Why SUSA's Pennsylvania poll isn't worth fretting about

[BeyondGeography]

Some excerpts from a good diary at Kos on how Obama's floor in Pa. won't go below 44%:

The arithmetic is very simple.

At a minimum 20% of the democratic primary voters will be African Americans and at maximum 80% of the democratic will be White. Now at a minimum Obama will win 80% of the African American voters and at a minimum he will win 35% of the White voters.

Doing the very simple arithmetic:

Obama vote = <(% African American voters) x (% African Americans for Obama) + (% White Voters) x (% Whites for Obama)> / 100

Obama vote = <(20) x (80) + (80) x (35)>/100 = 44%

Assuming 20% black turnout in a state where the black population is only 11% or 12% sounds risky, but this is a closed Democratic primary, so blacks will be over-represented as a % of the total. Moreover, the SUSA poll only puts black turnout at 14% and they give Hillary an abnormally high number of black votes. The 14% figure is obviously low. Blacks are around 12%-13% of the total population in Ohio and they represented 17% of the total vote there, and that was an open primary. Here's a strong comment on the whole thing:

SUSA had the following breakdown in their poll:

Black voters were 14% of respondents, and they went 74%-24% for Obama.

White voters were 82% of respondents, and they went 61%-32% for Clinton.

Given the 597 likely voters they talked to, that's 84 black people surveyed. Of those 84 black people, 62 were for Obama, 20 were for Clinton, 1 was undecided, and 1 was "other".

I'm no polling expert, but my guess is that the 84 person sample size has a pretty massive margin of error? Looks like SUSA undercounts his black support (only 14% of respondents does seem low for a closed primary).

http://www.dailykos.com/story/2008/4/8/13493/86829/1005/492258

The full poll with crosstabs:
http://www.surveyusa.com/client/PollReport.aspx?g=c79e5bab-a424-49f6-86d6-50c61cf729b7

[thevoiceofreason]

1. Any cross-tabs regarding oversampling of the black vote to verify?

If not, I agree - garbage in, garbage out.

[BeyondGeography]

3. Added...thanks

[Barack_America]

2. Thanks, I was looking for those crosstabs...

Those results did just seem off.

Around here, it also makes a HUGE difference where you're polling (i.e. which county), so yeah, I would imagine that a sample size as small as 84 would have enormous error.

[democrattotheend]

4. I disagree with that post. AAs should be about 15% of the vote

But I do expect the African American vote to become more polarized, so Obama should do a few points better than the SUSA poll. Otherwise the poll looks dead on to me, unfortunately.

[BeyondGeography]

6. One poster said blacks were 18% of the total vote in PA in 2004

and that was a general election. Of that's true, 20% is well within reach for a closed primary. Can't find a link for that though.

Why 15%?

[MadBadger]

5. I disagree, AA vote will most likely be below 20%.

[DemocratSinceBirth]

7. They Are 10.7 Of The PN Population

To get to 20% you have to assume they are all Democratic and they will all vote in the primary...

[thevoiceofreason]

9. They are motivated. They are over 95% democratic

So you really do get close to 20%. Safe side - 16 - 18%

[thevoiceofreason]

8. Crosstabs show no oversampling of AA vote

High MOE with only 84 AA surveyed - likely in the 10% range. Thus, AA vote could be 84-14 to 64-34. Garbage.

What causes a swing of 18% or so with one age group? Nothing. Let's see what these rocket scientists do next week.

[DerekJ]

10. Why are you guys upset it about it?!!

I have to say that lowering expectations are crucial at the moment.

Having said that, in a state like PA, I believe a sample of 597 people is pretty low.

You need at least 1200 - 1400. While mathematically the margin of error my drop only a point or two (compared to the 597). Demographically it could mean a swing of over 10 points.

Meh.. just my 2 cents