🚀 go-pugleaf

A colleague and I are working on a time series experiment that includes
both instantaneous event (eye-blinks) and duration (speech hesitation)
data.  We are trying to determine the best way to turn these data into
comparable series to allow time domain analysis, and are aware of two
techniques for doing so.  One uses a non-overlapping window of observational
units (e.g., count number of blinks in units 1-50, 51-100, 101-150, etc.)
and the other uses a sliding window (count in units 1-50, 2-51, 3-52, etc.).

We can see some potential drawbacks with each, namely that the former would
be especially prone to aliasing and the latter could introduce a moving
average artifact into the series.  However, we have been able to find little
expert analysis of the pros and cons of each method.

Could someone comment on the relative merits of these techniques, or direct
us to a source that discusses same?  If there are other possibilities for
generating the series we'd appreciate hearing about those too.

Thanks...............SC

=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
Steve Corman             |                            | OFFICE (602) 965-3830
Dept. of Communication   |       "No cerveza,         | DEPT.  (602) 965-5095
Arizona State University |        no trabajo."        |
TemPE, AZ  85287-1205    |                            | BITNET:ATSWC@ASUACAD
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=