012 RCP: Cantwell +20.4
Final margin: Cantwell +20.9
Bias: R+0.5

2012 RCP: Inslee +1.0
Final margin: Inslee +3.0
Bias: R+2.0

Average Republican bias: 3 points

Again, 2008-2012 is a consequence of keeping Rasmussen and sometimes Strategic Vision in the average while sometimes excluding the Washington Poll. True, the elections from 2004 and 2006 (excluding the gubernatorial race, which you excluded for some reason) seem to have shown a polling bias in favor of Republicans, but it would seem to undermine the argument that this is attributable to vote-by-mail if these bias has diminished in recent years.

[SPC]

These were the numbers I got.

It seems the polling in Washington is pretty consistently more Republican leaning than the final results.

2004 RCP: Kerry +4.5
Final margin: Kerry +7.2
Bias: R+2.7

2006 RCP: Cantwell +13.3
Final margin: Cantwell +17.0
Bias: R+3.7

2008 RCP: Obama +13.0
Final margin: Obama +17.2
Bias: R+4.2

2008 RCP: Gregoire +3.4
Final margin: Gregoire +6.4
Bias: R+3.0

2010 RCP: Murray +0.3
Final margin: Murray +3.8
Bias: R+3.5

2012 RCP: Obama +10.5
Final margin: Obama +14.9
Bias: R+4.4

2012 RCP: Cantwell +20.4
Final margin: Cantwell +20.9
Bias: R+0.5

2012 RCP: Inslee +1.0
Final margin: Inslee +3.0
Bias: R+2.0

Average Republican bias: 3 points

Again, 2008-2012 is a consequence of keeping Rasmussen and sometimes Strategic Vision in the average while sometimes excluding the Washington Poll. True, the elections from 2004 and 2006 (excluding the gubernatorial race, which you excluded for some reason) seem to have shown a polling bias in favor of Republicans, but it would seem to undermine the argument that this is attributable to vote-by-mail if these bias has diminished in recent years.

There was no gubernatorial race in Washington in 2006. If you're referring to the one in 2004, I didn't include it because RCP didn't have an average for it. Same for the Senate race in 2004. Doing my own makeshift average, the gubernatorial race would've had a 1.8 point Dem bias and the Senate race would've had a 3.7 point Republican bias, so little impact on the overall margin anyway. You can exclude certain polls from the averages to reduce the bias as you have, but all of the aggregation and projection sites will not be doing so. True, some like Silver will attempt to correct for this house effect, but most people will be taking them at face value.

Hence why I advocate simply taking a Rasmussen-free arithmetic average of recent polls to obtain a forecast, rather than relying on Dean Chambers-esque methods to concoct a rosy picture.