Map of life expectancy at birth from Global Education Project.

Wednesday, July 20, 2005

Too much to talk about

First, a track back to the Dharma Bums, and specifically this post.

Rexroth's Daughter is on the list for those Zogby Internet polls, and this one basically asks some questions about drinking water treatment the answers to which are completely obvious. Although there's a bit of obfuscation, the question boils down to, "Would you prefer to drink a glass of water that contains harmful byproducts of disinfection and/or may be contaminated with cryptosporidium or mercury; or one that doesn't have any of those contaminants?" Then it asks whether your city should sell you potentially contaminated water or not, and whether the EPA should allow drinking water to be contaminated. Well duhhh.

I said that this is not exactly a push poll because they aren't using a script to steer people to specific answers. Instead, they know in advance what the answers will be without having to push.

In fact, this is not a poll at all. It is an exercise intended to support a marketing and advertising claim. Someone is manufacturing a system for treating water, and they want to be able to claim that 99% of people polled support their system. We all get bogus polls like this in the mail. (I mostly get them from the Democratic Senatorial Campaign Committee and similar organizations.) The main purpose in direct mail campaigns is to get people to open the envelope and send in a contribution, but they may also use the "results" (fully known in advance) to make claims about "respondents to a mail survey." In these cases, the wording of the questions is also preposterously tendentious.

Real polling requires that:

  1. Respondents are selected at randomfrom a known population. Or, more precisely, that every member of that population has either an equal, or a known probability of being selected.
  2. The wording of questions (called "items") does not suggest to respondents what the pollster thinks the answer ought to be, or what the "right" answer is.
  3. The response categories that are offered are exhaustive (every possible answer is available, or at least you get to opt out with "not applicable" or "don't know") and mutually exclusive (if the pollster doesn't offer the opportunity to pick more than one answer, then it must be impossible for more than one answer to be true.)
  4. The order of questions is such that an earlier question is not likely to influence the answer to a later question.

There are various other considerations, such as so-called "socially desirable response bias" -- i.e., it isn't very useful to ask "Do you think Black people are inferior" because most people, even if they do think that, won't say so.

But the first requirement -- the random, or probability sample -- is essential or all the rest is meaningless. The purpose of a real poll is to estimate the percentage of people in a defined population who hold certain opinions or beliefs, or who report certain experiences or circumstances. Actually "survey" is a broader term. By a "poll" we usually mean a specific kind of survey that focuses on opinions. But surveys are used to study health conditions, health care access, hazardous exposures, and all sorts of other issues in public health.

So, up soon, a discussion of concepts of probability and statistics that are important in public health.