George,

Polls that attempt to predict voting results in New Hampshire primaries are notoriously inaccurate. I have observed the these polls over the years in many primary elections and they almost always miss the final voting results by fairly wide margins.

In any poll, error can occur due to sample size and sampling procedures.

For example, the CNN/WMUR/UNH poll (which had Obama ahead of Clinton 39% to 29%) interviewed only 268 Democratic respondents. Furthermore, this poll was conducted in just two days -– Jan. 5-6. That is not enough time to have completed a full sample. (People may not be at home when first called and need to be called back several times on different days.) The fact that there were only a few days between Iowa and New Hampshire for pollsters to take surveys probably exacerbated this sampling problem this year. (Also the fact the residents were being deluged with calls from candidates meant the many probably stopped answering their phone.)

A lot of polling error occurs in primaries because many respondents keep changing their minds. An hour after they are interviewed, they may have switched to someone else. Also, the fact that undeclared (independent) voters can vote in either the Republican and Democratic primaries in New Hampshire makes it very difficult for pollsters. Which primary to vote in is often a last minute decision. There is evidence that this was particularly pronounced in New Hampshire this year. Many independents ended up voting in the Republican primary for McCain. Those votes might otherwise have gone to Obama.

In short, polls in New Hampshire primaries are certainly not “hard” data. Obama’s lead was probably an error ridden polling illusion. Why waste time trying to figure out why actuality did not match this illusion?

David RePass (Electronic mail, January 9, 2008)