Texas polls differ depending upon minority sampling ...

[Alhena]

On mydd.com, NJ Independent made an interesting observation comparing the SUSA Texas poll and the Insider Advantage Texas poll (which showed Hillary with a 4 point lead):

The IA crosstabs seem wrong to me. They have nearly as many Hispanic voters as White (38% vs 37%), with Blacks far behind (22%). By contrast, SUSA shows a much larger proportion of Whites relative to Hispanics (47/28/21). On the other hand, IA only has 52% women whereas SUSA has 56%. I'll defer to Ralph of IVR-Polls for a definitive answer on this, but my inclination is that the Survey USA crosstabs are probably closer to reality.

- end quote -

It seems to me that these polls are fairly meaningless without a better indication of what the true minority sampling will be in the race. If you have the Insider Advantage poll with 37% Hispanics and another with 28% Hispanics, there's no reason whatsoever to expect comparable results.

[NJSecularist]

1. I pointed that out before - and really, that is the case with all polls

It all depends on how each demographic is sampled. In 2004, Latinos consisted of 24% of the electorate in the Texas Democratic primary. Even with a huge increase in Latino turnout, I could see 28% of 29% of the Latino being electorate, which is what Survey USA is protecting. But 37% of Latinos? That is just a sampling error, in my opinion. There is no way we'll see that many Latinos at the polls.

SurveyUSA has absolutely nailed the % of each demographic in most of their polls. Look at their California poll, where they correctly predicted a larger than average Latino turnout when nobody else did. They have finetuned their process and I take their results more seriously than most because of their reputation.

I throw out the Insider Advantage poll simply because of the fact that a) they vastly oversample Latinos and b) they have a very shoddy record this primary season.

[maseman]

2. You must remember something

Polling companies do what is called weighting. It is impossible without using quota sampling (calling a certain demo until you reach your goal) to hit an age demographic, gender demographic, ethnic demographic, etc.

Let me give you and example. When I do our research projects to project results to a population we look at age, gender, ethnicity, geography by county and sometimes household income.

If the ethnic breakdown in a market is 50% other (basically what we call all non black or Hispanic ethnicities), 30% black and 20% hispanic. We might get back random sampling that is 45% other, 25% black and 30% hisapnic. When this happens we "weight" the data to mirror the population.

We would make the "other" surveys worth more to match the 50% population estimate. We would make the "black" sample worth more to mirror the population and the "hispanic" sample would be worth less to mirror the population.

Nearly all survey research uses weighting to mirror the population estimates. If you do a true random sampling the numbers will be always off. Therefore you must weight the data in order to make it work out to mirror the population.

[NJSecularist]

3. With any type of weighting, I'm puzzled how they reached the conclusion that we would see 37% Latino

turnout.

[maseman]

5. I don't know either

Seems very off to me as well. If you randomly select people for polling or surveying it shoudl be fairly close to the population estimates.

[TexasObserver]

7. They think Latinos will vote a much higher % than ever.

They will, but it won't be that much higher, because whites and blacks are also voting in greater numbers.

I expect both blacks and Hispanics to be about 24-27% each of the Democratic vote. Blacks vote 85-90% Dem, while for hispanics it is more like 60-65% voting in the Dem primary.

[TexasObserver]

4. The Texas Demographics for these groups

1. There will be more women than men vote in each demo.

2. There will be more whites than black or Hispanics.

3. There will be similar numbers of blacks and Hispanics voting.

4. Women will give the edge to Clinton.

5. Men will give a larger edge to Obama.

6. Old will go Clinton, middle aged will be split pretty even, under 35 will go Obama.

7. Cities generally to Obama, except South Texas. Rural generally to Clinton, except East/Coastal Texas.

8. Brownsville, McAllen, Harlingen, Laredo, Del Rio, Eagle Pass, El Paso to Hillary.

9. Houston, Beaumont, Port Arthur, Galveston, Austin, Dallas, Ft. Worth to Obama.

10. San Antonio and Corpus Christi - close to a split.

11. West Texas and panhandle to Hillary.

12. Delegate count goes to Obama.

Roughly, it will be 25% black, 25% Hispanic, and 50% White/Asian/Other.

In delegate counts, all things equal, the edge goes to Obama, because the black districts get the most delegates and the Hispanic and white stronghold districts get the fewest delegates.

[NJSecularist]

8. Good summary.

Thanks for the summary. I'll bookmark your post for future reference, as I'm not really familar with Texas's demographics. I've been focusing more on Ohio and what Obama needs to do to win there.

[TexasObserver]

10. Thanks. As I say, I don't try to call any state but Texas.

The place where I think most of the polls being run here are deficient are the following:

*Over represent in their samples those who voted in the last Democratic primary/primaries.

*Badly under represent young, first time, and indie/crossover voters.

*Too much landline polling, too little or no polling of cell phone users.

*Over representing older.

Add it all up, and they're underestimating Obama by 5-14 points.

[adoraz]

6. very interesting, hey since you know the tabs can you figure out

what IA's overall percentage would of looked like if they did 47/28/21 ratios instead of the ones they have now?

[NJSecularist]

9. I have 46-44 in favor of Hillary

Obviously not that much of a difference, but a higher than expected black turnout in addition to him bridging the gap between him and Hillary in the Latino vote could give the state to Obama.