Another interesting article out in today's Hill newspaper. Democratic pollster Mark Mellman, who has worked for Senator John Kerry, among others, debunks the myth that polling is an exact science, but rather a methodology not understood by many.

The article is a must read (I know I've had a lot of must reads lately, but bear with me) as it passes along a lot of important information. Although journalists and pundits alike love quoting polls ad nauseum, most do not possess sufficient understanding of statistics to properly analyze, let alone read the multitude of polls that come out each month. What is more, Mellman explains that though there is significantly more polling available today than at any previous time, more doesn't necessarily mean better or more exact.

The two main critiques made by the author are as follows. First, there is disagreement over use of "likely" or "registered" voters (or if there is a significant difference between the two). The second, and I think more important point is that people do not understand what the margin of error means in polling (i.e. that there is a MOE for each candidate, not the poll itself). Selected quotes follow:

On the proliferation of polling:

"In 1948 and 1956, Gallup conducted seven national horse-race polls on the presidential election between March and July. In 2004, during those same months, 92 different surveys measured the race between Kerry and Bush. That is an average of nearly five per week."

On "likely" v. "registered":

"Some would argue that while some polls survey likely voters, others focus just on registered voters.

Studies show no systematic differences in the accuracy of so-called “likely voter” polls compared with polls of registered voters. Others point to the shoddy workmanship of some survey organizations."

On margins of error:

"Not to go statistical, but the margin of error exists on each number (the Bush vote and the Kerry vote, independently). Thus, the margin of error on the 'margin' between the candidates is twice the margin of error for the poll. For example, a poll with a margin of error of four points will have a margin of error on the margin between the candidates of plus or minus eight points."

I think this last quote is the gem of the entire piece. Although I had learned this in a college class my freshman year, most Americans (and more importantly and unfortunately most pundits) don't understand that the MOE is for each candidate, not just for the poll. At best, this is a good point to bring up in conversation, especially when some talking head begins professing that one candidate is surging ahead in the polls after a shift of a couple points; at the least, it is good fodder for my skeptical mind.