From the Editor: Lies, darned lies and polls
If you go by the polls, we’re in for a governor election that’s tighter than a fat man’s trousers after lunch.
Someone has to say stuff like that now that Dan Rather isn’t around on election night anymore.
At any rate, the race appears to be close, according to a couple of recent polls. A My People poll released on Halloween gives incumbent Democrat John Bel Edwards 50% support and 47% to Republican challenger Eddie Rispone with a margin of error of plus or minus 3.6%.
Another poll, taken a few days earlier by JMC Analytics, had the race at 48% for Edwards and 46% for Rispone. The reporting I’ve seen doesn’t list a margin of error for the JMC poll. It’s one of the things to watch for in poll reporting.
If the lead by one candidate is smaller than the margin of error, it’s not really a lead. It’s a dead heat.
That’s what margin of error means. It’s a measure of the limit of statistical accuracy.
If we read or report polls incorrectly, they may become self-fulfilling prophecies.
Polls are like weather forecasts. The meteorologist can nail the forecast 47 days in a row, but we remember the times when we got caught in the rain on what was supposed to be a sunny day.
The most famous example of wildly inaccurate polling is the 1948 presidential race, in which the polls had President Harry Truman losing badly to Republican Thomas Dewey. Truman won, giving us the famous photo showing a big Truman grin as he holds up a “Dewey defeats Truman” newspaper headline.
A history teacher once told my American history class that the polls were wrong because the pollsters relied on phone interviews. But not everyone had phones in 1948.
Who didn’t have phones? Farmers, minorities, lots of people in the South, people in other rural areas across the country, and low-income people generally. That was the Democratic base.
The pollsters violated a principle: Accurate polls require everyone in the population you claim to sample to have an equal chance of being polled, and respondents have to be chosen at random.
Pollsters faced the same challenge early in the cellphone era, when there was no easy way for phone banks to include cellphone numbers. Results would tend to under-represent people who relied on mobiles instead of landlines.
Now I get a robo-call once a week trying to sell me a service contract on a car I got rid of five years ago. We can assume the cellphone hurdle has been cleared.
The other example of bad polling that many people would point to is the 2016 election, when Republican Donald Trump was reputed to be a certain loser to Democrat Hillary Clinton.
Lots of TV talking heads were talking that way, but they didn’t understand the polls.
The Real Clear Politics collection of major presidential polls from the last few days before the 2016 election showed Clinton with leads within or a shade higher than the margin of error. The exception was a Los Angeles Times/USC tracking poll that had Trump ahead by three points.
Sure enough, Clinton won the popular vote. But the commonly cited polls were national polls, and we don’t elect presidents according to the national popular vote. Trump picked up enough votes to win some key swing states and gather enough electoral votes to become president.
So it’s important to view the interpretation of polls with a skeptical eye. And you should know the margin of error.
The margin is the square root of the sample size — the number of people in the area the poll is supposed to represent — divided into 1. So a margin of error of 3.5%, which is common in good polls, requires responses from something like 820 people.
The weird thing is it doesn’t matter whether you’re polling St. Mary Parish with 50,000 people or the United States with a population of 320 million. You still need about 820 responses to get a 3.5% margin of error.
Sometimes major polls will include many more responses in order to get a good margin for subgroups such as women, Republicans or Democrat, or minorities.
To really understand a poll, see who paid for it and how the questions were asked.
“Do you favor handsome, energetic Jones or that drunken idiot Smith?” is not likely to yield an accurate result.
Bill Decker is managing editor of The Daily Review.