[R] acf(): meaning of the blue horizontal lines

[Oliver Bandel]

Zitat von David Scott <d.scott at auckland.ac.nz>:

> On Tue, 28 Oct 2008, Oliver Bandel wrote:
>
> > Hello,
> >
> > what are they meaning? It could be something that
> > would show a threshhold above which the result is
> > indicating different meanings then just random noise.
> > But there is no description on the definition of those lines,
> > so it means nothing, if it is not clearly defined.
> >
> > Where can I find a detailed definiton?
> >
> >
> > Ciao,
> >   Oliver
> >
>
> Look at
>
> ?plot.acf
[...]

OK, this was interesting.
It was what I have thought that it could be the case,
but what I was not sure about.




>
> for which there is a pointer on the help to acf:
> 'The generic function plot has a method for objects of class "acf". '

class acf?! I have not heard of acf-class before.


>
> Note that if you can't guess what the lines are on the acf plot you
> don't
> know enough about time series to be using an acf plot and had best
> get a
> book on time series and read it.

I could guess the mening, but I didn't want to rely on my guesses.
I also read something about confience intervals of the acf(), but
never before had heard of confidence intervals in respect to
acf(), so I wasn't sure how to interpret it. Also a confidence interval
normally can be choosen (5% or 1% or something else).
I didn't found it in the manpages.

I guessed that the blue line could indicate spikes that are
"more than pure noise", but when you have a continues
acf(), there would be only at tau = 0 a peak and no other
values other than 0.
That in numerical acf() this might be different is clear,
but I didn't wanted to rely on just guessing.

when I look in ?plot.acf I also find, that this line should be
seen with appropriate caution.

Ok, so it is just a hint to peaks in the acf() which - if above the line
-
could indicate non-purely-random/correlated parts of a signal.

OK, then it match my guesses.
But now it's more then just guessing.


Thanks to all who has answered,
                                            Oliver