Everyone does straw polls
Straw polls need to be kept in perspective. To a professional who takes pride in producing highly accurate polling results, or to a pollster's client who is relying on the data to be correct, straw polls are an abomination. In contrast, for a casual observer of the world, straw polls can be fun and perhaps even somewhat informative.
We all conduct straw polls on a small scale as we chat with friends and relatives, at home and at work, at the bar and bowling alley, at the health club and hair dresser. The subject of the poll may be the weather, the local baseball team, or the way the president is doing his job. The topic need not be momentous to be of at least passing interest.
Radio call-in programs are very popular. Listeners enjoy hearing what others have to say, and occasionally call the program themselves. Everybody understands that they are not hearing the views of a randomly selected sample of the entire listening area, and that is just fine with them. Straw polls on a larger scale ask people to dial one or another 900 number, or vote at an Internet site, to express their preference on a particular issue, such as the best running back in the history of professional football. Millions of people have spent 50 cents a call in these polls to add their voice to the chorus of what they consider to be right-thinking Americans. They vote, not because they think the outcome matters, but because they find straw polls an engaging pastime.
Random sampling is based scientific principles
Just as we all conduct straw polls, each of us also does random sampling. We understand intuitively that the nineteenth-century Belgian statistician Adolphe Quetelet was right when he said it is not necessary to drink the whole bottle of wine to know how it tastes. Everyone knows that blood tests require (fortunately) just a sample of the body's blood.
Another oft-cited example is the cook taking a sip of soup to see how it tastes. It is best to stir the pot first to be sure the flavors of all the ingredients are sampled. If that is done, it does not matter whether the soup is in a small pot or a large cauldron; a single sip is enough to determine how the soup tastes.
The same principle applies for choosing a sample from a group of people. The universe could be Utica, New York, or the United States of America. If the soup is properly stirred it does not matter how big the pot is, because the same rules of random selection apply.
Rules of probability
Those rules are based on computations of probability. Flip a coin and there is a 50-50 chance it will come up heads. Flip it a second time and there is still a 50-50 chance it will come up heads. The laws of probability say that each time a fair coin is flipped, the chances are 50-50 that it will come up heads, even if it has already come up tails five times in a row. The laws of probability cannot predict individual occurrences, but they can help calculate the odds over the long run. After just five tosses there may be five tails and no heads. But if the coin is flipped enough times, in the long run the result will be 50% heads and 50% tails.
Random sampling works on the same principle. No matter how large the universe may be, if every member of that universe has an equal chance of being selected on each draw, then if a large enough sample is drawn, the characteristics of the sample will be very much like those of the entire universe.
How large a sample is large enough? It can be demonstrated mathematically that for a randomly selected sample of 400, the sample will come within plus or minus five percent of the actual percentage that exists in the universe. This ±5% is called the sampling error or margin of error. It does not mean that mistakes were made in drawing the sample. Rather it acknowledges that the laws of probability at work here are based on chance. Flipping the coin 400 times should produce heads 50% of the time. But because chance is involved, the result might be anywhere between 45% and 55%.
The same will be true of a random sample of people. If 50% of the U.S. population prefer hamburgers and 50% prefer hotdogs, any random sample of 400 should turn up roughly those proportions. However, any given sample could be off by as much as five percent in either direction.