In article <[EMAIL PROTECTED]>,
Jeff Goslin <[EMAIL PROTECTED]> wrote:
>Just so you know, I've only taken a few stats courses, and they were a few years
>ago now, so I'm a little rusty. I've got the extreme basics down no
>problem(mean, median, variance, stdev, etal), but anything beyond that is iffy.
The POPULATION mean, variance, etc., may be of some value in your
work, but not the sample stuff. Also, nothing of importance in your
problem is normally distributed.
>I have a "Stats 101" type textbook if any response requires me to look up what
>would be considered a first year stats type question, but I don't know where to
>*START* looking to answer this one...
>Anyways, I have a set of data, specifically blood glucose measurements of a
>diabetic over a 24 hour period. That data is measured at 2 hour intervals, but
>I need to be able to get from that data a "best guess" as to the minute by
>minute blood sugar of a patient, based on those measurements. Basically, I need
>to "fill in the blanks", the minutes in between the two hour intervals of time
>where the blood glucoses are actually measured, with guesses that may or may not
>be totally accurate, but are close enough for me to use as data in another
>application.
Time series packages assume too much stationarity, or at least
slow local variation of the systematic component.
I suggest that you use your knowledge of biology to produce a
more intelligent model than will be treated by any standard
time series package. But even this is not going to do too
much good with the information you have; interpolation
procedures are not able to cope well with relatively frequent
transients. Being a diabetic myself, I am not even convinced
that measurements at 10 minute intervals are enough to do it,
although a good model, and the damping effects of the biological
mechanisms involved, might manage it.
Knowing the times of eating, taking medication, sleep, and
exercise is of great importance in what you ask. If your
measurements are just after a meal, it is unlikely that the
one-hour post-prandial spike can be estimated from the rest
of the data.
>I don't have any super duper stats programs at my disposal, the best I can
>muster is Excel to perform any required calculations.
I see no way that anything that crude can do what you need.
>I thought I would create a "best fit curve graph" or some such thing, using the
>points that I actually have to create a curve on the graph. Then, I would use
>the intersection points of the graph to extrapolate the actual points of data at
>a given moment in time. If anyone has a better idea, or a simple equation that
>would produce the same results, or any advice, really, I'd be most appreciative.
Some of the packages might CLAIM to do what you want, but
statistics is not a collection of procedures into which one
inserts data and deduces the state of the universe.
Cookbook statistics, and this is all which will be found in
the packages, is no good if the recipe is inappropriate for
the material to be cooked. This is your situation.
--
This address is for information only. I do not claim that these views
are those of the Statistics Department or of Purdue University.
Herman Rubin, Dept. of Statistics, Purdue Univ., West Lafayette IN47907-1399
[EMAIL PROTECTED] Phone: (765)494-6054 FAX: (765)494-0558
=================================================================
Instructions for joining and leaving this list and remarks about
the problem of INAPPROPRIATE MESSAGES are available at
http://jse.stat.ncsu.edu/
=================================================================
