Tuesday, February 21, 2006

Good Luck Hilary

In a recent poll that has been splashing the newspapers yesterday and this morning, it was revealed that the United States is apparently ready for a Female President. According to these sources; the S.F. Chronicle amongst others, 79 percent of the participants in this survey were willing to vote for a Female candidate. What they so mention but fail to extrapolate upon is that the sample size (number of people who participated in this survey) is astronomically small. In most polls there are three things that are encouraged; randomness, wording and size.

The randomness issue is in reference to the sample; the participants are randomly selected from a larger group known as the population and the smaller selection is known as the sample. For example, the sample is meant to represent an entire state (the population) and the testers/pollsters select 100 people from one major city in that state (the sample). In this example, it is a very poor sample to represent the population the test is generated for, a random sample would have included other areas. You then come across terms like “blind” or double blind, normally this is generally referring to a study or poll where there is some sort of experiment. There are usually two groups, one where you apply the experiment to and one where you don't. (In medical drug trials it is usually something like Group A gets a drug and Group B gets a sugar pill) The unadulterated one is known as the control group and the other the test group. In a blind test the participant is unfamiliar which group he or she is in and in a double blind test neither do the testers. In polling this is hard; the only real “test” you are giving are the questions to see the reaction of the participants, however this leads us to the second issue - wording.

The wording of the questions are very important. Most pollsters and polling groups are paid by parties or candidates and the pollsters know what their varying bosses want to hear. So they will tweak the wording of the questions to generate the results that they want, either consciously or unconsciously. The general term for this is measurement error, where there is some error in the results as a result of something wrong in the test itself. For example, if I am a staunch conservative and want to prove that the U.S. Population is against abortion, my polling could have a question like; “Are you for or against the murder of unborn children?” where a question worded simply, “what is your opinion on abortion?” with a small list of possible responses could have been used. This is an extreme example but even in certain polls the decision to use one simple word or grammatical phrasing can alter results entirely. Other ridiculousness that comes about in the polling world surrounds the fact that you can change a poll any way you want to prove what you want. In certain polls those that examine the data will change the responses. Let us say that in a certain poll those being questioned were given the option to provide five responses; Strongly Favor, Favor, Undecided, Oppose, and Strongly Oppose. If the results are too similar or not providing the gravity that the pollsters want it to, they will simply group the Strongly Favors with the Favors, and the Opposes with the Strongly Opposes to only three possible responses. Not only does this alter the significance of the responses, it causes another error; those who responded to these questions could have answered differently had these been the original options. For example, an individual could have answered “Favor” to signify “Slightly Favor” but could have been so unsure that if the options were only three, he may have been geared more towards the middle ground since it now had more significance to the question – with five options there is more room with only three the answers mean more. It is a little hard to explain what I truly mean here, so I will try to put it another way; in a number response system as in 1 to 5, where one is Strongly Favor and five is Strongly Oppose, there is more of a gradient than for 1 to 3. In reference to size, it only makes sense that the larger the sample size of the population the better it represents that population.

If the population is say 295,734,134 and the sample size is 1,120, that is a rather small sample, that is approximately 0.0004%, not even a half of a percent. For those who wonder where I pulled those numbers, they are the totally population of the United States and the size of the sample used in the poll these papers are referring to. The sample was all from New York and grossly inadequate to make such claims as the Newspapers are attempting to. The wording I cannot comment on for I cannot find a copy of the poll, but, based on these other facts, it would seem that this poll does not show that America is ready for a female to run America, but that New York seems to be ready.

Here is the point; in the words of Public Enemy, don't believe the hype. If you see a poll or a survey out there remember to look at three things; randomness, wording and size. All those talking heads on Fox or MSNBC or other shows that have “call in polls,” the thing you should remember about those is that the population that they are sampling is not America, but the viewers of their programing, and the sample they get are those with enough time on their hands to call into the show and (in some cases) waste their money on the cost of the call.

Here is the disclaimer; I would love to see a female President, however, I would also love to have the media not treat us like idiotic children that will believe anything with the words “poll” in it.

No comments:

Post a Comment