Wednesday, October 18, 2006

The death of polling?

'Tis the season for polls, here in the U.S., what with the mid-term elections coming up in a couple of weeks. The other day there I received a poll by e-mail asking my political views. I snicker every time they ask me, as I can't vote. I used to get lots of stuff from the Republicans, but that was when I belonged to the Military Book Club. Ironically, Alana is registered as a Republican and she never got anything. Ironic, because she's mostly against the Republicans these days and she can vote, so she is someone they should be writing to.

A couple of months ago we got a letter in the mail saying that Nielson — the TV ratings people — wanted us to be a Nielson family. I was excited about it. I figure that my TV viewing would probably flag as an anomaly, but hey, I liked the idea of having my opinions matter (and my favourite shows receiving a slight ratings boost). And then I got to the kicker: they were going to phone us to finalize the deal. Phone us on our home phone.

One problem: we no longer have a home phone.

You see, back in January, 1995 we decided to get a broadband internet connection. When we ran the numbers, the cheapest option was to have our telephone disconnected and get internet through our cable TV company. Everyone pretty much phoned us on our cell phones anyway, so this wasn't a huge deal. So, we had the home phone disconnected.

The Nielson ratings thing got me thinking about the efficacy of phone polls. A small, but significant and growing, segment of the population are disconnecting their land lines and sticking entirely with cell phones. These people are dropping out of the population, poll wise.

Polling is based on a random selection of the population. Actually, many polls are not exactly random. Polling companies spend a lot of time breaking down regions by demographics. When they poll they are usually asking the opinions of a specific geographic areas with specific demographics. Within that group, the people polled are selected randomly.

Therein lies the problem. The national no-call list stops companies from calling folk at home (or, at least that's the idea). This doesn't apply to political polling. Hey, why would politicians limit their intelligence gathering efforts? However, no one can poll people on their cell phones. The reason is simple: in North America you pay a flat fee for unlimited telephone access on land lines, but you pay by the minute for cell phone calls. Even if you have a plan with minutes, you are still paying for a set number of minutes. So, since it costs you, the cell phone owner, for your time, polling is not allowed. What's more, there isn't (yet!) a central list of cell phone numbers freely (or cheaply) available to telemarketers.

(This immediately brings up the question: how do they do telephone polling in Europe and the rest of the world where even land lines don't have flat rates? I think, in Britain anyway, that the recipient's time is not charged, it is the caller that pays. When phoning a cell phone from a land line in North America, the opposite is true.)

It's likely that the people disconnecting their land lines are not randomly spread across the political and economic landscape. If that's the case, the people left over with land lines are no longer a representative sample. Thus, the phone polls are less accurate.

Funny enough, you still see the same margins of error quoted in poll results: typically somewhere around 3% to 5% error, valid 19 times out of 20. You don't see much difference, probably indicating that there aren't a lot of people — as yet — who haven't dropped their land lines. Or, the polling companies are lazy and simply assume that the old rules are still in effect.

My theory is that we've passed the point of maximum accuracy in telephone polls. I also include Nielson on this, as they require a land line (they generate phone numbers randomly, skipping cell phone numbers). From here on, telephone polls will become progressively less accurate. If this is the case, what will replace them?

That's a good question. A number of companies are doing internet polling. Internet polls are not accurate because they are not random. Now, there are errors in telephone polling. There are demographic groups that simply don't respond to polls the same way as others. On the internet it's even worse. Unsolicited e-mails are likely to get caught in spam filters. If an e-mail does get through, it requires the recipient to choose to take part in the poll. This skews the results, as people with a deep desire to get their opinion across will be more likely to take place. In short, internet polls are anything but accurate. (This doesn't stop television news from quoting internet polls, even after they add a disclaimer.)

There's probably another way around it. It should be technlogically possible to credit time spent on particular calls and charge the person doing the calling, but it would also require legislative changes. It also increases the cost of polling, so you'd probably see fewer objective polls. The polls you would see would be those commissioned by political parties, and probably after they massaged the data.

For now, our family is out of the opinion loop. That's probably just as well: two adults, one child, one of the adults is a legal alien, and with liberal leanings in a distinctly conservative congressional district... we're not exactly a traditional nuclear family. If anything, not polling us probably decreases polling error.


JAM said...

That's pretty cool thought, that you a least were asked to be a Nielsen watcher. This is all very interesting to me. I've often wondered about polling, and I learned more here than ever. I have had a life long wondering about the Nielsens. As in, who the heck likes THAT show sitting at number 3? I've had cable all my adult life and don't watch very much network stuff. I tend to watch, BBC America, Science Channel, History Channel (called the Hitler channel around our home), and SciFi Channel. I would love to have a Nielsen setup. The network ratings would drop because of us.

Also, it seems wierd to me that on the news, a poll can BE the news item. That still seems screwed up to me. I'm not exactly sure why. It just seems like a self-eating watermelon and not really a true news item.

Allan Goodall said...

I've often wondered about polling, and I learned more here than ever.

Thanks! I learned a little bit about telemarketing and polling when I worked at a newspaper.

I have had a life long wondering about the Nielsens. As in, who the heck likes THAT show sitting at number 3?

I've often thought the same thing. Alana, too. She has a history of watching TV shows that soon get canceled.

Also, it seems wierd to me that on the news, a poll can BE the news item.

I learned an interesting acronym that suits your comment: QFT (Quoted For Truth), a quick way of saying, "I agree!"

The main reason for polls, of course, is to get a jump on the results. That, and knowing poll results does help you figure out why the political parties are saying certain things.

Today was interesting. The News Hour on PBS mentioned a recent poll and analyzed it. They then said how mid-term election polls were more accurate than presidential polls at predicting the results. But, they continued by explaining why this year was different and why the poll results may not be telling the whole truth. Finally, the guests said that there will be a change in congress, but no one knows by how much.

To paraphrase, "We're not sure what's going to happen."

Oh, and I've had it with news readers saying that because the gap between two candidates is less than the margin of error, they are in a statistical dead heat. No, it does not! When the gap is less than the margin of error, that does not mean they are statistically tied. That does not mean that there is no lead! It means that you can't tell who is in the lead! Of course, no one in journalism likes to say, "Uh, we can't tell."