1: \begin{abstract}

2: Recent work has suggested that in highly

3: correlated systems, such as sandpiles, turbulent fluids, ignited trees in forest fires and

4: magnetization in a ferromagnet close to a critical point,

5: the probability distribution of a global quantity 

6: (i.e. total energy dissipation, magnetization  and so forth) that has been

7: normalized to the first two moments follows a specific non Gaussian

8:  curve.  

9: This curve follows a form suggested by extremum statistics, 

10: which is specified by a single parameter $a$ ($a=1$ corresponds to the

11: Fisher-Tippett Type I (``Gumbel") distribution.)

12: 

13: Here, we present a framework for testing for extremal statistics in  

14: a global observable. In any given system, we wish to obtain $a$ in order

15: to distinguish between

16: the different Fisher Tippett asymptotes, and to compare with the above work. 

17:  The normalizations of the extremal curves are obtained as a function

18: of $a$. We find that for realistic ranges of data, 

19:   the various extremal distributions when normalized to the first two moments

20:   are difficult to distinguish. In addition, the convergence to the limiting

21:   extremal distributions for finite datasets is both slow and  

22: varies with the asymptote.

23: However, when the third moment is expressed as a function of $a$ this is found

24: to be a more sensitive method.

25: \end{abstract}

26: