From the press release for the Marist poll on New Jersey,
we are treated with the following (emphasis mine):
When looking at the combined results from this two day poll conducted on Wednesday and Thursday of this week, Democrat Jon Corzine leads Republican Doug Forrester by 9 points among New Jersey voters likely to vote next Tuesday. Corzine receives the support of 49% of likely voters compared with 40% for Forrester. 11% are undecided. However, a comparison of the results from Wednesday to Thursday reveals the race has taken a decidedly different turn. On Wednesday, Jon Corzine had a comfortable lead of 13 points over Doug Forrester among likely voters. On Thursday, after the flurry of campaign ads and accusations, Corzine has the support of 46% of likely voters to 42% for Forrester.
Wow--the race really tightened up there, didn't it? However, being the sort of nerd that I am, I went to the bottom of hte press release to see what the size of the sample was for Thursday likely voters. I didn't find much (emphasis again mine):
This survey was conducted on November 2nd and 3rd, 2005. 781 registered voters in New Jersey were interviewed in proportion to the enrollment in each county and adjusted for turnout in comparable statewide elections. The results are statistically significant at ±3.5%. There are 430 likely voters and the results for this sub-sample are statistically significant at ±5%. 408 registered voters were interviewed on Wednesday and 373 were interviewed on Thursday. The results for these sub-samples are statistically significant at ±5%. The margin of error increases for cross-tabulations.
Wait a second here--Marist does not even list the margin for error of the sub-sample they trumpet at the top of their press release? They tell you what the sample error was for registered voters on each day. They also tell you what the sample error was for likely voters for both days combined. However, they don't even bother to list what the likely voter sample error was for Wednesday alone and Thursday alone. Importantly, this is after they used a comparison of the two said sub-samples as the main selling point of their poll.
You know why they don't tell you? Because the two samples they are comparing have around 200 people in each. That is a margin of error at around 8 or 9 percent. In other words, their poll does not show nearly as much movement as it implies.
This is a really bad case of media poll whoring. Before you trumpet a sub-sample as the main selling point of your poll, at least tell the public what the margin of error is for that sub-sample. This sort of thing has been my long running complaint about polling firms that are contracted by news outlets: they seem to use methodologies that will cook up spectacular, wildly shifting results that have a tendency to grab headlines. This is, of course, exactly what news outlets want: spectacular, wildly shifting poll results that make a good headline and get people to tune in / buy the periodical.
Marist should be ashamed of itself. This one will get the big headlines, but what you want hear as much about is that there were two other polls out today on the race, one from FDU that showed a narrow Corzine lead at 43-41, and one from Monmouth that still showed Corzine comfortably up 47-38. Just like the two new polls on Wednesday, these two polls conflict.
Considering everything, the race is probably a little closer than it was earlier in the week. However, considering not only this poll but also their past record in this race, it is just as clear that no one should ever pay attention to the Marist poll again.
