Articles and Analysis


POLL: InsiderAdvantage Pennsylvania Dems


Clinton 48, Obama 38



Ok - now which Insider Advantage do we believe? The one released 5 days ago was only 3% difference and this one is 10%?

I don't know which one to believe but - the crosstabs show wildly different sampling universes:

30-44 age bracket: Obama/Clinton was 46%/41% on April 3 but was 36%/49% on April 8 - that's a 10 point loss for Obama in 5 days and an 8 point gain for Clinton.

Black age brackets are wildly different between the two polls - as an example:

White/Black percentages for 18-29 year olds April 3 77%/20% - April 8 57%/41% - but Obama did slightly better in that "heavily white" age group on April 3 (70%) than in the "leaning white" same age group on April 8 (67%)

And Male/Female age brackets are wildly different between the two polls for age brackets 18-29 & 65+ yet not much change in the preference percentages.

So, my question is - which sample universe (for those in PA that would know) is more correct?



If all the polls agreed, we wouldn't need or be interested in polls. Outliers are good for the industry as a whole in that they keep the interest level up. Still, the individual pollsters ought to know that a disciplined methodology is their best approach.

I also get a little irritated when the pollsters editorialize or try to draw conclusions from the data they've just gathered. It makes me think Heisenberg is alive and well.

The situation in PA is really unique, and after reading some of the comments on the latest Rasmussen post, I have to agree that we tend to think (mistakenly) that the choice of a candidate is a rational process, and the numbers can be directly extrapolated from events in the campaign, press coverage, etc.

What makes this all so interesting are the imponderables. E.g., how many tired, blue-collar Democrats in PA might just stay home -- if they feel that voting for Hillary isn't going to make a difference - while the energetic Obamanites are more likely to make it to the polls if they see the gap open too wide. Or vice versa.

Among the recent polls, my vote goes to Quinnipiac, for sample size, familiarity with the region, distribution of respondents, and their apparent lack of agenda.



It is confusing too.

USElectionAtlas.org had PA in the dark red (strong clinton) for the longest time. Then with that week of polls showing sub-5 point leads, a tie and an apparent 2 point Obama lead it went from light red (slight clinton) to gray (toss up). Then last week a couple new polls came out showing Clinton in the 2-5 point margin and it went back to light red, and then the recent 10 and 18 point polls put it back into the dark red (strong clinton).

I think the problem is that they are OVER polling PA. PA is the new Iowa and a new poll comes out every day. There is always something new in the news about the candidates so I think you are bound to get a few converts every few days.
Wright deal - minus Obama, plus clinton
Obama speech - plus Obama, nochange clinton
Bosnia scandal - nochange Obama, minus clinton
Clinton tax returns - nochange Obama, slightly plus clinton
etc. etc. etc.

There is just too much crap that goes on from day to day to really take these daily PA polls seriously.



Have to echo a couple of the comments above. I find it annoying to have a pollster release a poll that is clearly an outlier, followed by another that is within a reasonable range of others' results (or vice versa) and claim that massive real change has taken place in a week. A bit less hubris on the part of individual polling organizations would go a long way toward establishing their credibility with me. (And the Heisenberg reference was especially apt and amusing, by the way.)



Speaking of PA, a new Strategic Vision poll conducted April 4-6 finds McCain ahead of Obama 48%-41% in this state.



I think we are missing the forest for the trees here ...

My (original) question still stands ... two InsiderAdvantage polls in PA have come out in the past week.

April 3 poll had this 45-42 for Clinton
April 8 poll has this 48-38 for Clinton

I understand, for all relevance they are both within each's margin of error (of 3.6).

My problem is the sampling of subgroups are quite different when you look at the tabs. I'm not saying either is more correct than the other. What I am asking is - which sampling universe FROM THE SAME POLLSTER is more correct, since they are quite different. Examples:

18-29 Black sample for 4/3 was 20%
Obama had 70% support

18-29 Black sample for 4/8 is 41%
Obama has 67% support (Black percentage doubles and he loses 3 points?)

18-29 Female and 65+ Female samples also are quite different between the two polls.



Im not sure what to tell you Thatcher. I looked at the PDF data for both polls, and for the most part most of the values are rather similar. And those which differ seem to go AGAINST the trend you would expect. For example

4/3 - AA aged 45-64 - 35.3% (40 people)
4/9 - AA aged 45-64 - 23.0% (26 people)
(I would expect fewer older AA's would favor Obama)

4/3 - Of the 72 people aged 18-29, 77.7% white, 20.2% AA (56 and 15 people respectively(
4/9 - Of the 75 people aged 18-29, 57.1% white, 41.1% AA (43 and 31 people respectively)
(Again, I would expect more young AA's, two demographics which favor Obama, would swing the poll in Obama's favor)

Finally, the only other stat which really stands out is Male/Female ratio among certain demographics. Im guessing this is why you have the 7 point overall poll swing (+3 to +10):
Aged 18-29:
4/3 - 60.9% female (41 people)
4/9 - 34.9% female (26 people)
Aged 65+:
4/3 - 57.9% female (99 people)
4/9 - 62.6% female (111 people)
4/3 - 57.8% female (65 people)
4/9 - 45.8% female (53 people)

It looks to me that they used fewer young females and more older females, overall - and older females seem to be consistent in their support of Clinton.
a 20 person swing (meaning +10 votes for one candidate, -10 for the other) is a 3% difference.
a 30 person swing (meaning +15 for one, -15 for the other) is a 4.5% difference
a 40 person swing (meaning +20 for one, -20 for the other) is a 6% difference.

This polling company does seem to leave out some important demographics, at least as far as PA is concerned. When Franklin & Marshall college polled me over the phone, they asked what county I'm in, my annual household income, my level of education and my religion: http://edisk.fandm.edu/FLI/keystone/pdf/keymar08_1.pdf
I think the most important being region. If they polled people mostly in Southeast PA in the 4/3 poll, then I think you would expect to see higher Obama numbers. On the other hand, if they polled more people from rural PA, Pennsyltuckey for lack of a better term, you would expect to see better numbers for Clinton.

If you look at that F&M poll I linked, you will see:
19% - Central PA
16% - Southeastern
16% - Allegheny (Western)
14% - Northeastern
13% - Philadelphia
12% - Southeastern
10% - Northwestern

Considering that SE PA encompasses both the immediate outskirts of Philadelphia, which I would say is a 10-15 mile radius into Montgomery, Delware and Bucks counties (which is not even half of these counties population wise) and the Clinton-friendly areas which are the remaining areas of those three counties which are borderline rural, you cant just put all 16% of SE PA in the Obama-favorable column. In other words, i would put this poll from F&M as something like 80-20, 80% of polling data taken from Clinton-friendly areas, and 20% from Obama-friendly.

The point im trying to make is that it is quite possible that the 4/3 poll was maybe something like 60-40 (Clinton-friendly areas to Obama-frienly) where as the 4/9 poll was more like 70-30. Since the regional information is not included, or probably not even asked, we will never know.



In March Franklin and Marshall did define its sample as tokar describes:

19% Central
16% Southeast
16% Allegheny
14% Northeast
13% Philadelphia
12% Southwest
10% Northwest

Previously in January they defined it this way:

33% Central
19% Southeast
12% Northeast
11% Allegheny
11% Northwest
9% Southwest
5% Philadelphia

In January they had Clinton at 40%, Obama at 20%.

In March, they had Clinton at 51%, Obama at 35%

By contrast - PPP, the north carolina pollster,
defined its early april sample this way:

northeast = 9%
southeast (Philly and the counties topar is talking about) - 44%
south central - 11%
west central - 7%
southwest - 29%

They don't list their second April sample.

In early april clinton was 62/26 in the northeast
she was 37/53 in all of the southeast.
53/30 in the west central
and 45/39 in the southwest.

If you agree that 70 percent of the Pennsylvania vote will come from the southwest and southeast you can see why PPP thinks the race will be close and Obama has a shot at winning.

If you think that 70 percent is too high, you get the numbers SurveyUSA is getting.

Now that Obama is going out to Levittown and the Northeast, it will be interesting to see what happens to the SurveyUSA poll and its assumptions about turnout.


Post a comment

Please be patient while your comment posts - sometimes it takes a minute or two. To check your comment, please wait 60 seconds and click your browser's refresh button. Note that comments with three or more hyperlinks will be held for approval.