The American Spectator

Reader Mail

That's Hot

8.7.08 @ 12:01AM

(Page 3 of 5)

Thank you, Robert Stacy McCain, for your article on "statistical certainty." It appears that you know what you're talking about. Well done.

Goodness knows you don't need my help in this, but let me try anyway.

The election of a president is not one election, but more than 50. In addition, it's not votes or voters who matter, but electoral votes [and electoral college voters aren't polled. and can't be because they are bound to what the electorate tells them to do]. Therein lies the first error in polling for a national election for president.

Let us assume, for argument's sake [for humor's sake, more like], that a pollster decides to do 50 state polls, then aggregate them.

Let's examine just the 50 states in more detail.

Each state has its unique political characteristics, demographics, and, sometimes most important, other sets of political races and/or referenda -- each of which can influence outcomes in the presidential race, each of which varies from year to year, and none of which are considered in preparing the sample. No "national" poll which pretends to be random can take all of that into consideration. Nor can a "statewide" poll, for that matter.

Assume that our intrepid pollster pushes on. Immediately upon deciding which of the thousands of telephone directories in each state he will use, the pollster destroys randomness. Skew.

Then, upon picking up each chosen telephone directory to "randomly" select people to call, the pollster has skewed his results toward people who have a telephone AND a listed number. Skew.

He then has to ask whether the person called wishes to participate -- another skew. He then has to decide to take whoever answers the telephone, or to screen out non-registered voters, or registered voters who tell him they don't plan to vote. Skew, skew, skew.

He then has to depend on the veracity of the called person. Skew.

Heaven forbid that the pollster avoids these problems by setting up a database from which to select called persons. Let's see, how do I indicate a frenzied shouting of SKEW!

Then he has to introduce all of the adjustments the pollster silently makes in order to make his non-random sample seem random or unbiased.

Howard Hirsch wrote in a letter published on August 1st: "Of course, the trick is to make the sample as random as possible, or at least stratified according to known characteristics. In that regard, sadly, most current polling falls far short." Well said, Mr. Hirsch.

Something is either "random" or it's not "random." There's no "as random as possible." "Stratification" further destroys randomness. And, according to Mr. Hirsch, pollster don't even do those things well.

Skewed is skewed, whether it is accidental or intentional.

Letter to the Editor

topics:
Health Care, John McCain, Television, Abortion, Hollywood, Law, NATO, Energy, Oil