A couple of weeks ago I did the Tahiti-Darwin correlation coefficient calculation that John is talking about here:

The number I get using Mathematica is -0.55, in line with John's interpretation of it being negative negative number because the dipole is anti-correlated.

In the figure below, I take the negative of the Tahiti measure to make it positive (I also multiply the CC by 100 for visual checking of numbers).

![SOI](http://imagizer.imageshack.us/a/img539/1412/48fe30.gif)

It is possible that Dara is doing something similar by way of taking the complement and just losing track of the sign.

The number of -0.55 does not look particularly strong and is a weaker anti-correlation than the -0.58, but this number is also extremely sensitive to the amount of high-frequency noise in the time series data. The higher frequencies can rapidly reduce the anti-correlation because any slight amount of phase shifting cause the two waves to rapidly lose alignment according to the CC formula.

I was hoping to find some other measure that considers the amplitude error caused by slight phase shifts and takes this into account for a "goodness of fit" criteria. In other words the error would be the distance in amplitude and time, instead of just amplitude differences at a particular time. I am not sure that such a beast exists, perhaps related to a variation of a Mahalanobis distance measure?

The number I get using Mathematica is -0.55, in line with John's interpretation of it being negative negative number because the dipole is anti-correlated.

In the figure below, I take the negative of the Tahiti measure to make it positive (I also multiply the CC by 100 for visual checking of numbers).

![SOI](http://imagizer.imageshack.us/a/img539/1412/48fe30.gif)

It is possible that Dara is doing something similar by way of taking the complement and just losing track of the sign.

The number of -0.55 does not look particularly strong and is a weaker anti-correlation than the -0.58, but this number is also extremely sensitive to the amount of high-frequency noise in the time series data. The higher frequencies can rapidly reduce the anti-correlation because any slight amount of phase shifting cause the two waves to rapidly lose alignment according to the CC formula.

I was hoping to find some other measure that considers the amplitude error caused by slight phase shifts and takes this into account for a "goodness of fit" criteria. In other words the error would be the distance in amplitude and time, instead of just amplitude differences at a particular time. I am not sure that such a beast exists, perhaps related to a variation of a Mahalanobis distance measure?