The Phantom's Polling Analysis Thread

[BluePhantom]

Also when I wrote the demographics breakdown last night I wrote the wrong figures for Overall Independents and Overall Democrats. Sorry, I was very tired. It has been fixed now to represent a 79 point Obama advantage and a 3 point Romney advantage respectively.

 

[snakedoc]

Allow me to do readers digest condensed version of all this it is a close race. Did I get it right?

 

[BluePhantom]

Allow me to do readers digest condensed version of all this it is a close race. Did I get it right?

If you want to sum it up in three words I would suggest: "It's a tie".

 

[jillian]

Allow me to do readers digest condensed version of all this it is a close race. Did I get it right?

If you want to sum it up in three words I would suggest: "It's a tie".

I don't think it is, though. But we'll just have to keep watching. I do, however, think the race naturally tightened due to a bit of a bump for Romney for, apparently, getting the GOP nod and being the last man standing

 

[snakedoc]

Allow me to do readers digest condensed version of all this it is a close race. Did I get it right?

If you want to sum it up in three words I would suggest: "It's a tie".

That's a good way to do it as well.

 

[BluePhantom]

Allow me to do readers digest condensed version of all this it is a close race. Did I get it right?

If you want to sum it up in three words I would suggest: "It's a tie".

I don't think it is, though. But we'll just have to keep watching. I do, however, think the race naturally tightened due to a bit of a bump for Romney for, apparently, getting the GOP nod and being the last man standing

I think that's probably true. The Rasmussen tracking poll has Obama +2 today and Romney was +5 a week ago with them. Whether that's just noise from an outlier poll in their averages or a trend is yet to be seen. Since it's a three day rolling average my guess is the latter, but we'll see. If I was a betting man I would guess that Romney got a quick little bump as you suggested and that might be starting to dry up. I imagine it will bump around for a few more weeks before it gets settled down into a recognizable pattern.

Obviously the data for the week ending 4/24 is yet to be collected but from what I have collected Romney currently holds a 0.36% advantage in the professional polls for the two week average compared to Obama's 0.77% lead in the combined two week average (pro and media). That's including today's Rasmussen poll and adjusted for sample size.

Even if we were to argue the RV vs. LV factor, which I am conservatively estimating would give a 3-4 point bump to Romney, we're still within the margin of error.

IMHO, I think this is about as close to a dead heat as we can imagine right now.

 

[jillian]

If you want to sum it up in three words I would suggest: "It's a tie".

I don't think it is, though. But we'll just have to keep watching. I do, however, think the race naturally tightened due to a bit of a bump for Romney for, apparently, getting the GOP nod and being the last man standing

I think that's probably true. The Rasmussen tracking poll has Obama +2 today and Romney was +5 a week ago with them. Whether that's just noise from an outlier poll in their averages or a trend is yet to be seen. Since it's a three day rolling average my guess is the latter, but we'll see. If I was a betting man I would guess that Romney got a quick little bump as you suggested and that might be starting to dry up. I imagine it will bump around for a few more weeks before it gets settled down into a recognizable pattern.

Obviously the data for the week ending 4/24 is yet to be collected but from what I have collected Romney currently holds a 0.36% advantage in the professional polls for the two week average compared to Obama's 0.77% lead in the combined two week average (pro and media). That's including today's Rasmussen poll and adjusted for sample size.

Even if we were to argue the RV vs. LV factor, which I am conservatively estimating would give a 3-4 point bump to Romney, we're still within the margin of error.

IMHO, I think this is about as close to a dead heat as we can imagine right now.

fwiw, i think your average is a bit off and is being dragged down by gallop and fox which do seem to be the outliers right now. CNN and ABC have the president at +9 and +7 respectively. ithink those overestimate the president's numbers for whatever reason. so if we remove the bottom two and top two poll results, we end up with the president at about +4. i think that's probably about right given the demographic results.

 

[BluePhantom]

fwiw, i think your average is a bit off and is being dragged down by gallop and fox which do seem to be the outliers right now. CNN and ABC have the president at +9 and +7 respectively. ithink those overestimate the president's numbers for whatever reason. so if we remove the bottom two and top two poll results, we end up with the president at about +4. i think that's probably about right given the demographic results.

Well let's have a look at that. Basically I am calculating the national polls in three primary ways with three subcategories for each. I know...I can't just make it simple.

The three primary ways are as follows:

Base Average: All I am doing here is taking the percentages for Obama and Romney and averaging them out with equal weight attached to each poll regardless of the sample size. The data in post #2 were all Base Averages. The problem is that because the sample sizes vary dramatically from poll to poll, those polls with lower sample sizes get over-evaluated. A poll of 1000 people, for example, counts the same as a poll of 500 people. Therefore, the poll with the smaller sample, in that example, carries twice the weight of the larger poll. This is how RCP calculates the average, but I believe that is a pretty questionable way to do it.

Adjusted Average: What I do here is apply the percentage of support to each candidate against the sample size and recalculate according to the total sample. So (let's do easy math here) Poll A is 1000 people with a 60/40 split for Obama, and Poll B is 500 people with a 50/50 tie. I will count 850 votes for Obama [(60%*1000)+(500*50%_=850] and 650 for Romney and then average them against 1,500 to give a hypothetical 57/43 spread for Obama. I think this is better because it more accurately reflects the sample sizes but it's not perfect because it gives extra weight to the question format of the larger poll. So if the larger poll is designed like shit we end up with deceptive results.

Two Week Rolling Average: I will do this in both Base Average and Adjusted Average formats, but it's essentially just using a two week rolling average to filter out statistical noise.

For each of those I calculate the sub-categories of professional pollsters, media polls, and combined averages. Now for my personal use I do a lot more than that, but for the purposes of this thread I think that about as deep as we want to get. So let's toss out the top and bottom two and recalculate and see what we get.

Base Average
Professional: Obama 46.5, Romney 46.5; tie
Media: Obama 48, Romney 43; Obama +5
Combined: Obama 47.25, Romney 44.75; Obama +2.5

Adjusted Average
Professional: Obama 45.88, Romney 46.63; Romney +0.75
Media: Obama 47.72, Romney 43; Obama +4.72
Combined: Obama 46.55, Romney 45.29; Obama +1.26

Two Week Rolling Average
Not enough data to calculate.

So taken as a whole (ignoring RV vs. LV for the moment) it's probably around Obama 47, Romney 45; Obama +2 So if we apply the margin or error we are essentially saying reality could be anywhere from Obama +5.5 to Romney +1.5

In other words it's probably relatively close to what you suggest, yes. Now for my personal use I will focus on the combined adjusted average (Obama +1.26) and apply an RV factor changing it to Romney +2.24 but that's beyond the calculations we are considering on this thread.


Jeez...all that typing to say "for the most part I tend to agree"

 

[jillian]

fwiw, i think your average is a bit off and is being dragged down by gallop and fox which do seem to be the outliers right now. CNN and ABC have the president at +9 and +7 respectively. ithink those overestimate the president's numbers for whatever reason. so if we remove the bottom two and top two poll results, we end up with the president at about +4. i think that's probably about right given the demographic results.

Well let's have a look at that. Basically I am calculating the national polls in three primary ways with three subcategories for each. I know...I can't just make it simple.

The three primary ways are as follows:

Base Average: All I am doing here is taking the percentages for Obama and Romney and averaging them out with equal weight attached to each poll regardless of the sample size. The data in post #2 were all Base Averages. The problem is that because the sample sizes vary dramatically from poll to poll, those polls with lower sample sizes get over-evaluated. A poll of 1000 people, for example, counts the same as a poll of 500 people. Therefore, the poll with the smaller sample, in that example, carries twice the weight of the larger poll. This is how RCP calculates the average, but I believe that is a pretty questionable way to do it.

Adjusted Average: What I do here is apply the percentage of support to each candidate against the sample size and recalculate according to the total sample. So (let's do easy math here) Poll A is 1000 people with a 60/40 split for Obama, and Poll B is 500 people with a 50/50 tie. I will count 850 votes for Obama [(60%*1000)+(500*50%_=850] and 650 for Romney and then average them against 1,500 to give a hypothetical 57/43 spread for Obama. I think this is better because it more accurately reflects the sample sizes but it's not perfect because it gives extra weight to the question format of the larger poll. So if the larger poll is designed like shit we end up with deceptive results.

Two Week Rolling Average: I will do this in both Base Average and Adjusted Average formats, but it's essentially just using a two week rolling average to filter out statistical noise.

For each of those I calculate the sub-categories of professional pollsters, media polls, and combined averages. Now for my personal use I do a lot more than that, but for the purposes of this thread I think that about as deep as we want to get. So let's toss out the top and bottom two and recalculate and see what we get.

Base Average
Professional: Obama 46.5, Romney 46.5; tie
Media: Obama 48, Romney 43; Obama +5
Combined: Obama 47.25, Romney 44.75; Obama +2.5

Adjusted Average
Professional: Obama 45.88, Romney 46.63; Romney +0.75
Media: Obama 47.72, Romney 43; Obama +4.72
Combined: Obama 46.55, Romney 45.29; Obama +1.26

Two Week Rolling Average
Not enough data to calculate.

So taken as a whole (ignoring RV vs. LV for the moment) it's probably around Obama 47, Romney 45; Obama +2 So if we apply the margin or error we are essentially saying reality could be anywhere from Obama +5.5 to Romney +1.5

In other words it's probably relatively close to what you suggest, yes. Now for my personal use I will focus on the combined adjusted average (Obama +1.26) and apply an RV factor changing it to Romney +2.24 but that's beyond the calculations we are considering on this thread.


Jeez...all that typing to say "for the most part I tend to agree"

 

[BluePhantom]

Weekly Breakdown for 4/24/12

A very disappointing week for poll geeks. There were no free standing polls released during the period in question. Only Gallup and Rasmussen's daily tracking polls gave any indication of the attitudes of the electorate and their results are....well.....confusing. Gallup (who uses a five day rolling average) started the week showing a three point Romney lead. That maxed out at five points mid-way through the cycle, turned to a two point Obama lead by the 23rd, and then skyrocketed to a seven point Obama lead by the 24th. Rasmussen (who uses a three day rolling average) was just the opposite. Starting with a two point Romney lead, by mid-week they were showing a three point Obama lead, before turning 180 degrees and finishing with a four point Romney lead.

Why is this confusing? Well aside from the fact that they showed the exact opposite trend by the end of the week they both showed massive boosts in the final two days for completely different candidates. So what happened was that a poll that was holding back their overall average dropped off their rotation, they each had a massive outlier poll on one or more days (probably the last two), or a combination of both.

Regardless of the reason, it's frustrating because without other polls to confirm one or the other we really have no idea which one is accurate and which is the outlier. With that said let's look at the statistics:

4/18/12 - 4/24/12

Base Average
Professional Polls
Rasmussen (LV - tracking): Obama 44, Romney 48
Gallup (RV - tracking): Obama 49, Romney 42

Average: Obama 46.50, Romney 45.00; Obama +1.50

Media Polls
None

Overall Base Average: Obama 46.50, Romney 45.00; Obama +1.50


Adjusted Average
Professional Polls
Rasmussen (LV - tracking): Obama 660, Romney 720
Gallup (RV - tracking): Obama 1078, Romney 924

Total: Obama 1738, Romney 1644

Average: Obama 46.97, Romney 44.43; Obama +2.54

Media Polls
None

Overall Adjusted Average: Obama 46.97, Romney 44.43; Obama +2.54


Two Week Rolling Average

Base Average
Professional Polls

Obama 46.00, Romney 45.50; Obama +0.50

Media Polls

Obama 48.50, Romney 43.75; Obama +4.75

Combined Average

Obama 47.00, Romney 44.80; Obama +2.20


Adjusted Average
Professional Polls

Obama 45.90, Romney 45.06; Obama +0.84

Media Polls

Obama 48.49, Romney 43.81; Obama +4.68

Combined Average

Obama 46.48, Romney 44.78; Obama +1.70


In state polls we got more of the same. Polls were released for Arizona, Texas, and New Hampshire. The New Hampshire poll was started on 4/6/12, however, and is therefore disqualified for consideration. We are not tracking Arizona or Texas as we are not considering them "swing states" at this time. If more polls begin to show a tight race we will start to include them.

There will not be a demographic breakdown this week as there is no new data to analyze.

 

[BluePhantom]

Weekly Breakdown for 5/1/12

Another rather frustrating week for us polling geeks. Only four polls were taken last week and one of them (the National Journal poll) was so useless that I was tempted to completely discard it. For one thing it’s an A poll which is notoriously unreliable and will always favor the Democratic candidate by a good 6 points at least…usually more, and they also didn’t provide even the most minute amount of statistical breakdown. Despite the temptation to ignore the poll due to its multiple problems, I included it in the calculations as I don’t want to get into the habit of choosing which polls to keep and which one to pitch. For my own personal use I will do that, but for the purposes of this thread it’s a slippery slope I have no desire to embark on.

The Gallup and Rasmussen tracking polls showed similar noise throughout this week as they did last week. They bounced around quite a bit ranging from big Romney leads back to big Obama leads before both of them settling at an even tie on the first of May. I am still not quite sure how to interpret this except to suggest that it’s only been three weeks since the race was, for all intents and purposes, defined as Romney vs. Obama for certain and the numbers just haven’t quite settled down yet.
With that said let's look at the statistics:

4/25/12 - 5/1/12

Base Average
Professional Polls
Rasmussen (LV - tracking): Obama 46, Romney 46
Gallup (RV - tracking): Obama 46, Romney 46

Average: Obama 46.00, Romney 46.00; Tie

Media Polls
Fox News (RV): Obama 46, Romney 46
National Journal (A): Obama 47, Romney 39

Average: Obama 46.50, Romney 42.50; Obama +4

Overall Base Average: Obama 46.25, Romney 44.25; Obama +2.00


Adjusted Average
Professional Polls
Rasmussen (LV - tracking): Obama 690, Romney 690
Gallup (RV - tracking): Obama 1012, Romney 1012

Total: Obama 1702, Romney 1702

Average: Obama 46.00, Romney 46.00; Tie

Media Polls
Fox News (RV): Obama 420.9, Romney 420.9
National Journal (A): Obama 471.88, Romney 391.56

Total: Obama 892.78, Romney 812.46

Average: Obama 46.52, Romney 42.34; Obama +4.19
Overall Adjusted Average: Obama 46.18, Romney 44.75; Obama +1.43


Two Week Rolling Average

Base Average
Professional Polls

Obama 46.25, Romney 45.50; Obama +0.75

Media Polls

Obama 46.50, Romney 42.50; Obama +4.00

Combined Average

Obama 46.38, Romney 44.63; Obama +1.75

Adjusted Average
Professional Polls

Obama 46.49, Romney 45.22; Obama +1.27

Media Polls

Obama 46.52, Romney 42.34; Obama +4.19

Combined Average

Obama 46.49, Romney 44.62; Obama +1.87

Only the Fox News poll provided any demographics information. That data will be crunched and released by tomorrow at the latest.

In state polls there was some bad news for Romney. A PPP(D) RV poll showed Obama with a 51/43 lead in the state of Virginia but Rasmussen showed a 44/45 Romney advantage in its LV poll taken a week prior. Averaged out over two weeks it breaks down to Obama 47.50, Romney 44.00; Obama +3.50.

In North Carolina we got a surprising poll from SUSA who did an RV poll instead of their usual LV polls. Why is anyone’s guess but we can speculate that whoever hired them to do the analysis wanted a cheaper route. Anyhow, they showed a 47/43 four point Obama lead but with no recent data to compare it to it’s difficult to take that as gospel. Still SUSA is generally deadly accurate so there’s certainly cause for cheers from Obama supporters on that one.

In Florida Rasmussen’s LV poll showed a 45/46 Romney advantage but as with the SUSA poll of North Carolina there are no other recent state polls to provide a basis for comparison. Still Rasmussen is also pretty accurate and given that it’s an LV poll, Romney supporters should be feeling pretty good about what is happening there as three weeks ago Obama held a 3.50 advantage in Florida.

In New Mexico PPP(D) released an RV poll showing a 54/40 split for Obama. Sounds like great news for Obama supporters but it’s worth nothing that Obama’s lead in New Mexico has dropped 5 points in the last two months. We will keep an eye on New Mexico to see if the trend remains but unless a tightening trend continues I will likely pull New Mexico from the swing state category in the next month.

A Trend Report will be released later today.

 

[BluePhantom]

5/1/2012 Demographics and Trends Analysis


With only one poll showing crosstabs for the week in question (Fox News - RV), we have to be a little careful with the data and take it with a slight grain of salt. The crosstabs did not supply a verifiable breakdown of the number of men and women in the sample, or the number of white/black/Hispanic/other etc. So I had to use the formula described earlier to estimate the breakdowns. As such this data is Estimated Data and without verifiable numbers it means we have to be careful not to get too giddy or too depressed with what they say.

One of the things to really keep in mind is that the poll sample showed 43% were Democrats, 37% were Republicans, and 17% were Independents. According to the standard 32/32/36 LV model of D/R/I it means that Democrats were WAY over-sampled, Republicans were slightly under-sampled, and the number of Independents wasn't even in the ballpark. If we were to weight the results of the poll according to the standard LV model the final poll results would go from a 46/46 tie to a 41/48 advantage in favor of Romney.

In unweighted results, among women Obama holds a 49/42 advantage. This shows a 2 point drop in support for Obama and a 2 point gain in support for Romney (a four point swing to Romney's side) since the last demographics breakdown three weeks ago. The data suggests a trend among women toward Romney as the swing is outside the margin of error, but it's slight trend and it remains to be seen whether it will hold and whether other polls will verify the trend.

According to Ideology, Independents have moved away from Obama since the last demographics report. Three weeks ago we saw a 42/45 split in favor of Romney. The most recent data shows that gap has increased to a 33/46 breakdown in Romney's favor...a ten point swing in the last three weeks. Note that Romney's support has remained steady at 45-46 percent while Obama's is, at least according to the data on hand, tanking. Again a ten point swing is outside the margin of error and so it appears there is a trend away from Obama among Independents but again we really need other polls to verify that...which right now we don't have.

According to race it was very difficult to extract any data I feel very confident about because of the way the crosstabs were written. It gave a breakdown between "white" and "non-white". In this thread we are primarily concerned with white and Hispanic. Currently among white voters this poll shows a 37/53 split for Romney and a 72/24 split for "non-white" voters. If we apply the standard model discussed in earlier posts to that "non-white" category it breaks down to almost exactly what we saw three weeks ago with Obama holding a 65/26 advantage among Hispanics. I want to stress though that this demographic is really just a guess based on projections of standard models against this particular poll, and the fact that when they are applied it matches perfectly with what the breakdown was three weeks ago. In other words there doesn't appear to be any movement according to race but I can't prove it.

In state trends

Ohio appears to be showing a trend away from Obama according to professional pollsters over the last two weeks. Romney's support has remained steady at 42% but Obama has dropped from 46% to 44%. This two point swing is within the margin of error so it could just be statistical noise. More data is required.

Florida seems to be reinforcing the Ohio results , however, as Obama has gone from a 3.50 point advantage three weeks ago to a one point disadvantage as of May 1st according to base averages. That's a 4.50 swing in Romney's favor. As this is outside the MOE and multiple polls show a similar trend we can say with a lot more confidence that this is an actual "trend" in Florida. Whether it holds of course remains to be seen.

There was not enough data to look for trends in any other state.


So I must emphasize again that the demographics data for this week was taken from one poll and compared to multiple polls over the last three weeks. The reason why was because, again, there was only one poll that provided demographic data last week so there was nothing else to use for comparison. That means we have to take those demographics trends very cautiously as there is simply not enough evidence from other polling agencies to support or refute that particular data set.

 

[PLYMCO_PILGRIM]

So after 2-3 weeks of obama gains Romney is having a good week it seems.

 

[BluePhantom]

So after 2-3 weeks of obama gains Romney is having a good week it seems.

It appears that way, yes. Now that 10 point swing among Independents is quite a change....personally I find it hard to believe that it swung quite that much but that's my opinion and in the "official reports" I am just posting the data that is there. We can discuss the data further as we wish, of course, and debate its validity as we choose to.

But I think what we should take from all this is a couple things:

1) I personally wouldn't get too worked up over this 2 point swing, and that 4 point swing in specific areas right now. It's too early and there's just not enough data to feel very secure about it.

2) What we can see though is that on the whole it appears to be trending in Romney's favor to some degree.

Now personally I am not surprised by this and I believe the trend will continue to develop. The GOP is still stinging a bit from the primaries. The wounds have not yet healed and currently there are still quite a few GOP voters who have not yet come aboard because they still haven't let go of Santorum or Gingrich or whoever. In time they will but it will take a few more weeks at least and the closer the election gets the more that will happen.

Now whether it was a ten point swing among Independents or not there was definitely some movement away from Obama. It could be in reaction to the recent reports on the economy and unemployment (which were dismal) but that's speculation of course.

 

[Montrovant]

It's funny how the polling can be more interesting than the actual political race.

 

[BluePhantom]

Electoral Vote Projection for week ending 5/1/2012

Given current data here's what I have

Secure
Obama: California (55), Connecticut (7), D.C. (3), Delaware (3), Hawaii (4), Illinois (20), Maine (4), Maryland (10), Massachusetts (11), Michigan (16), Minnesota (10), New Jersey (14), New York (29), Oregon (7), Rhode Island (4), Vermont (3), Washington (12)

Romney: Alabama (9), Alaska (3), Arkansas (6), Georgia (16), Idaho (4), Indiana (11), Kansas (6), Kentucky (8), Louisiana (8), Mississippi (6), Montana (3), Nebraska (5), North Dakota (3), Oklahoma (7), South Carolina (9), South Dakota (3), Tennessee (11), Texas (38), Utah (6), West Virginia (5), Wyoming (3)

Obama Secure Total: 212
Romney Secure Total: 170

Leaning
Obama: New Mexico (5), Pennsylvania (20), Wisconsin (10)

Romney: Arizona (11), Florida (29), Missouri (10), New Hampshire (4), North Carolina (15)

Obama Leaning Total: 35
Romney Leaning Total: 69


Obama Grand Total: 247
Romney Grand Total:239


Too Close to Call

Colorado (9), Iowa (6), Nevada (6), Ohio (18), Virginia (13)

Total TCTC: 52



As a side note: for practical consideration a tie will go to Romney as the tie is broken by the House of Representatives with each state getting one vote. In that scenario Obama would likely get a minimum of 21 votes and a maximum of 24. Romney would get a minimum of 26 and a maximum of 29.

 

[BluePhantom]

It's funny how the polling can be more interesting than the actual political race.

Sometimes, yeah.

 

[BluePhantom]

Obama Grand Total: 247
Romney Grand Total:239


Too Close to Call

Colorado (9), Iowa (6), Nevada (6), Ohio (18), Virginia (13)

Total TCTC: 52



As a side note: for practical consideration a tie will go to Romney as the tie is broken by the House of Representatives with each state getting one vote. In that scenario Obama would likely get a minimum of 21 votes and a maximum of 24. Romney would get a minimum of 26 and a maximum of 29.

Let's take a look at this for a second. The common wisdom is that the election will come down to six states (Nevada, Ohio, Florida, Virginia, Iowa, and New Hampshire), but if we accept the EV breakdown I have, which seems reasonable enough, then at the end of the day Obama would need 23 electoral votes to win and Romney would need 31 from those states I listed as "too close to call". Now let's give Obama Colorado, Nevada, and Iowa. That's only 21 EVs for a total of 268. That means Obama MUST take either Ohio or Virginia. Romney can give up CO, NV, and IA and reach 270 if he takes both Ohio and Virginia. Ohio is trending toward Romney right now and Obama's lead in Virginia is within the MOE. Obama's lead in Ohio is also within the MOE. In other words this election will really come down to Ohio and probably Virginia.

This is really good news for Romney supporters and it should absolutely terrify Obama supporters.

 

[PLYMCO_PILGRIM]

It's funny how the polling can be more interesting than the actual political race.

At least the polling isn't filled with all the political B.S.....or should I say the way phantom is running this thread isn't filled with political BS. Its just facts about what the polls say, nothing more nothing less....they say what they say.

I, for one, appreciate it.

 

[Foxfyre]

An excellent thread Phantom, and I hope we can keep it active all the way to election.

I think polling is instructive to the candidates at this time as they let them know how this or that rhetoric or tactic is playing in Peoria, and the polls do have a bearing on those tactics and the jargon we will hear in the campaign.

Polling in April and May isn't that useful as to who is likely to win in November as that will largely be determined on a number of things including the economy, whether any major scandal surfaces, or whether either candidate makes a major tactical blunder.