There has recently been considerable discussion on comp.lang.smalltalk and comp.object on the merits of static versus dynamic type checking. The static type advocates claim that static type checking eliminates an entire category of errors while the dynamic typing advocates claim that those errors rarely occur.
I'm working on a VisualAge/Smalltalk and GemStone/S project that has been in production for several years. In that time, the system has captured all the unhandled exceptions along with tracebacks for us to analyze and find bugs. These logs have been captured for over three years.
I cannot release the results of the analysis of these logs (for reasons of security), but I categorized 95% of them and found only one single report could have been detected by static typing. It was a "mustBeBoolean" exception because a method returned the receiver instead of a boolean. The problem was fixed immediately after the log was recorded and it never recurred.
Interestingly, some of the errors would have been masked by static typing because they involved performing arithmetic on uninitialized variables. In most static typing languages, integer variables are initialized to 0 which isn't always the right default. For us, they initialized to nil and raised an exception when we tried to use them in a calculation instead of quietly giving us the wrong answer.