A Pollster Responds

Bernood Yost, the director of Franklin & Marshall College's Floyd Institute for Public Policy, whose poll I flagged a couple of days ago as having a potentially suspect likely voter model, wrote me a kind note a couple of days ago explaining his process; it is reproduced in full below.

I’m writing to respond to your post about likely voter models. I am not going to quibble about with your assessment of likely voter modeling; in fact, the title of your article should probably be, All Likely Voter Models are Suspect. Most pollsters have built their models through careful study by focusing on both intentions and past behaviors, which has worked well in the past. But in 2008 we are all concerned about how all of the new voters, who will likely be disproportionately young and minority voters favoring Senator Obama, will affect those models. No one knows for sure and this requires everyone to make untested assumptions about what is going to happen.

The piece of data we looked at closely in our survey to corroborate our model was a question about whether the respondent had already voted. Seven percent of our sample reported they had already cast a ballot and Obama’s lead over McCain was 46 percent to 41 percent among these early voters—five points. Those who reported voting early in our sample were more likely to be over 55 years of age (53%), male (53%), and non-Hispanic whites (75%). The same proportion of Democrats and Republicans (40% each) reported early voting in our sample.

See, this is all I'm asking for -- evidence that the pollster has thought about what they are doing, rather than simply acting reflexively. We might disagree about certain things (in this case, the predictive value of a very small sample of early voters), but that's all in the spirit of a good discussion.

By contrast, you probably aren't likely to see John Zogby offer a coherent explanation of why he his poll contains as many Republicans as Democrats in a political environment where Democrats have a 10-point generic ballot advantage.