Random Digit Dialing is for the birds

So Quinnipiac released a new survey today showing Trump down 6 points in Pennsylvania. Now, on the one hand, kudos to them for avoiding herding (the practice where pollsters only release polls that are around the polling average), but there’s no forgiveness for the fact that they are still using the practice of random digit dialing.

You may ask, what is random digit dialing? Well it is the old-school practice of randomly dialing phone numbers to conduct your survey. Back in the days of 100% land lines and high response rates this was the main way survey research was done. However, the political world switched to list-based surveys years ago. The truth is there is no major pollster in the political world still using RDD.

Why?

Because we need to reach a representative sample of voters, not just adults, and the states provide lists of registered voters. In today’s world campaigns use lists to produce the numbers to call and then have additional weighting, quotas, stratifications, etc. to try to get a representative sample.

Now Quinnipiac doesn’t release any of their stratifications. They don’t release what party people are registered in, only what they identify as, and they released no education data.

They say they weight to the census which is comical since the census is no where near what elections look like. For example, by the census Philadelphia will produce more votes than Allegheny County, but we’re now many years into the opposite being true.

Looking through the VERY sparse cross tabs they provide it appears the sample is pretty substantially over educated and perhaps a bit too Democratic.

Quinnipiac is probably the worst public pollster of the bunch and their last PA survey in 2020 was a 10+ point miss so I probably shouldn’t expect better, but here we are.

I suspect the pollsters over at Quinnipiac should read my article on why public polling is off.