Explanation by Tom Jensen from PPP:
"Three days ago we said Barack Obama was up four in Colorado. Today Rasmussen says John McCain is up a point. I don't know who's right but I can tell you one reason for the difference in our polls.
Rasmussen's numbers are based on 500 interviews conducted Wednesday night.
We did 500 interviews last Tuesday night in Colorado, and based on those McCain was up by a point as well. But we also called everyone in the sample who we didn't reach Tuesday night again Wednesday morning, Wednesday afternoon, Wednesday night, and Thursday afternoon. We got almost another 500 respondents and they went for Obama by a good amount, pushing him to a four point lead.
As far as I can tell Rasmussen just makes calls until they get 500 respondents. When we buy a sample we attempt to reach every number in it at least four times over a period of several days and go with whatever number of respondents we get based on that. And we almost always find that the further into conducting the poll we go, the better Obama does. I don't know if it's that his supporters are harder to get on the line or what, but we're finding that trend in nearly every state we poll.
I'm not saying that one way or another is better- Rasmussen has an outstanding track record- just pointing out that differences in the way we conduct our polls may help explain some of the differences in the numbers we produce. It's not as simple as just saying Rasmussen is Republican and PPP is Democratic."
http://publicpolicypolling.blogspot.com/2008/08/ppp-rasmussen-and-colorado.html
Or is it that PPP goes and tries to find Obama votes so that his numbers appear better.
It seems that is what he is saying, if he's saying they have a bigger sample and they think there are people who for some non-random reason are less likely to be receivable at night time, or just less likely to pick up the phone, but will be there during the day or will pick up if you attempt many times.
That's not at all what he's saying. Read it again. They did a sample size of about 1000 (pre-selected) and got 500 respondents on their first night of calling. This crew was for McCain by one point. They then continued to try and reach the other 500 in the pre-selected sample. When they did, they found Obama was up over McCain in the original, pre-selected sample of 1000. They did not go looking for "Obama" votes to make his numbers appear better. The question here is which sampling strategy better represents the population (and produces more turthful responses) of actual voters on November 4th.
I was responding to the bold text, not Rowan's comment. Sorry for the confusion.
And yes, the reason why the PPP strategy might better represent the population is that there are people who for some non-random reason are less likely to be receivable at night time, or just less likely to pick up the phone, but will be there during the day or will pick up if you attempt many times. I was not arguing that they went looking for Obama votes. It could simply be that the Obama supporters for some reason are less likely to pick up the phone at night. Getting a truthful answer is a different question. You could have a very good, representative sample but be getting bad answers from them. One thing the robocall method of interviewing is going to filter out (for the most part) is respondent bias based on social embarrassment or the demographic of the caller.
