http://www.democraticunderground.com/discuss/duboard.php?az=show_mesg&forum=132&topic_id=6984237&mesg_id=6984751Surveys are WEIGHTED. The exact # of youth called doesn't matter only the % split of their vote.
Surveys are weighted based on age (and other factors). It will not affect the survey 1% if too little or too many young voters are called.
Looking at Gallup data (they only give weekly for free, and smaller breakdowns cost $).
Obama vs McCain (Aug 25-31)
18-29 54%-37%
30-49 48%-45%
50-64 49%-42%
65+ 44%-43%
How does Gallup (or any pollster) make sure they get the "right" amount of Youth vote? They don't need to.
Let's say 18-22 demographic makes up 10% of voting block.
When they conduct a survey of 1000, statistically they should get on average 100 18-24 year olds (10% of 1000). Let's look at two scenarios.
A) Too little of sub population.
==================================
Say on one survey they only get 65 18-24. Those 65 people vote 48 for Obama. That works out to a 74%-26% split.
B) Too much of a sub population
=================================
On another survey they get 118 18-24. Those 118 people vote 87 for Obama. That works out to a 74%-26% split.
Now you may think well in survey A Obama will do worse. Right? Well that would be Wrong. He would have the EXACT same final result in both surveys.
Surveys are weighted it doesn't matter the individual # of votes but instead the % (which remains the same) of what each sub-pop thinks.
Surveys get baseline data on (race, gender, age, marital status, zipcode which is used for rural vs urban, religion, party ID).
Each sub pop vote is converted to a %.
The sub pops are then weighted by how much they make up the electorate.
If 18% of voters are Black and they vote 94% for Obama then that 94% get multiplied by the 18%. This is done for each race to get a statistical representation of voting based on race. Once that is done the same thing is done for each other group (age, marital, etc). The final result is the cross product of each sub-pop (properly weighted).
The idea that if they call 1000 people and get too many women, not enough young people. and too many Hispanics they just publish that is laughable.
If you wanted to do a COMPLETELY RANDOM SURVEY with no weighting or sub-pops you CAN DO IT. However the smaller the sample size the larger the error. 1000 people out of 120 million on a 50/50 issue sampled will have a margin of error of +/- 18%. The survey would have no meaning. On the other hand if you wanted to get +/-4% with completely ranom survey you would need to call OVER 46K people.
All surveys used weighted sub pops to reduce the sample size to something manageable. They have been doing that for 60 years for elections (and 300+ years for other forms of statistical analysis).