October 22, 2004
Perhaps because I spend more time writing a blog these days and less time reading them, but I have seen very little attention given to this fact: In every poll which has released results for some definition of "battleground" states, John Kerry runs slightly stronger against George Bush in those toss-up states than he does nationally.
Ordinarily, I would advise caution in interpreting subgroup findings, as the smaller sample sizes come with considerably more sampling error. However, when we see a consistent pattern across multiple surveys, we can have a lot more confidence in the statistical significance of the finding.
Here is the list of what I have been able to cobble together (note that the definition of battleground states varies from poll to poll, from 12 to 20 states):
The pattern is consistent: In every case Kerry runs better in the "battleground states" than he does in the overall electorate, although in some cases the difference is quite small. Another survey that did not release specific numbers also showed the same pattern: John Gorman of Opinion Dynamics noted that his survey for Fox News showed a similar result: "One odd factor is that much of the lead is concentrated in the so-called 'red states,' which were pretty much conceded to Bush at the beginning. Thus his national lead does not reflect a big lead in the battleground states that will decide the election. We may well be facing a situation, as we did in 2000, where the popular vote and the electoral vote produce different results."
These findings raise a lot of interesting questions that I just don't have time to consider carefully. Do these findings suggest "the possibility," as the analysts at Harris put it, "that the popular vote and the electoral college vote may divide differently, as they did in 2000?" I also wonder how much of this result is concentrated in a few big swing states, like Pennsylvania, Ohio and Florida?
Any thoughts? Post 'em! As Dick Gephardt might say, "this is your blog too!"
Update: Alert reader RE sends this paragraph from a 10/18 release by Gallup: "Despite the lead Bush has opened up nationally, the two candidates are essentially tied among likely voters in the 16 competitive showdown states in which the race for electoral votes is being fought. This suggests that some of Bush's current national gains may in essence be 'lost' in states where the election outcome is fairly certain to be strongly for Bush or strongly for Kerry."
Thanks to IPSOS, ICR, Hart/McInturff and CBS for providing data.
10/22 - 11:21pm: Corrected erroneous Harris from original post
10-25 - Replaced original table with image file. Links now on the jump
Your last sentence echoes my thoughts: Could it be that Kerry's overall swing-state lead is mostly the result of a bigger lead in a bluish state like Pennsylvania outweighing small Bush margins in possibly reddish states like Wisconsin, Iowa, and New Mexico? I think I would be more convinced of Kerry's swing-state fortunes if I saw state-by-state data that showed him consistently up in the majority of these states -- if not beyond the margin of error than at least over a period of time. But those numbers probably don't exist. After all, there's a reason they're called swing states.
Posted by: Craig | Oct 22, 2004 4:03:40 PM
The more people see Kerry, the more they are reassured that he's presidential material. He's spending a lot of time in the swing states... therefore, they see him more, and consequently they are more reassured than people are outside of the swing states. That reassurance puts him in a better competitive position versus Bush. It's not much of a theory but it's mine...
Posted by: Anand | Oct 22, 2004 4:34:35 PM
If true, I wonder if there is a significant number of post-9/11 Democrats in the area around NYC (e.g. NJ, which isn't in play but is closer than the CW had it) combined with a small Governator-bump in CA (plausible but less likely) to partially explain this. It would indicate smaller margins for KEdwards in a few populous blue states without affecting the swing states (I know it's a reach, but you never know). I agree with Craig in that I am somewhat skeptical of the assertion that KEdwards is up generally in battleground states because a 3-5 point lead in MI and PN can outweigh BCheney leads in CO, NM, IA.
After all, there's a reason they're called swing states. Exactly. Bush seems to be up nationally, but not by a large margin. KEdwards seems to be leading the aggregate vote in the battleground states, but those by definition are the states that are close, and a two point lead in PN probably outweighs a close lead in NM and IA. Of course, that's why PN has more electoral college votes than NM or IA.
Anand may have a point. However, this assumes by "see Kerry" he means the people see KEdwards campaign ads, as opposed to debates which everyone everywhere in the country saw (no campaign ads in my corner of southwest VA). This would mean that his poll numbers going up should correlate with (1) the amount of time spend advertising in a state, and (2) the absolute amount of advertising in the state. I am not sure that is the case, as it seems like (and I could be wrong - correct me if so) KEdwards has had the aggregate lead in the battleground states for a while (excluding Rep post-convention bounce). Thus, the original explanation that PA and MI are greater than small Bush leads in OH and the smaller southwest states (FL seems to be a wash at this point).
Posted by: John Branch | Oct 22, 2004 4:48:17 PM
It seems to me that the implications for the future of the electoral college of another popular vote/electoral vote divergence would be profound.
Posted by: Steven Buss | Oct 22, 2004 5:04:51 PM
It's not so much that Kerry's such a great guy that seeing him more makes you want to vote for him. It's that nobody is as awful as one's political opponents say they are, so approval numbers tend to be higher among people who have met a candidate and noticed the absence of horns and a tail.
What does the divergence tell us? Campaigning matters more for an unknown person than for a well-known person.
I'm a little disappointed that the answer is so pedantic, too.
Posted by: Emcee Fleshy (D-Atlanta) | Oct 22, 2004 5:24:12 PM
Here's a link to a thorough examination of the separate polling in each of several swing states, with an eye toward how close Bush is coming to the 50% level, which is conventionally considered a threshold for an incumbent to be secure:
Posted by: Alan R. | Oct 22, 2004 5:25:28 PM
Your Harris Data are posted incorrectly, the correct data should read:
Bush Kerry Bush Kerryy
Har. #1 48 46 44 51
Har. #2 51 43 47 47
Posted by: PBJ Diddy | Oct 22, 2004 5:39:01 PM
IMHO, isn't the best approach here is to take the national pollsters' national values and the state polls' state values? Preferably with BOTH LV and RV data supplied.
Or if we're going to learn something from national poll breakdowns, why don't they list ALL respondents by party by state so we can learn something instead of coming up with a lump of around TWENTY "battleground states."
More nutty are designations by region. I've seen Kerry blowing out Bush in the West and Bush close to Kerry in the West. What do pollsters call the West? Is NM in the West? How about UT? CO? WY?
Again my opinion -- good state pollsters understand regional state dynamics and account for them by dialing numbers in areas with vastly different preferences.
For example, a pollster who just balances by party calling Florida might have a tough time with variance because a D in Miami is not the same as a D in the panhandle.
If a national poll of 600 gets 40 respondents from Florida, how helpful is that? Even if we take 10 national polls, my guesstimated FL sample size of 400 is less than ONE dedicated state poll.
The state polls of these "battleground states" are coming fast and furious, are done by many companies trying to balance party, region and other demographics, and have dramatically less MOE than these.
And they tell a much LESS clear picture -- those polls say that this election in the remaining, true battleground states individually could go either way. That's not the story a "17% lead" tells us.
Posted by: ER | Oct 22, 2004 7:41:54 PM
You got Harris LV#2 Battleground numbers wrong, it's Bush 44% Kerry 51%, you have it the other way around right now.
Posted by: dvo | Oct 22, 2004 7:50:35 PM
I think the 911 effect is certainly a factor in CT, NJ, NY. I think that takes around 10 points from Kerry straight away, although he should still take NY easily and NJ and CT. These states account for a significant chunk of US population, so that alone could account for 1-2 pts. Maybe the same in California, but state polls haven't shown it close either.
The overall idea of a battleground state poll is also a little confusing. Are AZ and TN considered battleground ? I don't think the are, especially TN ? What about WA and Oregon ? I don't think they are, especially WA.
Still, my persual of state polls tells me that the election is a lot closer than national polls would indicate. If the electoral map were akin to 2000, I would give an edge to Kerry since he only has to flip one small state (like NH). Now, given the census reapportionment, I think Bush has an edge.
Reent state polls I've seen show KErry with a small edge in PA, a miniscule lead in OH. Bush has a small edge in FL. Kerry has a small edge in MN and NH, Bush has one in NM. IA looks like a solid edge for Bush. Michigan looks like a good edge for Kerry (with 1 or 2 divergent polls). Wisconsin looks like a true tossup.
The election thus seems a lot closer than the national polls would indicate. -
Posted by: erg | Oct 22, 2004 8:23:11 PM
I consider it quite likely that the popular vote and the electoral results will be different. I've thought that for a long time, ever since Bush showed that in spite of what was said during the 2000 campaign and what he claimed just after the he took office that in fact he was going to govern for the political right instead of all Americans. I think this will also be the case in the 2008 contest between Jeb Bush and Hillary Clinton (assuming that they don't figure out a way to either run Dubya again or eliminate the election).
Posted by: Jim S | Oct 22, 2004 8:58:26 PM
If Dubya loses, he can run again in 2008. A divergence of electoral/popular vote probably makes that more likely, since it might create the impression that "he was robbed". Even moreso if the election goes to the courts.
Posted by: Elliot | Oct 22, 2004 9:03:41 PM
I don't think the aggregate numbers as shown are a reliable way to look at the situation: they hide too much information.
At the national level, Kerry does not enjoy the levels of support in CA and NY that Gore did. I live in CA and I think Bush is much closer to Kerry than any polls show. I doubt he'll win the state, but I wouldn't be surprised if he did. [As an aside, I live in the SF Bay Area and I'm not seeing the fervor among Democrats I did over the recall election or anything like it; the Democrats may be being complacent here and if they are - if turn out is on the low side - they may be surprised]. This, together with the fact he is running stronger in his core 2000 states, is boosting Bush’s national numbers.
But as for battleground states bucking the national trend, I think the numbers may be deceptive. The races are obviously closer here or they wouldn’t be "battleground states", but it seems to me that Bush is running stronger everywhere. It’s harder to tell at the state level, because there is more poll volatility and fewer polls than for the national data, and there are some clear outliers. If one were to rationalize the battleground state data and take into account the margin of error, I’m not sure the appearance that Kerry is running better in the battleground states than he is nationally would bear up. Looking at the RCP data, it appears that Bush has a lead large enough to be considered significant in 8 battleground states, Kerry has a significant lead in 5, and 4 are ties. Kerry has one state in which the trends may be against him, MI, and in FL, Bush has a shown a barely significant but stable lead in some polls, while others have bounced around, making trends hard to discern without going into much more detail. If you look just at Mason-Dixon, the most accurate pollster of 2002, the break down is: 8 Bush leads, 6 ties, 3 Kerry leads. Looking at the polls either way suggests that battleground state races look more like the national race — Bush with a small but significant lead — than the aggregate data suggest.
Posted by: C. Owen Johnson | Oct 23, 2004 5:45:53 AM
The comments by C. Owen Johnson regarding "battleground states" seem to be focusing precisely on what this column has said should not be the focus - that is, the polling margin. Instead, if one looks at the level of Bush support, it seems clear that Bush is in trouble in virtually all of these states.
Sam Wang's statistical meta-analysis of these polls indicates that although Bush leads in EV based on the poll margins, a shift of only 0.3% resulting from undecideds breaking for Kerry or from turnout, would change the result. See,
Posted by: Brent | Oct 23, 2004 11:35:09 AM
Given the current polling, I find it very unlikely that Bush will win the overall popular vote, either, but I think he may well lose by less overall than he does in the "swing" states.
Posted by: Nick Simmonds | Oct 23, 2004 1:03:25 PM
(slightly OT) MP, I found this recent piece on a CEA survey on cell phone use and poll response interesting - you might want to comment on it.
Posted by: bnh | Oct 23, 2004 2:25:10 PM
It is pretty apparent that all the polls do is reflect the last election. All of the weighting of likely voters, subgroups, regional voting, is based on the last election and the pollsters are trying to get their numbers to match the last election.
Could it be that in a race where Kerry's support is down in blue states as compared to Gore (e.g., NJ, Hawaii Iowa, Minn., even Cal and NY) and Bush's are up in red states that Bush will only have a 1 or 2 point lead in the national polls. There is something that does not make sense here. And it doesn't make sense because the pollsters are polling the last election.
On November 3 they will say that numbers changed because of some event that occured in the last week. The good pollsters will say we need to find a way to poll so that we detect the important changes to the electorate that occured since the last election.
And further affiant says nothing.
Posted by: Phil Dayton | Oct 23, 2004 4:37:09 PM
Popular/EC Disparity -- the 1860 model
If the two main parties are roughly evenly distributed nationwide, or at least if they both have roughly the same number of states very strongly partisan in their favor, then the EC result will tend to be the same as the popular result -- only more so. Even marginal popular results will be magnified by the winner-take-all rule for EC votes.
But as the parties become heavily regional, as they both become "national parties no more", as Zell Miller said of the Dems, the EC/popular relationship becomes unstable, and can produce highly disparate results.
The extreme case was the election of 1860. Lincoln got only 39.9% of the popular vote, but took 180 EC votes to the combined Dems' 123.
You sometimes here this result attributed to the fact that the Dems split 3 ways that year, with the implication that Lincoln squeaked by on mere pluralities in many or most states to amass his outsized EC lead. But this is completely untrue. The three Dems' support was itself so regionally splintered that only one of them tended to have any strength in any given state. Only in OR and CA did Dem splintering give Lincoln the plurality, and thus the EC votes for those states. Douglas, with 60.1% of the popular vote, would have lost 129 EC votes to 173, even if he had been the Dems' only candidate, and gotten every vote actually cast for himself, Breckenridge and Bell.
This happened because the Reps in 1860 were the marginal majority in about 3/5 of the country, but had very close to zero support in the other 2/5. The Dems were the only game in town in 2/5, and competitive, but not a majority, in the other 3/5. No EC cigars no matter how close in popular vote they were in several midwest states. The winner-take-all feature of the EC system meant most Dem popular votes were wasted, while the Reps had no popular votes to speak of in 2/5 of the country, that were vulnerable to "loss" to the winner take all rule.
As noted, the red states this year are deep red, thus reducing the Dem popular votes "wasted" in these states, whereas the Dem is leading marginally in the battleground states, "wasting" many Rep popular votes there. Perhaps enough for us to have, in this one year, both a hefty Rep popular win, and a hefty Dem EC win.
Posted by: Glen Tomkins | Oct 23, 2004 9:37:55 PM
The problem with your analysis is that the # of LV in swing states in the sample size is 20-30% of 600-1000 LV. So MOE is very high. I don't know how you can make any claims about the swing states voters just based on nationwide polls. By the way, any of the pollsters who say there are 17-20 swing states is not keeping up with current news (like Ipsos and Marist).
It makes more sense to look at the various individual swing state polls. You can aggregate them into a larger swing state poll if you want. But what you did to me is just plainly bad.
Posted by: Cableguy | Oct 24, 2004 1:22:40 AM
Regarding Brent's comment above: I see Sam Wang's statistical meta-analysis of these polls as supporting my contention rather than suggesting Bush is in trouble. As a physicist and engineer, I see some problems in some of the assumptions underlying Wang's methodology, but I'm willing to take his word [since I don't have time to do an independent analysis of his method] that his method is robust.
My point is that the polls are mix of apples and oranges, and suffer from systemic errors that can't be reduced, and this makes conclusions based on aggregated numbers unreliable. My approach was to try to get to some apples to apples comparison and use recent history and other correlated indicators to try to estimate the systemic errors. Also, I don't think the "incumbent rule" applies to this race because the data on which it is based are not relevant this year. So I also considered the "incumbent rule" on a state-by-state basis, using past polling data in considering how the "undecideds" might break.
Being the good engineer I am, I did this rather back-of-the-envelope, but the conclusion was that looking at the data as presented underestimates Bush's level of support by about 2 pts in key battleground states [FL, OH, MI, MN, PA, & WI] and by 3 or more nationally. Those are my conservative numbers.
Note: I don't have any reliable data on how new registration efforts might effect turnout in this election, so I assumed it would be weakly correlated. As has been widely pointed out, the Dems did better in 2000 than expected and the Pubs did better in 2004 than expected. Wang, for one seems to feel that activism will drive Dems to the polls this year in greater numbers, but others feel the opposite. I, for one, have noted Dem "campaign fatigue" where I live coupled with a general low-level distaste for Kerry that may be making voting a "chore" for Dems. The Pubs don't seem to feel that way. But what the feeling is like in key states is tough to know. So to be conservative, I took my numbers down a bit to reflect either a slight edge in Dem turnout or my estimation of the Incumbent Rule being off.
Posted by: C. Owen Johnson | Oct 24, 2004 5:43:39 AM
So, Owen, what was your methodology on the back of your envelope. How did you reach the conclusion that Bush's support is understated?
Posted by: Brent | Oct 24, 2004 9:53:29 AM
I don't think that most of these battleground states are really that undecided. You say there are sixteen, but I think the only real questionmarks are Florida, Ohio, New Mexico, Minnesota, Iowa, and Wisconsin. It looks like Kerry will grab NM, and possibly OH, while Bush will probably get WI and possibly FL. That leaves only a couple up in the air, really.
Posted by: Randomscrub | Oct 24, 2004 11:07:45 AM
Meta-analysis isn't simple aggregation
Simple aggregation of data (add up all the Bush vs all the Kerry results, figure your aggregate MOE from the total sample size) is a reasonable approach only if the studies you are aggregating are roughly equal in size, and have roughly the same methodology. Insofar as the actual conditions stray from that model, you should use the more careful and conservative aproach of meta-analysis, despite its much more complicated calculations.
Meta-analysis treats every study, or poll, in this case, as a separate testament to the underlying reality. Data isn't simply aggregated, because then the results of smaller polls would be inappropriately swamped by the larger studies. Larger polls do get a greater weight, because their random variance from the underlying reality is less than smaller studies, but this weight is far less than would obtain with simple aggregation. The MOE definitely gets smaller with each poll brought into the meta-analysis, but here again, the reduction in MOE is more conservative than you would get with the statistic applicable to simple aggregation.
Whatever the merits of Dr. Wang's assumptions about assigning undecided voters, his use of meta-analysis, rather than simple aggregation, is clearly a conservative measure, that tends to understate the strength of his conclusions, compared to computing the results of those assumptions about undecided voters by simple aggregation.
Posted by: Glen Tomkins | Oct 24, 2004 2:13:22 PM
What about the Nader factor? It could again be decisive. Nader may have long-standing gripes with Democrats, but another Bush victory will continue to undue all that Nader has striven for. Someone from either the Nader or Kerry teams should start the ball rolling to have Nader bow out gracefully. Alternatively you can send e-mails to Response5@votenader.org. I think that's where his team records negative comments. Don't curse him, that won't help.
Posted by: Steverino1 | Oct 25, 2004 10:27:47 AM
I think that part of what you're missing is that there are more "Kerry Battleground States" than there are "Bush Battleground States", and that is not good for Kerry.
Bush's Electoral College "base" is greater than Kerry's, so Kerry has to win more of the "battleground" than Bush does. Further, from what I've seen the majority (population wise) of the battleground states are states Gore took in 2000.
Saying that Kerry is leading in Gore states is nice for Kerry, but it's not sufficient. Because if they take the same states as last time, Bush wins by 18.
Posted by: Greg D | Oct 26, 2004 2:37:28 PM
The comments to this entry are closed.