On Sun, 08 Sep 2002 00:17:52 +0000, jim clark wrote: >> To be more specific on the setup: I have data from a hospital. During a >> period of four years, a number of patient were checked once or twice >> every year. During that period, the operator has changed. The changes in >> the measurement for the healthy patient are supposedly due to ageing of >> the patient, the operator that does the measurements and possibly the >> time at the moment of measurement (compare for example the length of a >> patient; if he/she ages, he/she gets shorter, but you also get shorter >> during the day). Maybe the changes are also influenced by the age of the >> patient (for an older patient, the changes may be slower/faster). > Might it be possible to set it up as a regression with indicator variables > for patients, operators, and perhaps time. Whether time needs to be coded > as indicators of some meaningful kind or simply treated as a linear > predictor would depend on the nature of the effects expected (e.g., aging, > as mentioned above). If you have too many patients to use indicator > variables, patient means can be used as a predictor to get the > corresponding SSs, but dfs need to be adjusted if statistical tests are > done.
I think time needs to be treated as a linear predictor. The next step would be to determine whether or not a specific change after some time is pathological or not. So, I need to set a bound on the changes that are considered 'normal'. But I've read up on some statistics books, and while most discuss balanced setups, I still have plenty things to try. What program would you think would be best for this kind of thing? Or should I just write my own Matlab-routines? > Building on Rich's point, might the variability associated with patients > "controlling statistically" for operators and times (i.e., a multiple > regression) provide a way to get at the reliability of measurement? I hope so. The changes in time are quite small (an effect will not be noticed when sampling is repeated within months), so if measurements are repeated within a reasonable amount of time, one may assume the time has not changed. As I stated, I have read some books and am trying to fit my problem into one of the discussed methods. I will do some more reading and thinking and then come back to the group with what I hope is a correct description for my problem. Thanks for the help! Regards, Koen . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
