Reply #10: More on the sample size. [View All]

[skids]

http://nashuaadvocate.blogspot.com/2005/01/news-election-2004-nep-adds-hundreds.html#comments

In addition, it should be noted that the number of entries in the raw data release for the national poll is not 13660, it is 13719.

Another note that should be made is that it was stipulated by NEP that some of the entries were counted more than once, because their techs were lazy. I don't remember the exact group that this was, but I think it may have been telephone surveys. They counted them more than once because of a shortsighted design: they had a flip-switch in the data which told which version of the survey a person took. However, some people (again I think it may have been some of the telephone surveys) took a survey which included questions from more than one of those survey classes, so instead of fixing it right, they made duplicate entries in the data for these people -- one for each survey type. However I do not remember which poll this effected -- could have been the national, or could have been the states.

As an aside, the _All file which contains the merged results from the states and the national poll contains 77006 entries. In both files, all entries have a nonzero weight, so none are zeroed out, though some are weighted as low as .035 (they are worth 3.5% of a person) and as high as 11.48 (they are worth 11 and a half people)

Now as to whether all this matters -- is there a statement from NEP that when the additional surveys were added that they did not re-weight the older surveys at the same point? Because if not, then these bets are off. That the addition of the new surveys cooincided with a reweighting of the entire dataset provides them with plausible deniability on this count.