In addition to the good suggestions already given, you could consider
measuring the coupling of the code. The simplest (least coupled code)
would have no static or global variables anywhere. The coupling
metric would get bigger which each REFERENCE to a variable that is
in any way static (that is, defined logically at a higher level than
the current function's stack).

A more accurate measure of coupling would increase the metric more
for variables that are more global. For example, in various object
oriented languages like Ada, C++, Java, etc. there are ways of
making a variable global to a class where each method in the class
can have access to that global variable. C++ programmers actually
tend to call this level of globality "local variables" which makes
it very hard to measure how global a variable is. However, by
considering everything to be global (at some numerical level)
that is not purely local (that is purely ADT, or purely reentrant)
an accurate measure of coupling can be derived. This coupling number
can be used to see the number of steps it takes to analyze the
impact of changes to the code or to write or exectue test plans
procedures for the cascading effects of changing variables.

Mike Brenner   [log in to unmask]