Dear members,
I need to detect trends in time series. To remove the effect of "Lag-1
serial correlation", it is suggested to use either Yue&Pilon or Zhang
method. Both methods are available in "zyp" package. The package uses
"kendall" package for trend analysis. ?
Based on Yue&Pilon (2002), if the lag-1 serial correlation is significant,
TFPW method will remove the effects of it prior to the?trend test; otherwise
trend test will be applied on original time series.?
I've compared the results of a sample time series with non-significant lag-1
serial correlation, using both zyp & kendall packages.?"yuepilon"
method in "zyp" gives me the following results:tau: 0.075 & sig:
0.388
while "kendall"?package gives me this:?
tau: 0.109 & sig: 0.216
The question is :?Does "zyp"?change the significance of the trend?in
this case as well? Is this a malfunction or did I miss something?
I've checked the script and it is mentioned (ln 65)?:?# Prewhiten the
original series
c <- acf(data,lag.max=1,plot=FALSE,na.action=na.pass)$acf[2]
Thank you for your consideration.
Best regards,
Morteza
<!--#yiv5285967173 P {margin-top:0;margin-bottom:0;}-->
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 1-10-2016 6-09-04 PM.png
Type: image/png
Size: 6799 bytes
Desc: not available
URL:
<https://stat.ethz.ch/pipermail/r-help/attachments/20160114/54f347d5/attachment.png>