Ay.

Stan, my friend, let Nutter's supporters be happy for a day.

Are you a budget hawk or a champion of social equity? You've been articulating the position that Fattah's "flights of fancy" airport leasing scheme is the best thing since chai lattes for weeks. Now you're saying that if it doesn't work, at least we're solid with the finances? What happened to all of that passion for the program and the folks it would supposedly benefit? FYI, you know that Mayors must always balance the budget, no deficit spending...hard to practice voodoo economics in this context. However, I do suspect that you've got a Nutter voodoo doll that you have stuck a few pins into :)

Because internal polls are done differently, and since the campaign did not release the actual questionnaire, I'm guessing it was selectively edited for public release.

I'm too tired to give examples of all the different ways an internal poll can be manipulated, but basically you ask the questions a dozen different ways until you get the numbers you like - and those are the ones you release.

Someone earlier in the thread said it must be right, because why would a campaign release misleading information? Well, it's the money, honey. Simulating an apparent surge stimulates fundraising. Simple strategic move on their part, and will probably work until the next "outside" poll.

I assumed your numbers compared the low extreme of the first poll with the high extreme of the second poll and then vice versa. I'm pretty sure that my probabilities are right if both polls had a 95% confidence level (and sound methodologies and normal distributions, etc.). All my numbers are in my first post.

18% means there is a 2.5% chance true support is less than 13%, 95% chance true support is between 13% and 23%, and a 2.5% true support is greater than 23%. The most likely true support is 18%.

Your example is so unlikely because true support being at the low extreme of the first poll is unlikely, and true support being at the high extreme of the second poll is unlikely. True support actually being both at the low extreme of the first poll and then at the high extreme of the second poll is very, very, very unlikely.

Polls don't have to publish their results with a 95% confidence level. The exact same poll could publish the results with a margin of error of only +/- 2%, but with a much, much lower confidence level.

The most important thing to remember is that all values within the margin of error are not equally likely. If a candidate polls (in a methodologically sound poll) at 12% +/-5%, it is far, far more likely that true support of the candidate is 11% (12% - 1%) than 16% (12% + 4%), even though both are within the margin of error.

anyone know how many people they would have had to ask to bring the MOE down to 2%?

Depends on what confidence level you want. I don't know the sample size required, but Wikipedia can give you a bunch of gobbledy-gook that somehow contains the method to your answer.

----
I support Michael Nutter for Mayor.

In response to Alex's question, if we use 900,000 for the number of registered voters in Philadelphia, the sample size would need to be about 2,400 to get the margin or error down to 2 percent. If we use 700,000 for the number of registered democratic voters the sample size remains at about 2,400 to get the margin of error down to 2 percent. As a matter of fact, we would have to drop down to about 400,000 voters to have the margin of error be less than 2 percent when using a sample size of 2,400. (Note: the registered voter numbers above are estimates for purposes of doing the calculations.)

Using the poll results and the margin of error of 5 percent, the ranges of possible results in terms of percentages would be the ranges in the parenthesis as follows:

Tom Knox: 27 (22 - 32)
Michael Nutter: 18 (13 - 23)
Chaka Fattah: 15 (10 - 20)
Bob Brady: 11 (6 - 16)
Dwight Evans: 6 (1 - 11)
Undecided: 23 (18 - 28).

Using a confidence level of 95 percent (95 percent is the general confidence level used by most researchers) and the above table, one could be 95 percent certain that the true voting results at the time of the poll based on the poll questions and responses would be in the ranges noted (e.g. 95 percent certain that 22 - 32 percent of the voters would vote for Knox, and 95 percent certain that Nutter would get between 13 - 23 percent of the votes). If there was normal distribution, and a 95 percent confidence level, 95 percent of groups of 403 people surveyed (the number used for the Nutter poll) would fall within 1.96 standard deviations of the mean (18 being the mean in the case of Nutter). Under a normal distribution, there would be a higher likelihood of 18 percent and a lower likelihood of amounts less than or greater than 18 percent in the case of Nutter. 95 percent of the normal distribution range would be from about 14.1 to 21.9 percent; 70 percent of the normal distribution range would be from about 16 to 20 percent.