time series and regions of change

Preface:  this is a statistical question more than an R question.

I have a vector of numbers (assume a regular time series).  Within
this time series, I have a set of regions of interest (all of
different lengths) that I want to compare against a "baseline" (which
is known).  There is some autocorrelation involved.  I would like to
determine the "significance" of the regions as judged by their length
and relative height from the baseline.  I could do a simple t-test or
something like that, but this seems to be too sensitive (probably due
to dependency between adjacent observations).  I have thought about
Hidden Markov Models, but I don't know the number of states.  Any
other ideas?
If the regions of interest are known (i.e., not determined in some way
from the data), then you could use a simple ANOVA using a covariance
matrix estimate that is robust to serial correlation. The package
sandwich provides such estimators (e.g., vcovHAC and kernHAC) that can
be plugged into the function waldtest() in the package lmtest. This
corresponds to conduncting an anova() with a different covariance matrix
estimate. See the sandwich vignette for some examples.

If the regions of interest are unknown and have to be estimated, the
function breakpoints() in package strucchange might be useful. See
help(breakpoints) for some examples and further references.

hth,
Z
Thanks,
Sean

______________________________________________
R-help at stat.math.ethz.ch mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide!
http://www.R-project.org/posting-guide.html

time series and regions of change

Thread (2 messages)