home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: sci.math.stat
- Path: sparky!uunet!sun-barr!cs.utexas.edu!qt.cs.utexas.edu!yale.edu!cs.yale.edu!engelson-sean
- From: engelson-sean@CS.YALE.EDU (Sean Philip Engelson)
- Subject: Correlation test (sort of)?
- Message-ID: <1992Jul31.135844.21061@cs.yale.edu>
- Sender: news@cs.yale.edu (Usenet News)
- Nntp-Posting-Host: fridge.ai.cs.yale.edu
- Organization: Yale AI Mobile Robotics Project
- Date: Fri, 31 Jul 1992 13:58:44 GMT
- Lines: 28
-
-
- Please forgive my statistically untutored way of speaking, but I have what I
- think should be a relatively simple hypothesis testing problem, which I (not
- being a statistician) don't quite know how to approach.
-
- There is a process (unmodelable) which I can measure in two different ways,
- giving me values, call them X and Y, for different states of the underlying
- process. I would like to make statements of the form:
-
- When X is high, Y is (usually) low. ("high" and "low" can be interchanged)
- (note the asymmetry, if X is low, we don't care about Y)
-
- I need a statistical testing method that gives me a measure of my confidence in
- such a statement. If necessary, I can put in threshholds defining "high" and
- "low", but I'd prefer not to (or to get them out of the method). I have a
- reasonable number of data points (about 300). Doing a regression analysis
- seems inappropriate, since (a) different values of X can be fundamentally (ie,
- ignoring noise) associated with the same value of Y and v.v. (since the
- underlying process state in multidimensional) and (b) the choice of regression
- model is not clear (since I have no real model for the process). Any and all
- suggestions would be most helpful, especially references.
-
- Thanks very much.
- --
- Sean Philip (Shlomo) Engelson Im eshkachekh Yerushalayim,
- Yale Department of Computer Science Tishkach yemini;
- Box 2158 Yale Station Tidbak leshoni l.chiki
- New Haven, CT 06520 Im lo ezkerekhi...
-