You are working with the text-only light edition of "H.Lohninger: Teach/Me Data Analysis, Springer-Verlag, Berlin-New York-Tokyo, 1999. ISBN 3-540-14743-8". Click here for further information.

Index
See also: introduction to autocorrelation 

Autocorrelation Function

The autocorrelation is calculated in the same way as the correlation between two variables (just using the same variable twice). If we consider time signals, the values of the autocorrelation may be calculated by shifting one of the two formal variables by a certain amount  Dt. If we plot the results of these calculations against the time shift, we obtain the autocorrelation function, or autocorrelogram. Here's an which shows the principal function of an autocorrelogram.

The idea is that a time-series yt (with the measurements x1, x2, x3, ..., xn) is correlated with itself. In the first step, the pairs to be correlated are:
 

First Series x1 x2 x3 ...  xn
Second Series x1 x2 x3 ...  xn

The correlation coefficient is obviously 1. In the next step, the second series is shifted one time-step to the right:
 

First Series x1 x2 ...  xn-1
Second Series x2 x3 ...  xn
and so on.

Here's another which shows the difference between an uncorrelated signal and a signal exhibiting strong autocorrelation.
 

Last Update: 2004-Jul-03