House Polling

I thought this was a fascinating post over at electoral-vote.com, so I'm going to put part of it here on why House of Representatives polling doesnt work very well.

The introduction of Internet telephony (VoIP) services, such Skype and Vonage, wreaked havoc with this scheme. VoIP customers can usually choose any area code they want. For example, a man in Omaha might choose Florida area code 561 so his mother in Florida could call him as a free local call. It also means that a pollster randomly calling 561 numbers might get someone who doesn't live in Florida. Since most people still have area codes that correctly designate which state they live in, for Senate polls, the problem is still manageable.

However, for House polls the problem is substantial. The layout of the area codes and exchanges do not align with congressional districts at all. While 914-949-xxxx numbers all lie entirely within NY-18, other exchanges straddle congressional district boundaries, especially when the CD has been gerrymandered into a pretzel. As a consequence, a pollster assigned to poll for some House race may have to call multiple area codes and exchanges, some of whose numbers lie within the district and some of whose numbers lie outside the district. Reverse lookup of the number about to be called is not always possible because many people have unlisted numbers.

As a consequence, some of the people polled may, in fact, not live in the district in question and some people who do live there may be missed. Of course the first question could something like be "Are you a registered voter in congressional district IN-07?" However, most voters probably don't know their CD number and some may be put off by such a question and hang up. Starting with "Hi, I'm doing a poll from the XYZ company. What's your zipcode?" is definitely a nonstarter.

I suppose you could ask zipcode at the end of the poll and just throw out any results that lie outside of the district, of course this would frustrate the hell out of the people making the calls if its a human administered survey, but it seems like it would work fine on an automated one. Whatever the solution I thought it to be an interesting and amusing problem in modern polling.