On 15 Jan 2003 08:12:36 -0800, [EMAIL PROTECTED] (Conn, Judith) wrote: > Thank you for your response, Donald. Not sure how this data fits- same > persons and (think) same question repeatedly asked every year over about > 8-10 year time frame. So think the repeated measures would be better. Do > you agree? Judy Conn
That's too short to worry about power-spectrum. But it is long enough that there's probably concern about irregularities of data collection, such as missing years or different lengths of gaps, if you were to try to jam the data into the conventional "repeated measures" design. If the data *have* been well controlled, the above might be a description of (say) labor surveys, where certain samples are followed for longer periods of time, and there is also a cross-sectional element. There are some fancy programs that have been used for the first-mentioned longitudinal designs. Programs such as SAS Mixed allow for fields that are missing-at-random, and for the specification of "covariance structures." I think - it is 'my humble opinion' - SAS-Mixed is extremely difficult to use correctly, and hard to understand after it *has* been used correctly. My own approach would be: Generate hypotheses that don't name a particular time; and create pragmatic scores that can accommodate Missing. You can compare 'early' averages versus 'late', for instance. [ snip] -- Rich Ulrich, [EMAIL PROTECTED] http://www.pitt.edu/~wpilib/index.html . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
