Wednesday, October 11, 2006

A *Little* Margin Of Error

Have you seen this? A study (done by a sample poll) determined that between 600,000 and 650,000 citizens have died a violent death in Iraq since March of 2003. However, other sources list the deaths at around 200,000 for one, and Iraq Body Count has the estimates between 44,000 and 49,000. That's a very LARGE discrepency. How did that come about? Simple. The sampling technique.

The researchers found 629 deaths after sampling 1,849 houses averaging 7 people per household. That's 12,943 people out of the approx 25.5 million people who live there. That means only 0.05078% of the population was sampled. That means that every single error that they make in their calculations is multiplied 1,970 times. So it doesn't take long out of those 12,943 people to have the number of deaths sky rocket.

Another error that this survey makes is what I like to call the "Accountant error." This happens when you ask an accountant to do a study to see the outcome. The accountant then says, "How do you WANT it to come out?" In other words, the study itself is arranged to show the results that suit your needs. For example, I could go to 1,900 families in the non-violent regions of Iraq and get a severely low estimated number.

Liberals have this weird thought that sample polls are absolutely right, absolutely all of the time. The same thing happened in 2004 when exit polls showed that Kerry would have won by a slim margin, but in actuallity the results were the opposite. People then claimed that the election was stolen, and we still hear the rants and raves about it now. The fact is that sample polls are not as accurate as they would lead you to believe. Variances in different areas that were not sampled can change the outcome dramatically. I would like to see the areas that those researchers surveyed to see if they were spread about the entire country, or concentrated in one specific area.

The study can be found here, and check this out. 12 of the 47 clusters they studied were in Baghdad, and 25 of the 47 clusters were in the 4 most violent of 19 sections of Iraq. 53% of the people surveyed were in 21% of the country. Seems a bit shady to me.

Cross Posted at GOP and College

No comments: