Dear Allstaters,
I have 6 months of daily counts that can range from 0 some days to a maximum value of 10. I would like to establish some kind of upper limit that would indicate unusual behavior for new observations. I am especially interested in any robust non-parametric methods to do this.
I though about SPC type limits however the daily counts are not independent as the number of counts on one day could be affected by counts on the previous day.
Another thought is to assume the data are Poisson and fit such a model however the variance > mean. This again does not seem to take into account the correlated nature of the data.
Is is possible to use some kind of Markov Chain where the counts are the states and you compute the probability of getting N counts at time t+1 given you saw M counts at time t?
Any thoughts or suggestions on approaches I might investigate are most welcome.
Happy new year
Mary.
---------------------------------
Looking for last minute shopping deals? Find them fast with Yahoo! Search.
|