I’m not really a political junkie, but I do pay a lot of attention when election season rolls around. We’re just two years away from a unique election cycle, when neither a sitting president nor vice-president will be running for president.
As with most of the things I’m intensely interested in, I have a project I’m working on for it. In this case, it’s a ranking of the potential nominees from each party based on their chances of winning the nomination. Positions on the issues play no role in this; I base it entirely on polls and fundraising.
And right now, both are failing me. The FEC’s web site doesn’t yet contain any financial data for the current election cycle. As for polling, it works very well near the top but is worthless at the very bottom.
Consider this ABC-Washington Post poll. Note that there are six Republican candidates that got 1% in the poll and three that got 0%. The sample size of Republicans is 344, so 1.72 would be the number of respondents that represents .5% of the poll, anything below which shows up here as 0%. How am I supposed to separate those three at 0% when they either got 0 or 1 person saying their name?
It gets worse. The threshhold for 1.5% would be 5.16 respondents. Therefore all those people at 1% got 2, 3, 4, or 5 respondents saying their name. I am left to assume that the poll results are sub-sorted by how many respondents said a name, but ties still exist, and worse, if they’re in alphabetical order, I don’t know which comparisons of two back-to-back candidates represent ties and which represent a different number of respondents! And it all reflects the luck of the draw! I’m ignoring margin of error in my rankings but even I can’t ignore this!
This poll was conducted on a national sample of 1000 adults. That’s how many should be polled from each party. The poll’s total sample should be closer to 2500.
Then I got an idea. Perhaps we could combine the results from several polls, thus adding to the sample size and lowering the margin of error. The chances of two polls contacting the same person are astronomical, so it’s like taking one big poll. For example, there are three similar polls from this month in the same field: the Gallup Poll has 412 Republican respondents, and the Zogby Poll has 301 Republican respondents. All have, ultimately, the same problem, but when you add their sample size together you get 344+412+301=1057 respondents in the sample. That means 5.285 respondents represent .5%, enough for some separation, weak though it may seem; meanwhile, 15.855 respondents represent 1.5%, enough to rest easy that six candidates would have at least some separation.
I would love to be the person to create this “superpoll”, which would be important far beyond this context, but unfortunately, the sort of raw data of pure numbers of respondents is treated as fairly proprietary. Either I have to get into a subscription service to get them (always for a fee) or they don’t offer it at all. Why, I’m not sure. I could guesstimate it by weighting the results of the various polls, but it’s an inexact science to say the least.
Which leaves nothing for me to work with, at least in the back of the field, but the analysis of others. I know it’s early and a lot can change, but predicting the future isn’t my priority so much as determining what’s going on right now, despite my emphasis on fundraising. Judging by polls from 2004, the sample size of polls won’t be increasing from here, though it might see a little more separation. It probably won’t get there very quickly, though – not with a field of this size.