human vs. computer political surveys

I'm not sure I would have cared much about this until I was a candidate, but SurveyUSA has done most of the polling in my two campaigns, so this-- including their use of computers and their accuracy-- caught my eye.

Then, there's the potential for race bias-- the extent to which people will say they'll vote for Obama when they do not intend to do so.

Anyway, from Carl Bialik in the WSJ...

There are some presidential polling numbers you won't see on the nightly network news broadcasts. Yet, they have proved themselves to be every bit as accurate as other, widely reported polls -- in some cases, more so.

These shunned polls, however, are conducted by computer rather than by a person, so they don't make the cut with many of the big mainstream media, nor with polling experts. One prominent polling textbook, by Paul J. Lavrakas and Michael Traugott, refers to these surveys as Computerized Response Automated Polls -- insulting acronym intended.

The critics have legitimate complaints about such polls, including that a 12-year-old boy can convince a computer, but probably not a live interviewer, that he's a 37-year-old woman. But in these times of slashed media-polling budgets, declining response rates and the migration to cellphones, most polls are far from theoretically pure. Watching the survey sausage get made isn't pretty. Excluding only computer-assisted polling numbers seems arbitrary and leaves gaps in our knowledge about the presidential election....

The automated polls, or IVRs for interactive voice response, work like this: Respondents hear a recorded voice -- sometimes of a local TV-news anchor, sometimes of a professional actor -- that greets them and asks if they're willing to take part in a quick survey. Then they're asked to enter their political preferences and demographic information using their keypad...

Automated polls can cost as little as one-tenth the equivalent, live-interview phone poll. The cost advantage builds when a poll is repeated, identically, to track opinion over time...

As a result, automated polls are beginning to crowd out the rest....

Their accuracy record in the primaries -- such as it was -- was roughly equivalent to the live-interviewer surveys. Each missed the final margin by an average of about seven points in these races...

SurveyUSA, which pioneered these polls, has an impressive record for accuracy....

Recorded polls, however, offer several advantages. Interviewers are selected because their voices inspire trust (SurveyUSA uses local TV anchors; other automated pollsters use actors or, in the case of Rasmussen Reports, women 30 to 40 years old with Midwestern accents). Politicians' names are pronounced correctly and identically each time, and responses entered correctly are recorded correctly.

There also is evidence that automated polls inspire honesty, particularly on sensitive topics....

Meanwhile, conventional polls are hardly reaching a truly random sample these days. Response rates have fallen below 20% in many cases, and it's hard to know whether the other 80% who aren't home or refuse participation are like those who do respond. Most pollsters aren't dialing cellphones. And traditional pollsters don't always randomly select respondents from within households....

Until this election cycle, the Capitol Hill newspaper Roll Call wouldn't publish results from automated polls. Now it's commissioning polls from SurveyUSA.