A Note on the “Margin of Error”

November 2, 2004

I was watching Brit Hume’s show on Fox last night and he was interviewing some expert on polls, margins of error, etc. and it didn’t all ring true to me. I am no statistician, but common sense tells me that what this expert was saying (and what many others seem to say) is just plain wrong. Brit was asking him why there was so much variation in the polls and what the significance was. The guy basically said (forgive the oversimplification) that these polls are all within the margin of error and therefore their results are statistically insignificant.

If I understood him correctly he was essentially saying that a poll showing Bush ahead three points was statistically the same as a poll showing Bush behind three points.

Now that just can’t be true. I think what the margin of error means is that the results “could” vary the amount of the margin of error, not that they are likely to and not that the raw numbers are irrelevant. In other words, I think two separate polling outfits with their own polls, each with a margin of error of three points and each with Bush ahead by three points are statistically equivalent, obviously.

I don’t think that a poll with a three point MOE showing Bush ahead by 3 points is equivalent to a poll with a three point MOE showing Bush behind by 3 points. If in one poll Bush has 51% that means he could have anywhere from 54% to 48%. But if the second poll shows him at 48%, doesn’t that mean he could have anywhere from 45% to 51%? I have to be right about this.

Another thing on this. Bush has been running consistently ahead by 2 or 3 points for some time now. If the “experts” theory were accurate, i.e., that his lead is statistically insignificant, wouldn’t statistical probability lead to Kerry being ahead half the time too? If these small, but consistent leads, made no difference, why is it that Bush was almost always ahead? I think the answer is that he was and is and this psychobabble on the MOE is liberal media spin — I’m not here referring to my main man Brit Hume now.

But I just think the company line about the poll results not mattering if within the MOE is B.S. I’ll guaranty you this. John Kerry would rather have had George Bush’s poll numbers the last few months than the other way around. Wouldn’t anyone rather be at 51% than 48% even if the MOE is three? Of course. Now there may be some technical flaws in my analysis, since I have no training in statistics, but I bet what I’m saying is a lot closer to the truth than what the “experts” who say these results are statistically insignificant are saying.