[R] multivariate problem

[David Winsemius]

On Jan 8, 2012, at 3:01 AM, Iasonas Lamprianou wrote:

> Dear all,
> I am not sure if this is the right place to ask this question, but I  
> will have a go. Please redirect me to a different place if this is  
> not the right one!
>
> I have a (relatively) simple problem which causes me some  
> frustration because I cannot find the solution. I measure ten  
> variables (var1 to var10) every day, they are all continuous  
> (linear) and most of them are correlated. Some days, for any reason,  
> the relationship between these variables may change. They are still  
> correlated, but their correlation may change slightly but  
> practically this is important. Or, one of the variables may increase  
> its value significantly suddenly and keep this high value for a few  
> days and then come back to the normal level. I am using R. Is there  
> any function I can use to help me identify these strange days when  
> the relationship between these variables changes? For example, if  
> DayX is such a strange day, factor analyzing the data before DayX  
> and after DayX separately would give me different factors (princial  
> components). But how can I identify such a daym without trial and  
> error?

The zoo package has `rollapply`. You would of course be required to be  
much more specific in defining your problem than you have been so far.

-- 
David Winsemius, MD
Heritage Laboratories
West Hartford, CT