A call to skepticism

Just got a couple of new polls. Let’s see…hmmm. Ever heard of Keystone Mercadeo Táctico? Of Top Data? No? OK, let’s look at their websites. Oh wait, they...

An ocean of fakes out there…

Just got a couple of new polls. Let’s see…hmmm. Ever heard of Keystone Mercadeo Táctico? Of Top Data? No?

OK, let’s look at their websites. Oh wait, they have no websites! Funny that…

Well, maybe they’re down for maintenance. So, let’s see how have these firms done in previous elections. What’s that you say? Keystone is not on record having ever polled in Venezuela before? TopData only sparsely, in the MUD Primaries, and on occasion was hysterically wrong?

That’s all I need to know.

For me, it’s pretty simple: three strikes you’re out. No name, no website, thin to no track record? Chances are these “firms” don’t actually exist outside the confines of their PowerPoint slides.

That said, for the real obsessives, I’d add one more thing. Though the three-strike test above is enough to label TopData as likely fake and Keystone as almost certainly fake, there are other elements that give them away too.

Keystone Cops Mercadeo Táctico claims a sample of 10,183 registered voters, touting it as “the biggest sample in Venezuela.” They think this bolsters their credibility. In fact, it all but guarantees it’s a fake.

Let’s see why. On a population of 19 million voters, a standard sample size of 1300 gives you a margin of error of 2.72%, 19 out of every 20 times. (That is, assuming the sample is well selected, the real situation is within 2.72% of what the poll says it is 95% of the time.) A sample size of 10,183 gives you a margin of error of 0.97%.

In other words, you multiply the cost of carrying out the poll by almost eight in order to improve its accuracy by less than three. This makes zero business sense: no pollster anywhere in the world can stay in business long with polls that are eight times more expensive than the industry standard.

It’s a bit like finding a burger joint directly across the street from a McDonald’s trying to sell a burger slightly nicer than, but really very similar to, a BigMac for $25 each. A-It won’t last. B-It’s so obvious it won’t last, you can’t help but wonder about the quiquirigüiqui behind it.

There’s a reason all established pollsters work on samples ranging from 700 to 2,000 or so – that’s the sweet spot where reasonable accuracy criss-crosses commercial viability.

When you see a poll claiming a sample size that’s substantially above or below that sweet spot, you should already have alarm bells going off in your head.

And when a firm you’ve never heard of, with no website and no track record claims to have performed five times the number of interviews at the upper reaches of the sweet spot, it’s pretty certain you’re looking at a fake.