Pollsters use a totally wrong, unscientific method. Typically, they survey between 700 and 1,000 people chosen randomly. Let's say these 1,000 people get chosen by equal distribution, that is, 20 in each state.
No one can project the results from a mere 20 persons to the universe of voters in any state. Thus, the number of surveys completed would be too small. Every statistician worth his salt should know this truth.
However, there is a far more damning reason why you should denounce all polls being foisted upon you. Because of the electoral college, the presidential election is fifty seven (57) separate elections and requires 57 surveys for full coverage to predict the full results. An accurate poll must account for the electoral college since presidential election doesn't get decided by the popular vote.
There is one election in each of 48 states that have winner-take-all elections for the electoral college electors. Two states, Maine and Nebraska, lack winner-take-all elections. Those states combine for nine elections.
According to The U.S. National Archives and Records Administration
The founding fathers established [the Electoral College] in the Constitution as a compromise between election of the President by a vote in Congress and election of the President by a popular vote of qualified citizens.
The Electoral College process consists of the selection of the electors, the meeting of the electors where they vote for President and Vice President, and the counting of the electoral votes by Congress.
The Electoral College consists of 538 electors. A majority of 270 electoral votes is required to elect the President. Your state’s entitled allotment of electors equals the number of members in its Congressional delegation: one for each member in the House of Representatives plus two for your Senators.
Under the 23rd Amendment of the Constitution, the District of Columbia is allocated 3 electors and treated like a state for purposes of the Electoral College. For this reason, in the following discussion, the word “state” also refers to the District of Columbia.
For results that have a margin of error ±4% at the 95% confidence anyone surveying would need to complete 44,688 surveys and all on the same day! Otherwise, you don't have a scientifically conducted poll with results that can get projected to the universe of voters.
The math: 784 surveys × 57 = 44,688
The United States Election Assistance Commission, which Congress established by the Help America Vote Act of 2002 (HAVA) as an independent, bipartisan commission, serves as a national clearinghouse of information on election administration and produces the foremost data collection effort related to voter registration, election administration, and the ways in which Americans cast their ballots.
The EAC defines two concepts:
- Voting Age Population (VAP): People who are 18 years of age or older, regardless of whether they are eligible to register to vote.
- Citizen Voting Age Population (CVAP): Persons who are citizens and of voting age (18 years or older). These numbers are estimates generated by the U.S. Census Bureau's American Community Survey.
We only care about the CVAP numbers. With the CVAP numbers, for each state, we can adjust the number of needed completed questionnaires to the count of legal voters in each state and still maintain our targeted margin of error with confidence. As it turns out, we can cut down the number of completed questionnaires to 600 for each state, except Wyoming, which we can cut down by one less to 599.
Here is what the completed questionnaire counts should look like:
Also, if some of the survey respondents complete a questionnaire before a candidate said something controversial, say freeze all immigration from countries with Mohammadans, while other survey respondents complete\ the same questionnaire afterward, the results cannot be used since the opinions are collected on different facts. Time-sensitive surveys like voting surveys must be done on the same day for all respondents to be accurate.
Now, during the 2012 presidential election campaign, a pollster would have needed to complete 9,000 questionnaires all on the same day to predict Obama the winner.
Nearly all presidential election polls have completed survey counts of 500 to 1,000. Thus, the results from such polls cannot be projected to the universe of likely voters, hence such pollsters report wrong results. In short, do not believe any of those polls.
Also, every time another variable gets added, doubling of completed surveys must happen. So if a legit pollster wants to know how Latina females plan to vote, such a pollster would need to complete a whopping 89,376 surveys on the same day within a tight news window!
Every poll ever reported on television and every poll ever printed in a newspaper regarding presidential races have been fabricated. Such polls were never scientific. Until polls adopt the scientific method of polling, all future presidential polls will be wrong.
Disclosure: I used to work as the director of public opinion polling and marketing research for a Dow Jones owned newspaper. My grad work at S.I. Newhouse School of Public Communications included public opinion polling.