- I've seen some other replies, and I have something to add to my first reply, concerning "independence."
On Mon, 07 Jul 2003 12:33:05 -0400, Rich Ulrich <[EMAIL PROTECTED]> wrote: > On 7 Jul 2003 09:03:02 -0700, [EMAIL PROTECTED] (KR) wrote: > > > If I have two different surveys from which I calculate values of P1 > > and P2, would Cov[P1, P2]=0? The two surveys were performed in > > different years and have different sample sizes. > RU > > As near as I can figure - > you have "Missing data" and no computation. - snip, details - Did someone ask, "Were these independently sampled?" If so, that question was not directed at P1 and P2, but rather, at the *selection." For having a valid and useful survey, it would be proper for every individual in the original population to have an equal chance to be in either sample. If the original population is "huge" compared to the sample Ns, that implies the overlap in individuals should be practically zero. - Sometimes, surveyors will force that number to be zero, because they will assume that a bias could be introduced when a person answers the questions not-naively. But, in a certain general sense of testing of testing whether the samples are random and "independent," you would look at the 2x2 table, for the counts of how many were included in both years, neither, or which one -- compared to the total N. The members of one sample might or might not be "independent" of each other, too. They would not be independent, relative to the outcome, if more than a chance-number of them shared a characteristic that was meaningful to what is being surveyed. -- Rich Ulrich, [EMAIL PROTECTED] http://www.pitt.edu/~wpilib/index.html "Taxes are the price we pay for civilization." Justice Holmes. . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
