Re: test time series?

On Dec 11, 2:17 pm, Francogrex <fra...@xxxxxxxx> wrote:
Hello, I have a series of rates over a period of time (years) from
1975 to 1996, this is an example:
c(0, 0, 0, 0.92, 0.87, 0.695, 1.2175, 0.34, 0.3, 0.5475, 1.0575,
1.5325, 2.4375, 3.005, 2.635, 5.78, 2.7225, 3.9975, 2.6575, 3.6,
1.88, 1.9025)

What test can I do to see if any point (rate) or points are
significantly different from others? Or to highlight that there is
some outlier at a certain year? Visually looking at the data is fine,
but can it be backed up by a test? PS: the rates are X number of
successes out of N trials.
Thanks

Dave put it very nicely. Let me put it less nicely. If it were
possible to detect outliers in a time series without knowing anything
else about the phenomena that produced the signal and the outliers,
this would imply that the set of signals and the set of outliers were
somehow completely disjoint in their behaviour. This is logically
impossible, since one person's outlier is another person's signal.
Ergo, one must know more about the phenomena involved than the data
alone, i.e. as Dave says, you need models of the signal and the
outliers. Clearly the models will depend on what the phenomena are,
given that time series data can arise from anything from radioactive
decay to the growth of a human being.

The only thing about Dave's post with which I do not agree is that he
claims that the only types of outlier behaviour one can have are those
for which his software has a model.

illywhacker;
.

Relevant Pages

  • Re: Estimating mean of a set containing outliers.
    ... The offset is typically around 150 counts/frame, ... background estimation to work properly. ... affected by outliers as much as the mean is. ... is that bursts of signals are what contain the most exciting data. ...
    (sci.stat.math)
  • Re: test time series?
    ... Mr. illywhacker, ... original posters questions and appear to me to be more ad hominum ... possible to detect outliers in a time series without knowing anything ... this would imply that the set of signals and the set of outliers were ...
    (sci.stat.math)