It feels good to be right the next day. Today’s Chronicle reports that the runoff race survey had a few problems. I reported on the bias that I detected, yesterday.

I’m responsible for one of the survey problems and I can now understand the other problem.

The survey was done with a computer voice giving respondents their choices. “If you plan to vote for Matt Gonzalez for mayor press one, now, on your touch pad.” I’m responsible for that.

In 1966, I was doing political polling in San Francisco and tried out the idea of doing interviews by computer. Computers couldn’t do a survey in those days but the public didn’t know that. I used a tape message and I would manually stop the tape message to allow for the dial response (most phones in those days were rotary so the respondents choice had to be 1 for the first choice or 9 for the second choice). My reason for the pseudo computer experiment was scientific. I wanted each interview to be identical, because I was concerned about interviewer variation and resulting potential bias.

I did about 50 interviews with my pseudo computer and 50 with a real interviewer. I reported, in a published paper, to the Pacific Chapter of the American Public Opinion Research Association that: the response rate was the same for the pseudo-computer and the live interviewer and the results were the same. Which was a positive outcome.

The computer questioning part of the runoff race survey doesn’t bother me.

The problem is the surveyor’s answer to a question: “….was based on a telephone survey of 1,012 people in San Francisco who indicated they were registered and certain to vote on Dec. 9.” That my friends is nonsense.

No question or series of questions gives you “certain to vote” information. The best you can get is a probability of voting. This guy doesn’t know what he is talking about and the probability of voting issue is the key flaw with this survey.

My advice: Ignore the survey. There may be bias problems but the technical quality of the survey is too low to worry about the sampling bias problem.