In yesterday's newspaper, an article about a statewide political race made me think of Chapter 5 and some of the issues raised there about samples and populations. Just in case you didn't see the article, here are the 4 key paragraphs:
Polling is getting to be a difficult profession, and polling in our state is about as hard as it gets. It's hard to figure out who's really leading among the people who will show up at the voting booth. Here are some of the problems.
Traditionally, the Republican eastern section of the state has more voters than either of the other two grand divisions [the middle and western sections]. If you do a random sample of the entire state in which you make as many calls in the Democrat-rich west as in the east, you skew the results toward Democrats. You may recall that polls conducted that way had the two main candidates for Governor even the last month of the most recent gubernatorial campaign. But our current Governor won by 10 percentage points. To get accurate results, you have to sample more voters in East than West.
But. Who answers the phone? When I speak to a group of people over 50, I ask them how many of them answer the phone when it rings. Virtually every hand goes up. If you talk to younger people, you will find that many of them screen their calls. If you look at the internal data on polling in our state, you often find that a disproportionate number of the people polled are over 60. They are home, and they answer the phone. Demographically, older voters in the middle and western parts of our state tend to be Democrats; they grew up that way. So, again, a statewide poll will tend to skew toward Democrats.
Are people telling the truth? Pity the poor pollster trying to discover if the person on the phone is really going to vote. Virtually no one admits that he or she doesn't intend to vote or didn't vote in the last election. Using actual voting rolls to identify likely voters is often cost-prohibitive for most polls. Good pollsters ask screening questions about who people have voted for in the past that can sometimes weed out the nonvoter, but it doesn't always work, and it isn't always done. So, is the poll a poll of the population, a poll of people registered to vote, or a poll of people who will actually vote?
OK. Those are the 4 key paragraphs from the newspaper article. What went through your mind as you read them?
As for me, I very much liked the writer's comments about polls that rely on phone calls (and his summary statement that, for my state, "to get accurate results, you have to sample more voters in East than West"). And I also thought he was "on the money" to raise the issue of "truthfulness." (Isn't the last sentence in the 4th paragraph simply wonderful!)
However, I finished the article wondering (1) if the writer truly understands what a "simple random sample" is and (2) if he's knows the meaning of a “stratified random sample." As for the 1st of these concerns, go back and read the 2nd sentence in the 2nd paragraph. (Doing what he suggests would FAIL to give each resident an equal chance of being in the poll, and the resulting sample would not be a random sample.) As for my 2nd concern, isn't the writer calling for a stratified random sample--stratified by geographical section of the state--when he asserts that "to get accurate results, you have to sample more voters in East than West"?
Copyright © 2012
Schuyler W. Huck