Errata |
|
Clarifications concerning my paper "Cross Wavelet Analysis. Significance Testing and Pitfalls" |
The approximative formula for WCO95:
The formula is only valid for nvoice approx. 33 scale values per
octave (200 values for 6 octaves, see caption of Fig. 3). This value
is unnecessarily high. For other values of nvoice the values for the
smoothing window length should be rescaled (e.g. for nvoice=20: a
chosen w=10 means inserting w=16.7 into the formula). This is already
accounted for in my software package.
The Monte Carlo simulations:
The dependency of the critical values on the two processes under
investigation is very weak (e.g. 2% difference between white noise
and an almost nonstationary AR1-process). Thus the approximative
formula may be utilized in all cases and no Monte Carlo simulations
are necessary.
Spurious Peaks:
I stated, that patches, significant on a 95% level, covering more than
5% of the total area, might be considered as being significant. In
fact, this is right, but a bit more complicated: The measure defining
the area of a patch depends on scale. In other words, a naivly spoken
"large" patch on a high scale might be smaller in relation to the
measure as a "smaller" patch on a lower scale. This is the basic idea
for my areawise significance test. However, when weighting every area
with the scale dependent measure, the 5% rule is correct again.
|
|
|